[AArch64][SME]: Scalarize masked gather/scatter in streaming mode

[hassnaa-arm]

Summary

To generate code compatible to streaming mode:

• Force scalarizing masked gather/scatter.

Testing files:

• expand-masked-gather.ll
• expand-masked-scatter.ll

[paulwalker-arm]

Are code generation tests all that useful for this patch? The change really affects the ScalarizeMaskedMem pass so the code generator never sees a call to gather or scatter intrinsics, which is presumably why they're no existing NEON code generation tests for these intrinsics. Perhaps instead you want to create an AArch64 version of llvm/test/Transforms/ScalarizeMaskedMemIntrin/X86/expand-masked-gather.ll (and expand-masked-scatter.ll)? Given the code change is independent of the vector length I also doubt you need to go so wide with your the number of vector lengths tested. Perhaps a test per element type and vary the vector length across those tests is enough.

There might still be an issue where some other code can create a GATHER ISD node during code generation but none of the current tests test such a scenario and so we can worry about that later. This patch is all about ensuring ScalarizeMaskedMem does what we need when in streaming mode.

[hassnaa-arm]

Add testing files for scalarizing gather/scatter
Remove masked gather/scatter testing files related to code generation.

[david-arm]

LGTM!

[sdesmalen]

llvm/test/Transforms/ScalarizeMaskedMemIntrin/AArch64/expand-masked-gather.ll

1 ↗

(On Diff #478618)

Is it worth moving both tests to a file called streaming-compatible-expand-masked-gather-scatter.ll ?

33–64 ↗

(On Diff #478618)

These tests don't add much value in testing the expansion of the intrinsic with the code you've written, so can just as well be removed.

llvm/test/Transforms/ScalarizeMaskedMemIntrin/AArch64/expand-masked-scatter.ll

4 ↗

(On Diff #478618)

can you use opaque pointers, so <2 x ptr> ?

31–62 ↗

(On Diff #478618)

These tests don't add much value in testing the expansion of the intrinsic with the code you've written, so can just as well be removed.

[hassnaa-arm]

Use opaque PTRs, Remove not needed tests.

[hassnaa-arm]

Merge testing files gather & scatter into one gather-scatter file

[sdesmalen]

llvm/test/Transforms/ScalarizeMaskedMemIntrin/AArch64/expand-masked-gather.ll

1 ↗

(On Diff #478946)

For this patch, I don't think you have to add any files in a precursory patch. There's not much value in that since we're not looking at a subtle differences in assembly.
In that case, there is also no need to 'delete' any files.

[hassnaa-arm]

Drop the precursory patch

[sdesmalen]

Thanks!

[DavidSpickett]

This test is failing on our bot that only builds the ARM target - https://lab.llvm.org/buildbot/#/builders/245/builds/1637

Is that command line option only available when you have AArch64?

[david-arm]

In D138588#3962977, @DavidSpickett wrote:

This test is failing on our bot that only builds the ARM target - https://lab.llvm.org/buildbot/#/builders/245/builds/1637

Is that command line option only available when you have AArch64?

Hi @DavidSpickett, hopefully I've fixed this with commit a043fcf92fa3

[DavidSpickett]

That did the trick, thanks.

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h

bool isLegalMaskedLoad(Type *DataType, Align Alignment) {
     return isLegalMaskedLoadStore(DataType, Alignment);
   }
 
   bool isLegalMaskedStore(Type *DataType, Align Alignment) {
     return isLegalMaskedLoadStore(DataType, Alignment);
   }
 
   bool isLegalMaskedGatherScatter(Type *DataType) const {
-    if (!ST->hasSVE())
+    if (!ST->hasSVE() || ST->forceStreamingCompatibleSVE())
       return false;
 
     // For fixed vectors, scalarize if not using SVE for them.
     auto *DataTypeFVTy = dyn_cast<FixedVectorType>(DataType);
     if (DataTypeFVTy && (!ST->useSVEForFixedLengthVectors() ||
                          DataTypeFVTy->getNumElements() < 2))
       return false;
 

llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-masked-gather.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -force-streaming-compatible-sve < %s | FileCheck %s
 
 target triple = "aarch64-unknown-linux-gnu"
 
 ;
 ; LD1B
 ;
 
 define void @masked_gather_v2i8(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v2i8:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    ldrb w9, [x0, #1]
+; CHECK-NEXT:    adrp x8, .LCPI0_0
+; CHECK-NEXT:    ptrue p0.s, vl2
+; CHECK-NEXT:    str w9, [sp, #4]
+; CHECK-NEXT:    ldr d0, [x8, :lo12:.LCPI0_0]
 ; CHECK-NEXT:    ldrb w8, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    str w8, [sp]
+; CHECK-NEXT:    ldr d1, [sp]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
 ; CHECK-NEXT:    ldr q1, [x1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    ldrb w8, [x0, #1]
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    cmeq v0.2s, v0.2s, #0
-; CHECK-NEXT:    sshll v0.2d, v0.2s, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1b { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    ptrue p0.s, vl2
-; CHECK-NEXT:    xtn v0.2s, v0.2d
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbz w9, #0, .LBB0_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldrb w9, [x9]
+; CHECK-NEXT:    fmov s0, w9
+; CHECK-NEXT:  .LBB0_2: // %else
+; CHECK-NEXT:    tbz w8, #1, .LBB0_4
+; CHECK-NEXT:  // %bb.3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.b }[4], [x8]
+; CHECK-NEXT:  .LBB0_4: // %else2
 ; CHECK-NEXT:    st1b { z0.s }, p0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <2 x i8>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = icmp eq <2 x i8> %cval, zeroinitializer
   %vals = call <2 x i8> @llvm.masked.gather.v2i8(<2 x ptr> %ptrs, i32 8, <2 x i1> %mask, <2 x i8> undef)
   store <2 x i8> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v4i8(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v4i8:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI1_0
 ; CHECK-NEXT:    ldr s0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    ushll v0.8h, v0.8b, #0
-; CHECK-NEXT:    cmeq v0.4h, v0.4h, #0
-; CHECK-NEXT:    sunpklo z0.s, z0.h
-; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    cmpne p1.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1b { z0.d }, p1/z, [z1.d]
-; CHECK-NEXT:    st1b { z0.d }, p0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    ptrue p0.h, vl4
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI1_0]
+; CHECK-NEXT:    uunpklo z0.h, z0.b
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z0.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z1.h, z0.h[2]
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z0.h, z0.h[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB1_6
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB1_7
+; CHECK-NEXT:  .LBB1_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB1_8
+; CHECK-NEXT:  .LBB1_3: // %else5
+; CHECK-NEXT:    tbz w8, #3, .LBB1_5
+; CHECK-NEXT:  .LBB1_4: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.b }[6], [x8]
+; CHECK-NEXT:  .LBB1_5: // %else8
+; CHECK-NEXT:    st1b { z0.h }, p0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB1_6: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldrb w9, [x9]
+; CHECK-NEXT:    fmov s0, w9
+; CHECK-NEXT:    tbz w8, #1, .LBB1_2
+; CHECK-NEXT:  .LBB1_7: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[2], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB1_3
+; CHECK-NEXT:  .LBB1_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[4], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB1_4
+; CHECK-NEXT:    b .LBB1_5
   %cval = load <4 x i8>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = icmp eq <4 x i8> %cval, zeroinitializer
   %vals = call <4 x i8> @llvm.masked.gather.v4i8(<4 x ptr> %ptrs, i32 8, <4 x i1> %mask, <4 x i8> undef)
   store <4 x i8> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v8i8(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_gather_v8i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI2_0
+; CHECK-NEXT:    ldr d0, [x0]
+; CHECK-NEXT:    ptrue p0.b, vl8
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI2_0]
+; CHECK-NEXT:    cmpeq p0.b, p0/z, z0.b, z1.b
+; CHECK-NEXT:    mov z0.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.b, z0.b[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    mov z2.b, z0.b[2]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z3.b, z0.b[3]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.b, z0.b[4]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    mov z5.b, z0.b[5]
+; CHECK-NEXT:    mov z1.b, z0.b[6]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    bfi w9, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    mov z6.b, z0.b[7]
+; CHECK-NEXT:    bfi w9, w8, #4, #1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    and w8, w11, #0x1
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    orr w8, w9, w8, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbnz w9, #0, .LBB2_10
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB2_11
+; CHECK-NEXT:  .LBB2_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB2_12
+; CHECK-NEXT:  .LBB2_3: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB2_13
+; CHECK-NEXT:  .LBB2_4: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB2_14
+; CHECK-NEXT:  .LBB2_5: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB2_15
+; CHECK-NEXT:  .LBB2_6: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB2_16
+; CHECK-NEXT:  .LBB2_7: // %else17
+; CHECK-NEXT:    tbz w8, #7, .LBB2_9
+; CHECK-NEXT:  .LBB2_8: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.b }[7], [x8]
+; CHECK-NEXT:  .LBB2_9: // %else20
+; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB2_10: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr b0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB2_2
+; CHECK-NEXT:  .LBB2_11: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB2_3
+; CHECK-NEXT:  .LBB2_12: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB2_4
+; CHECK-NEXT:  .LBB2_13: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB2_5
+; CHECK-NEXT:  .LBB2_14: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB2_6
+; CHECK-NEXT:  .LBB2_15: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB2_7
+; CHECK-NEXT:  .LBB2_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB2_8
+; CHECK-NEXT:    b .LBB2_9
   %cval = load <8 x i8>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = icmp eq <8 x i8> %cval, zeroinitializer
   %vals = call <8 x i8> @llvm.masked.gather.v8i8(<8 x ptr> %ptrs, i32 8, <8 x i1> %mask, <8 x i8> undef)
   store <8 x i8> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v16i8(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v16i8:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI3_0
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmeq v0.16b, v0.16b, #0
-; CHECK-NEXT:    sunpklo z0.h, z0.b
-; CHECK-NEXT:    sunpklo z0.s, z0.h
-; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1b { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
-; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
-; CHECK-NEXT:    uzp1 z0.b, z0.b, z0.b
+; CHECK-NEXT:    ptrue p0.b, vl16
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI3_0]
+; CHECK-NEXT:    cmpeq p0.b, p0/z, z0.b, z1.b
+; CHECK-NEXT:    mov z0.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.b, z0.b[1]
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    mov z2.b, z0.b[2]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z3.b, z0.b[3]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z4.b, z0.b[4]
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.b, z0.b[5]
+; CHECK-NEXT:    bfi w10, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    mov z6.b, z0.b[6]
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z7.b, z0.b[7]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    bfi w10, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    bfi w10, w8, #4, #1
+; CHECK-NEXT:    fmov w8, s7
+; CHECK-NEXT:    mov z16.b, z0.b[8]
+; CHECK-NEXT:    bfi w10, w9, #5, #1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    mov z17.b, z0.b[9]
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z18.b, z0.b[10]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #6
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w10, w8, lsl #7
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    mov z1.b, z0.b[11]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z2.b, z0.b[12]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #9
+; CHECK-NEXT:    mov z19.b, z0.b[13]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #10
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z3.b, z0.b[14]
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z0.b, z0.b[15]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    tbnz w9, #0, .LBB3_18
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB3_19
+; CHECK-NEXT:  .LBB3_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB3_20
+; CHECK-NEXT:  .LBB3_3: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB3_21
+; CHECK-NEXT:  .LBB3_4: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB3_22
+; CHECK-NEXT:  .LBB3_5: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB3_23
+; CHECK-NEXT:  .LBB3_6: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB3_24
+; CHECK-NEXT:  .LBB3_7: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB3_25
+; CHECK-NEXT:  .LBB3_8: // %else20
+; CHECK-NEXT:    ldr q1, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB3_26
+; CHECK-NEXT:  .LBB3_9: // %else23
+; CHECK-NEXT:    tbnz w8, #9, .LBB3_27
+; CHECK-NEXT:  .LBB3_10: // %else26
+; CHECK-NEXT:    ldr q1, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB3_28
+; CHECK-NEXT:  .LBB3_11: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB3_29
+; CHECK-NEXT:  .LBB3_12: // %else32
+; CHECK-NEXT:    ldr q1, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB3_30
+; CHECK-NEXT:  .LBB3_13: // %else35
+; CHECK-NEXT:    tbnz w8, #13, .LBB3_31
+; CHECK-NEXT:  .LBB3_14: // %else38
+; CHECK-NEXT:    ldr q1, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB3_32
+; CHECK-NEXT:  .LBB3_15: // %else41
+; CHECK-NEXT:    tbz w8, #15, .LBB3_17
+; CHECK-NEXT:  .LBB3_16: // %cond.load43
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.b }[15], [x8]
+; CHECK-NEXT:  .LBB3_17: // %else44
 ; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB3_18: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr b0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB3_2
+; CHECK-NEXT:  .LBB3_19: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB3_3
+; CHECK-NEXT:  .LBB3_20: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB3_4
+; CHECK-NEXT:  .LBB3_21: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB3_5
+; CHECK-NEXT:  .LBB3_22: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB3_6
+; CHECK-NEXT:  .LBB3_23: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB3_7
+; CHECK-NEXT:  .LBB3_24: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[6], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB3_8
+; CHECK-NEXT:  .LBB3_25: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[7], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB3_9
+; CHECK-NEXT:  .LBB3_26: // %cond.load22
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[8], [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB3_10
+; CHECK-NEXT:  .LBB3_27: // %cond.load25
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[9], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #80]
+; CHECK-NEXT:    tbz w8, #10, .LBB3_11
+; CHECK-NEXT:  .LBB3_28: // %cond.load28
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[10], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB3_12
+; CHECK-NEXT:  .LBB3_29: // %cond.load31
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[11], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB3_13
+; CHECK-NEXT:  .LBB3_30: // %cond.load34
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[12], [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB3_14
+; CHECK-NEXT:  .LBB3_31: // %cond.load37
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[13], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB3_15
+; CHECK-NEXT:  .LBB3_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[14], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB3_16
+; CHECK-NEXT:    b .LBB3_17
   %cval = load <16 x i8>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = icmp eq <16 x i8> %cval, zeroinitializer
   %vals = call <16 x i8> @llvm.masked.gather.v16i8(<16 x ptr> %ptrs, i32 8, <16 x i1> %mask, <16 x i8> undef)
   store <16 x i8> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v32i8(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v32i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.b, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1b { z0.b }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.b, p0/z, z0.b, #0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1b { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1b { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI4_0
+; CHECK-NEXT:    ptrue p0.b, vl16
+; CHECK-NEXT:    ldp q3, q0, [x0]
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI4_0]
+; CHECK-NEXT:    cmpeq p1.b, p0/z, z0.b, z2.b
+; CHECK-NEXT:    cmpeq p0.b, p0/z, z3.b, z2.b
+; CHECK-NEXT:    mov z0.b, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    mov z1.b, z0.b[1]
+; CHECK-NEXT:    mov z4.b, z0.b[2]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z5.b, z0.b[3]
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z6.b, z0.b[4]
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z7.b, z0.b[5]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s6
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    mov z16.b, z0.b[6]
+; CHECK-NEXT:    bfi w9, w11, #19, #1
+; CHECK-NEXT:    mov z17.b, z0.b[7]
+; CHECK-NEXT:    bfi w9, w8, #20, #1
+; CHECK-NEXT:    fmov w8, s16
+; CHECK-NEXT:    mov z18.b, z0.b[8]
+; CHECK-NEXT:    bfi w9, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z19.b, z0.b[9]
+; CHECK-NEXT:    mov z20.b, z0.b[10]
+; CHECK-NEXT:    mov z1.b, z0.b[11]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w9, w8, lsl #22
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #23
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    fmov w9, s19
+; CHECK-NEXT:    mov z4.b, z0.b[12]
+; CHECK-NEXT:    mov z5.b, z0.b[13]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z3.b, z2.b[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w12, s2
+; CHECK-NEXT:    mov z6.b, z2.b[5]
+; CHECK-NEXT:    mov z7.b, z2.b[6]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z3.b, z2.b[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z4.b, z2.b[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z5.b, z2.b[4]
+; CHECK-NEXT:    bfi w12, w11, #1, #1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z16.b, z2.b[7]
+; CHECK-NEXT:    bfi w12, w9, #2, #1
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    bfi w12, w10, #3, #1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    bfi w12, w11, #4, #1
+; CHECK-NEXT:    mov z17.b, z2.b[8]
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    bfi w12, w9, #5, #1
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    mov z18.b, z2.b[9]
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    mov z19.b, z2.b[10]
+; CHECK-NEXT:    orr w9, w12, w9, lsl #6
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z20.b, z2.b[11]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z21.b, z2.b[12]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #7
+; CHECK-NEXT:    fmov w11, s19
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    mov z22.b, z2.b[13]
+; CHECK-NEXT:    orr w9, w9, w12, lsl #9
+; CHECK-NEXT:    fmov w12, s21
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z23.b, z2.b[14]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z1.b, z0.b[14]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    orr w9, w9, w11, lsl #12
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    fmov w12, s1
+; CHECK-NEXT:    mov z24.b, z2.b[15]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z0.b, z0.b[15]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s0
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB4_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr b0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB4_3
+; CHECK-NEXT:    b .LBB4_4
+; CHECK-NEXT:  .LBB4_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB4_4
+; CHECK-NEXT:  .LBB4_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[1], [x9]
+; CHECK-NEXT:  .LBB4_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB4_20
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB4_21
+; CHECK-NEXT:  .LBB4_6: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB4_22
+; CHECK-NEXT:  .LBB4_7: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB4_23
+; CHECK-NEXT:  .LBB4_8: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB4_24
+; CHECK-NEXT:  .LBB4_9: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB4_25
+; CHECK-NEXT:  .LBB4_10: // %else20
+; CHECK-NEXT:    ldr q1, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB4_26
+; CHECK-NEXT:  .LBB4_11: // %else23
+; CHECK-NEXT:    tbnz w8, #9, .LBB4_27
+; CHECK-NEXT:  .LBB4_12: // %else26
+; CHECK-NEXT:    ldr q1, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB4_28
+; CHECK-NEXT:  .LBB4_13: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB4_29
+; CHECK-NEXT:  .LBB4_14: // %else32
+; CHECK-NEXT:    ldr q1, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB4_30
+; CHECK-NEXT:  .LBB4_15: // %else35
+; CHECK-NEXT:    tbnz w8, #13, .LBB4_31
+; CHECK-NEXT:  .LBB4_16: // %else38
+; CHECK-NEXT:    ldr q1, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB4_32
+; CHECK-NEXT:  .LBB4_17: // %else41
+; CHECK-NEXT:    tbnz w8, #15, .LBB4_33
+; CHECK-NEXT:  .LBB4_18: // %else44
+; CHECK-NEXT:    ldr q2, [x1, #128]
+; CHECK-NEXT:    tbz w8, #16, .LBB4_34
+; CHECK-NEXT:  .LBB4_19: // %cond.load46
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #17, .LBB4_35
+; CHECK-NEXT:    b .LBB4_36
+; CHECK-NEXT:  .LBB4_20: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB4_6
+; CHECK-NEXT:  .LBB4_21: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB4_7
+; CHECK-NEXT:  .LBB4_22: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB4_8
+; CHECK-NEXT:  .LBB4_23: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB4_9
+; CHECK-NEXT:  .LBB4_24: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[6], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB4_10
+; CHECK-NEXT:  .LBB4_25: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[7], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB4_11
+; CHECK-NEXT:  .LBB4_26: // %cond.load22
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[8], [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB4_12
+; CHECK-NEXT:  .LBB4_27: // %cond.load25
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[9], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #80]
+; CHECK-NEXT:    tbz w8, #10, .LBB4_13
+; CHECK-NEXT:  .LBB4_28: // %cond.load28
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[10], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB4_14
+; CHECK-NEXT:  .LBB4_29: // %cond.load31
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[11], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB4_15
+; CHECK-NEXT:  .LBB4_30: // %cond.load34
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[12], [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB4_16
+; CHECK-NEXT:  .LBB4_31: // %cond.load37
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[13], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB4_17
+; CHECK-NEXT:  .LBB4_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[14], [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB4_18
+; CHECK-NEXT:  .LBB4_33: // %cond.load43
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.b }[15], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #128]
+; CHECK-NEXT:    tbnz w8, #16, .LBB4_19
+; CHECK-NEXT:  .LBB4_34:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #17, .LBB4_36
+; CHECK-NEXT:  .LBB4_35: // %cond.load49
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[1], [x9]
+; CHECK-NEXT:  .LBB4_36: // %else50
+; CHECK-NEXT:    ldr q2, [x1, #144]
+; CHECK-NEXT:    tbnz w8, #18, .LBB4_52
+; CHECK-NEXT:  // %bb.37: // %else53
+; CHECK-NEXT:    tbnz w8, #19, .LBB4_53
+; CHECK-NEXT:  .LBB4_38: // %else56
+; CHECK-NEXT:    ldr q2, [x1, #160]
+; CHECK-NEXT:    tbnz w8, #20, .LBB4_54
+; CHECK-NEXT:  .LBB4_39: // %else59
+; CHECK-NEXT:    tbnz w8, #21, .LBB4_55
+; CHECK-NEXT:  .LBB4_40: // %else62
+; CHECK-NEXT:    ldr q2, [x1, #176]
+; CHECK-NEXT:    tbnz w8, #22, .LBB4_56
+; CHECK-NEXT:  .LBB4_41: // %else65
+; CHECK-NEXT:    tbnz w8, #23, .LBB4_57
+; CHECK-NEXT:  .LBB4_42: // %else68
+; CHECK-NEXT:    ldr q2, [x1, #192]
+; CHECK-NEXT:    tbnz w8, #24, .LBB4_58
+; CHECK-NEXT:  .LBB4_43: // %else71
+; CHECK-NEXT:    tbnz w8, #25, .LBB4_59
+; CHECK-NEXT:  .LBB4_44: // %else74
+; CHECK-NEXT:    ldr q2, [x1, #208]
+; CHECK-NEXT:    tbnz w8, #26, .LBB4_60
+; CHECK-NEXT:  .LBB4_45: // %else77
+; CHECK-NEXT:    tbnz w8, #27, .LBB4_61
+; CHECK-NEXT:  .LBB4_46: // %else80
+; CHECK-NEXT:    ldr q2, [x1, #224]
+; CHECK-NEXT:    tbnz w8, #28, .LBB4_62
+; CHECK-NEXT:  .LBB4_47: // %else83
+; CHECK-NEXT:    tbnz w8, #29, .LBB4_63
+; CHECK-NEXT:  .LBB4_48: // %else86
+; CHECK-NEXT:    ldr q2, [x1, #240]
+; CHECK-NEXT:    tbnz w8, #30, .LBB4_64
+; CHECK-NEXT:  .LBB4_49: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB4_51
+; CHECK-NEXT:  .LBB4_50: // %cond.load91
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ld1 { v1.b }[15], [x8]
+; CHECK-NEXT:  .LBB4_51: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB4_52: // %cond.load52
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[2], [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB4_38
+; CHECK-NEXT:  .LBB4_53: // %cond.load55
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #160]
+; CHECK-NEXT:    tbz w8, #20, .LBB4_39
+; CHECK-NEXT:  .LBB4_54: // %cond.load58
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[4], [x9]
+; CHECK-NEXT:    tbz w8, #21, .LBB4_40
+; CHECK-NEXT:  .LBB4_55: // %cond.load61
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[5], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #176]
+; CHECK-NEXT:    tbz w8, #22, .LBB4_41
+; CHECK-NEXT:  .LBB4_56: // %cond.load64
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[6], [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB4_42
+; CHECK-NEXT:  .LBB4_57: // %cond.load67
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[7], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #192]
+; CHECK-NEXT:    tbz w8, #24, .LBB4_43
+; CHECK-NEXT:  .LBB4_58: // %cond.load70
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[8], [x9]
+; CHECK-NEXT:    tbz w8, #25, .LBB4_44
+; CHECK-NEXT:  .LBB4_59: // %cond.load73
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[9], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #208]
+; CHECK-NEXT:    tbz w8, #26, .LBB4_45
+; CHECK-NEXT:  .LBB4_60: // %cond.load76
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[10], [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB4_46
+; CHECK-NEXT:  .LBB4_61: // %cond.load79
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[11], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #224]
+; CHECK-NEXT:    tbz w8, #28, .LBB4_47
+; CHECK-NEXT:  .LBB4_62: // %cond.load82
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[12], [x9]
+; CHECK-NEXT:    tbz w8, #29, .LBB4_48
+; CHECK-NEXT:  .LBB4_63: // %cond.load85
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[13], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #240]
+; CHECK-NEXT:    tbz w8, #30, .LBB4_49
+; CHECK-NEXT:  .LBB4_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.b }[14], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB4_50
+; CHECK-NEXT:    b .LBB4_51
   %cval = load <32 x i8>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = icmp eq <32 x i8> %cval, zeroinitializer
   %vals = call <32 x i8> @llvm.masked.gather.v32i8(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x i8> undef)
   store <32 x i8> %vals, ptr %a
   ret void
 }
 
 ;
 ; LD1H
 ;
 
 define void @masked_gather_v2i16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v2i16:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    ldrh w9, [x0, #2]
+; CHECK-NEXT:    adrp x8, .LCPI5_0
+; CHECK-NEXT:    ptrue p0.s, vl2
+; CHECK-NEXT:    str w9, [sp, #4]
+; CHECK-NEXT:    ldr d0, [x8, :lo12:.LCPI5_0]
 ; CHECK-NEXT:    ldrh w8, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    str w8, [sp]
+; CHECK-NEXT:    ldr d1, [sp]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
 ; CHECK-NEXT:    ldr q1, [x1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    ldrh w8, [x0, #2]
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    cmeq v0.2s, v0.2s, #0
-; CHECK-NEXT:    sshll v0.2d, v0.2s, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1h { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    ptrue p0.s, vl2
-; CHECK-NEXT:    xtn v0.2s, v0.2d
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbz w9, #0, .LBB5_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldrh w9, [x9]
+; CHECK-NEXT:    fmov s0, w9
+; CHECK-NEXT:  .LBB5_2: // %else
+; CHECK-NEXT:    tbz w8, #1, .LBB5_4
+; CHECK-NEXT:  // %bb.3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x8]
+; CHECK-NEXT:  .LBB5_4: // %else2
 ; CHECK-NEXT:    st1h { z0.s }, p0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <2 x i16>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = icmp eq <2 x i16> %cval, zeroinitializer
   %vals = call <2 x i16> @llvm.masked.gather.v2i16(<2 x ptr> %ptrs, i32 8, <2 x i1> %mask, <2 x i16> undef)
   store <2 x i16> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v4i16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v4i16:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI6_0
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmeq v0.4h, v0.4h, #0
-; CHECK-NEXT:    sunpklo z0.s, z0.h
-; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1h { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
-; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
+; CHECK-NEXT:    ptrue p0.h, vl4
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI6_0]
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z1.h, z0.h[2]
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z0.h, z0.h[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB6_6
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB6_7
+; CHECK-NEXT:  .LBB6_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB6_8
+; CHECK-NEXT:  .LBB6_3: // %else5
+; CHECK-NEXT:    tbz w8, #3, .LBB6_5
+; CHECK-NEXT:  .LBB6_4: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.h }[3], [x8]
+; CHECK-NEXT:  .LBB6_5: // %else8
 ; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB6_6: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB6_2
+; CHECK-NEXT:  .LBB6_7: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB6_3
+; CHECK-NEXT:  .LBB6_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB6_4
+; CHECK-NEXT:    b .LBB6_5
   %cval = load <4 x i16>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = icmp eq <4 x i16> %cval, zeroinitializer
   %vals = call <4 x i16> @llvm.masked.gather.v4i16(<4 x ptr> %ptrs, i32 8, <4 x i1> %mask, <4 x i16> undef)
   store <4 x i16> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v8i16(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_gather_v8i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI7_0
+; CHECK-NEXT:    ldr q0, [x0]
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI7_0]
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z3.h, z0.h[3]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.h, z0.h[4]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    mov z5.h, z0.h[5]
+; CHECK-NEXT:    mov z1.h, z0.h[6]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    bfi w9, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    mov z6.h, z0.h[7]
+; CHECK-NEXT:    bfi w9, w8, #4, #1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    and w8, w11, #0x1
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    orr w8, w9, w8, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbnz w9, #0, .LBB7_10
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB7_11
+; CHECK-NEXT:  .LBB7_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB7_12
+; CHECK-NEXT:  .LBB7_3: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB7_13
+; CHECK-NEXT:  .LBB7_4: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB7_14
+; CHECK-NEXT:  .LBB7_5: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB7_15
+; CHECK-NEXT:  .LBB7_6: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB7_16
+; CHECK-NEXT:  .LBB7_7: // %else17
+; CHECK-NEXT:    tbz w8, #7, .LBB7_9
+; CHECK-NEXT:  .LBB7_8: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.h }[7], [x8]
+; CHECK-NEXT:  .LBB7_9: // %else20
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB7_10: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB7_2
+; CHECK-NEXT:  .LBB7_11: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB7_3
+; CHECK-NEXT:  .LBB7_12: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB7_4
+; CHECK-NEXT:  .LBB7_13: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB7_5
+; CHECK-NEXT:  .LBB7_14: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB7_6
+; CHECK-NEXT:  .LBB7_15: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB7_7
+; CHECK-NEXT:  .LBB7_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB7_8
+; CHECK-NEXT:    b .LBB7_9
   %cval = load <8 x i16>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = icmp eq <8 x i16> %cval, zeroinitializer
   %vals = call <8 x i16> @llvm.masked.gather.v8i16(<8 x ptr> %ptrs, i32 8, <8 x i1> %mask, <8 x i16> undef)
   store <8 x i16> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v16i16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v16i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl16
-; CHECK-NEXT:    ptrue p1.d, vl16
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, #0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1h { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1h { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI8_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI8_0]
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z2.h, z1.h
+; CHECK-NEXT:    mov z0.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    mov z3.h, z0.h[3]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z4.h, z0.h[4]
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    mov z5.h, z0.h[5]
+; CHECK-NEXT:    mov z6.h, z0.h[6]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z7.h, z0.h[7]
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #6
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z3.h, z0.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    orr w8, w8, w10, lsl #8
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z4.h, z0.h[4]
+; CHECK-NEXT:    mov z5.h, z0.h[5]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #9
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    mov z6.h, z0.h[6]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    orr w8, w8, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    tbz w9, #0, .LBB8_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB8_3
+; CHECK-NEXT:    b .LBB8_4
+; CHECK-NEXT:  .LBB8_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB8_4
+; CHECK-NEXT:  .LBB8_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[1], [x9]
+; CHECK-NEXT:  .LBB8_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB8_12
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB8_13
+; CHECK-NEXT:  .LBB8_6: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB8_14
+; CHECK-NEXT:  .LBB8_7: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB8_15
+; CHECK-NEXT:  .LBB8_8: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB8_16
+; CHECK-NEXT:  .LBB8_9: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB8_17
+; CHECK-NEXT:  .LBB8_10: // %else20
+; CHECK-NEXT:    ldr q2, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB8_18
+; CHECK-NEXT:  .LBB8_11: // %cond.load22
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB8_19
+; CHECK-NEXT:    b .LBB8_20
+; CHECK-NEXT:  .LBB8_12: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB8_6
+; CHECK-NEXT:  .LBB8_13: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB8_7
+; CHECK-NEXT:  .LBB8_14: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB8_8
+; CHECK-NEXT:  .LBB8_15: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB8_9
+; CHECK-NEXT:  .LBB8_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[6], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB8_10
+; CHECK-NEXT:  .LBB8_17: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[7], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB8_11
+; CHECK-NEXT:  .LBB8_18:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #9, .LBB8_20
+; CHECK-NEXT:  .LBB8_19: // %cond.load25
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[1], [x9]
+; CHECK-NEXT:  .LBB8_20: // %else26
+; CHECK-NEXT:    ldr q2, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB8_28
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB8_29
+; CHECK-NEXT:  .LBB8_22: // %else32
+; CHECK-NEXT:    ldr q2, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB8_30
+; CHECK-NEXT:  .LBB8_23: // %else35
+; CHECK-NEXT:    tbnz w8, #13, .LBB8_31
+; CHECK-NEXT:  .LBB8_24: // %else38
+; CHECK-NEXT:    ldr q2, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB8_32
+; CHECK-NEXT:  .LBB8_25: // %else41
+; CHECK-NEXT:    tbz w8, #15, .LBB8_27
+; CHECK-NEXT:  .LBB8_26: // %cond.load43
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ld1 { v1.h }[7], [x8]
+; CHECK-NEXT:  .LBB8_27: // %else44
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB8_28: // %cond.load28
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB8_22
+; CHECK-NEXT:  .LBB8_29: // %cond.load31
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB8_23
+; CHECK-NEXT:  .LBB8_30: // %cond.load34
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB8_24
+; CHECK-NEXT:  .LBB8_31: // %cond.load37
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[5], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB8_25
+; CHECK-NEXT:  .LBB8_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB8_26
+; CHECK-NEXT:    b .LBB8_27
   %cval = load <16 x i16>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = icmp eq <16 x i16> %cval, zeroinitializer
   %vals = call <16 x i16> @llvm.masked.gather.v16i16(<16 x ptr> %ptrs, i32 8, <16 x i1> %mask, <16 x i16> undef)
   store <16 x i16> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v32i16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v32i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, #0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1h { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1h { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI9_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q1, q2, [x0, #32]
+; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI9_0]
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z1.h, z0.h
+; CHECK-NEXT:    mov z3.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z2.h, z0.h
+; CHECK-NEXT:    fmov w8, s3
+; CHECK-NEXT:    mov z5.h, z3.h[1]
+; CHECK-NEXT:    mov z6.h, z3.h[2]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    mov z7.h, z3.h[3]
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z16.h, z3.h[4]
+; CHECK-NEXT:    mov z18.h, z3.h[6]
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z17.h, z3.h[5]
+; CHECK-NEXT:    mov z19.h, z3.h[7]
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s19
+; CHECK-NEXT:    bfi w8, w12, #20, #1
+; CHECK-NEXT:    mov z2.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    bfi w8, w9, #21, #1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z3.h, z2.h[1]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #22
+; CHECK-NEXT:    mov z5.h, z2.h[2]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    ldp q4, q1, [x0]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z6.h, z2.h[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z7.h, z2.h[4]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z4.h, z0.h
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    mov z4.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z5.h, z4.h[1]
+; CHECK-NEXT:    fmov w12, s4
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.h, z4.h[2]
+; CHECK-NEXT:    mov z6.h, z4.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov w12, s6
+; CHECK-NEXT:    bfi w10, w11, #1, #1
+; CHECK-NEXT:    mov z7.h, z4.h[4]
+; CHECK-NEXT:    mov z18.h, z4.h[6]
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z17.h, z4.h[5]
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z19.h, z4.h[7]
+; CHECK-NEXT:    bfi w10, w11, #4, #1
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z1.h, z0.h
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w10, w9, #5, #1
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    orr w9, w10, w11, lsl #6
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    fmov w11, s0
+; CHECK-NEXT:    mov z16.h, z2.h[5]
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    mov z4.h, z0.h[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.h, z0.h[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z6.h, z0.h[4]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z7.h, z0.h[5]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.h, z0.h[6]
+; CHECK-NEXT:    orr w8, w8, w12, lsl #29
+; CHECK-NEXT:    fmov w12, s6
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z3.h, z2.h[6]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov w12, s3
+; CHECK-NEXT:    mov z17.h, z0.h[7]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z2.h, z2.h[7]
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB9_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB9_3
+; CHECK-NEXT:    b .LBB9_4
+; CHECK-NEXT:  .LBB9_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB9_4
+; CHECK-NEXT:  .LBB9_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[1], [x9]
+; CHECK-NEXT:  .LBB9_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB9_12
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB9_13
+; CHECK-NEXT:  .LBB9_6: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB9_14
+; CHECK-NEXT:  .LBB9_7: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB9_15
+; CHECK-NEXT:  .LBB9_8: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB9_16
+; CHECK-NEXT:  .LBB9_9: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB9_17
+; CHECK-NEXT:  .LBB9_10: // %else20
+; CHECK-NEXT:    ldr q2, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB9_18
+; CHECK-NEXT:  .LBB9_11: // %cond.load22
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB9_19
+; CHECK-NEXT:    b .LBB9_20
+; CHECK-NEXT:  .LBB9_12: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB9_6
+; CHECK-NEXT:  .LBB9_13: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB9_7
+; CHECK-NEXT:  .LBB9_14: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB9_8
+; CHECK-NEXT:  .LBB9_15: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB9_9
+; CHECK-NEXT:  .LBB9_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[6], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB9_10
+; CHECK-NEXT:  .LBB9_17: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[7], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB9_11
+; CHECK-NEXT:  .LBB9_18:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #9, .LBB9_20
+; CHECK-NEXT:  .LBB9_19: // %cond.load25
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[1], [x9]
+; CHECK-NEXT:  .LBB9_20: // %else26
+; CHECK-NEXT:    ldr q2, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB9_28
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB9_29
+; CHECK-NEXT:  .LBB9_22: // %else32
+; CHECK-NEXT:    ldr q2, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB9_30
+; CHECK-NEXT:  .LBB9_23: // %else35
+; CHECK-NEXT:    tbnz w8, #13, .LBB9_31
+; CHECK-NEXT:  .LBB9_24: // %else38
+; CHECK-NEXT:    ldr q2, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB9_32
+; CHECK-NEXT:  .LBB9_25: // %else41
+; CHECK-NEXT:    tbnz w8, #15, .LBB9_33
+; CHECK-NEXT:  .LBB9_26: // %else44
+; CHECK-NEXT:    ldr q3, [x1, #128]
+; CHECK-NEXT:    tbz w8, #16, .LBB9_34
+; CHECK-NEXT:  .LBB9_27: // %cond.load46
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #17, .LBB9_35
+; CHECK-NEXT:    b .LBB9_36
+; CHECK-NEXT:  .LBB9_28: // %cond.load28
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB9_22
+; CHECK-NEXT:  .LBB9_29: // %cond.load31
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB9_23
+; CHECK-NEXT:  .LBB9_30: // %cond.load34
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB9_24
+; CHECK-NEXT:  .LBB9_31: // %cond.load37
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[5], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB9_25
+; CHECK-NEXT:  .LBB9_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[6], [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB9_26
+; CHECK-NEXT:  .LBB9_33: // %cond.load43
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[7], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #128]
+; CHECK-NEXT:    tbnz w8, #16, .LBB9_27
+; CHECK-NEXT:  .LBB9_34:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #17, .LBB9_36
+; CHECK-NEXT:  .LBB9_35: // %cond.load49
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[1], [x9]
+; CHECK-NEXT:  .LBB9_36: // %else50
+; CHECK-NEXT:    ldr q3, [x1, #144]
+; CHECK-NEXT:    tbnz w8, #18, .LBB9_44
+; CHECK-NEXT:  // %bb.37: // %else53
+; CHECK-NEXT:    tbnz w8, #19, .LBB9_45
+; CHECK-NEXT:  .LBB9_38: // %else56
+; CHECK-NEXT:    ldr q3, [x1, #160]
+; CHECK-NEXT:    tbnz w8, #20, .LBB9_46
+; CHECK-NEXT:  .LBB9_39: // %else59
+; CHECK-NEXT:    tbnz w8, #21, .LBB9_47
+; CHECK-NEXT:  .LBB9_40: // %else62
+; CHECK-NEXT:    ldr q3, [x1, #176]
+; CHECK-NEXT:    tbnz w8, #22, .LBB9_48
+; CHECK-NEXT:  .LBB9_41: // %else65
+; CHECK-NEXT:    tbnz w8, #23, .LBB9_49
+; CHECK-NEXT:  .LBB9_42: // %else68
+; CHECK-NEXT:    ldr q4, [x1, #192]
+; CHECK-NEXT:    tbz w8, #24, .LBB9_50
+; CHECK-NEXT:  .LBB9_43: // %cond.load70
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #25, .LBB9_51
+; CHECK-NEXT:    b .LBB9_52
+; CHECK-NEXT:  .LBB9_44: // %cond.load52
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB9_38
+; CHECK-NEXT:  .LBB9_45: // %cond.load55
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[3], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #160]
+; CHECK-NEXT:    tbz w8, #20, .LBB9_39
+; CHECK-NEXT:  .LBB9_46: // %cond.load58
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #21, .LBB9_40
+; CHECK-NEXT:  .LBB9_47: // %cond.load61
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[5], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #176]
+; CHECK-NEXT:    tbz w8, #22, .LBB9_41
+; CHECK-NEXT:  .LBB9_48: // %cond.load64
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[6], [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB9_42
+; CHECK-NEXT:  .LBB9_49: // %cond.load67
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[7], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #192]
+; CHECK-NEXT:    tbnz w8, #24, .LBB9_43
+; CHECK-NEXT:  .LBB9_50:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #25, .LBB9_52
+; CHECK-NEXT:  .LBB9_51: // %cond.load73
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[1], [x9]
+; CHECK-NEXT:  .LBB9_52: // %else74
+; CHECK-NEXT:    ldr q4, [x1, #208]
+; CHECK-NEXT:    tbnz w8, #26, .LBB9_60
+; CHECK-NEXT:  // %bb.53: // %else77
+; CHECK-NEXT:    tbnz w8, #27, .LBB9_61
+; CHECK-NEXT:  .LBB9_54: // %else80
+; CHECK-NEXT:    ldr q4, [x1, #224]
+; CHECK-NEXT:    tbnz w8, #28, .LBB9_62
+; CHECK-NEXT:  .LBB9_55: // %else83
+; CHECK-NEXT:    tbnz w8, #29, .LBB9_63
+; CHECK-NEXT:  .LBB9_56: // %else86
+; CHECK-NEXT:    ldr q4, [x1, #240]
+; CHECK-NEXT:    tbnz w8, #30, .LBB9_64
+; CHECK-NEXT:  .LBB9_57: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB9_59
+; CHECK-NEXT:  .LBB9_58: // %cond.load91
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    ld1 { v3.h }[7], [x8]
+; CHECK-NEXT:  .LBB9_59: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB9_60: // %cond.load76
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB9_54
+; CHECK-NEXT:  .LBB9_61: // %cond.load79
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[3], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #224]
+; CHECK-NEXT:    tbz w8, #28, .LBB9_55
+; CHECK-NEXT:  .LBB9_62: // %cond.load82
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #29, .LBB9_56
+; CHECK-NEXT:  .LBB9_63: // %cond.load85
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[5], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #240]
+; CHECK-NEXT:    tbz w8, #30, .LBB9_57
+; CHECK-NEXT:  .LBB9_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB9_58
+; CHECK-NEXT:    b .LBB9_59
   %cval = load <32 x i16>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = icmp eq <32 x i16> %cval, zeroinitializer
   %vals = call <32 x i16> @llvm.masked.gather.v32i16(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x i16> undef)
   store <32 x i16> %vals, ptr %a
   ret void
 }
 
 ;
 ; LD1W
 ;
 
 define void @masked_gather_v2i32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v2i32:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI10_0
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ptrue p0.s, vl2
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI10_0]
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
 ; CHECK-NEXT:    ldr q1, [x1]
-; CHECK-NEXT:    cmeq v0.2s, v0.2s, #0
-; CHECK-NEXT:    sshll v0.2d, v0.2s, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    xtn v0.2s, v0.2d
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbz w9, #0, .LBB10_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:  .LBB10_2: // %else
+; CHECK-NEXT:    tbz w8, #1, .LBB10_4
+; CHECK-NEXT:  // %bb.3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x8]
+; CHECK-NEXT:  .LBB10_4: // %else2
 ; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <2 x i32>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = icmp eq <2 x i32> %cval, zeroinitializer
   %vals = call <2 x i32> @llvm.masked.gather.v2i32(<2 x ptr> %ptrs, i32 8, <2 x i1> %mask, <2 x i32> undef)
   store <2 x i32> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v4i32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v4i32:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI11_0
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmeq v0.4s, v0.4s, #0
-; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI11_0]
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z1.s, z0.s[2]
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB11_6
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB11_7
+; CHECK-NEXT:  .LBB11_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB11_8
+; CHECK-NEXT:  .LBB11_3: // %else5
+; CHECK-NEXT:    tbz w8, #3, .LBB11_5
+; CHECK-NEXT:  .LBB11_4: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x8]
+; CHECK-NEXT:  .LBB11_5: // %else8
 ; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB11_6: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB11_2
+; CHECK-NEXT:  .LBB11_7: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB11_3
+; CHECK-NEXT:  .LBB11_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB11_4
+; CHECK-NEXT:    b .LBB11_5
   %cval = load <4 x i32>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = icmp eq <4 x i32> %cval, zeroinitializer
   %vals = call <4 x i32> @llvm.masked.gather.v4i32(<4 x ptr> %ptrs, i32 8, <4 x i1> %mask, <4 x i32> undef)
   store <4 x i32> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v8i32(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_gather_v8i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI12_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI12_0]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z2.s, z1.s
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov z3.s, z0.s[2]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z4.s, z0.s[3]
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    mov z0.s, z1.s[1]
+; CHECK-NEXT:    mov z2.s, z1.s[2]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    mov z3.s, z1.s[3]
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbz w9, #0, .LBB12_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB12_3
+; CHECK-NEXT:    b .LBB12_4
+; CHECK-NEXT:  .LBB12_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB12_4
+; CHECK-NEXT:  .LBB12_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB12_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB12_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB12_9
+; CHECK-NEXT:  .LBB12_6: // %else8
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB12_10
+; CHECK-NEXT:  .LBB12_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB12_11
+; CHECK-NEXT:    b .LBB12_12
+; CHECK-NEXT:  .LBB12_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB12_6
+; CHECK-NEXT:  .LBB12_9: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB12_7
+; CHECK-NEXT:  .LBB12_10:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #5, .LBB12_12
+; CHECK-NEXT:  .LBB12_11: // %cond.load13
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:  .LBB12_12: // %else14
+; CHECK-NEXT:    ldr q2, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB12_16
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    tbz w8, #7, .LBB12_15
+; CHECK-NEXT:  .LBB12_14: // %cond.load19
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ld1 { v1.s }[3], [x8]
+; CHECK-NEXT:  .LBB12_15: // %else20
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB12_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB12_14
+; CHECK-NEXT:    b .LBB12_15
   %cval = load <8 x i32>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = icmp eq <8 x i32> %cval, zeroinitializer
   %vals = call <8 x i32> @llvm.masked.gather.v8i32(<8 x ptr> %ptrs, i32 8, <8 x i1> %mask, <8 x i32> undef)
   store <8 x i32> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v16i32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v16i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl16
-; CHECK-NEXT:    ptrue p1.d, vl16
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, #0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI13_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI13_0]
+; CHECK-NEXT:    ldp q4, q3, [x0, #32]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z2.s, z1.s
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov z5.s, z0.s[2]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z6.s, z0.s[3]
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    mov z5.s, z0.s[2]
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z6.s, z0.s[3]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z4.s, z1.s
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    fmov w11, s0
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    mov z4.s, z0.s[2]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #7
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    fmov w11, s0
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z3.s, z1.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    mov z3.s, z0.s[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #9
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #10
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    tbz w9, #0, .LBB13_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB13_3
+; CHECK-NEXT:    b .LBB13_4
+; CHECK-NEXT:  .LBB13_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB13_4
+; CHECK-NEXT:  .LBB13_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB13_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB13_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB13_9
+; CHECK-NEXT:  .LBB13_6: // %else8
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB13_10
+; CHECK-NEXT:  .LBB13_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB13_11
+; CHECK-NEXT:    b .LBB13_12
+; CHECK-NEXT:  .LBB13_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB13_6
+; CHECK-NEXT:  .LBB13_9: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB13_7
+; CHECK-NEXT:  .LBB13_10:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #5, .LBB13_12
+; CHECK-NEXT:  .LBB13_11: // %cond.load13
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:  .LBB13_12: // %else14
+; CHECK-NEXT:    ldr q2, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB13_16
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB13_17
+; CHECK-NEXT:  .LBB13_14: // %else20
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB13_18
+; CHECK-NEXT:  .LBB13_15: // %cond.load22
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB13_19
+; CHECK-NEXT:    b .LBB13_20
+; CHECK-NEXT:  .LBB13_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB13_14
+; CHECK-NEXT:  .LBB13_17: // %cond.load19
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[3], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB13_15
+; CHECK-NEXT:  .LBB13_18:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #9, .LBB13_20
+; CHECK-NEXT:  .LBB13_19: // %cond.load25
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB13_20: // %else26
+; CHECK-NEXT:    ldr q3, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB13_24
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB13_25
+; CHECK-NEXT:  .LBB13_22: // %else32
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB13_26
+; CHECK-NEXT:  .LBB13_23: // %cond.load34
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #13, .LBB13_27
+; CHECK-NEXT:    b .LBB13_28
+; CHECK-NEXT:  .LBB13_24: // %cond.load28
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB13_22
+; CHECK-NEXT:  .LBB13_25: // %cond.load31
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB13_23
+; CHECK-NEXT:  .LBB13_26:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #13, .LBB13_28
+; CHECK-NEXT:  .LBB13_27: // %cond.load37
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:  .LBB13_28: // %else38
+; CHECK-NEXT:    ldr q4, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB13_32
+; CHECK-NEXT:  // %bb.29: // %else41
+; CHECK-NEXT:    tbz w8, #15, .LBB13_31
+; CHECK-NEXT:  .LBB13_30: // %cond.load43
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    ld1 { v3.s }[3], [x8]
+; CHECK-NEXT:  .LBB13_31: // %else44
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB13_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB13_30
+; CHECK-NEXT:    b .LBB13_31
   %cval = load <16 x i32>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = icmp eq <16 x i32> %cval, zeroinitializer
   %vals = call <16 x i32> @llvm.masked.gather.v16i32(<16 x ptr> %ptrs, i32 8, <16 x i1> %mask, <16 x i32> undef)
   store <16 x i32> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v32i32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v32i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, #0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI14_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q1, q2, [x0, #64]
+; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI14_0]
+; CHECK-NEXT:    ldp q4, q3, [x0, #96]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z5.s, z1.s[1]
+; CHECK-NEXT:    mov z6.s, z1.s[2]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    mov z7.s, z1.s[3]
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z5.s, z2.s[2]
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    mov z1.s, z2.s[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z4.s, z0.s
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    mov z5.s, z4.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    mov z6.s, z4.s[2]
+; CHECK-NEXT:    mov z7.s, z4.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z3.s, z0.s
+; CHECK-NEXT:    ldp q16, q4, [x0]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    mov z5.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z16.s, z0.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    mov z17.s, z16.s[1]
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z17.s, z16.s[2]
+; CHECK-NEXT:    mov z18.s, z16.s[3]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z4.s, z0.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.s, z5.s[1]
+; CHECK-NEXT:    fmov w13, s4
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z16.s, z4.s[2]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z7.s, z5.s[2]
+; CHECK-NEXT:    ldp q2, q1, [x0, #32]
+; CHECK-NEXT:    mov z3.s, z5.s[3]
+; CHECK-NEXT:    mov z5.s, z4.s[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    fmov w13, s16
+; CHECK-NEXT:    fmov w12, s5
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z17.s, z4.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    orr w9, w10, w13, lsl #6
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s4
+; CHECK-NEXT:    mov z5.s, z2.s[2]
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov w12, s0
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z4.s, z0.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB14_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB14_3
+; CHECK-NEXT:    b .LBB14_4
+; CHECK-NEXT:  .LBB14_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB14_4
+; CHECK-NEXT:  .LBB14_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB14_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB14_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB14_9
+; CHECK-NEXT:  .LBB14_6: // %else8
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB14_10
+; CHECK-NEXT:  .LBB14_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB14_11
+; CHECK-NEXT:    b .LBB14_12
+; CHECK-NEXT:  .LBB14_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB14_6
+; CHECK-NEXT:  .LBB14_9: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB14_7
+; CHECK-NEXT:  .LBB14_10:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #5, .LBB14_12
+; CHECK-NEXT:  .LBB14_11: // %cond.load13
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:  .LBB14_12: // %else14
+; CHECK-NEXT:    ldr q2, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB14_16
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB14_17
+; CHECK-NEXT:  .LBB14_14: // %else20
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB14_18
+; CHECK-NEXT:  .LBB14_15: // %cond.load22
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB14_19
+; CHECK-NEXT:    b .LBB14_20
+; CHECK-NEXT:  .LBB14_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB14_14
+; CHECK-NEXT:  .LBB14_17: // %cond.load19
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[3], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB14_15
+; CHECK-NEXT:  .LBB14_18:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #9, .LBB14_20
+; CHECK-NEXT:  .LBB14_19: // %cond.load25
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB14_20: // %else26
+; CHECK-NEXT:    ldr q3, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB14_24
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB14_25
+; CHECK-NEXT:  .LBB14_22: // %else32
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB14_26
+; CHECK-NEXT:  .LBB14_23: // %cond.load34
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #13, .LBB14_27
+; CHECK-NEXT:    b .LBB14_28
+; CHECK-NEXT:  .LBB14_24: // %cond.load28
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB14_22
+; CHECK-NEXT:  .LBB14_25: // %cond.load31
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB14_23
+; CHECK-NEXT:  .LBB14_26:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #13, .LBB14_28
+; CHECK-NEXT:  .LBB14_27: // %cond.load37
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:  .LBB14_28: // %else38
+; CHECK-NEXT:    ldr q4, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB14_32
+; CHECK-NEXT:  // %bb.29: // %else41
+; CHECK-NEXT:    tbnz w8, #15, .LBB14_33
+; CHECK-NEXT:  .LBB14_30: // %else44
+; CHECK-NEXT:    ldr q5, [x1, #128]
+; CHECK-NEXT:    tbz w8, #16, .LBB14_34
+; CHECK-NEXT:  .LBB14_31: // %cond.load46
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #17, .LBB14_35
+; CHECK-NEXT:    b .LBB14_36
+; CHECK-NEXT:  .LBB14_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB14_30
+; CHECK-NEXT:  .LBB14_33: // %cond.load43
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[3], [x9]
+; CHECK-NEXT:    ldr q5, [x1, #128]
+; CHECK-NEXT:    tbnz w8, #16, .LBB14_31
+; CHECK-NEXT:  .LBB14_34:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    tbz w8, #17, .LBB14_36
+; CHECK-NEXT:  .LBB14_35: // %cond.load49
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[1], [x9]
+; CHECK-NEXT:  .LBB14_36: // %else50
+; CHECK-NEXT:    ldr q5, [x1, #144]
+; CHECK-NEXT:    tbnz w8, #18, .LBB14_40
+; CHECK-NEXT:  // %bb.37: // %else53
+; CHECK-NEXT:    tbnz w8, #19, .LBB14_41
+; CHECK-NEXT:  .LBB14_38: // %else56
+; CHECK-NEXT:    ldr q6, [x1, #160]
+; CHECK-NEXT:    tbz w8, #20, .LBB14_42
+; CHECK-NEXT:  .LBB14_39: // %cond.load58
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #21, .LBB14_43
+; CHECK-NEXT:    b .LBB14_44
+; CHECK-NEXT:  .LBB14_40: // %cond.load52
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB14_38
+; CHECK-NEXT:  .LBB14_41: // %cond.load55
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[3], [x9]
+; CHECK-NEXT:    ldr q6, [x1, #160]
+; CHECK-NEXT:    tbnz w8, #20, .LBB14_39
+; CHECK-NEXT:  .LBB14_42:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:    tbz w8, #21, .LBB14_44
+; CHECK-NEXT:  .LBB14_43: // %cond.load61
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[1], [x9]
+; CHECK-NEXT:  .LBB14_44: // %else62
+; CHECK-NEXT:    ldr q6, [x1, #176]
+; CHECK-NEXT:    tbnz w8, #22, .LBB14_48
+; CHECK-NEXT:  // %bb.45: // %else65
+; CHECK-NEXT:    tbnz w8, #23, .LBB14_49
+; CHECK-NEXT:  .LBB14_46: // %else68
+; CHECK-NEXT:    ldr q7, [x1, #192]
+; CHECK-NEXT:    tbz w8, #24, .LBB14_50
+; CHECK-NEXT:  .LBB14_47: // %cond.load70
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #25, .LBB14_51
+; CHECK-NEXT:    b .LBB14_52
+; CHECK-NEXT:  .LBB14_48: // %cond.load64
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB14_46
+; CHECK-NEXT:  .LBB14_49: // %cond.load67
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[3], [x9]
+; CHECK-NEXT:    ldr q7, [x1, #192]
+; CHECK-NEXT:    tbnz w8, #24, .LBB14_47
+; CHECK-NEXT:  .LBB14_50:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    tbz w8, #25, .LBB14_52
+; CHECK-NEXT:  .LBB14_51: // %cond.load73
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[1], [x9]
+; CHECK-NEXT:  .LBB14_52: // %else74
+; CHECK-NEXT:    ldr q7, [x1, #208]
+; CHECK-NEXT:    tbnz w8, #26, .LBB14_56
+; CHECK-NEXT:  // %bb.53: // %else77
+; CHECK-NEXT:    tbnz w8, #27, .LBB14_57
+; CHECK-NEXT:  .LBB14_54: // %else80
+; CHECK-NEXT:    ldr q16, [x1, #224]
+; CHECK-NEXT:    tbz w8, #28, .LBB14_58
+; CHECK-NEXT:  .LBB14_55: // %cond.load82
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #29, .LBB14_59
+; CHECK-NEXT:    b .LBB14_60
+; CHECK-NEXT:  .LBB14_56: // %cond.load76
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB14_54
+; CHECK-NEXT:  .LBB14_57: // %cond.load79
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[3], [x9]
+; CHECK-NEXT:    ldr q16, [x1, #224]
+; CHECK-NEXT:    tbnz w8, #28, .LBB14_55
+; CHECK-NEXT:  .LBB14_58:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    tbz w8, #29, .LBB14_60
+; CHECK-NEXT:  .LBB14_59: // %cond.load85
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:  .LBB14_60: // %else86
+; CHECK-NEXT:    ldr q16, [x1, #240]
+; CHECK-NEXT:    tbnz w8, #30, .LBB14_64
+; CHECK-NEXT:  // %bb.61: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB14_63
+; CHECK-NEXT:  .LBB14_62: // %cond.load91
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x8, d16
+; CHECK-NEXT:    ld1 { v7.s }[3], [x8]
+; CHECK-NEXT:  .LBB14_63: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    stp q4, q5, [x0, #64]
+; CHECK-NEXT:    stp q6, q7, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB14_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB14_62
+; CHECK-NEXT:    b .LBB14_63
   %cval = load <32 x i32>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = icmp eq <32 x i32> %cval, zeroinitializer
   %vals = call <32 x i32> @llvm.masked.gather.v32i32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x i32> undef)
   store <32 x i32> %vals, ptr %a
   ret void
 }
 
; CHECK-NEXT:    ret
   %vals = call <1 x i64> @llvm.masked.gather.v1i64(<1 x ptr> %ptrs, i32 8, <1 x i1> %mask, <1 x i64> undef)
   store <1 x i64> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v2i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v2i64:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI16_0
 ; CHECK-NEXT:    ldr q0, [x0]
 ; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI16_0]
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    mov z0.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x8, d1
 ; CHECK-NEXT:    ldr q1, [x1]
-; CHECK-NEXT:    cmeq v0.2d, v0.2d, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [z1.d]
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbz w9, #0, .LBB16_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:  .LBB16_2: // %else
+; CHECK-NEXT:    tbz w8, #1, .LBB16_4
+; CHECK-NEXT:  // %bb.3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x8]
+; CHECK-NEXT:  .LBB16_4: // %else2
 ; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <2 x i64>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = icmp eq <2 x i64> %cval, zeroinitializer
   %vals = call <2 x i64> @llvm.masked.gather.v2i64(<2 x ptr> %ptrs, i32 8, <2 x i1> %mask, <2 x i64> undef)
   store <2 x i64> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v4i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v4i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmpeq p1.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1d { z0.d }, p1/z, [z1.d]
-; CHECK-NEXT:    st1d { z0.d }, p0, [x0]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI17_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI17_0]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z2.d, z1.d
+; CHECK-NEXT:    mov z0.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z3.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x8, d3
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    mov z0.d, z1.d[1]
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbz w9, #0, .LBB17_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB17_3
+; CHECK-NEXT:    b .LBB17_4
+; CHECK-NEXT:  .LBB17_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB17_4
+; CHECK-NEXT:  .LBB17_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB17_4: // %else2
+; CHECK-NEXT:    ldr q2, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB17_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB17_7
+; CHECK-NEXT:    b .LBB17_8
+; CHECK-NEXT:  .LBB17_6:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #3, .LBB17_8
+; CHECK-NEXT:  .LBB17_7: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ld1 { v1.d }[1], [x8]
+; CHECK-NEXT:  .LBB17_8: // %else8
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <4 x i64>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = icmp eq <4 x i64> %cval, zeroinitializer
   %vals = call <4 x i64> @llvm.masked.gather.v4i64(<4 x ptr> %ptrs, i32 8, <4 x i1> %mask, <4 x i64> undef)
   store <4 x i64> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v8i64(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_gather_v8i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI18_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldp q3, q4, [x0, #32]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI18_0]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    mov z0.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z2.d, z1.d
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z3.d, z1.d
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z4.d, z1.d
+; CHECK-NEXT:    mov z4.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z0.d, z2.d[1]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov x11, d0
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z3.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    mov z0.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov x10, d0
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov x11, d3
+; CHECK-NEXT:    mov z0.d, z3.d[1]
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov x10, d0
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbz w9, #0, .LBB18_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB18_3
+; CHECK-NEXT:    b .LBB18_4
+; CHECK-NEXT:  .LBB18_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB18_4
+; CHECK-NEXT:  .LBB18_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB18_4: // %else2
+; CHECK-NEXT:    ldr q2, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB18_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB18_7
+; CHECK-NEXT:    b .LBB18_8
+; CHECK-NEXT:  .LBB18_6:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #3, .LBB18_8
+; CHECK-NEXT:  .LBB18_7: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[1], [x9]
+; CHECK-NEXT:  .LBB18_8: // %else8
+; CHECK-NEXT:    ldr q3, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB18_10
+; CHECK-NEXT:  // %bb.9: // %cond.load10
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB18_11
+; CHECK-NEXT:    b .LBB18_12
+; CHECK-NEXT:  .LBB18_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #5, .LBB18_12
+; CHECK-NEXT:  .LBB18_11: // %cond.load13
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[1], [x9]
+; CHECK-NEXT:  .LBB18_12: // %else14
+; CHECK-NEXT:    ldr q4, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB18_14
+; CHECK-NEXT:  // %bb.13: // %cond.load16
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB18_15
+; CHECK-NEXT:    b .LBB18_16
+; CHECK-NEXT:  .LBB18_14:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #7, .LBB18_16
+; CHECK-NEXT:  .LBB18_15: // %cond.load19
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    ld1 { v3.d }[1], [x8]
+; CHECK-NEXT:  .LBB18_16: // %else20
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %cval = load <8 x i64>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = icmp eq <8 x i64> %cval, zeroinitializer
   %vals = call <8 x i64> @llvm.masked.gather.v8i64(<8 x ptr> %ptrs, i32 8, <8 x i1> %mask, <8 x i64> undef)
   store <8 x i64> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v16i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v16i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmpeq p1.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1d { z0.d }, p1/z, [z1.d]
-; CHECK-NEXT:    st1d { z0.d }, p0, [x0]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI19_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q1, q2, [x0]
+; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI19_0]
+; CHECK-NEXT:    ldp q3, q4, [x0, #32]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z1.d, z0.d
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z2.d, z0.d
+; CHECK-NEXT:    mov z6.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z3.d, z0.d
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov x11, d1
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    mov z2.d, z5.d[1]
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z4.d, z0.d
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldr q5, [x0, #64]
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov x11, d4
+; CHECK-NEXT:    bfi w8, w9, #5, #1
+; CHECK-NEXT:    ldp q1, q3, [x0, #80]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z5.d, z0.d
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #6
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z1.d, z0.d
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z3.d, z0.d
+; CHECK-NEXT:    fmov x11, d1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    ldr q2, [x0, #112]
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #9
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z2.d, z0.d
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    mov z3.d, z1.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z0.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x11, d0
+; CHECK-NEXT:    mov z1.d, z0.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    tbz w9, #0, .LBB19_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB19_3
+; CHECK-NEXT:    b .LBB19_4
+; CHECK-NEXT:  .LBB19_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB19_4
+; CHECK-NEXT:  .LBB19_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB19_4: // %else2
+; CHECK-NEXT:    ldr q2, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB19_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB19_7
+; CHECK-NEXT:    b .LBB19_8
+; CHECK-NEXT:  .LBB19_6:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #3, .LBB19_8
+; CHECK-NEXT:  .LBB19_7: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[1], [x9]
+; CHECK-NEXT:  .LBB19_8: // %else8
+; CHECK-NEXT:    ldr q3, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB19_10
+; CHECK-NEXT:  // %bb.9: // %cond.load10
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB19_11
+; CHECK-NEXT:    b .LBB19_12
+; CHECK-NEXT:  .LBB19_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #5, .LBB19_12
+; CHECK-NEXT:  .LBB19_11: // %cond.load13
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[1], [x9]
+; CHECK-NEXT:  .LBB19_12: // %else14
+; CHECK-NEXT:    ldr q4, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB19_14
+; CHECK-NEXT:  // %bb.13: // %cond.load16
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB19_15
+; CHECK-NEXT:    b .LBB19_16
+; CHECK-NEXT:  .LBB19_14:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #7, .LBB19_16
+; CHECK-NEXT:  .LBB19_15: // %cond.load19
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[1], [x9]
+; CHECK-NEXT:  .LBB19_16: // %else20
+; CHECK-NEXT:    ldr q5, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB19_18
+; CHECK-NEXT:  // %bb.17: // %cond.load22
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB19_19
+; CHECK-NEXT:    b .LBB19_20
+; CHECK-NEXT:  .LBB19_18:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    tbz w8, #9, .LBB19_20
+; CHECK-NEXT:  .LBB19_19: // %cond.load25
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.d }[1], [x9]
+; CHECK-NEXT:  .LBB19_20: // %else26
+; CHECK-NEXT:    ldr q6, [x1, #80]
+; CHECK-NEXT:    tbz w8, #10, .LBB19_22
+; CHECK-NEXT:  // %bb.21: // %cond.load28
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #11, .LBB19_23
+; CHECK-NEXT:    b .LBB19_24
+; CHECK-NEXT:  .LBB19_22:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:    tbz w8, #11, .LBB19_24
+; CHECK-NEXT:  .LBB19_23: // %cond.load31
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.d }[1], [x9]
+; CHECK-NEXT:  .LBB19_24: // %else32
+; CHECK-NEXT:    ldr q7, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB19_26
+; CHECK-NEXT:  // %bb.25: // %cond.load34
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #13, .LBB19_27
+; CHECK-NEXT:    b .LBB19_28
+; CHECK-NEXT:  .LBB19_26:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    tbz w8, #13, .LBB19_28
+; CHECK-NEXT:  .LBB19_27: // %cond.load37
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.d }[1], [x9]
+; CHECK-NEXT:  .LBB19_28: // %else38
+; CHECK-NEXT:    ldr q16, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB19_30
+; CHECK-NEXT:  // %bb.29: // %cond.load40
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB19_31
+; CHECK-NEXT:    b .LBB19_32
+; CHECK-NEXT:  .LBB19_30:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    tbz w8, #15, .LBB19_32
+; CHECK-NEXT:  .LBB19_31: // %cond.load43
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x8, d16
+; CHECK-NEXT:    ld1 { v7.d }[1], [x8]
+; CHECK-NEXT:  .LBB19_32: // %else44
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    stp q4, q5, [x0, #64]
+; CHECK-NEXT:    stp q6, q7, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <16 x i64>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = icmp eq <16 x i64> %cval, zeroinitializer
   %vals = call <16 x i64> @llvm.masked.gather.v16i64(<16 x ptr> %ptrs, i32 8, <16 x i1> %mask, <16 x i64> undef)
   store <16 x i64> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v32i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v32i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmpeq p1.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1d { z0.d }, p1/z, [z1.d]
-; CHECK-NEXT:    st1d { z0.d }, p0, [x0]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI20_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q5, q7, [x0, #128]
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI20_0]
+; CHECK-NEXT:    ldp q4, q3, [x0, #64]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z5.d, z2.d
+; CHECK-NEXT:    mov z16.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z7.d, z2.d
+; CHECK-NEXT:    fmov x8, d16
+; CHECK-NEXT:    mov z17.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z18.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q19, q7, [x0, #160]
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z20.d, z18.d[1]
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    fmov x11, d20
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z19.d, z2.d
+; CHECK-NEXT:    mov z18.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z7.d, z2.d
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    mov z7.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z19.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    fmov x11, d7
+; CHECK-NEXT:    ldp q17, q16, [x0, #192]
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    mov z18.d, z7.d[1]
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z17.d, z2.d
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z17.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z16.d, z2.d
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    mov z18.d, z17.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z16.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q19, q7, [x0, #224]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    fmov x11, d16
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    mov z17.d, z16.d[1]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z19.d, z2.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    ldp q18, q16, [x0]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z18.d, z2.d
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z19.d, z17.d[1]
+; CHECK-NEXT:    ldp q6, q5, [x0, #32]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z17.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    mov z18.d, z17.d[1]
+; CHECK-NEXT:    fmov x11, d17
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z16.d, z2.d
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z16.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z17.d, z16.d[1]
+; CHECK-NEXT:    fmov x12, d16
+; CHECK-NEXT:    bfi w11, w10, #1, #1
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z6.d, z2.d
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z6.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z5.d, z2.d
+; CHECK-NEXT:    bfi w11, w12, #2, #1
+; CHECK-NEXT:    mov z16.d, z6.d[1]
+; CHECK-NEXT:    fmov x12, d6
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    bfi w11, w10, #3, #1
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    bfi w11, w12, #4, #1
+; CHECK-NEXT:    mov z6.d, z5.d[1]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z4.d, z2.d
+; CHECK-NEXT:    bfi w11, w9, #5, #1
+; CHECK-NEXT:    mov z4.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    fmov x12, d4
+; CHECK-NEXT:    mov z5.d, z4.d[1]
+; CHECK-NEXT:    orr w9, w11, w9, lsl #6
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z7.d, z2.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z4.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z3.d, z2.d
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov x11, d3
+; CHECK-NEXT:    mov z5.d, z3.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z1.d, z2.d
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, z2.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov x11, d5
+; CHECK-NEXT:    fmov x12, d1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    mov z3.d, z1.d[1]
+; CHECK-NEXT:    mov z0.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.d, z4.d[1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov x12, d0
+; CHECK-NEXT:    mov z2.d, z0.d[1]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov x11, d4
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    orr w9, w9, w12, lsl #14
+; CHECK-NEXT:    fmov x12, d1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w12, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB20_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB20_3
+; CHECK-NEXT:    b .LBB20_4
+; CHECK-NEXT:  .LBB20_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB20_4
+; CHECK-NEXT:  .LBB20_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_4: // %else2
+; CHECK-NEXT:    ldr q2, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB20_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB20_7
+; CHECK-NEXT:    b .LBB20_8
+; CHECK-NEXT:  .LBB20_6:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #3, .LBB20_8
+; CHECK-NEXT:  .LBB20_7: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_8: // %else8
+; CHECK-NEXT:    ldr q3, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB20_10
+; CHECK-NEXT:  // %bb.9: // %cond.load10
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB20_11
+; CHECK-NEXT:    b .LBB20_12
+; CHECK-NEXT:  .LBB20_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #5, .LBB20_12
+; CHECK-NEXT:  .LBB20_11: // %cond.load13
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_12: // %else14
+; CHECK-NEXT:    ldr q4, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB20_14
+; CHECK-NEXT:  // %bb.13: // %cond.load16
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB20_15
+; CHECK-NEXT:    b .LBB20_16
+; CHECK-NEXT:  .LBB20_14:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #7, .LBB20_16
+; CHECK-NEXT:  .LBB20_15: // %cond.load19
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_16: // %else20
+; CHECK-NEXT:    ldr q5, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB20_18
+; CHECK-NEXT:  // %bb.17: // %cond.load22
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB20_19
+; CHECK-NEXT:    b .LBB20_20
+; CHECK-NEXT:  .LBB20_18:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    tbz w8, #9, .LBB20_20
+; CHECK-NEXT:  .LBB20_19: // %cond.load25
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_20: // %else26
+; CHECK-NEXT:    ldr q6, [x1, #80]
+; CHECK-NEXT:    tbz w8, #10, .LBB20_22
+; CHECK-NEXT:  // %bb.21: // %cond.load28
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #11, .LBB20_23
+; CHECK-NEXT:    b .LBB20_24
+; CHECK-NEXT:  .LBB20_22:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:    tbz w8, #11, .LBB20_24
+; CHECK-NEXT:  .LBB20_23: // %cond.load31
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_24: // %else32
+; CHECK-NEXT:    ldr q7, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB20_26
+; CHECK-NEXT:  // %bb.25: // %cond.load34
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #13, .LBB20_27
+; CHECK-NEXT:    b .LBB20_28
+; CHECK-NEXT:  .LBB20_26:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    tbz w8, #13, .LBB20_28
+; CHECK-NEXT:  .LBB20_27: // %cond.load37
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_28: // %else38
+; CHECK-NEXT:    ldr q16, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB20_30
+; CHECK-NEXT:  // %bb.29: // %cond.load40
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB20_31
+; CHECK-NEXT:    b .LBB20_32
+; CHECK-NEXT:  .LBB20_30:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    tbz w8, #15, .LBB20_32
+; CHECK-NEXT:  .LBB20_31: // %cond.load43
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_32: // %else44
+; CHECK-NEXT:    ldr q17, [x1, #128]
+; CHECK-NEXT:    tbz w8, #16, .LBB20_34
+; CHECK-NEXT:  // %bb.33: // %cond.load46
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v16.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #17, .LBB20_35
+; CHECK-NEXT:    b .LBB20_36
+; CHECK-NEXT:  .LBB20_34:
+; CHECK-NEXT:    // implicit-def: $q16
+; CHECK-NEXT:    tbz w8, #17, .LBB20_36
+; CHECK-NEXT:  .LBB20_35: // %cond.load49
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v16.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_36: // %else50
+; CHECK-NEXT:    ldr q18, [x1, #144]
+; CHECK-NEXT:    tbz w8, #18, .LBB20_38
+; CHECK-NEXT:  // %bb.37: // %cond.load52
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v17.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #19, .LBB20_39
+; CHECK-NEXT:    b .LBB20_40
+; CHECK-NEXT:  .LBB20_38:
+; CHECK-NEXT:    // implicit-def: $q17
+; CHECK-NEXT:    tbz w8, #19, .LBB20_40
+; CHECK-NEXT:  .LBB20_39: // %cond.load55
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v17.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_40: // %else56
+; CHECK-NEXT:    ldr q19, [x1, #160]
+; CHECK-NEXT:    tbz w8, #20, .LBB20_42
+; CHECK-NEXT:  // %bb.41: // %cond.load58
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v18.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #21, .LBB20_43
+; CHECK-NEXT:    b .LBB20_44
+; CHECK-NEXT:  .LBB20_42:
+; CHECK-NEXT:    // implicit-def: $q18
+; CHECK-NEXT:    tbz w8, #21, .LBB20_44
+; CHECK-NEXT:  .LBB20_43: // %cond.load61
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v18.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_44: // %else62
+; CHECK-NEXT:    ldr q20, [x1, #176]
+; CHECK-NEXT:    tbz w8, #22, .LBB20_46
+; CHECK-NEXT:  // %bb.45: // %cond.load64
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #23, .LBB20_47
+; CHECK-NEXT:    b .LBB20_48
+; CHECK-NEXT:  .LBB20_46:
+; CHECK-NEXT:    // implicit-def: $q19
+; CHECK-NEXT:    tbz w8, #23, .LBB20_48
+; CHECK-NEXT:  .LBB20_47: // %cond.load67
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_48: // %else68
+; CHECK-NEXT:    ldr q21, [x1, #192]
+; CHECK-NEXT:    tbz w8, #24, .LBB20_50
+; CHECK-NEXT:  // %bb.49: // %cond.load70
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v20.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #25, .LBB20_51
+; CHECK-NEXT:    b .LBB20_52
+; CHECK-NEXT:  .LBB20_50:
+; CHECK-NEXT:    // implicit-def: $q20
+; CHECK-NEXT:    tbz w8, #25, .LBB20_52
+; CHECK-NEXT:  .LBB20_51: // %cond.load73
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v20.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_52: // %else74
+; CHECK-NEXT:    ldr q22, [x1, #208]
+; CHECK-NEXT:    tbz w8, #26, .LBB20_54
+; CHECK-NEXT:  // %bb.53: // %cond.load76
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v21.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #27, .LBB20_55
+; CHECK-NEXT:    b .LBB20_56
+; CHECK-NEXT:  .LBB20_54:
+; CHECK-NEXT:    // implicit-def: $q21
+; CHECK-NEXT:    tbz w8, #27, .LBB20_56
+; CHECK-NEXT:  .LBB20_55: // %cond.load79
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v21.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_56: // %else80
+; CHECK-NEXT:    ldr q23, [x1, #224]
+; CHECK-NEXT:    tbz w8, #28, .LBB20_58
+; CHECK-NEXT:  // %bb.57: // %cond.load82
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v22.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #29, .LBB20_59
+; CHECK-NEXT:    b .LBB20_60
+; CHECK-NEXT:  .LBB20_58:
+; CHECK-NEXT:    // implicit-def: $q22
+; CHECK-NEXT:    tbz w8, #29, .LBB20_60
+; CHECK-NEXT:  .LBB20_59: // %cond.load85
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v22.d }[1], [x9]
+; CHECK-NEXT:  .LBB20_60: // %else86
+; CHECK-NEXT:    ldr q24, [x1, #240]
+; CHECK-NEXT:    tbz w8, #30, .LBB20_62
+; CHECK-NEXT:  // %bb.61: // %cond.load88
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v23.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB20_63
+; CHECK-NEXT:    b .LBB20_64
+; CHECK-NEXT:  .LBB20_62:
+; CHECK-NEXT:    // implicit-def: $q23
+; CHECK-NEXT:    tbz w8, #31, .LBB20_64
+; CHECK-NEXT:  .LBB20_63: // %cond.load91
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x8, d24
+; CHECK-NEXT:    ld1 { v23.d }[1], [x8]
+; CHECK-NEXT:  .LBB20_64: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    stp q4, q5, [x0, #64]
+; CHECK-NEXT:    stp q6, q7, [x0, #96]
+; CHECK-NEXT:    stp q16, q17, [x0, #128]
+; CHECK-NEXT:    stp q18, q19, [x0, #160]
+; CHECK-NEXT:    stp q20, q21, [x0, #192]
+; CHECK-NEXT:    stp q22, q23, [x0, #224]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <32 x i64>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = icmp eq <32 x i64> %cval, zeroinitializer
   %vals = call <32 x i64> @llvm.masked.gather.v32i64(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x i64> undef)
   store <32 x i64> %vals, ptr %a
   ret void
 }
 
 ;
 ; LD1H (float)
 ;
 
 define void @masked_gather_v2f16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v2f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr s1, [x0]
-; CHECK-NEXT:    movi d0, #0000000000000000
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    fcmeq v1.4h, v1.4h, #0.0
-; CHECK-NEXT:    sshll v1.4s, v1.4h, #0
-; CHECK-NEXT:    fmov w8, s1
-; CHECK-NEXT:    mov w9, v1.s[1]
-; CHECK-NEXT:    ldr q1, [x1]
-; CHECK-NEXT:    mov v0.h[0], w8
-; CHECK-NEXT:    mov v0.h[1], w9
-; CHECK-NEXT:    shl v0.4h, v0.4h, #15
-; CHECK-NEXT:    cmlt v0.4h, v0.4h, #0
-; CHECK-NEXT:    sunpklo z0.s, z0.h
-; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1h { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
-; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
-; CHECK-NEXT:    str s0, [x0]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI21_0
+; CHECK-NEXT:    ldr s0, [x0]
+; CHECK-NEXT:    ptrue p0.h, vl4
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI21_0]
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    uunpklo z0.s, z0.h
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbz w9, #0, .LBB21_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:  .LBB21_2: // %else
+; CHECK-NEXT:    tbz w8, #1, .LBB21_4
+; CHECK-NEXT:  // %bb.3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ldr h1, [x8]
+; CHECK-NEXT:    str h0, [sp]
+; CHECK-NEXT:    str h1, [sp, #2]
+; CHECK-NEXT:    ldr d0, [sp]
+; CHECK-NEXT:  .LBB21_4: // %else2
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    str w8, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <2 x half>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = fcmp oeq <2 x half> %cval, zeroinitializer
   %vals = call <2 x half> @llvm.masked.gather.v2f16(<2 x ptr> %ptrs, i32 8, <2 x i1> %mask, <2 x half> undef)
   store <2 x half> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v4f16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v4f16:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI22_0
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq v0.4h, v0.4h, #0.0
-; CHECK-NEXT:    sunpklo z0.s, z0.h
-; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1h { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
-; CHECK-NEXT:    uzp1 z0.h, z0.h, z0.h
+; CHECK-NEXT:    ptrue p0.h, vl4
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI22_0]
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z1.h, z0.h[2]
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z0.h, z0.h[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB22_6
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB22_7
+; CHECK-NEXT:  .LBB22_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB22_8
+; CHECK-NEXT:  .LBB22_3: // %else5
+; CHECK-NEXT:    tbz w8, #3, .LBB22_5
+; CHECK-NEXT:  .LBB22_4: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.h }[3], [x8]
+; CHECK-NEXT:  .LBB22_5: // %else8
 ; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB22_6: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB22_2
+; CHECK-NEXT:  .LBB22_7: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB22_3
+; CHECK-NEXT:  .LBB22_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB22_4
+; CHECK-NEXT:    b .LBB22_5
   %cval = load <4 x half>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = fcmp oeq <4 x half> %cval, zeroinitializer
   %vals = call <4 x half> @llvm.masked.gather.v4f16(<4 x ptr> %ptrs, i32 8, <4 x i1> %mask, <4 x half> undef)
   store <4 x half> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v8f16(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_gather_v8f16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI23_0
+; CHECK-NEXT:    ldr q0, [x0]
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI23_0]
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z3.h, z0.h[3]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.h, z0.h[4]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    mov z5.h, z0.h[5]
+; CHECK-NEXT:    mov z1.h, z0.h[6]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    bfi w9, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    mov z6.h, z0.h[7]
+; CHECK-NEXT:    bfi w9, w8, #4, #1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    and w8, w11, #0x1
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    orr w8, w9, w8, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbnz w9, #0, .LBB23_10
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB23_11
+; CHECK-NEXT:  .LBB23_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB23_12
+; CHECK-NEXT:  .LBB23_3: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB23_13
+; CHECK-NEXT:  .LBB23_4: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB23_14
+; CHECK-NEXT:  .LBB23_5: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB23_15
+; CHECK-NEXT:  .LBB23_6: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB23_16
+; CHECK-NEXT:  .LBB23_7: // %else17
+; CHECK-NEXT:    tbz w8, #7, .LBB23_9
+; CHECK-NEXT:  .LBB23_8: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.h }[7], [x8]
+; CHECK-NEXT:  .LBB23_9: // %else20
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB23_10: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB23_2
+; CHECK-NEXT:  .LBB23_11: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB23_3
+; CHECK-NEXT:  .LBB23_12: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB23_4
+; CHECK-NEXT:  .LBB23_13: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB23_5
+; CHECK-NEXT:  .LBB23_14: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB23_6
+; CHECK-NEXT:  .LBB23_15: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB23_7
+; CHECK-NEXT:  .LBB23_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB23_8
+; CHECK-NEXT:    b .LBB23_9
   %cval = load <8 x half>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = fcmp oeq <8 x half> %cval, zeroinitializer
   %vals = call <8 x half> @llvm.masked.gather.v8f16(<8 x ptr> %ptrs, i32 8, <8 x i1> %mask, <8 x half> undef)
   store <8 x half> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v16f16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v16f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl16
-; CHECK-NEXT:    ptrue p1.d, vl16
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1h { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1h { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI24_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI24_0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z2.h, z1.h
+; CHECK-NEXT:    mov z0.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    mov z3.h, z0.h[3]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z4.h, z0.h[4]
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    mov z5.h, z0.h[5]
+; CHECK-NEXT:    mov z6.h, z0.h[6]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z7.h, z0.h[7]
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #6
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z3.h, z0.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    orr w8, w8, w10, lsl #8
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z4.h, z0.h[4]
+; CHECK-NEXT:    mov z5.h, z0.h[5]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #9
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    mov z6.h, z0.h[6]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    orr w8, w8, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    tbz w9, #0, .LBB24_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB24_3
+; CHECK-NEXT:    b .LBB24_4
+; CHECK-NEXT:  .LBB24_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB24_4
+; CHECK-NEXT:  .LBB24_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[1], [x9]
+; CHECK-NEXT:  .LBB24_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB24_12
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB24_13
+; CHECK-NEXT:  .LBB24_6: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB24_14
+; CHECK-NEXT:  .LBB24_7: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB24_15
+; CHECK-NEXT:  .LBB24_8: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB24_16
+; CHECK-NEXT:  .LBB24_9: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB24_17
+; CHECK-NEXT:  .LBB24_10: // %else20
+; CHECK-NEXT:    ldr q2, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB24_18
+; CHECK-NEXT:  .LBB24_11: // %cond.load22
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB24_19
+; CHECK-NEXT:    b .LBB24_20
+; CHECK-NEXT:  .LBB24_12: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB24_6
+; CHECK-NEXT:  .LBB24_13: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB24_7
+; CHECK-NEXT:  .LBB24_14: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB24_8
+; CHECK-NEXT:  .LBB24_15: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB24_9
+; CHECK-NEXT:  .LBB24_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[6], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB24_10
+; CHECK-NEXT:  .LBB24_17: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[7], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB24_11
+; CHECK-NEXT:  .LBB24_18:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #9, .LBB24_20
+; CHECK-NEXT:  .LBB24_19: // %cond.load25
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[1], [x9]
+; CHECK-NEXT:  .LBB24_20: // %else26
+; CHECK-NEXT:    ldr q2, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB24_28
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB24_29
+; CHECK-NEXT:  .LBB24_22: // %else32
+; CHECK-NEXT:    ldr q2, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB24_30
+; CHECK-NEXT:  .LBB24_23: // %else35
+; CHECK-NEXT:    tbnz w8, #13, .LBB24_31
+; CHECK-NEXT:  .LBB24_24: // %else38
+; CHECK-NEXT:    ldr q2, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB24_32
+; CHECK-NEXT:  .LBB24_25: // %else41
+; CHECK-NEXT:    tbz w8, #15, .LBB24_27
+; CHECK-NEXT:  .LBB24_26: // %cond.load43
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ld1 { v1.h }[7], [x8]
+; CHECK-NEXT:  .LBB24_27: // %else44
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB24_28: // %cond.load28
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB24_22
+; CHECK-NEXT:  .LBB24_29: // %cond.load31
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB24_23
+; CHECK-NEXT:  .LBB24_30: // %cond.load34
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB24_24
+; CHECK-NEXT:  .LBB24_31: // %cond.load37
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[5], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB24_25
+; CHECK-NEXT:  .LBB24_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB24_26
+; CHECK-NEXT:    b .LBB24_27
   %cval = load <16 x half>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = fcmp oeq <16 x half> %cval, zeroinitializer
   %vals = call <16 x half> @llvm.masked.gather.v16f16(<16 x ptr> %ptrs, i32 8, <16 x i1> %mask, <16 x half> undef)
   store <16 x half> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v32f16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v32f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1h { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1h { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI25_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q1, q2, [x0, #32]
+; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI25_0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z1.h, z0.h
+; CHECK-NEXT:    mov z3.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z2.h, z0.h
+; CHECK-NEXT:    fmov w8, s3
+; CHECK-NEXT:    mov z5.h, z3.h[1]
+; CHECK-NEXT:    mov z6.h, z3.h[2]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    mov z7.h, z3.h[3]
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z16.h, z3.h[4]
+; CHECK-NEXT:    mov z18.h, z3.h[6]
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z17.h, z3.h[5]
+; CHECK-NEXT:    mov z19.h, z3.h[7]
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s19
+; CHECK-NEXT:    bfi w8, w12, #20, #1
+; CHECK-NEXT:    mov z2.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    bfi w8, w9, #21, #1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z3.h, z2.h[1]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #22
+; CHECK-NEXT:    mov z5.h, z2.h[2]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    ldp q4, q1, [x0]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z6.h, z2.h[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z7.h, z2.h[4]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z4.h, z0.h
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    mov z4.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z5.h, z4.h[1]
+; CHECK-NEXT:    fmov w12, s4
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.h, z4.h[2]
+; CHECK-NEXT:    mov z6.h, z4.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov w12, s6
+; CHECK-NEXT:    bfi w10, w11, #1, #1
+; CHECK-NEXT:    mov z7.h, z4.h[4]
+; CHECK-NEXT:    mov z18.h, z4.h[6]
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z17.h, z4.h[5]
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z19.h, z4.h[7]
+; CHECK-NEXT:    bfi w10, w11, #4, #1
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z1.h, z0.h
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    mov z0.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w10, w9, #5, #1
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    orr w9, w10, w11, lsl #6
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    fmov w11, s0
+; CHECK-NEXT:    mov z16.h, z2.h[5]
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    mov z4.h, z0.h[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.h, z0.h[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z6.h, z0.h[4]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z7.h, z0.h[5]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.h, z0.h[6]
+; CHECK-NEXT:    orr w8, w8, w12, lsl #29
+; CHECK-NEXT:    fmov w12, s6
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z3.h, z2.h[6]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov w12, s3
+; CHECK-NEXT:    mov z17.h, z0.h[7]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z2.h, z2.h[7]
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB25_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr h0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB25_3
+; CHECK-NEXT:    b .LBB25_4
+; CHECK-NEXT:  .LBB25_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB25_4
+; CHECK-NEXT:  .LBB25_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[1], [x9]
+; CHECK-NEXT:  .LBB25_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB25_12
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB25_13
+; CHECK-NEXT:  .LBB25_6: // %else8
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB25_14
+; CHECK-NEXT:  .LBB25_7: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB25_15
+; CHECK-NEXT:  .LBB25_8: // %else14
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB25_16
+; CHECK-NEXT:  .LBB25_9: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB25_17
+; CHECK-NEXT:  .LBB25_10: // %else20
+; CHECK-NEXT:    ldr q2, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB25_18
+; CHECK-NEXT:  .LBB25_11: // %cond.load22
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB25_19
+; CHECK-NEXT:    b .LBB25_20
+; CHECK-NEXT:  .LBB25_12: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB25_6
+; CHECK-NEXT:  .LBB25_13: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[3], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB25_7
+; CHECK-NEXT:  .LBB25_14: // %cond.load10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB25_8
+; CHECK-NEXT:  .LBB25_15: // %cond.load13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[5], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB25_9
+; CHECK-NEXT:  .LBB25_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[6], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB25_10
+; CHECK-NEXT:  .LBB25_17: // %cond.load19
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.h }[7], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB25_11
+; CHECK-NEXT:  .LBB25_18:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #9, .LBB25_20
+; CHECK-NEXT:  .LBB25_19: // %cond.load25
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[1], [x9]
+; CHECK-NEXT:  .LBB25_20: // %else26
+; CHECK-NEXT:    ldr q2, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB25_28
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB25_29
+; CHECK-NEXT:  .LBB25_22: // %else32
+; CHECK-NEXT:    ldr q2, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB25_30
+; CHECK-NEXT:  .LBB25_23: // %else35
+; CHECK-NEXT:    tbnz w8, #13, .LBB25_31
+; CHECK-NEXT:  .LBB25_24: // %else38
+; CHECK-NEXT:    ldr q2, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB25_32
+; CHECK-NEXT:  .LBB25_25: // %else41
+; CHECK-NEXT:    tbnz w8, #15, .LBB25_33
+; CHECK-NEXT:  .LBB25_26: // %else44
+; CHECK-NEXT:    ldr q3, [x1, #128]
+; CHECK-NEXT:    tbz w8, #16, .LBB25_34
+; CHECK-NEXT:  .LBB25_27: // %cond.load46
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #17, .LBB25_35
+; CHECK-NEXT:    b .LBB25_36
+; CHECK-NEXT:  .LBB25_28: // %cond.load28
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB25_22
+; CHECK-NEXT:  .LBB25_29: // %cond.load31
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB25_23
+; CHECK-NEXT:  .LBB25_30: // %cond.load34
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB25_24
+; CHECK-NEXT:  .LBB25_31: // %cond.load37
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[5], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB25_25
+; CHECK-NEXT:  .LBB25_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[6], [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB25_26
+; CHECK-NEXT:  .LBB25_33: // %cond.load43
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.h }[7], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #128]
+; CHECK-NEXT:    tbnz w8, #16, .LBB25_27
+; CHECK-NEXT:  .LBB25_34:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #17, .LBB25_36
+; CHECK-NEXT:  .LBB25_35: // %cond.load49
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[1], [x9]
+; CHECK-NEXT:  .LBB25_36: // %else50
+; CHECK-NEXT:    ldr q3, [x1, #144]
+; CHECK-NEXT:    tbnz w8, #18, .LBB25_44
+; CHECK-NEXT:  // %bb.37: // %else53
+; CHECK-NEXT:    tbnz w8, #19, .LBB25_45
+; CHECK-NEXT:  .LBB25_38: // %else56
+; CHECK-NEXT:    ldr q3, [x1, #160]
+; CHECK-NEXT:    tbnz w8, #20, .LBB25_46
+; CHECK-NEXT:  .LBB25_39: // %else59
+; CHECK-NEXT:    tbnz w8, #21, .LBB25_47
+; CHECK-NEXT:  .LBB25_40: // %else62
+; CHECK-NEXT:    ldr q3, [x1, #176]
+; CHECK-NEXT:    tbnz w8, #22, .LBB25_48
+; CHECK-NEXT:  .LBB25_41: // %else65
+; CHECK-NEXT:    tbnz w8, #23, .LBB25_49
+; CHECK-NEXT:  .LBB25_42: // %else68
+; CHECK-NEXT:    ldr q4, [x1, #192]
+; CHECK-NEXT:    tbz w8, #24, .LBB25_50
+; CHECK-NEXT:  .LBB25_43: // %cond.load70
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #25, .LBB25_51
+; CHECK-NEXT:    b .LBB25_52
+; CHECK-NEXT:  .LBB25_44: // %cond.load52
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB25_38
+; CHECK-NEXT:  .LBB25_45: // %cond.load55
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[3], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #160]
+; CHECK-NEXT:    tbz w8, #20, .LBB25_39
+; CHECK-NEXT:  .LBB25_46: // %cond.load58
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #21, .LBB25_40
+; CHECK-NEXT:  .LBB25_47: // %cond.load61
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[5], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #176]
+; CHECK-NEXT:    tbz w8, #22, .LBB25_41
+; CHECK-NEXT:  .LBB25_48: // %cond.load64
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[6], [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB25_42
+; CHECK-NEXT:  .LBB25_49: // %cond.load67
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[7], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #192]
+; CHECK-NEXT:    tbnz w8, #24, .LBB25_43
+; CHECK-NEXT:  .LBB25_50:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #25, .LBB25_52
+; CHECK-NEXT:  .LBB25_51: // %cond.load73
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[1], [x9]
+; CHECK-NEXT:  .LBB25_52: // %else74
+; CHECK-NEXT:    ldr q4, [x1, #208]
+; CHECK-NEXT:    tbnz w8, #26, .LBB25_60
+; CHECK-NEXT:  // %bb.53: // %else77
+; CHECK-NEXT:    tbnz w8, #27, .LBB25_61
+; CHECK-NEXT:  .LBB25_54: // %else80
+; CHECK-NEXT:    ldr q4, [x1, #224]
+; CHECK-NEXT:    tbnz w8, #28, .LBB25_62
+; CHECK-NEXT:  .LBB25_55: // %else83
+; CHECK-NEXT:    tbnz w8, #29, .LBB25_63
+; CHECK-NEXT:  .LBB25_56: // %else86
+; CHECK-NEXT:    ldr q4, [x1, #240]
+; CHECK-NEXT:    tbnz w8, #30, .LBB25_64
+; CHECK-NEXT:  .LBB25_57: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB25_59
+; CHECK-NEXT:  .LBB25_58: // %cond.load91
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    ld1 { v3.h }[7], [x8]
+; CHECK-NEXT:  .LBB25_59: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB25_60: // %cond.load76
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[2], [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB25_54
+; CHECK-NEXT:  .LBB25_61: // %cond.load79
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[3], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #224]
+; CHECK-NEXT:    tbz w8, #28, .LBB25_55
+; CHECK-NEXT:  .LBB25_62: // %cond.load82
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[4], [x9]
+; CHECK-NEXT:    tbz w8, #29, .LBB25_56
+; CHECK-NEXT:  .LBB25_63: // %cond.load85
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[5], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #240]
+; CHECK-NEXT:    tbz w8, #30, .LBB25_57
+; CHECK-NEXT:  .LBB25_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB25_58
+; CHECK-NEXT:    b .LBB25_59
   %cval = load <32 x half>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = fcmp oeq <32 x half> %cval, zeroinitializer
   %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
   store <32 x half> %vals, ptr %a
   ret void
 }
 
 ;
 ; LD1W (float)
 ;
 
 define void @masked_gather_v2f32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v2f32:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI26_0
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ptrue p0.s, vl2
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI26_0]
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
 ; CHECK-NEXT:    ldr q1, [x1]
-; CHECK-NEXT:    fcmeq v0.2s, v0.2s, #0.0
-; CHECK-NEXT:    sshll v0.2d, v0.2s, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    xtn v0.2s, v0.2d
+; CHECK-NEXT:    // implicit-def: $d0
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbz w9, #0, .LBB26_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:  .LBB26_2: // %else
+; CHECK-NEXT:    tbz w8, #1, .LBB26_4
+; CHECK-NEXT:  // %bb.3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x8]
+; CHECK-NEXT:  .LBB26_4: // %else2
 ; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <2 x float>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = fcmp oeq <2 x float> %cval, zeroinitializer
   %vals = call <2 x float> @llvm.masked.gather.v2f32(<2 x ptr> %ptrs, i32 8, <2 x i1> %mask, <2 x float> undef)
   store <2 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v4f32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v4f32:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI27_0
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq v0.4s, v0.4s, #0.0
-; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    uzp1 z0.s, z0.s, z0.s
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI27_0]
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z1.s, z0.s[2]
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB27_6
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB27_7
+; CHECK-NEXT:  .LBB27_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB27_8
+; CHECK-NEXT:  .LBB27_3: // %else5
+; CHECK-NEXT:    tbz w8, #3, .LBB27_5
+; CHECK-NEXT:  .LBB27_4: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x8]
+; CHECK-NEXT:  .LBB27_5: // %else8
 ; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB27_6: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB27_2
+; CHECK-NEXT:  .LBB27_7: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB27_3
+; CHECK-NEXT:  .LBB27_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB27_4
+; CHECK-NEXT:    b .LBB27_5
   %cval = load <4 x float>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = fcmp oeq <4 x float> %cval, zeroinitializer
   %vals = call <4 x float> @llvm.masked.gather.v4f32(<4 x ptr> %ptrs, i32 8, <4 x i1> %mask, <4 x float> undef)
   store <4 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v8f32(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_gather_v8f32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI28_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI28_0]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z2.s, z1.s
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov z3.s, z0.s[2]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z4.s, z0.s[3]
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    mov z0.s, z1.s[1]
+; CHECK-NEXT:    mov z2.s, z1.s[2]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    mov z3.s, z1.s[3]
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbz w9, #0, .LBB28_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB28_3
+; CHECK-NEXT:    b .LBB28_4
+; CHECK-NEXT:  .LBB28_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB28_4
+; CHECK-NEXT:  .LBB28_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB28_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB28_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB28_9
+; CHECK-NEXT:  .LBB28_6: // %else8
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB28_10
+; CHECK-NEXT:  .LBB28_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB28_11
+; CHECK-NEXT:    b .LBB28_12
+; CHECK-NEXT:  .LBB28_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB28_6
+; CHECK-NEXT:  .LBB28_9: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB28_7
+; CHECK-NEXT:  .LBB28_10:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #5, .LBB28_12
+; CHECK-NEXT:  .LBB28_11: // %cond.load13
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:  .LBB28_12: // %else14
+; CHECK-NEXT:    ldr q2, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB28_16
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    tbz w8, #7, .LBB28_15
+; CHECK-NEXT:  .LBB28_14: // %cond.load19
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ld1 { v1.s }[3], [x8]
+; CHECK-NEXT:  .LBB28_15: // %else20
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB28_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB28_14
+; CHECK-NEXT:    b .LBB28_15
   %cval = load <8 x float>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = fcmp oeq <8 x float> %cval, zeroinitializer
   %vals = call <8 x float> @llvm.masked.gather.v8f32(<8 x ptr> %ptrs, i32 8, <8 x i1> %mask, <8 x float> undef)
   store <8 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v16f32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v16f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl16
-; CHECK-NEXT:    ptrue p1.d, vl16
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI29_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI29_0]
+; CHECK-NEXT:    ldp q4, q3, [x0, #32]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z1.s
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov z5.s, z0.s[2]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z6.s, z0.s[3]
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    mov z5.s, z0.s[2]
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z6.s, z0.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z1.s
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    mov z0.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    fmov w11, s0
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    mov z4.s, z0.s[2]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #7
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    fmov w11, s0
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z3.s, z1.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    mov z3.s, z0.s[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #9
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #10
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    tbz w9, #0, .LBB29_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB29_3
+; CHECK-NEXT:    b .LBB29_4
+; CHECK-NEXT:  .LBB29_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB29_4
+; CHECK-NEXT:  .LBB29_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB29_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB29_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB29_9
+; CHECK-NEXT:  .LBB29_6: // %else8
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB29_10
+; CHECK-NEXT:  .LBB29_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB29_11
+; CHECK-NEXT:    b .LBB29_12
+; CHECK-NEXT:  .LBB29_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB29_6
+; CHECK-NEXT:  .LBB29_9: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB29_7
+; CHECK-NEXT:  .LBB29_10:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #5, .LBB29_12
+; CHECK-NEXT:  .LBB29_11: // %cond.load13
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:  .LBB29_12: // %else14
+; CHECK-NEXT:    ldr q2, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB29_16
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB29_17
+; CHECK-NEXT:  .LBB29_14: // %else20
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB29_18
+; CHECK-NEXT:  .LBB29_15: // %cond.load22
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB29_19
+; CHECK-NEXT:    b .LBB29_20
+; CHECK-NEXT:  .LBB29_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB29_14
+; CHECK-NEXT:  .LBB29_17: // %cond.load19
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[3], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB29_15
+; CHECK-NEXT:  .LBB29_18:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #9, .LBB29_20
+; CHECK-NEXT:  .LBB29_19: // %cond.load25
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB29_20: // %else26
+; CHECK-NEXT:    ldr q3, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB29_24
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB29_25
+; CHECK-NEXT:  .LBB29_22: // %else32
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB29_26
+; CHECK-NEXT:  .LBB29_23: // %cond.load34
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #13, .LBB29_27
+; CHECK-NEXT:    b .LBB29_28
+; CHECK-NEXT:  .LBB29_24: // %cond.load28
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB29_22
+; CHECK-NEXT:  .LBB29_25: // %cond.load31
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB29_23
+; CHECK-NEXT:  .LBB29_26:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #13, .LBB29_28
+; CHECK-NEXT:  .LBB29_27: // %cond.load37
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:  .LBB29_28: // %else38
+; CHECK-NEXT:    ldr q4, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB29_32
+; CHECK-NEXT:  // %bb.29: // %else41
+; CHECK-NEXT:    tbz w8, #15, .LBB29_31
+; CHECK-NEXT:  .LBB29_30: // %cond.load43
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    ld1 { v3.s }[3], [x8]
+; CHECK-NEXT:  .LBB29_31: // %else44
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB29_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB29_30
+; CHECK-NEXT:    b .LBB29_31
   %cval = load <16 x float>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = fcmp oeq <16 x float> %cval, zeroinitializer
   %vals = call <16 x float> @llvm.masked.gather.v16f32(<16 x ptr> %ptrs, i32 8, <16 x i1> %mask, <16 x float> undef)
   store <16 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v32f32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v32f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI30_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q1, q2, [x0, #64]
+; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI30_0]
+; CHECK-NEXT:    ldp q4, q3, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z5.s, z1.s[1]
+; CHECK-NEXT:    mov z6.s, z1.s[2]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    mov z7.s, z1.s[3]
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z5.s, z2.s[2]
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    mov z1.s, z2.s[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z0.s
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    mov z5.s, z4.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    mov z6.s, z4.s[2]
+; CHECK-NEXT:    mov z7.s, z4.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z3.s, z0.s
+; CHECK-NEXT:    ldp q16, q4, [x0]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    mov z5.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z16.s, z0.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    mov z17.s, z16.s[1]
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z17.s, z16.s[2]
+; CHECK-NEXT:    mov z18.s, z16.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z0.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.s, z5.s[1]
+; CHECK-NEXT:    fmov w13, s4
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z16.s, z4.s[2]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z7.s, z5.s[2]
+; CHECK-NEXT:    ldp q2, q1, [x0, #32]
+; CHECK-NEXT:    mov z3.s, z5.s[3]
+; CHECK-NEXT:    mov z5.s, z4.s[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    fmov w13, s16
+; CHECK-NEXT:    fmov w12, s5
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z17.s, z4.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    orr w9, w10, w13, lsl #6
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s4
+; CHECK-NEXT:    mov z5.s, z2.s[2]
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov w12, s0
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z4.s, z0.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB30_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB30_3
+; CHECK-NEXT:    b .LBB30_4
+; CHECK-NEXT:  .LBB30_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB30_4
+; CHECK-NEXT:  .LBB30_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB30_4: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB30_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB30_9
+; CHECK-NEXT:  .LBB30_6: // %else8
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB30_10
+; CHECK-NEXT:  .LBB30_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB30_11
+; CHECK-NEXT:    b .LBB30_12
+; CHECK-NEXT:  .LBB30_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB30_6
+; CHECK-NEXT:  .LBB30_9: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    tbnz w8, #4, .LBB30_7
+; CHECK-NEXT:  .LBB30_10:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #5, .LBB30_12
+; CHECK-NEXT:  .LBB30_11: // %cond.load13
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:  .LBB30_12: // %else14
+; CHECK-NEXT:    ldr q2, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB30_16
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB30_17
+; CHECK-NEXT:  .LBB30_14: // %else20
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB30_18
+; CHECK-NEXT:  .LBB30_15: // %cond.load22
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB30_19
+; CHECK-NEXT:    b .LBB30_20
+; CHECK-NEXT:  .LBB30_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB30_14
+; CHECK-NEXT:  .LBB30_17: // %cond.load19
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[3], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    tbnz w8, #8, .LBB30_15
+; CHECK-NEXT:  .LBB30_18:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #9, .LBB30_20
+; CHECK-NEXT:  .LBB30_19: // %cond.load25
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB30_20: // %else26
+; CHECK-NEXT:    ldr q3, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB30_24
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB30_25
+; CHECK-NEXT:  .LBB30_22: // %else32
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB30_26
+; CHECK-NEXT:  .LBB30_23: // %cond.load34
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #13, .LBB30_27
+; CHECK-NEXT:    b .LBB30_28
+; CHECK-NEXT:  .LBB30_24: // %cond.load28
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB30_22
+; CHECK-NEXT:  .LBB30_25: // %cond.load31
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    tbnz w8, #12, .LBB30_23
+; CHECK-NEXT:  .LBB30_26:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #13, .LBB30_28
+; CHECK-NEXT:  .LBB30_27: // %cond.load37
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:  .LBB30_28: // %else38
+; CHECK-NEXT:    ldr q4, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB30_32
+; CHECK-NEXT:  // %bb.29: // %else41
+; CHECK-NEXT:    tbnz w8, #15, .LBB30_33
+; CHECK-NEXT:  .LBB30_30: // %else44
+; CHECK-NEXT:    ldr q5, [x1, #128]
+; CHECK-NEXT:    tbz w8, #16, .LBB30_34
+; CHECK-NEXT:  .LBB30_31: // %cond.load46
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #17, .LBB30_35
+; CHECK-NEXT:    b .LBB30_36
+; CHECK-NEXT:  .LBB30_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB30_30
+; CHECK-NEXT:  .LBB30_33: // %cond.load43
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[3], [x9]
+; CHECK-NEXT:    ldr q5, [x1, #128]
+; CHECK-NEXT:    tbnz w8, #16, .LBB30_31
+; CHECK-NEXT:  .LBB30_34:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    tbz w8, #17, .LBB30_36
+; CHECK-NEXT:  .LBB30_35: // %cond.load49
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[1], [x9]
+; CHECK-NEXT:  .LBB30_36: // %else50
+; CHECK-NEXT:    ldr q5, [x1, #144]
+; CHECK-NEXT:    tbnz w8, #18, .LBB30_40
+; CHECK-NEXT:  // %bb.37: // %else53
+; CHECK-NEXT:    tbnz w8, #19, .LBB30_41
+; CHECK-NEXT:  .LBB30_38: // %else56
+; CHECK-NEXT:    ldr q6, [x1, #160]
+; CHECK-NEXT:    tbz w8, #20, .LBB30_42
+; CHECK-NEXT:  .LBB30_39: // %cond.load58
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #21, .LBB30_43
+; CHECK-NEXT:    b .LBB30_44
+; CHECK-NEXT:  .LBB30_40: // %cond.load52
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB30_38
+; CHECK-NEXT:  .LBB30_41: // %cond.load55
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[3], [x9]
+; CHECK-NEXT:    ldr q6, [x1, #160]
+; CHECK-NEXT:    tbnz w8, #20, .LBB30_39
+; CHECK-NEXT:  .LBB30_42:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:    tbz w8, #21, .LBB30_44
+; CHECK-NEXT:  .LBB30_43: // %cond.load61
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[1], [x9]
+; CHECK-NEXT:  .LBB30_44: // %else62
+; CHECK-NEXT:    ldr q6, [x1, #176]
+; CHECK-NEXT:    tbnz w8, #22, .LBB30_48
+; CHECK-NEXT:  // %bb.45: // %else65
+; CHECK-NEXT:    tbnz w8, #23, .LBB30_49
+; CHECK-NEXT:  .LBB30_46: // %else68
+; CHECK-NEXT:    ldr q7, [x1, #192]
+; CHECK-NEXT:    tbz w8, #24, .LBB30_50
+; CHECK-NEXT:  .LBB30_47: // %cond.load70
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #25, .LBB30_51
+; CHECK-NEXT:    b .LBB30_52
+; CHECK-NEXT:  .LBB30_48: // %cond.load64
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB30_46
+; CHECK-NEXT:  .LBB30_49: // %cond.load67
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[3], [x9]
+; CHECK-NEXT:    ldr q7, [x1, #192]
+; CHECK-NEXT:    tbnz w8, #24, .LBB30_47
+; CHECK-NEXT:  .LBB30_50:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    tbz w8, #25, .LBB30_52
+; CHECK-NEXT:  .LBB30_51: // %cond.load73
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[1], [x9]
+; CHECK-NEXT:  .LBB30_52: // %else74
+; CHECK-NEXT:    ldr q7, [x1, #208]
+; CHECK-NEXT:    tbnz w8, #26, .LBB30_56
+; CHECK-NEXT:  // %bb.53: // %else77
+; CHECK-NEXT:    tbnz w8, #27, .LBB30_57
+; CHECK-NEXT:  .LBB30_54: // %else80
+; CHECK-NEXT:    ldr q16, [x1, #224]
+; CHECK-NEXT:    tbz w8, #28, .LBB30_58
+; CHECK-NEXT:  .LBB30_55: // %cond.load82
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #29, .LBB30_59
+; CHECK-NEXT:    b .LBB30_60
+; CHECK-NEXT:  .LBB30_56: // %cond.load76
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB30_54
+; CHECK-NEXT:  .LBB30_57: // %cond.load79
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[3], [x9]
+; CHECK-NEXT:    ldr q16, [x1, #224]
+; CHECK-NEXT:    tbnz w8, #28, .LBB30_55
+; CHECK-NEXT:  .LBB30_58:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    tbz w8, #29, .LBB30_60
+; CHECK-NEXT:  .LBB30_59: // %cond.load85
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:  .LBB30_60: // %else86
+; CHECK-NEXT:    ldr q16, [x1, #240]
+; CHECK-NEXT:    tbnz w8, #30, .LBB30_64
+; CHECK-NEXT:  // %bb.61: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB30_63
+; CHECK-NEXT:  .LBB30_62: // %cond.load91
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x8, d16
+; CHECK-NEXT:    ld1 { v7.s }[3], [x8]
+; CHECK-NEXT:  .LBB30_63: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    stp q4, q5, [x0, #64]
+; CHECK-NEXT:    stp q6, q7, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB30_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB30_62
+; CHECK-NEXT:    b .LBB30_63
   %cval = load <32 x float>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = fcmp oeq <32 x float> %cval, zeroinitializer
   %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
   store <32 x float> %vals, ptr %a
   ret void
 }
 
; CHECK-NEXT:    ret
   %vals = call <1 x double> @llvm.masked.gather.v1f64(<1 x ptr> %ptrs, i32 8, <1 x i1> %mask, <1 x double> undef)
   store <1 x double> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v2f64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v2f64:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI32_0
 ; CHECK-NEXT:    ldr q0, [x0]
 ; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI32_0]
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    mov z0.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x8, d1
 ; CHECK-NEXT:    ldr q1, [x1]
-; CHECK-NEXT:    fcmeq v0.2d, v0.2d, #0.0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [z1.d]
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbz w9, #0, .LBB32_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:  .LBB32_2: // %else
+; CHECK-NEXT:    tbz w8, #1, .LBB32_4
+; CHECK-NEXT:  // %bb.3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x8]
+; CHECK-NEXT:  .LBB32_4: // %else2
 ; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <2 x double>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = fcmp oeq <2 x double> %cval, zeroinitializer
   %vals = call <2 x double> @llvm.masked.gather.v2f64(<2 x ptr> %ptrs, i32 8, <2 x i1> %mask, <2 x double> undef)
   store <2 x double> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v4f64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v4f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    ld1d { z0.d }, p1/z, [z1.d]
-; CHECK-NEXT:    st1d { z0.d }, p0, [x0]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI33_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI33_0]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z2.d, z1.d
+; CHECK-NEXT:    mov z0.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z3.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x8, d3
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    mov z0.d, z1.d[1]
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbz w9, #0, .LBB33_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB33_3
+; CHECK-NEXT:    b .LBB33_4
+; CHECK-NEXT:  .LBB33_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB33_4
+; CHECK-NEXT:  .LBB33_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB33_4: // %else2
+; CHECK-NEXT:    ldr q2, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB33_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB33_7
+; CHECK-NEXT:    b .LBB33_8
+; CHECK-NEXT:  .LBB33_6:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #3, .LBB33_8
+; CHECK-NEXT:  .LBB33_7: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ld1 { v1.d }[1], [x8]
+; CHECK-NEXT:  .LBB33_8: // %else8
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <4 x double>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = fcmp oeq <4 x double> %cval, zeroinitializer
   %vals = call <4 x double> @llvm.masked.gather.v4f64(<4 x ptr> %ptrs, i32 8, <4 x i1> %mask, <4 x double> undef)
   store <4 x double> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v8f64(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_gather_v8f64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI34_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q0, q2, [x0]
+; CHECK-NEXT:    ldp q3, q4, [x0, #32]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI34_0]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    mov z0.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z2.d, z1.d
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z3.d, z1.d
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z4.d, z1.d
+; CHECK-NEXT:    mov z4.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z0.d, z2.d[1]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov x11, d0
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z3.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    mov z0.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov x10, d0
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov x11, d3
+; CHECK-NEXT:    mov z0.d, z3.d[1]
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov x10, d0
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbz w9, #0, .LBB34_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB34_3
+; CHECK-NEXT:    b .LBB34_4
+; CHECK-NEXT:  .LBB34_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB34_4
+; CHECK-NEXT:  .LBB34_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB34_4: // %else2
+; CHECK-NEXT:    ldr q2, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB34_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB34_7
+; CHECK-NEXT:    b .LBB34_8
+; CHECK-NEXT:  .LBB34_6:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #3, .LBB34_8
+; CHECK-NEXT:  .LBB34_7: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[1], [x9]
+; CHECK-NEXT:  .LBB34_8: // %else8
+; CHECK-NEXT:    ldr q3, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB34_10
+; CHECK-NEXT:  // %bb.9: // %cond.load10
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB34_11
+; CHECK-NEXT:    b .LBB34_12
+; CHECK-NEXT:  .LBB34_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #5, .LBB34_12
+; CHECK-NEXT:  .LBB34_11: // %cond.load13
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[1], [x9]
+; CHECK-NEXT:  .LBB34_12: // %else14
+; CHECK-NEXT:    ldr q4, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB34_14
+; CHECK-NEXT:  // %bb.13: // %cond.load16
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB34_15
+; CHECK-NEXT:    b .LBB34_16
+; CHECK-NEXT:  .LBB34_14:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #7, .LBB34_16
+; CHECK-NEXT:  .LBB34_15: // %cond.load19
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    ld1 { v3.d }[1], [x8]
+; CHECK-NEXT:  .LBB34_16: // %else20
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %cval = load <8 x double>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = fcmp oeq <8 x double> %cval, zeroinitializer
   %vals = call <8 x double> @llvm.masked.gather.v8f64(<8 x ptr> %ptrs, i32 8, <8 x i1> %mask, <8 x double> undef)
   store <8 x double> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v16f64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v16f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    ld1d { z0.d }, p1/z, [z1.d]
-; CHECK-NEXT:    st1d { z0.d }, p0, [x0]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI35_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q1, q2, [x0]
+; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI35_0]
+; CHECK-NEXT:    ldp q3, q4, [x0, #32]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z0.d
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z2.d, z0.d
+; CHECK-NEXT:    mov z6.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z3.d, z0.d
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov x11, d1
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    mov z2.d, z5.d[1]
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z4.d, z0.d
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldr q5, [x0, #64]
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov x11, d4
+; CHECK-NEXT:    bfi w8, w9, #5, #1
+; CHECK-NEXT:    ldp q1, q3, [x0, #80]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z5.d, z0.d
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #6
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z0.d
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z3.d, z0.d
+; CHECK-NEXT:    fmov x11, d1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    ldr q2, [x0, #112]
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #9
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z2.d, z0.d
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    mov z3.d, z1.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z0.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x11, d0
+; CHECK-NEXT:    mov z1.d, z0.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    tbz w9, #0, .LBB35_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB35_3
+; CHECK-NEXT:    b .LBB35_4
+; CHECK-NEXT:  .LBB35_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB35_4
+; CHECK-NEXT:  .LBB35_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB35_4: // %else2
+; CHECK-NEXT:    ldr q2, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB35_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB35_7
+; CHECK-NEXT:    b .LBB35_8
+; CHECK-NEXT:  .LBB35_6:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #3, .LBB35_8
+; CHECK-NEXT:  .LBB35_7: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[1], [x9]
+; CHECK-NEXT:  .LBB35_8: // %else8
+; CHECK-NEXT:    ldr q3, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB35_10
+; CHECK-NEXT:  // %bb.9: // %cond.load10
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB35_11
+; CHECK-NEXT:    b .LBB35_12
+; CHECK-NEXT:  .LBB35_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #5, .LBB35_12
+; CHECK-NEXT:  .LBB35_11: // %cond.load13
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[1], [x9]
+; CHECK-NEXT:  .LBB35_12: // %else14
+; CHECK-NEXT:    ldr q4, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB35_14
+; CHECK-NEXT:  // %bb.13: // %cond.load16
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB35_15
+; CHECK-NEXT:    b .LBB35_16
+; CHECK-NEXT:  .LBB35_14:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #7, .LBB35_16
+; CHECK-NEXT:  .LBB35_15: // %cond.load19
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[1], [x9]
+; CHECK-NEXT:  .LBB35_16: // %else20
+; CHECK-NEXT:    ldr q5, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB35_18
+; CHECK-NEXT:  // %bb.17: // %cond.load22
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB35_19
+; CHECK-NEXT:    b .LBB35_20
+; CHECK-NEXT:  .LBB35_18:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    tbz w8, #9, .LBB35_20
+; CHECK-NEXT:  .LBB35_19: // %cond.load25
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.d }[1], [x9]
+; CHECK-NEXT:  .LBB35_20: // %else26
+; CHECK-NEXT:    ldr q6, [x1, #80]
+; CHECK-NEXT:    tbz w8, #10, .LBB35_22
+; CHECK-NEXT:  // %bb.21: // %cond.load28
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #11, .LBB35_23
+; CHECK-NEXT:    b .LBB35_24
+; CHECK-NEXT:  .LBB35_22:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:    tbz w8, #11, .LBB35_24
+; CHECK-NEXT:  .LBB35_23: // %cond.load31
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.d }[1], [x9]
+; CHECK-NEXT:  .LBB35_24: // %else32
+; CHECK-NEXT:    ldr q7, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB35_26
+; CHECK-NEXT:  // %bb.25: // %cond.load34
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #13, .LBB35_27
+; CHECK-NEXT:    b .LBB35_28
+; CHECK-NEXT:  .LBB35_26:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    tbz w8, #13, .LBB35_28
+; CHECK-NEXT:  .LBB35_27: // %cond.load37
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.d }[1], [x9]
+; CHECK-NEXT:  .LBB35_28: // %else38
+; CHECK-NEXT:    ldr q16, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB35_30
+; CHECK-NEXT:  // %bb.29: // %cond.load40
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB35_31
+; CHECK-NEXT:    b .LBB35_32
+; CHECK-NEXT:  .LBB35_30:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    tbz w8, #15, .LBB35_32
+; CHECK-NEXT:  .LBB35_31: // %cond.load43
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x8, d16
+; CHECK-NEXT:    ld1 { v7.d }[1], [x8]
+; CHECK-NEXT:  .LBB35_32: // %else44
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    stp q4, q5, [x0, #64]
+; CHECK-NEXT:    stp q6, q7, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <16 x double>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = fcmp oeq <16 x double> %cval, zeroinitializer
   %vals = call <16 x double> @llvm.masked.gather.v16f64(<16 x ptr> %ptrs, i32 8, <16 x i1> %mask, <16 x double> undef)
   store <16 x double> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_v32f64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_v32f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    ld1d { z0.d }, p1/z, [z1.d]
-; CHECK-NEXT:    st1d { z0.d }, p0, [x0]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI36_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q5, q7, [x0, #128]
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI36_0]
+; CHECK-NEXT:    ldp q4, q3, [x0, #64]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z5.d, z2.d
+; CHECK-NEXT:    mov z16.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z7.d, z2.d
+; CHECK-NEXT:    fmov x8, d16
+; CHECK-NEXT:    mov z17.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z18.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q19, q7, [x0, #160]
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z20.d, z18.d[1]
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    fmov x11, d20
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z19.d, z2.d
+; CHECK-NEXT:    mov z18.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z7.d, z2.d
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    mov z7.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z19.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    fmov x11, d7
+; CHECK-NEXT:    ldp q17, q16, [x0, #192]
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    mov z18.d, z7.d[1]
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z17.d, z2.d
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z17.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z16.d, z2.d
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    mov z18.d, z17.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z16.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q19, q7, [x0, #224]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    fmov x11, d16
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    mov z17.d, z16.d[1]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z19.d, z2.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    ldp q18, q16, [x0]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z18.d, z2.d
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z19.d, z17.d[1]
+; CHECK-NEXT:    ldp q6, q5, [x0, #32]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z17.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    mov z18.d, z17.d[1]
+; CHECK-NEXT:    fmov x11, d17
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z16.d, z2.d
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z16.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z17.d, z16.d[1]
+; CHECK-NEXT:    fmov x12, d16
+; CHECK-NEXT:    bfi w11, w10, #1, #1
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z6.d, z2.d
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z6.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z5.d, z2.d
+; CHECK-NEXT:    bfi w11, w12, #2, #1
+; CHECK-NEXT:    mov z16.d, z6.d[1]
+; CHECK-NEXT:    fmov x12, d6
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    bfi w11, w10, #3, #1
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    bfi w11, w12, #4, #1
+; CHECK-NEXT:    mov z6.d, z5.d[1]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z4.d, z2.d
+; CHECK-NEXT:    bfi w11, w9, #5, #1
+; CHECK-NEXT:    mov z4.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    fmov x12, d4
+; CHECK-NEXT:    mov z5.d, z4.d[1]
+; CHECK-NEXT:    orr w9, w11, w9, lsl #6
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z7.d, z2.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z4.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z3.d, z2.d
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov x11, d3
+; CHECK-NEXT:    mov z5.d, z3.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z2.d
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, z2.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov x11, d5
+; CHECK-NEXT:    fmov x12, d1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    mov z3.d, z1.d[1]
+; CHECK-NEXT:    mov z0.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.d, z4.d[1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov x12, d0
+; CHECK-NEXT:    mov z2.d, z0.d[1]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov x11, d4
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    orr w9, w9, w12, lsl #14
+; CHECK-NEXT:    fmov x12, d1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w12, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB36_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    tbnz w8, #1, .LBB36_3
+; CHECK-NEXT:    b .LBB36_4
+; CHECK-NEXT:  .LBB36_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    tbz w8, #1, .LBB36_4
+; CHECK-NEXT:  .LBB36_3: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_4: // %else2
+; CHECK-NEXT:    ldr q2, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB36_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB36_7
+; CHECK-NEXT:    b .LBB36_8
+; CHECK-NEXT:  .LBB36_6:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    tbz w8, #3, .LBB36_8
+; CHECK-NEXT:  .LBB36_7: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_8: // %else8
+; CHECK-NEXT:    ldr q3, [x1, #32]
+; CHECK-NEXT:    tbz w8, #4, .LBB36_10
+; CHECK-NEXT:  // %bb.9: // %cond.load10
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #5, .LBB36_11
+; CHECK-NEXT:    b .LBB36_12
+; CHECK-NEXT:  .LBB36_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    tbz w8, #5, .LBB36_12
+; CHECK-NEXT:  .LBB36_11: // %cond.load13
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_12: // %else14
+; CHECK-NEXT:    ldr q4, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB36_14
+; CHECK-NEXT:  // %bb.13: // %cond.load16
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB36_15
+; CHECK-NEXT:    b .LBB36_16
+; CHECK-NEXT:  .LBB36_14:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    tbz w8, #7, .LBB36_16
+; CHECK-NEXT:  .LBB36_15: // %cond.load19
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_16: // %else20
+; CHECK-NEXT:    ldr q5, [x1, #64]
+; CHECK-NEXT:    tbz w8, #8, .LBB36_18
+; CHECK-NEXT:  // %bb.17: // %cond.load22
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #9, .LBB36_19
+; CHECK-NEXT:    b .LBB36_20
+; CHECK-NEXT:  .LBB36_18:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    tbz w8, #9, .LBB36_20
+; CHECK-NEXT:  .LBB36_19: // %cond.load25
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_20: // %else26
+; CHECK-NEXT:    ldr q6, [x1, #80]
+; CHECK-NEXT:    tbz w8, #10, .LBB36_22
+; CHECK-NEXT:  // %bb.21: // %cond.load28
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #11, .LBB36_23
+; CHECK-NEXT:    b .LBB36_24
+; CHECK-NEXT:  .LBB36_22:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:    tbz w8, #11, .LBB36_24
+; CHECK-NEXT:  .LBB36_23: // %cond.load31
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_24: // %else32
+; CHECK-NEXT:    ldr q7, [x1, #96]
+; CHECK-NEXT:    tbz w8, #12, .LBB36_26
+; CHECK-NEXT:  // %bb.25: // %cond.load34
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #13, .LBB36_27
+; CHECK-NEXT:    b .LBB36_28
+; CHECK-NEXT:  .LBB36_26:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    tbz w8, #13, .LBB36_28
+; CHECK-NEXT:  .LBB36_27: // %cond.load37
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_28: // %else38
+; CHECK-NEXT:    ldr q16, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB36_30
+; CHECK-NEXT:  // %bb.29: // %cond.load40
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB36_31
+; CHECK-NEXT:    b .LBB36_32
+; CHECK-NEXT:  .LBB36_30:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    tbz w8, #15, .LBB36_32
+; CHECK-NEXT:  .LBB36_31: // %cond.load43
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_32: // %else44
+; CHECK-NEXT:    ldr q17, [x1, #128]
+; CHECK-NEXT:    tbz w8, #16, .LBB36_34
+; CHECK-NEXT:  // %bb.33: // %cond.load46
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v16.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #17, .LBB36_35
+; CHECK-NEXT:    b .LBB36_36
+; CHECK-NEXT:  .LBB36_34:
+; CHECK-NEXT:    // implicit-def: $q16
+; CHECK-NEXT:    tbz w8, #17, .LBB36_36
+; CHECK-NEXT:  .LBB36_35: // %cond.load49
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v16.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_36: // %else50
+; CHECK-NEXT:    ldr q18, [x1, #144]
+; CHECK-NEXT:    tbz w8, #18, .LBB36_38
+; CHECK-NEXT:  // %bb.37: // %cond.load52
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v17.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #19, .LBB36_39
+; CHECK-NEXT:    b .LBB36_40
+; CHECK-NEXT:  .LBB36_38:
+; CHECK-NEXT:    // implicit-def: $q17
+; CHECK-NEXT:    tbz w8, #19, .LBB36_40
+; CHECK-NEXT:  .LBB36_39: // %cond.load55
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v17.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_40: // %else56
+; CHECK-NEXT:    ldr q19, [x1, #160]
+; CHECK-NEXT:    tbz w8, #20, .LBB36_42
+; CHECK-NEXT:  // %bb.41: // %cond.load58
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v18.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #21, .LBB36_43
+; CHECK-NEXT:    b .LBB36_44
+; CHECK-NEXT:  .LBB36_42:
+; CHECK-NEXT:    // implicit-def: $q18
+; CHECK-NEXT:    tbz w8, #21, .LBB36_44
+; CHECK-NEXT:  .LBB36_43: // %cond.load61
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v18.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_44: // %else62
+; CHECK-NEXT:    ldr q20, [x1, #176]
+; CHECK-NEXT:    tbz w8, #22, .LBB36_46
+; CHECK-NEXT:  // %bb.45: // %cond.load64
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #23, .LBB36_47
+; CHECK-NEXT:    b .LBB36_48
+; CHECK-NEXT:  .LBB36_46:
+; CHECK-NEXT:    // implicit-def: $q19
+; CHECK-NEXT:    tbz w8, #23, .LBB36_48
+; CHECK-NEXT:  .LBB36_47: // %cond.load67
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_48: // %else68
+; CHECK-NEXT:    ldr q21, [x1, #192]
+; CHECK-NEXT:    tbz w8, #24, .LBB36_50
+; CHECK-NEXT:  // %bb.49: // %cond.load70
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v20.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #25, .LBB36_51
+; CHECK-NEXT:    b .LBB36_52
+; CHECK-NEXT:  .LBB36_50:
+; CHECK-NEXT:    // implicit-def: $q20
+; CHECK-NEXT:    tbz w8, #25, .LBB36_52
+; CHECK-NEXT:  .LBB36_51: // %cond.load73
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v20.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_52: // %else74
+; CHECK-NEXT:    ldr q22, [x1, #208]
+; CHECK-NEXT:    tbz w8, #26, .LBB36_54
+; CHECK-NEXT:  // %bb.53: // %cond.load76
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v21.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #27, .LBB36_55
+; CHECK-NEXT:    b .LBB36_56
+; CHECK-NEXT:  .LBB36_54:
+; CHECK-NEXT:    // implicit-def: $q21
+; CHECK-NEXT:    tbz w8, #27, .LBB36_56
+; CHECK-NEXT:  .LBB36_55: // %cond.load79
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v21.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_56: // %else80
+; CHECK-NEXT:    ldr q23, [x1, #224]
+; CHECK-NEXT:    tbz w8, #28, .LBB36_58
+; CHECK-NEXT:  // %bb.57: // %cond.load82
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v22.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #29, .LBB36_59
+; CHECK-NEXT:    b .LBB36_60
+; CHECK-NEXT:  .LBB36_58:
+; CHECK-NEXT:    // implicit-def: $q22
+; CHECK-NEXT:    tbz w8, #29, .LBB36_60
+; CHECK-NEXT:  .LBB36_59: // %cond.load85
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v22.d }[1], [x9]
+; CHECK-NEXT:  .LBB36_60: // %else86
+; CHECK-NEXT:    ldr q24, [x1, #240]
+; CHECK-NEXT:    tbz w8, #30, .LBB36_62
+; CHECK-NEXT:  // %bb.61: // %cond.load88
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v23.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB36_63
+; CHECK-NEXT:    b .LBB36_64
+; CHECK-NEXT:  .LBB36_62:
+; CHECK-NEXT:    // implicit-def: $q23
+; CHECK-NEXT:    tbz w8, #31, .LBB36_64
+; CHECK-NEXT:  .LBB36_63: // %cond.load91
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x8, d24
+; CHECK-NEXT:    ld1 { v23.d }[1], [x8]
+; CHECK-NEXT:  .LBB36_64: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    stp q4, q5, [x0, #64]
+; CHECK-NEXT:    stp q6, q7, [x0, #96]
+; CHECK-NEXT:    stp q16, q17, [x0, #128]
+; CHECK-NEXT:    stp q18, q19, [x0, #160]
+; CHECK-NEXT:    stp q20, q21, [x0, #192]
+; CHECK-NEXT:    stp q22, q23, [x0, #224]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %cval = load <32 x double>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = fcmp oeq <32 x double> %cval, zeroinitializer
   %vals = call <32 x double> @llvm.masked.gather.v32f64(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x double> undef)
   store <32 x double> %vals, ptr %a
   ret void
 }
 
 ; The above tests test the types, the below tests check that the addressing
 ; modes still function
 
 define void @masked_gather_32b_scaled_sext_f16(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_gather_32b_scaled_sext_f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.s, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1h { z0.s }, p0/z, [x2, z1.s, sxtw #1]
-; CHECK-NEXT:    st1h { z0.s }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    adrp x8, .LCPI37_1
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q1, q19, [x0, #32]
+; CHECK-NEXT:    adrp x12, .LCPI37_0
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI37_1]
+; CHECK-NEXT:    ldp q20, q18, [x0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z1.h, z3.h
+; CHECK-NEXT:    mov z1.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    mov z2.h, z1.h[1]
+; CHECK-NEXT:    mov z21.h, z1.h[2]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z22.h, z1.h[3]
+; CHECK-NEXT:    fmov w10, s21
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z23.h, z1.h[4]
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s23
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    mov z25.h, z1.h[6]
+; CHECK-NEXT:    bfi w9, w11, #19, #1
+; CHECK-NEXT:    mov z24.h, z1.h[5]
+; CHECK-NEXT:    ldp q4, q0, [x1, #96]
+; CHECK-NEXT:    bfi w9, w8, #20, #1
+; CHECK-NEXT:    fmov w8, s25
+; CHECK-NEXT:    mov z26.h, z1.h[7]
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    ldp q7, q6, [x1, #64]
+; CHECK-NEXT:    ldp q5, q16, [x1, #32]
+; CHECK-NEXT:    ldp q2, q17, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    bfi w9, w10, #21, #1
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z19.h, z3.h
+; CHECK-NEXT:    orr w8, w9, w8, lsl #22
+; CHECK-NEXT:    fmov w9, s26
+; CHECK-NEXT:    mov z21.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z20.h, z3.h
+; CHECK-NEXT:    fmov w10, s21
+; CHECK-NEXT:    mov z22.h, z21.h[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    mov z23.h, z21.h[2]
+; CHECK-NEXT:    mov z24.h, z21.h[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z25.h, z21.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    mov z26.h, z21.h[5]
+; CHECK-NEXT:    mov z20.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldr q1, [x12, :lo12:.LCPI37_0]
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    mov z27.h, z21.h[6]
+; CHECK-NEXT:    mov z19.h, z21.h[7]
+; CHECK-NEXT:    fmov w11, s25
+; CHECK-NEXT:    mov z21.h, z20.h[1]
+; CHECK-NEXT:    fmov w12, s26
+; CHECK-NEXT:    fmov w14, s20
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w13, s21
+; CHECK-NEXT:    mov z21.h, z20.h[2]
+; CHECK-NEXT:    mov z22.h, z20.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    mov z23.h, z20.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    and w12, w14, #0x1
+; CHECK-NEXT:    fmov w14, s22
+; CHECK-NEXT:    mov z24.h, z20.h[5]
+; CHECK-NEXT:    bfi w12, w13, #1, #1
+; CHECK-NEXT:    fmov w13, s23
+; CHECK-NEXT:    bfi w12, w9, #2, #1
+; CHECK-NEXT:    fmov w9, s24
+; CHECK-NEXT:    bfi w12, w14, #3, #1
+; CHECK-NEXT:    mov z25.h, z20.h[6]
+; CHECK-NEXT:    mov z26.h, z20.h[7]
+; CHECK-NEXT:    bfi w12, w13, #4, #1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    bfi w12, w9, #5, #1
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z18.h, z3.h
+; CHECK-NEXT:    orr w8, w8, w11, lsl #29
+; CHECK-NEXT:    mov z3.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    fmov w13, s3
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z18.h, z3.h[1]
+; CHECK-NEXT:    mov z20.h, z3.h[2]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w12, w9, lsl #6
+; CHECK-NEXT:    and w12, w13, #0x1
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    orr w9, w9, w11, lsl #7
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    mov z21.h, z3.h[3]
+; CHECK-NEXT:    orr w9, w9, w12, lsl #8
+; CHECK-NEXT:    fmov w12, s20
+; CHECK-NEXT:    fmov w13, s21
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z22.h, z3.h[4]
+; CHECK-NEXT:    mov z23.h, z3.h[5]
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #9
+; CHECK-NEXT:    and w11, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #30
+; CHECK-NEXT:    orr w9, w9, w12, lsl #10
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    mov z24.h, z3.h[6]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    fmov w12, s24
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z25.h, z3.h[7]
+; CHECK-NEXT:    ldr q3, [sp]
+; CHECK-NEXT:    sunpklo z18.d, z2.s
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s25
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z1.d
+; CHECK-NEXT:    ext z2.b, z2.b, z2.b, #8
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    orr w8, w8, w10, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    sunpklo z2.d, z2.s
+; CHECK-NEXT:    movprfx z19, z2
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z1.d
+; CHECK-NEXT:    tbz w8, #0, .LBB37_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ldr h2, [x9]
+; CHECK-NEXT:    b .LBB37_3
+; CHECK-NEXT:  .LBB37_2:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:  .LBB37_3: // %else
+; CHECK-NEXT:    sunpklo z20.d, z17.s
+; CHECK-NEXT:    add z19.d, z3.d, z19.d
+; CHECK-NEXT:    tbz w8, #1, .LBB37_5
+; CHECK-NEXT:  // %bb.4: // %cond.load1
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v2.h }[1], [x9]
+; CHECK-NEXT:  .LBB37_5: // %else2
+; CHECK-NEXT:    ext z17.b, z17.b, z17.b, #8
+; CHECK-NEXT:    movprfx z18, z20
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z1.d
+; CHECK-NEXT:    tbz w8, #2, .LBB37_7
+; CHECK-NEXT:  // %bb.6: // %cond.load4
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v2.h }[2], [x9]
+; CHECK-NEXT:  .LBB37_7: // %else5
+; CHECK-NEXT:    sunpklo z17.d, z17.s
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #3, .LBB37_9
+; CHECK-NEXT:  // %bb.8: // %cond.load7
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v2.h }[3], [x9]
+; CHECK-NEXT:  .LBB37_9: // %else8
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #4, .LBB37_11
+; CHECK-NEXT:  // %bb.10: // %cond.load10
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v2.h }[4], [x9]
+; CHECK-NEXT:  .LBB37_11: // %else11
+; CHECK-NEXT:    sunpklo z19.d, z5.s
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #5, .LBB37_13
+; CHECK-NEXT:  // %bb.12: // %cond.load13
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v2.h }[5], [x9]
+; CHECK-NEXT:  .LBB37_13: // %else14
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    movprfx z18, z19
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z1.d
+; CHECK-NEXT:    tbz w8, #6, .LBB37_15
+; CHECK-NEXT:  // %bb.14: // %cond.load16
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v2.h }[6], [x9]
+; CHECK-NEXT:  .LBB37_15: // %else17
+; CHECK-NEXT:    sunpklo z5.d, z5.s
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #7, .LBB37_17
+; CHECK-NEXT:  // %bb.16: // %cond.load19
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v2.h }[7], [x9]
+; CHECK-NEXT:  .LBB37_17: // %else20
+; CHECK-NEXT:    movprfx z17, z5
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #8, .LBB37_19
+; CHECK-NEXT:  // %bb.18: // %cond.load22
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.h }[0], [x9]
+; CHECK-NEXT:    b .LBB37_20
+; CHECK-NEXT:  .LBB37_19:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:  .LBB37_20: // %else23
+; CHECK-NEXT:    sunpklo z19.d, z16.s
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #9, .LBB37_22
+; CHECK-NEXT:  // %bb.21: // %cond.load25
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.h }[1], [x9]
+; CHECK-NEXT:  .LBB37_22: // %else26
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    movprfx z18, z19
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z1.d
+; CHECK-NEXT:    tbz w8, #10, .LBB37_24
+; CHECK-NEXT:  // %bb.23: // %cond.load28
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v5.h }[2], [x9]
+; CHECK-NEXT:  .LBB37_24: // %else29
+; CHECK-NEXT:    sunpklo z16.d, z16.s
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #11, .LBB37_26
+; CHECK-NEXT:  // %bb.25: // %cond.load31
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v5.h }[3], [x9]
+; CHECK-NEXT:  .LBB37_26: // %else32
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z1.d
+; CHECK-NEXT:    tbz w8, #12, .LBB37_28
+; CHECK-NEXT:  // %bb.27: // %cond.load34
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.h }[4], [x9]
+; CHECK-NEXT:  .LBB37_28: // %else35
+; CHECK-NEXT:    sunpklo z17.d, z7.s
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #13, .LBB37_30
+; CHECK-NEXT:  // %bb.29: // %cond.load37
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.h }[5], [x9]
+; CHECK-NEXT:  .LBB37_30: // %else38
+; CHECK-NEXT:    ext z7.b, z7.b, z7.b, #8
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #14, .LBB37_32
+; CHECK-NEXT:  // %bb.31: // %cond.load40
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v5.h }[6], [x9]
+; CHECK-NEXT:  .LBB37_32: // %else41
+; CHECK-NEXT:    sunpklo z7.d, z7.s
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #15, .LBB37_34
+; CHECK-NEXT:  // %bb.33: // %cond.load43
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v5.h }[7], [x9]
+; CHECK-NEXT:  .LBB37_34: // %else44
+; CHECK-NEXT:    movprfx z16, z7
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z1.d
+; CHECK-NEXT:    tbz w8, #16, .LBB37_36
+; CHECK-NEXT:  // %bb.35: // %cond.load46
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.h }[0], [x9]
+; CHECK-NEXT:    b .LBB37_37
+; CHECK-NEXT:  .LBB37_36:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:  .LBB37_37: // %else47
+; CHECK-NEXT:    sunpklo z18.d, z6.s
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #17, .LBB37_39
+; CHECK-NEXT:  // %bb.38: // %cond.load49
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.h }[1], [x9]
+; CHECK-NEXT:  .LBB37_39: // %else50
+; CHECK-NEXT:    ext z6.b, z6.b, z6.b, #8
+; CHECK-NEXT:    movprfx z17, z18
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #18, .LBB37_41
+; CHECK-NEXT:  // %bb.40: // %cond.load52
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.h }[2], [x9]
+; CHECK-NEXT:  .LBB37_41: // %else53
+; CHECK-NEXT:    sunpklo z6.d, z6.s
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #19, .LBB37_43
+; CHECK-NEXT:  // %bb.42: // %cond.load55
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.h }[3], [x9]
+; CHECK-NEXT:  .LBB37_43: // %else56
+; CHECK-NEXT:    lsl z6.d, p0/m, z6.d, z1.d
+; CHECK-NEXT:    tbz w8, #20, .LBB37_45
+; CHECK-NEXT:  // %bb.44: // %cond.load58
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.h }[4], [x9]
+; CHECK-NEXT:  .LBB37_45: // %else59
+; CHECK-NEXT:    sunpklo z16.d, z4.s
+; CHECK-NEXT:    add z6.d, z3.d, z6.d
+; CHECK-NEXT:    tbz w8, #21, .LBB37_47
+; CHECK-NEXT:  // %bb.46: // %cond.load61
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.h }[5], [x9]
+; CHECK-NEXT:  .LBB37_47: // %else62
+; CHECK-NEXT:    ext z4.b, z4.b, z4.b, #8
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z1.d
+; CHECK-NEXT:    tbz w8, #22, .LBB37_49
+; CHECK-NEXT:  // %bb.48: // %cond.load64
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v7.h }[6], [x9]
+; CHECK-NEXT:  .LBB37_49: // %else65
+; CHECK-NEXT:    sunpklo z4.d, z4.s
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #23, .LBB37_51
+; CHECK-NEXT:  // %bb.50: // %cond.load67
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v7.h }[7], [x9]
+; CHECK-NEXT:  .LBB37_51: // %else68
+; CHECK-NEXT:    movprfx z6, z4
+; CHECK-NEXT:    lsl z6.d, p0/m, z6.d, z1.d
+; CHECK-NEXT:    tbz w8, #24, .LBB37_53
+; CHECK-NEXT:  // %bb.52: // %cond.load70
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[0], [x9]
+; CHECK-NEXT:    b .LBB37_54
+; CHECK-NEXT:  .LBB37_53:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:  .LBB37_54: // %else71
+; CHECK-NEXT:    sunpklo z17.d, z0.s
+; CHECK-NEXT:    add z6.d, z3.d, z6.d
+; CHECK-NEXT:    tbz w8, #25, .LBB37_56
+; CHECK-NEXT:  // %bb.55: // %cond.load73
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[1], [x9]
+; CHECK-NEXT:  .LBB37_56: // %else74
+; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #8
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #26, .LBB37_58
+; CHECK-NEXT:  // %bb.57: // %cond.load76
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v4.h }[2], [x9]
+; CHECK-NEXT:  .LBB37_58: // %else77
+; CHECK-NEXT:    sunpklo z16.d, z0.s
+; CHECK-NEXT:    add z0.d, z3.d, z17.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB37_65
+; CHECK-NEXT:  // %bb.59: // %else80
+; CHECK-NEXT:    lslr z1.d, p0/m, z1.d, z16.d
+; CHECK-NEXT:    tbnz w8, #28, .LBB37_66
+; CHECK-NEXT:  .LBB37_60: // %else83
+; CHECK-NEXT:    add z1.d, z3.d, z1.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB37_67
+; CHECK-NEXT:  .LBB37_61: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB37_68
+; CHECK-NEXT:  .LBB37_62: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB37_64
+; CHECK-NEXT:  .LBB37_63: // %cond.load91
+; CHECK-NEXT:    mov z0.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    ld1 { v4.h }[7], [x8]
+; CHECK-NEXT:  .LBB37_64: // %else92
+; CHECK-NEXT:    stp q2, q5, [x0]
+; CHECK-NEXT:    stp q7, q4, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB37_65: // %cond.load79
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v4.h }[3], [x9]
+; CHECK-NEXT:    lslr z1.d, p0/m, z1.d, z16.d
+; CHECK-NEXT:    tbz w8, #28, .LBB37_60
+; CHECK-NEXT:  .LBB37_66: // %cond.load82
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    ld1 { v4.h }[4], [x9]
+; CHECK-NEXT:    add z1.d, z3.d, z1.d
+; CHECK-NEXT:    tbz w8, #29, .LBB37_61
+; CHECK-NEXT:  .LBB37_67: // %cond.load85
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    ld1 { v4.h }[5], [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB37_62
+; CHECK-NEXT:  .LBB37_68: // %cond.load88
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v4.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB37_63
+; CHECK-NEXT:    b .LBB37_64
   %cvals = load <32 x half>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = sext <32 x i32> %idxs to <32 x i64>
   %ptrs = getelementptr half, ptr %base, <32 x i64> %ext
   %mask = fcmp oeq <32 x half> %cvals, zeroinitializer
   %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
   store <32 x half> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_32b_scaled_sext_f32(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_gather_32b_scaled_sext_f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    ld1w { z0.s }, p1/z, [x2, z1.s, sxtw #2]
-; CHECK-NEXT:    st1w { z0.s }, p0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    adrp x8, .LCPI38_1
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q2, q1, [x0, #64]
+; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI38_1]
+; CHECK-NEXT:    ldp q19, q4, [x0, #32]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z5.s, z3.s[1]
+; CHECK-NEXT:    fmov w8, s5
+; CHECK-NEXT:    mov z18.s, z3.s[2]
+; CHECK-NEXT:    mov z24.s, z3.s[3]
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z3.s, z1.s[1]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s18
+; CHECK-NEXT:    ldp q22, q20, [x0]
+; CHECK-NEXT:    mov z24.s, z1.s[2]
+; CHECK-NEXT:    mov z25.s, z1.s[3]
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    bfi w9, w8, #18, #1
+; CHECK-NEXT:    bfi w9, w10, #19, #1
+; CHECK-NEXT:    fmov w8, s3
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    ldp q6, q2, [x1, #96]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldp q16, q17, [x1, #64]
+; CHECK-NEXT:    ldp q5, q7, [x1, #32]
+; CHECK-NEXT:    ldp q18, q1, [x1]
+; CHECK-NEXT:    ldp q23, q21, [x0, #96]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z23.s, z0.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #22
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z21.s, z0.s
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z23.s, z3.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s23
+; CHECK-NEXT:    mov z24.s, z3.s[2]
+; CHECK-NEXT:    mov z25.s, z3.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #23
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    adrp x8, .LCPI38_0
+; CHECK-NEXT:    orr w9, w9, w11, lsl #24
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    orr w9, w9, w10, lsl #25
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w12, s3
+; CHECK-NEXT:    mov z23.s, z3.s[1]
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s25
+; CHECK-NEXT:    mov z24.s, z3.s[2]
+; CHECK-NEXT:    mov z21.s, z3.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #26
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI38_0]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z22.s, z0.s
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    mov z22.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w9, w11, lsl #27
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    mov z23.s, z22.s[1]
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z23.s, z22.s[2]
+; CHECK-NEXT:    mov z25.s, z22.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z20.s, z0.s
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w9, w10, #1, #1
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    fmov w12, s25
+; CHECK-NEXT:    mov z20.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z22.s, z20.s[1]
+; CHECK-NEXT:    mov z23.s, z20.s[2]
+; CHECK-NEXT:    fmov w13, s20
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    bfi w9, w12, #3, #1
+; CHECK-NEXT:    fmov w12, s23
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z24.s, z20.s[3]
+; CHECK-NEXT:    bfi w9, w13, #4, #1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z19.s, z0.s
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov w12, s24
+; CHECK-NEXT:    mov z19.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w13, s19
+; CHECK-NEXT:    orr w9, w9, w10, lsl #6
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z20.s, z19.s[1]
+; CHECK-NEXT:    mov z22.s, z19.s[2]
+; CHECK-NEXT:    mov z23.s, z19.s[3]
+; CHECK-NEXT:    and w10, w13, #0x1
+; CHECK-NEXT:    orr w9, w9, w12, lsl #7
+; CHECK-NEXT:    orr w8, w8, w11, lsl #30
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    orr w9, w9, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    fmov w12, s23
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z4.s, z0.s
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z4.s, z0.s[1]
+; CHECK-NEXT:    mov z19.s, z0.s[2]
+; CHECK-NEXT:    mov z20.s, z0.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    orr w9, w9, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s0
+; CHECK-NEXT:    fmov w12, s19
+; CHECK-NEXT:    ldr q4, [sp]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    sunpklo z0.d, z18.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ext z18.b, z18.b, z18.b, #8
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    sunpklo z18.d, z18.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s20
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    fmov w10, s21
+; CHECK-NEXT:    lsl z0.d, p0/m, z0.d, z3.d
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    add z20.d, z4.d, z0.d
+; CHECK-NEXT:    orr w8, w8, w10, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB38_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    b .LBB38_3
+; CHECK-NEXT:  .LBB38_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:  .LBB38_3: // %else
+; CHECK-NEXT:    sunpklo z19.d, z1.s
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #1, .LBB38_5
+; CHECK-NEXT:  // %bb.4: // %cond.load1
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB38_5: // %else2
+; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #2, .LBB38_7
+; CHECK-NEXT:  // %bb.6: // %cond.load4
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:  .LBB38_7: // %else5
+; CHECK-NEXT:    sunpklo z1.d, z1.s
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #3, .LBB38_9
+; CHECK-NEXT:  // %bb.8: // %cond.load7
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:  .LBB38_9: // %else8
+; CHECK-NEXT:    movprfx z18, z1
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #4, .LBB38_11
+; CHECK-NEXT:  // %bb.10: // %cond.load10
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    b .LBB38_12
+; CHECK-NEXT:  .LBB38_11:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:  .LBB38_12: // %else11
+; CHECK-NEXT:    sunpklo z20.d, z5.s
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #5, .LBB38_14
+; CHECK-NEXT:  // %bb.13: // %cond.load13
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:  .LBB38_14: // %else14
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    movprfx z19, z20
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #6, .LBB38_16
+; CHECK-NEXT:  // %bb.15: // %cond.load16
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:  .LBB38_16: // %else17
+; CHECK-NEXT:    sunpklo z5.d, z5.s
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #7, .LBB38_18
+; CHECK-NEXT:  // %bb.17: // %cond.load19
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v1.s }[3], [x9]
+; CHECK-NEXT:  .LBB38_18: // %else20
+; CHECK-NEXT:    movprfx z18, z5
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #8, .LBB38_20
+; CHECK-NEXT:  // %bb.19: // %cond.load22
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v5.s }[0], [x9]
+; CHECK-NEXT:    b .LBB38_21
+; CHECK-NEXT:  .LBB38_20:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:  .LBB38_21: // %else23
+; CHECK-NEXT:    sunpklo z20.d, z7.s
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #9, .LBB38_23
+; CHECK-NEXT:  // %bb.22: // %cond.load25
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v5.s }[1], [x9]
+; CHECK-NEXT:  .LBB38_23: // %else26
+; CHECK-NEXT:    ext z7.b, z7.b, z7.b, #8
+; CHECK-NEXT:    movprfx z19, z20
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #10, .LBB38_25
+; CHECK-NEXT:  // %bb.24: // %cond.load28
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.s }[2], [x9]
+; CHECK-NEXT:  .LBB38_25: // %else29
+; CHECK-NEXT:    sunpklo z7.d, z7.s
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #11, .LBB38_27
+; CHECK-NEXT:  // %bb.26: // %cond.load31
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.s }[3], [x9]
+; CHECK-NEXT:  .LBB38_27: // %else32
+; CHECK-NEXT:    movprfx z18, z7
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #12, .LBB38_29
+; CHECK-NEXT:  // %bb.28: // %cond.load34
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    b .LBB38_30
+; CHECK-NEXT:  .LBB38_29:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:  .LBB38_30: // %else35
+; CHECK-NEXT:    sunpklo z20.d, z16.s
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #13, .LBB38_32
+; CHECK-NEXT:  // %bb.31: // %cond.load37
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:  .LBB38_32: // %else38
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    movprfx z19, z20
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #14, .LBB38_34
+; CHECK-NEXT:  // %bb.33: // %cond.load40
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:  .LBB38_34: // %else41
+; CHECK-NEXT:    sunpklo z16.d, z16.s
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #15, .LBB38_36
+; CHECK-NEXT:  // %bb.35: // %cond.load43
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v7.s }[3], [x9]
+; CHECK-NEXT:  .LBB38_36: // %else44
+; CHECK-NEXT:    movprfx z18, z16
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #16, .LBB38_38
+; CHECK-NEXT:  // %bb.37: // %cond.load46
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v16.s }[0], [x9]
+; CHECK-NEXT:    b .LBB38_39
+; CHECK-NEXT:  .LBB38_38:
+; CHECK-NEXT:    // implicit-def: $q16
+; CHECK-NEXT:  .LBB38_39: // %else47
+; CHECK-NEXT:    sunpklo z20.d, z17.s
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #17, .LBB38_41
+; CHECK-NEXT:  // %bb.40: // %cond.load49
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v16.s }[1], [x9]
+; CHECK-NEXT:  .LBB38_41: // %else50
+; CHECK-NEXT:    ext z17.b, z17.b, z17.b, #8
+; CHECK-NEXT:    movprfx z19, z20
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #18, .LBB38_43
+; CHECK-NEXT:  // %bb.42: // %cond.load52
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v16.s }[2], [x9]
+; CHECK-NEXT:  .LBB38_43: // %else53
+; CHECK-NEXT:    sunpklo z17.d, z17.s
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #19, .LBB38_45
+; CHECK-NEXT:  // %bb.44: // %cond.load55
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v16.s }[3], [x9]
+; CHECK-NEXT:  .LBB38_45: // %else56
+; CHECK-NEXT:    movprfx z18, z17
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #20, .LBB38_47
+; CHECK-NEXT:  // %bb.46: // %cond.load58
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v17.s }[0], [x9]
+; CHECK-NEXT:    b .LBB38_48
+; CHECK-NEXT:  .LBB38_47:
+; CHECK-NEXT:    // implicit-def: $q17
+; CHECK-NEXT:  .LBB38_48: // %else59
+; CHECK-NEXT:    sunpklo z20.d, z6.s
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #21, .LBB38_50
+; CHECK-NEXT:  // %bb.49: // %cond.load61
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v17.s }[1], [x9]
+; CHECK-NEXT:  .LBB38_50: // %else62
+; CHECK-NEXT:    ext z6.b, z6.b, z6.b, #8
+; CHECK-NEXT:    movprfx z19, z20
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #22, .LBB38_52
+; CHECK-NEXT:  // %bb.51: // %cond.load64
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v17.s }[2], [x9]
+; CHECK-NEXT:  .LBB38_52: // %else65
+; CHECK-NEXT:    sunpklo z6.d, z6.s
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #23, .LBB38_54
+; CHECK-NEXT:  // %bb.53: // %cond.load67
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v17.s }[3], [x9]
+; CHECK-NEXT:  .LBB38_54: // %else68
+; CHECK-NEXT:    movprfx z18, z6
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #24, .LBB38_56
+; CHECK-NEXT:  // %bb.55: // %cond.load70
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v6.s }[0], [x9]
+; CHECK-NEXT:    b .LBB38_57
+; CHECK-NEXT:  .LBB38_56:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:  .LBB38_57: // %else71
+; CHECK-NEXT:    sunpklo z20.d, z2.s
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #25, .LBB38_59
+; CHECK-NEXT:  // %bb.58: // %cond.load73
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v6.s }[1], [x9]
+; CHECK-NEXT:  .LBB38_59: // %else74
+; CHECK-NEXT:    ext z2.b, z2.b, z2.b, #8
+; CHECK-NEXT:    movprfx z19, z20
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #26, .LBB38_61
+; CHECK-NEXT:  // %bb.60: // %cond.load76
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v6.s }[2], [x9]
+; CHECK-NEXT:  .LBB38_61: // %else77
+; CHECK-NEXT:    sunpklo z2.d, z2.s
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #27, .LBB38_63
+; CHECK-NEXT:  // %bb.62: // %cond.load79
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v6.s }[3], [x9]
+; CHECK-NEXT:  .LBB38_63: // %else80
+; CHECK-NEXT:    lslr z3.d, p0/m, z3.d, z2.d
+; CHECK-NEXT:    tbz w8, #28, .LBB38_65
+; CHECK-NEXT:  // %bb.64: // %cond.load82
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    add z3.d, z4.d, z3.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB38_66
+; CHECK-NEXT:    b .LBB38_67
+; CHECK-NEXT:  .LBB38_65:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    add z3.d, z4.d, z3.d
+; CHECK-NEXT:    tbz w8, #29, .LBB38_67
+; CHECK-NEXT:  .LBB38_66: // %cond.load85
+; CHECK-NEXT:    mov z4.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB38_67: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB38_71
+; CHECK-NEXT:  // %bb.68: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB38_70
+; CHECK-NEXT:  .LBB38_69: // %cond.load91
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x8, d3
+; CHECK-NEXT:    ld1 { v2.s }[3], [x8]
+; CHECK-NEXT:  .LBB38_70: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q5, q7, [x0, #32]
+; CHECK-NEXT:    stp q16, q17, [x0, #64]
+; CHECK-NEXT:    stp q6, q2, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB38_71: // %cond.load88
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB38_69
+; CHECK-NEXT:    b .LBB38_70
   %cvals = load <32 x float>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = sext <32 x i32> %idxs to <32 x i64>
   %ptrs = getelementptr float, ptr %base, <32 x i64> %ext
   %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
   %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
   store <32 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_32b_scaled_sext_f64(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_gather_32b_scaled_sext_f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1sw { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    ld1d { z0.d }, p1/z, [x2, z1.d, lsl #3]
-; CHECK-NEXT:    st1d { z0.d }, p0, [x0]
+; CHECK-NEXT:    adrp x8, .LCPI39_1
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q4, q2, [x0, #128]
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI39_1]
+; CHECK-NEXT:    ldp q21, q28, [x0, #192]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z4.d, z3.d
+; CHECK-NEXT:    mov z4.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z2.d, z3.d
+; CHECK-NEXT:    mov z5.d, z4.d[1]
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q5, q4, [x0, #160]
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z6.d, z2.d[1]
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z5.d, z3.d
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z4.d, z3.d
+; CHECK-NEXT:    mov z5.d, z2.d[1]
+; CHECK-NEXT:    fmov x11, d2
+; CHECK-NEXT:    bfi w8, w9, #18, #1
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w10, #19, #1
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    bfi w8, w11, #20, #1
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z21.d, z3.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    bfi w8, w9, #21, #1
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #22
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    ldp q16, q7, [x0, #64]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z28.d, z3.d
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    adrp x9, .LCPI39_0
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    mov z4.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q19, q6, [x1, #96]
+; CHECK-NEXT:    fmov x11, d4
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z28.d, z4.d[1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    ldp q20, q22, [x1, #64]
+; CHECK-NEXT:    ldp q5, q17, [x1, #32]
+; CHECK-NEXT:    ldp q21, q2, [x1]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    ldp q24, q23, [x0, #32]
+; CHECK-NEXT:    ldp q26, q25, [x0]
+; CHECK-NEXT:    ldp q27, q18, [x0, #224]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z27.d, z3.d
+; CHECK-NEXT:    ldr q4, [x9, :lo12:.LCPI39_0]
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z27.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    fmov x10, d27
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z26.d, z3.d
+; CHECK-NEXT:    orr w8, w8, w11, lsl #26
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z26.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z28.d, z27.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    mov z27.d, z26.d[1]
+; CHECK-NEXT:    fmov x11, d26
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    fmov x10, d27
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z25.d, z3.d
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z25.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z26.d, z25.d[1]
+; CHECK-NEXT:    fmov x12, d25
+; CHECK-NEXT:    bfi w11, w10, #1, #1
+; CHECK-NEXT:    fmov x10, d26
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z24.d, z3.d
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z24.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z23.d, z3.d
+; CHECK-NEXT:    bfi w11, w12, #2, #1
+; CHECK-NEXT:    mov z25.d, z24.d[1]
+; CHECK-NEXT:    fmov x12, d24
+; CHECK-NEXT:    mov z23.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    bfi w11, w10, #3, #1
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    bfi w11, w12, #4, #1
+; CHECK-NEXT:    mov z24.d, z23.d[1]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z16.d, z3.d
+; CHECK-NEXT:    bfi w11, w9, #5, #1
+; CHECK-NEXT:    mov z16.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d24
+; CHECK-NEXT:    fmov x12, d16
+; CHECK-NEXT:    mov z23.d, z16.d[1]
+; CHECK-NEXT:    orr w9, w11, w9, lsl #6
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z18.d, z3.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z7.d, z3.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    mov z7.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov x11, d7
+; CHECK-NEXT:    mov z18.d, z7.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z3.d
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, z3.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov x11, d18
+; CHECK-NEXT:    fmov x12, d1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    mov z7.d, z1.d[1]
+; CHECK-NEXT:    mov z0.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.d, z16.d[1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov x12, d0
+; CHECK-NEXT:    mov z3.d, z0.d[1]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov x11, d16
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    ldr q7, [sp]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    sunpklo z0.d, z21.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    lsl z0.d, p0/m, z0.d, z4.d
+; CHECK-NEXT:    orr w8, w8, w11, lsl #30
+; CHECK-NEXT:    ext z21.b, z21.b, z21.b, #8
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    orr w9, w9, w12, lsl #14
+; CHECK-NEXT:    fmov x12, d1
+; CHECK-NEXT:    add z3.d, z7.d, z0.d
+; CHECK-NEXT:    sunpklo z1.d, z21.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    movprfx z16, z1
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z4.d
+; CHECK-NEXT:    orr w8, w8, w12, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB39_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ldr d0, [x9]
+; CHECK-NEXT:    b .LBB39_3
+; CHECK-NEXT:  .LBB39_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:  .LBB39_3: // %else
+; CHECK-NEXT:    sunpklo z1.d, z2.s
+; CHECK-NEXT:    add z16.d, z7.d, z16.d
+; CHECK-NEXT:    tbz w8, #1, .LBB39_5
+; CHECK-NEXT:  // %bb.4: // %cond.load1
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v0.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_5: // %else2
+; CHECK-NEXT:    ext z2.b, z2.b, z2.b, #8
+; CHECK-NEXT:    movprfx z3, z1
+; CHECK-NEXT:    lsl z3.d, p0/m, z3.d, z4.d
+; CHECK-NEXT:    tbz w8, #2, .LBB39_7
+; CHECK-NEXT:  // %bb.6: // %cond.load4
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v1.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_8
+; CHECK-NEXT:  .LBB39_7:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:  .LBB39_8: // %else5
+; CHECK-NEXT:    sunpklo z2.d, z2.s
+; CHECK-NEXT:    add z3.d, z7.d, z3.d
+; CHECK-NEXT:    tbz w8, #3, .LBB39_10
+; CHECK-NEXT:  // %bb.9: // %cond.load7
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v1.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_10: // %else8
+; CHECK-NEXT:    movprfx z18, z2
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z4.d
+; CHECK-NEXT:    tbz w8, #4, .LBB39_12
+; CHECK-NEXT:  // %bb.11: // %cond.load10
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_13
+; CHECK-NEXT:  .LBB39_12:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:  .LBB39_13: // %else11
+; CHECK-NEXT:    sunpklo z16.d, z5.s
+; CHECK-NEXT:    add z18.d, z7.d, z18.d
+; CHECK-NEXT:    tbz w8, #5, .LBB39_15
+; CHECK-NEXT:  // %bb.14: // %cond.load13
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_15: // %else14
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z4.d
+; CHECK-NEXT:    tbz w8, #6, .LBB39_17
+; CHECK-NEXT:  // %bb.16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v3.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_18
+; CHECK-NEXT:  .LBB39_17:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:  .LBB39_18: // %else17
+; CHECK-NEXT:    sunpklo z5.d, z5.s
+; CHECK-NEXT:    add z16.d, z7.d, z16.d
+; CHECK-NEXT:    tbz w8, #7, .LBB39_20
+; CHECK-NEXT:  // %bb.19: // %cond.load19
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v3.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_20: // %else20
+; CHECK-NEXT:    movprfx z21, z5
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z4.d
+; CHECK-NEXT:    tbz w8, #8, .LBB39_22
+; CHECK-NEXT:  // %bb.21: // %cond.load22
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v5.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_23
+; CHECK-NEXT:  .LBB39_22:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:  .LBB39_23: // %else23
+; CHECK-NEXT:    sunpklo z18.d, z17.s
+; CHECK-NEXT:    add z21.d, z7.d, z21.d
+; CHECK-NEXT:    tbz w8, #9, .LBB39_25
+; CHECK-NEXT:  // %bb.24: // %cond.load25
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v5.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_25: // %else26
+; CHECK-NEXT:    ext z17.b, z17.b, z17.b, #8
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z4.d
+; CHECK-NEXT:    tbz w8, #10, .LBB39_27
+; CHECK-NEXT:  // %bb.26: // %cond.load28
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v16.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_28
+; CHECK-NEXT:  .LBB39_27:
+; CHECK-NEXT:    // implicit-def: $q16
+; CHECK-NEXT:  .LBB39_28: // %else29
+; CHECK-NEXT:    sunpklo z17.d, z17.s
+; CHECK-NEXT:    add z18.d, z7.d, z18.d
+; CHECK-NEXT:    tbz w8, #11, .LBB39_30
+; CHECK-NEXT:  // %bb.29: // %cond.load31
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v16.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_30: // %else32
+; CHECK-NEXT:    movprfx z23, z17
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z4.d
+; CHECK-NEXT:    tbz w8, #12, .LBB39_32
+; CHECK-NEXT:  // %bb.31: // %cond.load34
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v17.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_33
+; CHECK-NEXT:  .LBB39_32:
+; CHECK-NEXT:    // implicit-def: $q17
+; CHECK-NEXT:  .LBB39_33: // %else35
+; CHECK-NEXT:    sunpklo z21.d, z20.s
+; CHECK-NEXT:    add z23.d, z7.d, z23.d
+; CHECK-NEXT:    tbz w8, #13, .LBB39_35
+; CHECK-NEXT:  // %bb.34: // %cond.load37
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v17.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_35: // %else38
+; CHECK-NEXT:    ext z20.b, z20.b, z20.b, #8
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z4.d
+; CHECK-NEXT:    tbz w8, #14, .LBB39_37
+; CHECK-NEXT:  // %bb.36: // %cond.load40
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v18.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_38
+; CHECK-NEXT:  .LBB39_37:
+; CHECK-NEXT:    // implicit-def: $q18
+; CHECK-NEXT:  .LBB39_38: // %else41
+; CHECK-NEXT:    sunpklo z20.d, z20.s
+; CHECK-NEXT:    add z21.d, z7.d, z21.d
+; CHECK-NEXT:    tbz w8, #15, .LBB39_40
+; CHECK-NEXT:  // %bb.39: // %cond.load43
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v18.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_40: // %else44
+; CHECK-NEXT:    movprfx z24, z20
+; CHECK-NEXT:    lsl z24.d, p0/m, z24.d, z4.d
+; CHECK-NEXT:    tbz w8, #16, .LBB39_42
+; CHECK-NEXT:  // %bb.41: // %cond.load46
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v20.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_43
+; CHECK-NEXT:  .LBB39_42:
+; CHECK-NEXT:    // implicit-def: $q20
+; CHECK-NEXT:  .LBB39_43: // %else47
+; CHECK-NEXT:    sunpklo z23.d, z22.s
+; CHECK-NEXT:    add z24.d, z7.d, z24.d
+; CHECK-NEXT:    tbz w8, #17, .LBB39_45
+; CHECK-NEXT:  // %bb.44: // %cond.load49
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v20.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_45: // %else50
+; CHECK-NEXT:    ext z22.b, z22.b, z22.b, #8
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z4.d
+; CHECK-NEXT:    tbz w8, #18, .LBB39_47
+; CHECK-NEXT:  // %bb.46: // %cond.load52
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v21.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_48
+; CHECK-NEXT:  .LBB39_47:
+; CHECK-NEXT:    // implicit-def: $q21
+; CHECK-NEXT:  .LBB39_48: // %else53
+; CHECK-NEXT:    sunpklo z22.d, z22.s
+; CHECK-NEXT:    add z23.d, z7.d, z23.d
+; CHECK-NEXT:    tbz w8, #19, .LBB39_50
+; CHECK-NEXT:  // %bb.49: // %cond.load55
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v21.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_50: // %else56
+; CHECK-NEXT:    movprfx z25, z22
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z4.d
+; CHECK-NEXT:    tbz w8, #20, .LBB39_52
+; CHECK-NEXT:  // %bb.51: // %cond.load58
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v22.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_53
+; CHECK-NEXT:  .LBB39_52:
+; CHECK-NEXT:    // implicit-def: $q22
+; CHECK-NEXT:  .LBB39_53: // %else59
+; CHECK-NEXT:    sunpklo z24.d, z19.s
+; CHECK-NEXT:    add z25.d, z7.d, z25.d
+; CHECK-NEXT:    tbz w8, #21, .LBB39_55
+; CHECK-NEXT:  // %bb.54: // %cond.load61
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v22.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_55: // %else62
+; CHECK-NEXT:    ext z19.b, z19.b, z19.b, #8
+; CHECK-NEXT:    lsl z24.d, p0/m, z24.d, z4.d
+; CHECK-NEXT:    tbz w8, #22, .LBB39_57
+; CHECK-NEXT:  // %bb.56: // %cond.load64
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v23.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_58
+; CHECK-NEXT:  .LBB39_57:
+; CHECK-NEXT:    // implicit-def: $q23
+; CHECK-NEXT:  .LBB39_58: // %else65
+; CHECK-NEXT:    sunpklo z19.d, z19.s
+; CHECK-NEXT:    add z24.d, z7.d, z24.d
+; CHECK-NEXT:    tbz w8, #23, .LBB39_60
+; CHECK-NEXT:  // %bb.59: // %cond.load67
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v23.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_60: // %else68
+; CHECK-NEXT:    movprfx z26, z19
+; CHECK-NEXT:    lsl z26.d, p0/m, z26.d, z4.d
+; CHECK-NEXT:    tbz w8, #24, .LBB39_62
+; CHECK-NEXT:  // %bb.61: // %cond.load70
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v19.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_63
+; CHECK-NEXT:  .LBB39_62:
+; CHECK-NEXT:    // implicit-def: $q19
+; CHECK-NEXT:  .LBB39_63: // %else71
+; CHECK-NEXT:    sunpklo z25.d, z6.s
+; CHECK-NEXT:    add z26.d, z7.d, z26.d
+; CHECK-NEXT:    tbz w8, #25, .LBB39_65
+; CHECK-NEXT:  // %bb.64: // %cond.load73
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v19.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_65: // %else74
+; CHECK-NEXT:    ext z6.b, z6.b, z6.b, #8
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z4.d
+; CHECK-NEXT:    tbz w8, #26, .LBB39_67
+; CHECK-NEXT:  // %bb.66: // %cond.load76
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    ld1 { v24.d }[0], [x9]
+; CHECK-NEXT:    b .LBB39_68
+; CHECK-NEXT:  .LBB39_67:
+; CHECK-NEXT:    // implicit-def: $q24
+; CHECK-NEXT:  .LBB39_68: // %else77
+; CHECK-NEXT:    sunpklo z6.d, z6.s
+; CHECK-NEXT:    add z25.d, z7.d, z25.d
+; CHECK-NEXT:    tbz w8, #27, .LBB39_70
+; CHECK-NEXT:  // %bb.69: // %cond.load79
+; CHECK-NEXT:    mov z26.d, z26.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    ld1 { v24.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_70: // %else80
+; CHECK-NEXT:    lsl z6.d, p0/m, z6.d, z4.d
+; CHECK-NEXT:    tbz w8, #28, .LBB39_72
+; CHECK-NEXT:  // %bb.71: // %cond.load82
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v4.d }[0], [x9]
+; CHECK-NEXT:    add z6.d, z7.d, z6.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB39_73
+; CHECK-NEXT:    b .LBB39_74
+; CHECK-NEXT:  .LBB39_72:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    add z6.d, z7.d, z6.d
+; CHECK-NEXT:    tbz w8, #29, .LBB39_74
+; CHECK-NEXT:  .LBB39_73: // %cond.load85
+; CHECK-NEXT:    mov z7.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v4.d }[1], [x9]
+; CHECK-NEXT:  .LBB39_74: // %else86
+; CHECK-NEXT:    tbz w8, #30, .LBB39_76
+; CHECK-NEXT:  // %bb.75: // %cond.load88
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v7.d }[0], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB39_77
+; CHECK-NEXT:    b .LBB39_78
+; CHECK-NEXT:  .LBB39_76:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    tbz w8, #31, .LBB39_78
+; CHECK-NEXT:  .LBB39_77: // %cond.load91
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x8, d6
+; CHECK-NEXT:    ld1 { v7.d }[1], [x8]
+; CHECK-NEXT:  .LBB39_78: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    stp q5, q16, [x0, #64]
+; CHECK-NEXT:    stp q17, q18, [x0, #96]
+; CHECK-NEXT:    stp q20, q21, [x0, #128]
+; CHECK-NEXT:    stp q22, q23, [x0, #160]
+; CHECK-NEXT:    stp q19, q24, [x0, #192]
+; CHECK-NEXT:    stp q4, q7, [x0, #224]
+; CHECK-NEXT:    add sp, sp, #32
 ; CHECK-NEXT:    ret
   %cvals = load <32 x double>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = sext <32 x i32> %idxs to <32 x i64>
   %ptrs = getelementptr double, ptr %base, <32 x i64> %ext
   %mask = fcmp oeq <32 x double> %cvals, zeroinitializer
   %vals = call <32 x double> @llvm.masked.gather.v32f64(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x double> undef)
   store <32 x double> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_32b_scaled_zext(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_gather_32b_scaled_zext:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.s, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1h { z0.s }, p0/z, [x2, z1.s, uxtw #1]
-; CHECK-NEXT:    st1h { z0.s }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    adrp x8, .LCPI40_1
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q1, q19, [x0, #32]
+; CHECK-NEXT:    adrp x12, .LCPI40_0
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI40_1]
+; CHECK-NEXT:    ldp q20, q18, [x0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z1.h, z3.h
+; CHECK-NEXT:    mov z1.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    mov z2.h, z1.h[1]
+; CHECK-NEXT:    mov z21.h, z1.h[2]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z22.h, z1.h[3]
+; CHECK-NEXT:    fmov w10, s21
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z23.h, z1.h[4]
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s23
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    mov z25.h, z1.h[6]
+; CHECK-NEXT:    bfi w9, w11, #19, #1
+; CHECK-NEXT:    mov z24.h, z1.h[5]
+; CHECK-NEXT:    ldp q4, q0, [x1, #96]
+; CHECK-NEXT:    bfi w9, w8, #20, #1
+; CHECK-NEXT:    fmov w8, s25
+; CHECK-NEXT:    mov z26.h, z1.h[7]
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    ldp q7, q6, [x1, #64]
+; CHECK-NEXT:    ldp q5, q16, [x1, #32]
+; CHECK-NEXT:    ldp q2, q17, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    bfi w9, w10, #21, #1
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z19.h, z3.h
+; CHECK-NEXT:    orr w8, w9, w8, lsl #22
+; CHECK-NEXT:    fmov w9, s26
+; CHECK-NEXT:    mov z21.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z20.h, z3.h
+; CHECK-NEXT:    fmov w10, s21
+; CHECK-NEXT:    mov z22.h, z21.h[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    mov z23.h, z21.h[2]
+; CHECK-NEXT:    mov z24.h, z21.h[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z25.h, z21.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    mov z26.h, z21.h[5]
+; CHECK-NEXT:    mov z20.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldr q1, [x12, :lo12:.LCPI40_0]
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    mov z27.h, z21.h[6]
+; CHECK-NEXT:    mov z19.h, z21.h[7]
+; CHECK-NEXT:    fmov w11, s25
+; CHECK-NEXT:    mov z21.h, z20.h[1]
+; CHECK-NEXT:    fmov w12, s26
+; CHECK-NEXT:    fmov w14, s20
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w13, s21
+; CHECK-NEXT:    mov z21.h, z20.h[2]
+; CHECK-NEXT:    mov z22.h, z20.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    mov z23.h, z20.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    and w12, w14, #0x1
+; CHECK-NEXT:    fmov w14, s22
+; CHECK-NEXT:    mov z24.h, z20.h[5]
+; CHECK-NEXT:    bfi w12, w13, #1, #1
+; CHECK-NEXT:    fmov w13, s23
+; CHECK-NEXT:    bfi w12, w9, #2, #1
+; CHECK-NEXT:    fmov w9, s24
+; CHECK-NEXT:    bfi w12, w14, #3, #1
+; CHECK-NEXT:    mov z25.h, z20.h[6]
+; CHECK-NEXT:    mov z26.h, z20.h[7]
+; CHECK-NEXT:    bfi w12, w13, #4, #1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    bfi w12, w9, #5, #1
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z18.h, z3.h
+; CHECK-NEXT:    orr w8, w8, w11, lsl #29
+; CHECK-NEXT:    mov z3.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    fmov w13, s3
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z18.h, z3.h[1]
+; CHECK-NEXT:    mov z20.h, z3.h[2]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w12, w9, lsl #6
+; CHECK-NEXT:    and w12, w13, #0x1
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    orr w9, w9, w11, lsl #7
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    mov z21.h, z3.h[3]
+; CHECK-NEXT:    orr w9, w9, w12, lsl #8
+; CHECK-NEXT:    fmov w12, s20
+; CHECK-NEXT:    fmov w13, s21
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z22.h, z3.h[4]
+; CHECK-NEXT:    mov z23.h, z3.h[5]
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #9
+; CHECK-NEXT:    and w11, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #30
+; CHECK-NEXT:    orr w9, w9, w12, lsl #10
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    mov z24.h, z3.h[6]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    fmov w12, s24
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z25.h, z3.h[7]
+; CHECK-NEXT:    ldr q3, [sp]
+; CHECK-NEXT:    uunpklo z18.d, z2.s
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s25
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z1.d
+; CHECK-NEXT:    ext z2.b, z2.b, z2.b, #8
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    orr w8, w8, w10, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    uunpklo z2.d, z2.s
+; CHECK-NEXT:    movprfx z19, z2
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z1.d
+; CHECK-NEXT:    tbz w8, #0, .LBB40_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ldr h2, [x9]
+; CHECK-NEXT:    b .LBB40_3
+; CHECK-NEXT:  .LBB40_2:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:  .LBB40_3: // %else
+; CHECK-NEXT:    uunpklo z20.d, z17.s
+; CHECK-NEXT:    add z19.d, z3.d, z19.d
+; CHECK-NEXT:    tbz w8, #1, .LBB40_5
+; CHECK-NEXT:  // %bb.4: // %cond.load1
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v2.h }[1], [x9]
+; CHECK-NEXT:  .LBB40_5: // %else2
+; CHECK-NEXT:    ext z17.b, z17.b, z17.b, #8
+; CHECK-NEXT:    movprfx z18, z20
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z1.d
+; CHECK-NEXT:    tbz w8, #2, .LBB40_7
+; CHECK-NEXT:  // %bb.6: // %cond.load4
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v2.h }[2], [x9]
+; CHECK-NEXT:  .LBB40_7: // %else5
+; CHECK-NEXT:    uunpklo z17.d, z17.s
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #3, .LBB40_9
+; CHECK-NEXT:  // %bb.8: // %cond.load7
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v2.h }[3], [x9]
+; CHECK-NEXT:  .LBB40_9: // %else8
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #4, .LBB40_11
+; CHECK-NEXT:  // %bb.10: // %cond.load10
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v2.h }[4], [x9]
+; CHECK-NEXT:  .LBB40_11: // %else11
+; CHECK-NEXT:    uunpklo z19.d, z5.s
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #5, .LBB40_13
+; CHECK-NEXT:  // %bb.12: // %cond.load13
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v2.h }[5], [x9]
+; CHECK-NEXT:  .LBB40_13: // %else14
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    movprfx z18, z19
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z1.d
+; CHECK-NEXT:    tbz w8, #6, .LBB40_15
+; CHECK-NEXT:  // %bb.14: // %cond.load16
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v2.h }[6], [x9]
+; CHECK-NEXT:  .LBB40_15: // %else17
+; CHECK-NEXT:    uunpklo z5.d, z5.s
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #7, .LBB40_17
+; CHECK-NEXT:  // %bb.16: // %cond.load19
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v2.h }[7], [x9]
+; CHECK-NEXT:  .LBB40_17: // %else20
+; CHECK-NEXT:    movprfx z17, z5
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #8, .LBB40_19
+; CHECK-NEXT:  // %bb.18: // %cond.load22
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.h }[0], [x9]
+; CHECK-NEXT:    b .LBB40_20
+; CHECK-NEXT:  .LBB40_19:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:  .LBB40_20: // %else23
+; CHECK-NEXT:    uunpklo z19.d, z16.s
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #9, .LBB40_22
+; CHECK-NEXT:  // %bb.21: // %cond.load25
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.h }[1], [x9]
+; CHECK-NEXT:  .LBB40_22: // %else26
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    movprfx z18, z19
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z1.d
+; CHECK-NEXT:    tbz w8, #10, .LBB40_24
+; CHECK-NEXT:  // %bb.23: // %cond.load28
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v5.h }[2], [x9]
+; CHECK-NEXT:  .LBB40_24: // %else29
+; CHECK-NEXT:    uunpklo z16.d, z16.s
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #11, .LBB40_26
+; CHECK-NEXT:  // %bb.25: // %cond.load31
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v5.h }[3], [x9]
+; CHECK-NEXT:  .LBB40_26: // %else32
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z1.d
+; CHECK-NEXT:    tbz w8, #12, .LBB40_28
+; CHECK-NEXT:  // %bb.27: // %cond.load34
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.h }[4], [x9]
+; CHECK-NEXT:  .LBB40_28: // %else35
+; CHECK-NEXT:    uunpklo z17.d, z7.s
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #13, .LBB40_30
+; CHECK-NEXT:  // %bb.29: // %cond.load37
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v5.h }[5], [x9]
+; CHECK-NEXT:  .LBB40_30: // %else38
+; CHECK-NEXT:    ext z7.b, z7.b, z7.b, #8
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #14, .LBB40_32
+; CHECK-NEXT:  // %bb.31: // %cond.load40
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v5.h }[6], [x9]
+; CHECK-NEXT:  .LBB40_32: // %else41
+; CHECK-NEXT:    uunpklo z7.d, z7.s
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #15, .LBB40_34
+; CHECK-NEXT:  // %bb.33: // %cond.load43
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v5.h }[7], [x9]
+; CHECK-NEXT:  .LBB40_34: // %else44
+; CHECK-NEXT:    movprfx z16, z7
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z1.d
+; CHECK-NEXT:    tbz w8, #16, .LBB40_36
+; CHECK-NEXT:  // %bb.35: // %cond.load46
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.h }[0], [x9]
+; CHECK-NEXT:    b .LBB40_37
+; CHECK-NEXT:  .LBB40_36:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:  .LBB40_37: // %else47
+; CHECK-NEXT:    uunpklo z18.d, z6.s
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #17, .LBB40_39
+; CHECK-NEXT:  // %bb.38: // %cond.load49
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.h }[1], [x9]
+; CHECK-NEXT:  .LBB40_39: // %else50
+; CHECK-NEXT:    ext z6.b, z6.b, z6.b, #8
+; CHECK-NEXT:    movprfx z17, z18
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #18, .LBB40_41
+; CHECK-NEXT:  // %bb.40: // %cond.load52
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.h }[2], [x9]
+; CHECK-NEXT:  .LBB40_41: // %else53
+; CHECK-NEXT:    uunpklo z6.d, z6.s
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #19, .LBB40_43
+; CHECK-NEXT:  // %bb.42: // %cond.load55
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.h }[3], [x9]
+; CHECK-NEXT:  .LBB40_43: // %else56
+; CHECK-NEXT:    lsl z6.d, p0/m, z6.d, z1.d
+; CHECK-NEXT:    tbz w8, #20, .LBB40_45
+; CHECK-NEXT:  // %bb.44: // %cond.load58
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.h }[4], [x9]
+; CHECK-NEXT:  .LBB40_45: // %else59
+; CHECK-NEXT:    uunpklo z16.d, z4.s
+; CHECK-NEXT:    add z6.d, z3.d, z6.d
+; CHECK-NEXT:    tbz w8, #21, .LBB40_47
+; CHECK-NEXT:  // %bb.46: // %cond.load61
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.h }[5], [x9]
+; CHECK-NEXT:  .LBB40_47: // %else62
+; CHECK-NEXT:    ext z4.b, z4.b, z4.b, #8
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z1.d
+; CHECK-NEXT:    tbz w8, #22, .LBB40_49
+; CHECK-NEXT:  // %bb.48: // %cond.load64
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v7.h }[6], [x9]
+; CHECK-NEXT:  .LBB40_49: // %else65
+; CHECK-NEXT:    uunpklo z4.d, z4.s
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #23, .LBB40_51
+; CHECK-NEXT:  // %bb.50: // %cond.load67
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v7.h }[7], [x9]
+; CHECK-NEXT:  .LBB40_51: // %else68
+; CHECK-NEXT:    movprfx z6, z4
+; CHECK-NEXT:    lsl z6.d, p0/m, z6.d, z1.d
+; CHECK-NEXT:    tbz w8, #24, .LBB40_53
+; CHECK-NEXT:  // %bb.52: // %cond.load70
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[0], [x9]
+; CHECK-NEXT:    b .LBB40_54
+; CHECK-NEXT:  .LBB40_53:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:  .LBB40_54: // %else71
+; CHECK-NEXT:    uunpklo z17.d, z0.s
+; CHECK-NEXT:    add z6.d, z3.d, z6.d
+; CHECK-NEXT:    tbz w8, #25, .LBB40_56
+; CHECK-NEXT:  // %bb.55: // %cond.load73
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[1], [x9]
+; CHECK-NEXT:  .LBB40_56: // %else74
+; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #8
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #26, .LBB40_58
+; CHECK-NEXT:  // %bb.57: // %cond.load76
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v4.h }[2], [x9]
+; CHECK-NEXT:  .LBB40_58: // %else77
+; CHECK-NEXT:    uunpklo z16.d, z0.s
+; CHECK-NEXT:    add z0.d, z3.d, z17.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB40_65
+; CHECK-NEXT:  // %bb.59: // %else80
+; CHECK-NEXT:    lslr z1.d, p0/m, z1.d, z16.d
+; CHECK-NEXT:    tbnz w8, #28, .LBB40_66
+; CHECK-NEXT:  .LBB40_60: // %else83
+; CHECK-NEXT:    add z1.d, z3.d, z1.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB40_67
+; CHECK-NEXT:  .LBB40_61: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB40_68
+; CHECK-NEXT:  .LBB40_62: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB40_64
+; CHECK-NEXT:  .LBB40_63: // %cond.load91
+; CHECK-NEXT:    mov z0.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    ld1 { v4.h }[7], [x8]
+; CHECK-NEXT:  .LBB40_64: // %else92
+; CHECK-NEXT:    stp q2, q5, [x0]
+; CHECK-NEXT:    stp q7, q4, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB40_65: // %cond.load79
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v4.h }[3], [x9]
+; CHECK-NEXT:    lslr z1.d, p0/m, z1.d, z16.d
+; CHECK-NEXT:    tbz w8, #28, .LBB40_60
+; CHECK-NEXT:  .LBB40_66: // %cond.load82
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    ld1 { v4.h }[4], [x9]
+; CHECK-NEXT:    add z1.d, z3.d, z1.d
+; CHECK-NEXT:    tbz w8, #29, .LBB40_61
+; CHECK-NEXT:  .LBB40_67: // %cond.load85
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    ld1 { v4.h }[5], [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB40_62
+; CHECK-NEXT:  .LBB40_68: // %cond.load88
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v4.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB40_63
+; CHECK-NEXT:    b .LBB40_64
   %cvals = load <32 x half>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = zext <32 x i32> %idxs to <32 x i64>
   %ptrs = getelementptr half, ptr %base, <32 x i64> %ext
   %mask = fcmp oeq <32 x half> %cvals, zeroinitializer
   %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
   store <32 x half> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_32b_unscaled_sext(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_gather_32b_unscaled_sext:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.s, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1h { z0.s }, p0/z, [x2, z1.s, sxtw]
-; CHECK-NEXT:    st1h { z0.s }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    adrp x8, .LCPI41_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q1, q18, [x0, #32]
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI41_0]
+; CHECK-NEXT:    ldp q2, q0, [x1, #96]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z1.h, z3.h
+; CHECK-NEXT:    mov z1.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z16.h, z1.h[1]
+; CHECK-NEXT:    mov z19.h, z1.h[2]
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    ldp q20, q17, [x0]
+; CHECK-NEXT:    mov z21.h, z1.h[3]
+; CHECK-NEXT:    mov z22.h, z1.h[4]
+; CHECK-NEXT:    mov z23.h, z1.h[5]
+; CHECK-NEXT:    mov z24.h, z1.h[6]
+; CHECK-NEXT:    mov z25.h, z1.h[7]
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    ldp q6, q5, [x1, #64]
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    ldp q4, q7, [x1, #32]
+; CHECK-NEXT:    ldp q1, q16, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    bfi w8, w9, #19, #1
+; CHECK-NEXT:    fmov w9, s24
+; CHECK-NEXT:    bfi w8, w10, #20, #1
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z18.h, z3.h
+; CHECK-NEXT:    bfi w8, w11, #21, #1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z21.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s21
+; CHECK-NEXT:    mov z22.h, z21.h[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    orr w8, w8, w10, lsl #23
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z23.h, z21.h[2]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z20.h, z3.h
+; CHECK-NEXT:    mov z24.h, z21.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    mov z25.h, z21.h[4]
+; CHECK-NEXT:    mov z20.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z26.h, z21.h[5]
+; CHECK-NEXT:    mov z19.h, z21.h[6]
+; CHECK-NEXT:    mov z18.h, z21.h[7]
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z21.h, z20.h[1]
+; CHECK-NEXT:    fmov w12, s25
+; CHECK-NEXT:    fmov w14, s20
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w13, s21
+; CHECK-NEXT:    mov z21.h, z20.h[2]
+; CHECK-NEXT:    mov z22.h, z20.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    mov z23.h, z20.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    and w12, w14, #0x1
+; CHECK-NEXT:    fmov w14, s22
+; CHECK-NEXT:    mov z24.h, z20.h[5]
+; CHECK-NEXT:    bfi w12, w13, #1, #1
+; CHECK-NEXT:    fmov w13, s23
+; CHECK-NEXT:    bfi w12, w9, #2, #1
+; CHECK-NEXT:    fmov w9, s24
+; CHECK-NEXT:    bfi w12, w14, #3, #1
+; CHECK-NEXT:    mov z25.h, z20.h[6]
+; CHECK-NEXT:    mov z27.h, z20.h[7]
+; CHECK-NEXT:    bfi w12, w13, #4, #1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    bfi w12, w9, #5, #1
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z17.h, z3.h
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    mov z3.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    mov z17.h, z3.h[1]
+; CHECK-NEXT:    fmov w13, s3
+; CHECK-NEXT:    orr w9, w12, w9, lsl #6
+; CHECK-NEXT:    fmov w12, s17
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    mov z20.h, z3.h[2]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #7
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    mov z21.h, z3.h[3]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    orr w9, w9, w13, lsl #8
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    mov z22.h, z3.h[4]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #9
+; CHECK-NEXT:    fmov w11, s21
+; CHECK-NEXT:    fmov w12, s22
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z23.h, z3.h[5]
+; CHECK-NEXT:    mov z24.h, z3.h[6]
+; CHECK-NEXT:    mov z25.h, z3.h[7]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov w12, s19
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    ldr q3, [sp]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    sunpklo z17.d, z1.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    sunpklo z18.d, z1.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB41_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ldr h1, [x9]
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB41_3
+; CHECK-NEXT:    b .LBB41_4
+; CHECK-NEXT:  .LBB41_2:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #1, .LBB41_4
+; CHECK-NEXT:  .LBB41_3: // %cond.load1
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v1.h }[1], [x9]
+; CHECK-NEXT:  .LBB41_4: // %else2
+; CHECK-NEXT:    sunpklo z17.d, z16.s
+; CHECK-NEXT:    tbz w8, #2, .LBB41_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v1.h }[2], [x9]
+; CHECK-NEXT:  .LBB41_6: // %else5
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB41_15
+; CHECK-NEXT:  // %bb.7: // %else8
+; CHECK-NEXT:    sunpklo z16.d, z16.s
+; CHECK-NEXT:    tbnz w8, #4, .LBB41_16
+; CHECK-NEXT:  .LBB41_8: // %else11
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB41_17
+; CHECK-NEXT:  .LBB41_9: // %else14
+; CHECK-NEXT:    sunpklo z17.d, z4.s
+; CHECK-NEXT:    tbz w8, #6, .LBB41_11
+; CHECK-NEXT:  .LBB41_10: // %cond.load16
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v1.h }[6], [x9]
+; CHECK-NEXT:  .LBB41_11: // %else17
+; CHECK-NEXT:    ext z4.b, z4.b, z4.b, #8
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #7, .LBB41_13
+; CHECK-NEXT:  // %bb.12: // %cond.load19
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v1.h }[7], [x9]
+; CHECK-NEXT:  .LBB41_13: // %else20
+; CHECK-NEXT:    sunpklo z16.d, z4.s
+; CHECK-NEXT:    tbz w8, #8, .LBB41_18
+; CHECK-NEXT:  // %bb.14: // %cond.load22
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v4.h }[0], [x9]
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB41_19
+; CHECK-NEXT:    b .LBB41_20
+; CHECK-NEXT:  .LBB41_15: // %cond.load7
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v1.h }[3], [x9]
+; CHECK-NEXT:    sunpklo z16.d, z16.s
+; CHECK-NEXT:    tbz w8, #4, .LBB41_8
+; CHECK-NEXT:  .LBB41_16: // %cond.load10
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v1.h }[4], [x9]
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #5, .LBB41_9
+; CHECK-NEXT:  .LBB41_17: // %cond.load13
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v1.h }[5], [x9]
+; CHECK-NEXT:    sunpklo z17.d, z4.s
+; CHECK-NEXT:    tbnz w8, #6, .LBB41_10
+; CHECK-NEXT:    b .LBB41_11
+; CHECK-NEXT:  .LBB41_18:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #9, .LBB41_20
+; CHECK-NEXT:  .LBB41_19: // %cond.load25
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v4.h }[1], [x9]
+; CHECK-NEXT:  .LBB41_20: // %else26
+; CHECK-NEXT:    sunpklo z17.d, z7.s
+; CHECK-NEXT:    tbz w8, #10, .LBB41_22
+; CHECK-NEXT:  // %bb.21: // %cond.load28
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[2], [x9]
+; CHECK-NEXT:  .LBB41_22: // %else29
+; CHECK-NEXT:    ext z7.b, z7.b, z7.b, #8
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB41_31
+; CHECK-NEXT:  // %bb.23: // %else32
+; CHECK-NEXT:    sunpklo z7.d, z7.s
+; CHECK-NEXT:    tbnz w8, #12, .LBB41_32
+; CHECK-NEXT:  .LBB41_24: // %else35
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB41_33
+; CHECK-NEXT:  .LBB41_25: // %else38
+; CHECK-NEXT:    sunpklo z16.d, z6.s
+; CHECK-NEXT:    tbz w8, #14, .LBB41_27
+; CHECK-NEXT:  .LBB41_26: // %cond.load40
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v4.h }[6], [x9]
+; CHECK-NEXT:  .LBB41_27: // %else41
+; CHECK-NEXT:    ext z6.b, z6.b, z6.b, #8
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #15, .LBB41_29
+; CHECK-NEXT:  // %bb.28: // %cond.load43
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v4.h }[7], [x9]
+; CHECK-NEXT:  .LBB41_29: // %else44
+; CHECK-NEXT:    sunpklo z7.d, z6.s
+; CHECK-NEXT:    tbz w8, #16, .LBB41_34
+; CHECK-NEXT:  // %bb.30: // %cond.load46
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v6.h }[0], [x9]
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB41_35
+; CHECK-NEXT:    b .LBB41_36
+; CHECK-NEXT:  .LBB41_31: // %cond.load31
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[3], [x9]
+; CHECK-NEXT:    sunpklo z7.d, z7.s
+; CHECK-NEXT:    tbz w8, #12, .LBB41_24
+; CHECK-NEXT:  .LBB41_32: // %cond.load34
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v4.h }[4], [x9]
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbz w8, #13, .LBB41_25
+; CHECK-NEXT:  .LBB41_33: // %cond.load37
+; CHECK-NEXT:    mov z16.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[5], [x9]
+; CHECK-NEXT:    sunpklo z16.d, z6.s
+; CHECK-NEXT:    tbnz w8, #14, .LBB41_26
+; CHECK-NEXT:    b .LBB41_27
+; CHECK-NEXT:  .LBB41_34:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbz w8, #17, .LBB41_36
+; CHECK-NEXT:  .LBB41_35: // %cond.load49
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v6.h }[1], [x9]
+; CHECK-NEXT:  .LBB41_36: // %else50
+; CHECK-NEXT:    sunpklo z16.d, z5.s
+; CHECK-NEXT:    tbz w8, #18, .LBB41_38
+; CHECK-NEXT:  // %bb.37: // %cond.load52
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.h }[2], [x9]
+; CHECK-NEXT:  .LBB41_38: // %else53
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB41_47
+; CHECK-NEXT:  // %bb.39: // %else56
+; CHECK-NEXT:    sunpklo z5.d, z5.s
+; CHECK-NEXT:    tbnz w8, #20, .LBB41_48
+; CHECK-NEXT:  .LBB41_40: // %else59
+; CHECK-NEXT:    add z5.d, z3.d, z5.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB41_49
+; CHECK-NEXT:  .LBB41_41: // %else62
+; CHECK-NEXT:    sunpklo z7.d, z2.s
+; CHECK-NEXT:    tbz w8, #22, .LBB41_43
+; CHECK-NEXT:  .LBB41_42: // %cond.load64
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v6.h }[6], [x9]
+; CHECK-NEXT:  .LBB41_43: // %else65
+; CHECK-NEXT:    ext z2.b, z2.b, z2.b, #8
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbz w8, #23, .LBB41_45
+; CHECK-NEXT:  // %bb.44: // %cond.load67
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v6.h }[7], [x9]
+; CHECK-NEXT:  .LBB41_45: // %else68
+; CHECK-NEXT:    sunpklo z5.d, z2.s
+; CHECK-NEXT:    tbz w8, #24, .LBB41_50
+; CHECK-NEXT:  // %bb.46: // %cond.load70
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v2.h }[0], [x9]
+; CHECK-NEXT:    add z5.d, z3.d, z5.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB41_51
+; CHECK-NEXT:    b .LBB41_52
+; CHECK-NEXT:  .LBB41_47: // %cond.load55
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.h }[3], [x9]
+; CHECK-NEXT:    sunpklo z5.d, z5.s
+; CHECK-NEXT:    tbz w8, #20, .LBB41_40
+; CHECK-NEXT:  .LBB41_48: // %cond.load58
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v6.h }[4], [x9]
+; CHECK-NEXT:    add z5.d, z3.d, z5.d
+; CHECK-NEXT:    tbz w8, #21, .LBB41_41
+; CHECK-NEXT:  .LBB41_49: // %cond.load61
+; CHECK-NEXT:    mov z7.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.h }[5], [x9]
+; CHECK-NEXT:    sunpklo z7.d, z2.s
+; CHECK-NEXT:    tbnz w8, #22, .LBB41_42
+; CHECK-NEXT:    b .LBB41_43
+; CHECK-NEXT:  .LBB41_50:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    add z5.d, z3.d, z5.d
+; CHECK-NEXT:    tbz w8, #25, .LBB41_52
+; CHECK-NEXT:  .LBB41_51: // %cond.load73
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v2.h }[1], [x9]
+; CHECK-NEXT:  .LBB41_52: // %else74
+; CHECK-NEXT:    sunpklo z7.d, z0.s
+; CHECK-NEXT:    tbz w8, #26, .LBB41_54
+; CHECK-NEXT:  // %bb.53: // %cond.load76
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v2.h }[2], [x9]
+; CHECK-NEXT:  .LBB41_54: // %else77
+; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #8
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB41_61
+; CHECK-NEXT:  // %bb.55: // %else80
+; CHECK-NEXT:    sunpklo z0.d, z0.s
+; CHECK-NEXT:    tbnz w8, #28, .LBB41_62
+; CHECK-NEXT:  .LBB41_56: // %else83
+; CHECK-NEXT:    add z0.d, z3.d, z0.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB41_63
+; CHECK-NEXT:  .LBB41_57: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB41_64
+; CHECK-NEXT:  .LBB41_58: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB41_60
+; CHECK-NEXT:  .LBB41_59: // %cond.load91
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    ld1 { v2.h }[7], [x8]
+; CHECK-NEXT:  .LBB41_60: // %else92
+; CHECK-NEXT:    stp q1, q4, [x0]
+; CHECK-NEXT:    stp q6, q2, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB41_61: // %cond.load79
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v2.h }[3], [x9]
+; CHECK-NEXT:    sunpklo z0.d, z0.s
+; CHECK-NEXT:    tbz w8, #28, .LBB41_56
+; CHECK-NEXT:  .LBB41_62: // %cond.load82
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v2.h }[4], [x9]
+; CHECK-NEXT:    add z0.d, z3.d, z0.d
+; CHECK-NEXT:    tbz w8, #29, .LBB41_57
+; CHECK-NEXT:  .LBB41_63: // %cond.load85
+; CHECK-NEXT:    mov z3.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[5], [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB41_58
+; CHECK-NEXT:  .LBB41_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    ld1 { v2.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB41_59
+; CHECK-NEXT:    b .LBB41_60
   %cvals = load <32 x half>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = sext <32 x i32> %idxs to <32 x i64>
   %byte_ptrs = getelementptr i8, ptr %base, <32 x i64> %ext
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x half> %cvals, zeroinitializer
   %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
   store <32 x half> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_32b_unscaled_zext(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_gather_32b_unscaled_zext:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.s, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1h { z0.s }, p0/z, [x2, z1.s, uxtw]
-; CHECK-NEXT:    st1h { z0.s }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    adrp x8, .LCPI42_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q1, q18, [x0, #32]
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI42_0]
+; CHECK-NEXT:    ldp q2, q0, [x1, #96]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z1.h, z3.h
+; CHECK-NEXT:    mov z1.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z16.h, z1.h[1]
+; CHECK-NEXT:    mov z19.h, z1.h[2]
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    ldp q20, q17, [x0]
+; CHECK-NEXT:    mov z21.h, z1.h[3]
+; CHECK-NEXT:    mov z22.h, z1.h[4]
+; CHECK-NEXT:    mov z23.h, z1.h[5]
+; CHECK-NEXT:    mov z24.h, z1.h[6]
+; CHECK-NEXT:    mov z25.h, z1.h[7]
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    ldp q6, q5, [x1, #64]
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    ldp q4, q7, [x1, #32]
+; CHECK-NEXT:    ldp q1, q16, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    bfi w8, w9, #19, #1
+; CHECK-NEXT:    fmov w9, s24
+; CHECK-NEXT:    bfi w8, w10, #20, #1
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z18.h, z3.h
+; CHECK-NEXT:    bfi w8, w11, #21, #1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z21.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s21
+; CHECK-NEXT:    mov z22.h, z21.h[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    orr w8, w8, w10, lsl #23
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z23.h, z21.h[2]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z20.h, z3.h
+; CHECK-NEXT:    mov z24.h, z21.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    mov z25.h, z21.h[4]
+; CHECK-NEXT:    mov z20.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z26.h, z21.h[5]
+; CHECK-NEXT:    mov z19.h, z21.h[6]
+; CHECK-NEXT:    mov z18.h, z21.h[7]
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z21.h, z20.h[1]
+; CHECK-NEXT:    fmov w12, s25
+; CHECK-NEXT:    fmov w14, s20
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w13, s21
+; CHECK-NEXT:    mov z21.h, z20.h[2]
+; CHECK-NEXT:    mov z22.h, z20.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    mov z23.h, z20.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    and w12, w14, #0x1
+; CHECK-NEXT:    fmov w14, s22
+; CHECK-NEXT:    mov z24.h, z20.h[5]
+; CHECK-NEXT:    bfi w12, w13, #1, #1
+; CHECK-NEXT:    fmov w13, s23
+; CHECK-NEXT:    bfi w12, w9, #2, #1
+; CHECK-NEXT:    fmov w9, s24
+; CHECK-NEXT:    bfi w12, w14, #3, #1
+; CHECK-NEXT:    mov z25.h, z20.h[6]
+; CHECK-NEXT:    mov z27.h, z20.h[7]
+; CHECK-NEXT:    bfi w12, w13, #4, #1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    bfi w12, w9, #5, #1
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z17.h, z3.h
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    mov z3.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    mov z17.h, z3.h[1]
+; CHECK-NEXT:    fmov w13, s3
+; CHECK-NEXT:    orr w9, w12, w9, lsl #6
+; CHECK-NEXT:    fmov w12, s17
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    mov z20.h, z3.h[2]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #7
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    mov z21.h, z3.h[3]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    orr w9, w9, w13, lsl #8
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    mov z22.h, z3.h[4]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #9
+; CHECK-NEXT:    fmov w11, s21
+; CHECK-NEXT:    fmov w12, s22
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z23.h, z3.h[5]
+; CHECK-NEXT:    mov z24.h, z3.h[6]
+; CHECK-NEXT:    mov z25.h, z3.h[7]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov w12, s19
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    ldr q3, [sp]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    uunpklo z17.d, z1.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    uunpklo z18.d, z1.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB42_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ldr h1, [x9]
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB42_3
+; CHECK-NEXT:    b .LBB42_4
+; CHECK-NEXT:  .LBB42_2:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    add z18.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #1, .LBB42_4
+; CHECK-NEXT:  .LBB42_3: // %cond.load1
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v1.h }[1], [x9]
+; CHECK-NEXT:  .LBB42_4: // %else2
+; CHECK-NEXT:    uunpklo z17.d, z16.s
+; CHECK-NEXT:    tbz w8, #2, .LBB42_6
+; CHECK-NEXT:  // %bb.5: // %cond.load4
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v1.h }[2], [x9]
+; CHECK-NEXT:  .LBB42_6: // %else5
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB42_15
+; CHECK-NEXT:  // %bb.7: // %else8
+; CHECK-NEXT:    uunpklo z16.d, z16.s
+; CHECK-NEXT:    tbnz w8, #4, .LBB42_16
+; CHECK-NEXT:  .LBB42_8: // %else11
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB42_17
+; CHECK-NEXT:  .LBB42_9: // %else14
+; CHECK-NEXT:    uunpklo z17.d, z4.s
+; CHECK-NEXT:    tbz w8, #6, .LBB42_11
+; CHECK-NEXT:  .LBB42_10: // %cond.load16
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v1.h }[6], [x9]
+; CHECK-NEXT:  .LBB42_11: // %else17
+; CHECK-NEXT:    ext z4.b, z4.b, z4.b, #8
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbz w8, #7, .LBB42_13
+; CHECK-NEXT:  // %bb.12: // %cond.load19
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v1.h }[7], [x9]
+; CHECK-NEXT:  .LBB42_13: // %else20
+; CHECK-NEXT:    uunpklo z16.d, z4.s
+; CHECK-NEXT:    tbz w8, #8, .LBB42_18
+; CHECK-NEXT:  // %bb.14: // %cond.load22
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v4.h }[0], [x9]
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB42_19
+; CHECK-NEXT:    b .LBB42_20
+; CHECK-NEXT:  .LBB42_15: // %cond.load7
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v1.h }[3], [x9]
+; CHECK-NEXT:    uunpklo z16.d, z16.s
+; CHECK-NEXT:    tbz w8, #4, .LBB42_8
+; CHECK-NEXT:  .LBB42_16: // %cond.load10
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v1.h }[4], [x9]
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #5, .LBB42_9
+; CHECK-NEXT:  .LBB42_17: // %cond.load13
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v1.h }[5], [x9]
+; CHECK-NEXT:    uunpklo z17.d, z4.s
+; CHECK-NEXT:    tbnz w8, #6, .LBB42_10
+; CHECK-NEXT:    b .LBB42_11
+; CHECK-NEXT:  .LBB42_18:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #9, .LBB42_20
+; CHECK-NEXT:  .LBB42_19: // %cond.load25
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v4.h }[1], [x9]
+; CHECK-NEXT:  .LBB42_20: // %else26
+; CHECK-NEXT:    uunpklo z17.d, z7.s
+; CHECK-NEXT:    tbz w8, #10, .LBB42_22
+; CHECK-NEXT:  // %bb.21: // %cond.load28
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[2], [x9]
+; CHECK-NEXT:  .LBB42_22: // %else29
+; CHECK-NEXT:    ext z7.b, z7.b, z7.b, #8
+; CHECK-NEXT:    add z17.d, z3.d, z17.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB42_31
+; CHECK-NEXT:  // %bb.23: // %else32
+; CHECK-NEXT:    uunpklo z7.d, z7.s
+; CHECK-NEXT:    tbnz w8, #12, .LBB42_32
+; CHECK-NEXT:  .LBB42_24: // %else35
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB42_33
+; CHECK-NEXT:  .LBB42_25: // %else38
+; CHECK-NEXT:    uunpklo z16.d, z6.s
+; CHECK-NEXT:    tbz w8, #14, .LBB42_27
+; CHECK-NEXT:  .LBB42_26: // %cond.load40
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v4.h }[6], [x9]
+; CHECK-NEXT:  .LBB42_27: // %else41
+; CHECK-NEXT:    ext z6.b, z6.b, z6.b, #8
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbz w8, #15, .LBB42_29
+; CHECK-NEXT:  // %bb.28: // %cond.load43
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v4.h }[7], [x9]
+; CHECK-NEXT:  .LBB42_29: // %else44
+; CHECK-NEXT:    uunpklo z7.d, z6.s
+; CHECK-NEXT:    tbz w8, #16, .LBB42_34
+; CHECK-NEXT:  // %bb.30: // %cond.load46
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v6.h }[0], [x9]
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB42_35
+; CHECK-NEXT:    b .LBB42_36
+; CHECK-NEXT:  .LBB42_31: // %cond.load31
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[3], [x9]
+; CHECK-NEXT:    uunpklo z7.d, z7.s
+; CHECK-NEXT:    tbz w8, #12, .LBB42_24
+; CHECK-NEXT:  .LBB42_32: // %cond.load34
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v4.h }[4], [x9]
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbz w8, #13, .LBB42_25
+; CHECK-NEXT:  .LBB42_33: // %cond.load37
+; CHECK-NEXT:    mov z16.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v4.h }[5], [x9]
+; CHECK-NEXT:    uunpklo z16.d, z6.s
+; CHECK-NEXT:    tbnz w8, #14, .LBB42_26
+; CHECK-NEXT:    b .LBB42_27
+; CHECK-NEXT:  .LBB42_34:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbz w8, #17, .LBB42_36
+; CHECK-NEXT:  .LBB42_35: // %cond.load49
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v6.h }[1], [x9]
+; CHECK-NEXT:  .LBB42_36: // %else50
+; CHECK-NEXT:    uunpklo z16.d, z5.s
+; CHECK-NEXT:    tbz w8, #18, .LBB42_38
+; CHECK-NEXT:  // %bb.37: // %cond.load52
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.h }[2], [x9]
+; CHECK-NEXT:  .LBB42_38: // %else53
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    add z16.d, z3.d, z16.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB42_47
+; CHECK-NEXT:  // %bb.39: // %else56
+; CHECK-NEXT:    uunpklo z5.d, z5.s
+; CHECK-NEXT:    tbnz w8, #20, .LBB42_48
+; CHECK-NEXT:  .LBB42_40: // %else59
+; CHECK-NEXT:    add z5.d, z3.d, z5.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB42_49
+; CHECK-NEXT:  .LBB42_41: // %else62
+; CHECK-NEXT:    uunpklo z7.d, z2.s
+; CHECK-NEXT:    tbz w8, #22, .LBB42_43
+; CHECK-NEXT:  .LBB42_42: // %cond.load64
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v6.h }[6], [x9]
+; CHECK-NEXT:  .LBB42_43: // %else65
+; CHECK-NEXT:    ext z2.b, z2.b, z2.b, #8
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbz w8, #23, .LBB42_45
+; CHECK-NEXT:  // %bb.44: // %cond.load67
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v6.h }[7], [x9]
+; CHECK-NEXT:  .LBB42_45: // %else68
+; CHECK-NEXT:    uunpklo z5.d, z2.s
+; CHECK-NEXT:    tbz w8, #24, .LBB42_50
+; CHECK-NEXT:  // %bb.46: // %cond.load70
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v2.h }[0], [x9]
+; CHECK-NEXT:    add z5.d, z3.d, z5.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB42_51
+; CHECK-NEXT:    b .LBB42_52
+; CHECK-NEXT:  .LBB42_47: // %cond.load55
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.h }[3], [x9]
+; CHECK-NEXT:    uunpklo z5.d, z5.s
+; CHECK-NEXT:    tbz w8, #20, .LBB42_40
+; CHECK-NEXT:  .LBB42_48: // %cond.load58
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v6.h }[4], [x9]
+; CHECK-NEXT:    add z5.d, z3.d, z5.d
+; CHECK-NEXT:    tbz w8, #21, .LBB42_41
+; CHECK-NEXT:  .LBB42_49: // %cond.load61
+; CHECK-NEXT:    mov z7.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.h }[5], [x9]
+; CHECK-NEXT:    uunpklo z7.d, z2.s
+; CHECK-NEXT:    tbnz w8, #22, .LBB42_42
+; CHECK-NEXT:    b .LBB42_43
+; CHECK-NEXT:  .LBB42_50:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    add z5.d, z3.d, z5.d
+; CHECK-NEXT:    tbz w8, #25, .LBB42_52
+; CHECK-NEXT:  .LBB42_51: // %cond.load73
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v2.h }[1], [x9]
+; CHECK-NEXT:  .LBB42_52: // %else74
+; CHECK-NEXT:    uunpklo z7.d, z0.s
+; CHECK-NEXT:    tbz w8, #26, .LBB42_54
+; CHECK-NEXT:  // %bb.53: // %cond.load76
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v2.h }[2], [x9]
+; CHECK-NEXT:  .LBB42_54: // %else77
+; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #8
+; CHECK-NEXT:    add z7.d, z3.d, z7.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB42_61
+; CHECK-NEXT:  // %bb.55: // %else80
+; CHECK-NEXT:    uunpklo z0.d, z0.s
+; CHECK-NEXT:    tbnz w8, #28, .LBB42_62
+; CHECK-NEXT:  .LBB42_56: // %else83
+; CHECK-NEXT:    add z0.d, z3.d, z0.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB42_63
+; CHECK-NEXT:  .LBB42_57: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB42_64
+; CHECK-NEXT:  .LBB42_58: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB42_60
+; CHECK-NEXT:  .LBB42_59: // %cond.load91
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    ld1 { v2.h }[7], [x8]
+; CHECK-NEXT:  .LBB42_60: // %else92
+; CHECK-NEXT:    stp q1, q4, [x0]
+; CHECK-NEXT:    stp q6, q2, [x0, #32]
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB42_61: // %cond.load79
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v2.h }[3], [x9]
+; CHECK-NEXT:    uunpklo z0.d, z0.s
+; CHECK-NEXT:    tbz w8, #28, .LBB42_56
+; CHECK-NEXT:  .LBB42_62: // %cond.load82
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v2.h }[4], [x9]
+; CHECK-NEXT:    add z0.d, z3.d, z0.d
+; CHECK-NEXT:    tbz w8, #29, .LBB42_57
+; CHECK-NEXT:  .LBB42_63: // %cond.load85
+; CHECK-NEXT:    mov z3.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.h }[5], [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB42_58
+; CHECK-NEXT:  .LBB42_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    ld1 { v2.h }[6], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB42_59
+; CHECK-NEXT:    b .LBB42_60
   %cvals = load <32 x half>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = zext <32 x i32> %idxs to <32 x i64>
   %byte_ptrs = getelementptr i8, ptr %base, <32 x i64> %ext
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x half> %cvals, zeroinitializer
   %vals = call <32 x half> @llvm.masked.gather.v32f16(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x half> undef)
   store <32 x half> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_64b_scaled(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_gather_64b_scaled:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [x2, z1.d, lsl #2]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #64
+; CHECK-NEXT:    .cfi_def_cfa_offset 64
+; CHECK-NEXT:    stp d11, d10, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #48] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -24
+; CHECK-NEXT:    .cfi_offset b11, -32
+; CHECK-NEXT:    adrp x8, .LCPI43_1
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q1, q31, [x0, #64]
+; CHECK-NEXT:    adrp x12, .LCPI43_0
+; CHECK-NEXT:    ldr q28, [x8, :lo12:.LCPI43_1]
+; CHECK-NEXT:    ldp q3, q30, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z28.s
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z31.s, z28.s
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    mov z4.s, z1.s[1]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    mov z8.s, z1.s[2]
+; CHECK-NEXT:    mov z9.s, z1.s[3]
+; CHECK-NEXT:    mov z31.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s9
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z9.s, z31.s[2]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s8
+; CHECK-NEXT:    mov z8.s, z31.s[1]
+; CHECK-NEXT:    fmov w11, s31
+; CHECK-NEXT:    mov z10.s, z31.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z3.s, z28.s
+; CHECK-NEXT:    bfi w9, w8, #18, #1
+; CHECK-NEXT:    fmov w8, s8
+; CHECK-NEXT:    bfi w9, w10, #19, #1
+; CHECK-NEXT:    fmov w10, s9
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    and w8, w10, #0x1
+; CHECK-NEXT:    fmov w10, s10
+; CHECK-NEXT:    mov z31.s, z3.s[1]
+; CHECK-NEXT:    mov z8.s, z3.s[2]
+; CHECK-NEXT:    orr w8, w9, w8, lsl #22
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z9.s, z3.s[3]
+; CHECK-NEXT:    ldp q29, q27, [x0]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #23
+; CHECK-NEXT:    fmov w10, s8
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    fmov w9, s31
+; CHECK-NEXT:    fmov w11, s9
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z30.s, z28.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z30.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z29.s, z28.s
+; CHECK-NEXT:    mov z9.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z27.s, z28.s
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    mov z10.s, z9.s[1]
+; CHECK-NEXT:    fmov w10, s9
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s10
+; CHECK-NEXT:    mov z10.s, z9.s[2]
+; CHECK-NEXT:    mov z11.s, z9.s[3]
+; CHECK-NEXT:    ldr q3, [x12, :lo12:.LCPI43_0]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s11
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s10
+; CHECK-NEXT:    mov z27.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z31.s, z30.s[1]
+; CHECK-NEXT:    fmov w13, s27
+; CHECK-NEXT:    fmov w11, s30
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z9.s, z27.s[2]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z8.s, z30.s[2]
+; CHECK-NEXT:    ldp q26, q5, [x0, #32]
+; CHECK-NEXT:    mov z29.s, z30.s[3]
+; CHECK-NEXT:    mov z30.s, z27.s[1]
+; CHECK-NEXT:    fmov w9, s31
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    fmov w13, s9
+; CHECK-NEXT:    fmov w12, s30
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z10.s, z27.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z26.s, z28.s
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    orr w9, w10, w13, lsl #6
+; CHECK-NEXT:    fmov w10, s10
+; CHECK-NEXT:    mov z26.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z5.s, z28.s
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    mov z27.s, z26.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s27
+; CHECK-NEXT:    mov z30.s, z26.s[2]
+; CHECK-NEXT:    mov z31.s, z26.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    mov z5.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov w12, s5
+; CHECK-NEXT:    mov z26.s, z5.s[1]
+; CHECK-NEXT:    mov z27.s, z5.s[2]
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s31
+; CHECK-NEXT:    mov z28.s, z5.s[3]
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w12, s8
+; CHECK-NEXT:    ldp q0, q25, [x1]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    lsl z0.d, p0/m, z0.d, z3.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    ldp q24, q2, [x1, #32]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s28
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z3.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    ldp q23, q6, [x1, #64]
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q22, q16, [x1, #96]
+; CHECK-NEXT:    ldp q21, q18, [x1, #128]
+; CHECK-NEXT:    ldp q20, q19, [x1, #160]
+; CHECK-NEXT:    ldp q17, q7, [x1, #192]
+; CHECK-NEXT:    ldp q4, q1, [x1, #224]
+; CHECK-NEXT:    stp x2, x2, [sp]
+; CHECK-NEXT:    ldr q5, [sp]
+; CHECK-NEXT:    add z26.d, z5.d, z0.d
+; CHECK-NEXT:    tbz w8, #0, .LBB43_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    add z25.d, z5.d, z25.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB43_3
+; CHECK-NEXT:    b .LBB43_4
+; CHECK-NEXT:  .LBB43_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    add z25.d, z5.d, z25.d
+; CHECK-NEXT:    tbz w8, #1, .LBB43_4
+; CHECK-NEXT:  .LBB43_3: // %cond.load1
+; CHECK-NEXT:    mov z26.d, z26.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB43_4: // %else2
+; CHECK-NEXT:    lsl z24.d, p0/m, z24.d, z3.d
+; CHECK-NEXT:    tbnz w8, #2, .LBB43_9
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    add z24.d, z5.d, z24.d
+; CHECK-NEXT:    tbz w8, #3, .LBB43_7
+; CHECK-NEXT:  .LBB43_6: // %cond.load7
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:  .LBB43_7: // %else8
+; CHECK-NEXT:    movprfx z25, z2
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z3.d
+; CHECK-NEXT:    tbz w8, #4, .LBB43_10
+; CHECK-NEXT:  // %bb.8: // %cond.load10
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    add z25.d, z5.d, z25.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB43_11
+; CHECK-NEXT:    b .LBB43_12
+; CHECK-NEXT:  .LBB43_9: // %cond.load4
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    add z24.d, z5.d, z24.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB43_6
+; CHECK-NEXT:    b .LBB43_7
+; CHECK-NEXT:  .LBB43_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    add z25.d, z5.d, z25.d
+; CHECK-NEXT:    tbz w8, #5, .LBB43_12
+; CHECK-NEXT:  .LBB43_11: // %cond.load13
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB43_12: // %else14
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z3.d
+; CHECK-NEXT:    tbnz w8, #6, .LBB43_17
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    add z24.d, z5.d, z23.d
+; CHECK-NEXT:    tbz w8, #7, .LBB43_15
+; CHECK-NEXT:  .LBB43_14: // %cond.load19
+; CHECK-NEXT:    mov z23.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:  .LBB43_15: // %else20
+; CHECK-NEXT:    movprfx z23, z6
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z3.d
+; CHECK-NEXT:    tbz w8, #8, .LBB43_18
+; CHECK-NEXT:  // %bb.16: // %cond.load22
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v6.s }[0], [x9]
+; CHECK-NEXT:    add z23.d, z5.d, z23.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB43_19
+; CHECK-NEXT:    b .LBB43_20
+; CHECK-NEXT:  .LBB43_17: // %cond.load16
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    add z24.d, z5.d, z23.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB43_14
+; CHECK-NEXT:    b .LBB43_15
+; CHECK-NEXT:  .LBB43_18:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:    add z23.d, z5.d, z23.d
+; CHECK-NEXT:    tbz w8, #9, .LBB43_20
+; CHECK-NEXT:  .LBB43_19: // %cond.load25
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v6.s }[1], [x9]
+; CHECK-NEXT:  .LBB43_20: // %else26
+; CHECK-NEXT:    lsl z22.d, p0/m, z22.d, z3.d
+; CHECK-NEXT:    tbnz w8, #10, .LBB43_25
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    add z22.d, z5.d, z22.d
+; CHECK-NEXT:    tbz w8, #11, .LBB43_23
+; CHECK-NEXT:  .LBB43_22: // %cond.load31
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v6.s }[3], [x9]
+; CHECK-NEXT:  .LBB43_23: // %else32
+; CHECK-NEXT:    movprfx z23, z16
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z3.d
+; CHECK-NEXT:    tbz w8, #12, .LBB43_26
+; CHECK-NEXT:  // %bb.24: // %cond.load34
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v16.s }[0], [x9]
+; CHECK-NEXT:    add z23.d, z5.d, z23.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB43_27
+; CHECK-NEXT:    b .LBB43_28
+; CHECK-NEXT:  .LBB43_25: // %cond.load28
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v6.s }[2], [x9]
+; CHECK-NEXT:    add z22.d, z5.d, z22.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB43_22
+; CHECK-NEXT:    b .LBB43_23
+; CHECK-NEXT:  .LBB43_26:
+; CHECK-NEXT:    // implicit-def: $q16
+; CHECK-NEXT:    add z23.d, z5.d, z23.d
+; CHECK-NEXT:    tbz w8, #13, .LBB43_28
+; CHECK-NEXT:  .LBB43_27: // %cond.load37
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v16.s }[1], [x9]
+; CHECK-NEXT:  .LBB43_28: // %else38
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z3.d
+; CHECK-NEXT:    tbnz w8, #14, .LBB43_33
+; CHECK-NEXT:  // %bb.29: // %else41
+; CHECK-NEXT:    add z22.d, z5.d, z21.d
+; CHECK-NEXT:    tbz w8, #15, .LBB43_31
+; CHECK-NEXT:  .LBB43_30: // %cond.load43
+; CHECK-NEXT:    mov z21.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v16.s }[3], [x9]
+; CHECK-NEXT:  .LBB43_31: // %else44
+; CHECK-NEXT:    movprfx z21, z18
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z3.d
+; CHECK-NEXT:    tbz w8, #16, .LBB43_34
+; CHECK-NEXT:  // %bb.32: // %cond.load46
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v18.s }[0], [x9]
+; CHECK-NEXT:    add z21.d, z5.d, z21.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB43_35
+; CHECK-NEXT:    b .LBB43_36
+; CHECK-NEXT:  .LBB43_33: // %cond.load40
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v16.s }[2], [x9]
+; CHECK-NEXT:    add z22.d, z5.d, z21.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB43_30
+; CHECK-NEXT:    b .LBB43_31
+; CHECK-NEXT:  .LBB43_34:
+; CHECK-NEXT:    // implicit-def: $q18
+; CHECK-NEXT:    add z21.d, z5.d, z21.d
+; CHECK-NEXT:    tbz w8, #17, .LBB43_36
+; CHECK-NEXT:  .LBB43_35: // %cond.load49
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v18.s }[1], [x9]
+; CHECK-NEXT:  .LBB43_36: // %else50
+; CHECK-NEXT:    lsl z20.d, p0/m, z20.d, z3.d
+; CHECK-NEXT:    tbnz w8, #18, .LBB43_41
+; CHECK-NEXT:  // %bb.37: // %else53
+; CHECK-NEXT:    add z20.d, z5.d, z20.d
+; CHECK-NEXT:    tbz w8, #19, .LBB43_39
+; CHECK-NEXT:  .LBB43_38: // %cond.load55
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v18.s }[3], [x9]
+; CHECK-NEXT:  .LBB43_39: // %else56
+; CHECK-NEXT:    movprfx z21, z19
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z3.d
+; CHECK-NEXT:    tbz w8, #20, .LBB43_42
+; CHECK-NEXT:  // %bb.40: // %cond.load58
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.s }[0], [x9]
+; CHECK-NEXT:    add z21.d, z5.d, z21.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB43_43
+; CHECK-NEXT:    b .LBB43_44
+; CHECK-NEXT:  .LBB43_41: // %cond.load52
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v18.s }[2], [x9]
+; CHECK-NEXT:    add z20.d, z5.d, z20.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB43_38
+; CHECK-NEXT:    b .LBB43_39
+; CHECK-NEXT:  .LBB43_42:
+; CHECK-NEXT:    // implicit-def: $q19
+; CHECK-NEXT:    add z21.d, z5.d, z21.d
+; CHECK-NEXT:    tbz w8, #21, .LBB43_44
+; CHECK-NEXT:  .LBB43_43: // %cond.load61
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.s }[1], [x9]
+; CHECK-NEXT:  .LBB43_44: // %else62
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z3.d
+; CHECK-NEXT:    tbnz w8, #22, .LBB43_49
+; CHECK-NEXT:  // %bb.45: // %else65
+; CHECK-NEXT:    add z20.d, z5.d, z17.d
+; CHECK-NEXT:    tbz w8, #23, .LBB43_47
+; CHECK-NEXT:  .LBB43_46: // %cond.load67
+; CHECK-NEXT:    mov z17.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v19.s }[3], [x9]
+; CHECK-NEXT:  .LBB43_47: // %else68
+; CHECK-NEXT:    movprfx z17, z7
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z3.d
+; CHECK-NEXT:    tbz w8, #24, .LBB43_50
+; CHECK-NEXT:  // %bb.48: // %cond.load70
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    add z17.d, z5.d, z17.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB43_51
+; CHECK-NEXT:    b .LBB43_52
+; CHECK-NEXT:  .LBB43_49: // %cond.load64
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v19.s }[2], [x9]
+; CHECK-NEXT:    add z20.d, z5.d, z17.d
+; CHECK-NEXT:    tbnz w8, #23, .LBB43_46
+; CHECK-NEXT:    b .LBB43_47
+; CHECK-NEXT:  .LBB43_50:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    add z17.d, z5.d, z17.d
+; CHECK-NEXT:    tbz w8, #25, .LBB43_52
+; CHECK-NEXT:  .LBB43_51: // %cond.load73
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:  .LBB43_52: // %else74
+; CHECK-NEXT:    lsl z4.d, p0/m, z4.d, z3.d
+; CHECK-NEXT:    tbnz w8, #26, .LBB43_56
+; CHECK-NEXT:  // %bb.53: // %else77
+; CHECK-NEXT:    add z4.d, z5.d, z4.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB43_57
+; CHECK-NEXT:  .LBB43_54: // %else80
+; CHECK-NEXT:    lslr z3.d, p0/m, z3.d, z1.d
+; CHECK-NEXT:    tbz w8, #28, .LBB43_58
+; CHECK-NEXT:  .LBB43_55: // %cond.load82
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    add z3.d, z5.d, z3.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB43_59
+; CHECK-NEXT:    b .LBB43_60
+; CHECK-NEXT:  .LBB43_56: // %cond.load76
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:    add z4.d, z5.d, z4.d
+; CHECK-NEXT:    tbz w8, #27, .LBB43_54
+; CHECK-NEXT:  .LBB43_57: // %cond.load79
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v7.s }[3], [x9]
+; CHECK-NEXT:    lslr z3.d, p0/m, z3.d, z1.d
+; CHECK-NEXT:    tbnz w8, #28, .LBB43_55
+; CHECK-NEXT:  .LBB43_58:
+; CHECK-NEXT:    // implicit-def: $q1
+; CHECK-NEXT:    add z3.d, z5.d, z3.d
+; CHECK-NEXT:    tbz w8, #29, .LBB43_60
+; CHECK-NEXT:  .LBB43_59: // %cond.load85
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:  .LBB43_60: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB43_64
+; CHECK-NEXT:  // %bb.61: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB43_63
+; CHECK-NEXT:  .LBB43_62: // %cond.load91
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x8, d3
+; CHECK-NEXT:    ld1 { v1.s }[3], [x8]
+; CHECK-NEXT:  .LBB43_63: // %else92
+; CHECK-NEXT:    ldp d9, d8, [sp, #48] // 16-byte Folded Reload
+; CHECK-NEXT:    stp q0, q2, [x0]
+; CHECK-NEXT:    ldp d11, d10, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    stp q6, q16, [x0, #32]
+; CHECK-NEXT:    stp q18, q19, [x0, #64]
+; CHECK-NEXT:    stp q7, q1, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #64
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB43_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB43_62
+; CHECK-NEXT:    b .LBB43_63
   %cvals = load <32 x float>, ptr %a
   %idxs = load <32 x i64>, ptr %b
   %ptrs = getelementptr float, ptr %base, <32 x i64> %idxs
   %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
   %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
   store <32 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_64b_unscaled(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_gather_64b_unscaled:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [x2, z1.d]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #48
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    str d10, [sp, #16] // 8-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -32
+; CHECK-NEXT:    adrp x8, .LCPI44_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q2, q1, [x0, #64]
+; CHECK-NEXT:    ldr q26, [x8, :lo12:.LCPI44_0]
+; CHECK-NEXT:    ldp q30, q29, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z26.s
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z26.s
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z3.s, z2.s[1]
+; CHECK-NEXT:    fmov w8, s3
+; CHECK-NEXT:    mov z6.s, z2.s[2]
+; CHECK-NEXT:    mov z3.s, z2.s[3]
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z31.s, z1.s[2]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s6
+; CHECK-NEXT:    mov z3.s, z1.s[1]
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    mov z8.s, z1.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z30.s, z26.s
+; CHECK-NEXT:    bfi w9, w8, #18, #1
+; CHECK-NEXT:    fmov w8, s3
+; CHECK-NEXT:    bfi w9, w10, #19, #1
+; CHECK-NEXT:    fmov w10, s31
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    mov z30.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    mov z31.s, z30.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z9.s, z30.s[3]
+; CHECK-NEXT:    ldp q28, q27, [x0]
+; CHECK-NEXT:    orr w8, w9, w10, lsl #22
+; CHECK-NEXT:    fmov w9, s8
+; CHECK-NEXT:    fmov w10, s30
+; CHECK-NEXT:    mov z8.s, z30.s[2]
+; CHECK-NEXT:    fmov w11, s9
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z29.s, z26.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z29.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z28.s, z26.s
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s31
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s8
+; CHECK-NEXT:    mov z8.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z27.s, z26.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z9.s, z8.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z10.s, z8.s[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s8
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s9
+; CHECK-NEXT:    mov z9.s, z8.s[2]
+; CHECK-NEXT:    fmov w12, s10
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z27.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s9
+; CHECK-NEXT:    fmov w13, s27
+; CHECK-NEXT:    mov z30.s, z29.s[1]
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    mov z8.s, z27.s[2]
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z31.s, z29.s[2]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z28.s, z29.s[3]
+; CHECK-NEXT:    ldp q4, q0, [x0, #32]
+; CHECK-NEXT:    mov z29.s, z27.s[1]
+; CHECK-NEXT:    fmov w9, s30
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    fmov w13, s8
+; CHECK-NEXT:    fmov w12, s29
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z9.s, z27.s[3]
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z26.s
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    orr w9, w10, w13, lsl #6
+; CHECK-NEXT:    fmov w10, s9
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    mov z27.s, z4.s[1]
+; CHECK-NEXT:    fmov w12, s27
+; CHECK-NEXT:    mov z29.s, z4.s[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z30.s, z4.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z26.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w12, s0
+; CHECK-NEXT:    mov z4.s, z0.s[1]
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s30
+; CHECK-NEXT:    mov z26.s, z0.s[2]
+; CHECK-NEXT:    mov z27.s, z0.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w12, s31
+; CHECK-NEXT:    ldp q25, q24, [x1]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    ldp q2, q23, [x1, #32]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s28
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    ldp q5, q22, [x1, #64]
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q7, q21, [x1, #96]
+; CHECK-NEXT:    ldp q18, q20, [x1, #128]
+; CHECK-NEXT:    ldp q19, q17, [x1, #160]
+; CHECK-NEXT:    ldp q16, q6, [x1, #192]
+; CHECK-NEXT:    ldp q3, q1, [x1, #224]
+; CHECK-NEXT:    stp x2, x2, [sp]
+; CHECK-NEXT:    ldr q4, [sp]
+; CHECK-NEXT:    add z26.d, z4.d, z25.d
+; CHECK-NEXT:    tbz w8, #0, .LBB44_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    add z25.d, z4.d, z24.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB44_3
+; CHECK-NEXT:    b .LBB44_4
+; CHECK-NEXT:  .LBB44_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    add z25.d, z4.d, z24.d
+; CHECK-NEXT:    tbz w8, #1, .LBB44_4
+; CHECK-NEXT:  .LBB44_3: // %cond.load1
+; CHECK-NEXT:    mov z24.d, z26.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB44_4: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB44_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    add z24.d, z4.d, z2.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB44_9
+; CHECK-NEXT:  .LBB44_6: // %else8
+; CHECK-NEXT:    tbz w8, #4, .LBB44_10
+; CHECK-NEXT:  .LBB44_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    add z23.d, z4.d, z23.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB44_11
+; CHECK-NEXT:    b .LBB44_12
+; CHECK-NEXT:  .LBB44_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    add z24.d, z4.d, z2.d
+; CHECK-NEXT:    tbz w8, #3, .LBB44_6
+; CHECK-NEXT:  .LBB44_9: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #4, .LBB44_7
+; CHECK-NEXT:  .LBB44_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    add z23.d, z4.d, z23.d
+; CHECK-NEXT:    tbz w8, #5, .LBB44_12
+; CHECK-NEXT:  .LBB44_11: // %cond.load13
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB44_12: // %else14
+; CHECK-NEXT:    tbnz w8, #6, .LBB44_16
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    add z24.d, z4.d, z5.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB44_17
+; CHECK-NEXT:  .LBB44_14: // %else20
+; CHECK-NEXT:    tbz w8, #8, .LBB44_18
+; CHECK-NEXT:  .LBB44_15: // %cond.load22
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v5.s }[0], [x9]
+; CHECK-NEXT:    add z23.d, z4.d, z22.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB44_19
+; CHECK-NEXT:    b .LBB44_20
+; CHECK-NEXT:  .LBB44_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    add z24.d, z4.d, z5.d
+; CHECK-NEXT:    tbz w8, #7, .LBB44_14
+; CHECK-NEXT:  .LBB44_17: // %cond.load19
+; CHECK-NEXT:    mov z5.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #8, .LBB44_15
+; CHECK-NEXT:  .LBB44_18:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:    add z23.d, z4.d, z22.d
+; CHECK-NEXT:    tbz w8, #9, .LBB44_20
+; CHECK-NEXT:  .LBB44_19: // %cond.load25
+; CHECK-NEXT:    mov z22.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v5.s }[1], [x9]
+; CHECK-NEXT:  .LBB44_20: // %else26
+; CHECK-NEXT:    tbnz w8, #10, .LBB44_24
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    add z22.d, z4.d, z7.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB44_25
+; CHECK-NEXT:  .LBB44_22: // %else32
+; CHECK-NEXT:    tbz w8, #12, .LBB44_26
+; CHECK-NEXT:  .LBB44_23: // %cond.load34
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    add z21.d, z4.d, z21.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB44_27
+; CHECK-NEXT:    b .LBB44_28
+; CHECK-NEXT:  .LBB44_24: // %cond.load28
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v5.s }[2], [x9]
+; CHECK-NEXT:    add z22.d, z4.d, z7.d
+; CHECK-NEXT:    tbz w8, #11, .LBB44_22
+; CHECK-NEXT:  .LBB44_25: // %cond.load31
+; CHECK-NEXT:    mov z7.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v5.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #12, .LBB44_23
+; CHECK-NEXT:  .LBB44_26:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    add z21.d, z4.d, z21.d
+; CHECK-NEXT:    tbz w8, #13, .LBB44_28
+; CHECK-NEXT:  .LBB44_27: // %cond.load37
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:  .LBB44_28: // %else38
+; CHECK-NEXT:    tbnz w8, #14, .LBB44_32
+; CHECK-NEXT:  // %bb.29: // %else41
+; CHECK-NEXT:    add z22.d, z4.d, z18.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB44_33
+; CHECK-NEXT:  .LBB44_30: // %else44
+; CHECK-NEXT:    tbz w8, #16, .LBB44_34
+; CHECK-NEXT:  .LBB44_31: // %cond.load46
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v18.s }[0], [x9]
+; CHECK-NEXT:    add z21.d, z4.d, z20.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB44_35
+; CHECK-NEXT:    b .LBB44_36
+; CHECK-NEXT:  .LBB44_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:    add z22.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #15, .LBB44_30
+; CHECK-NEXT:  .LBB44_33: // %cond.load43
+; CHECK-NEXT:    mov z18.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v7.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #16, .LBB44_31
+; CHECK-NEXT:  .LBB44_34:
+; CHECK-NEXT:    // implicit-def: $q18
+; CHECK-NEXT:    add z21.d, z4.d, z20.d
+; CHECK-NEXT:    tbz w8, #17, .LBB44_36
+; CHECK-NEXT:  .LBB44_35: // %cond.load49
+; CHECK-NEXT:    mov z20.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v18.s }[1], [x9]
+; CHECK-NEXT:  .LBB44_36: // %else50
+; CHECK-NEXT:    tbnz w8, #18, .LBB44_40
+; CHECK-NEXT:  // %bb.37: // %else53
+; CHECK-NEXT:    add z20.d, z4.d, z19.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB44_41
+; CHECK-NEXT:  .LBB44_38: // %else56
+; CHECK-NEXT:    tbz w8, #20, .LBB44_42
+; CHECK-NEXT:  .LBB44_39: // %cond.load58
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.s }[0], [x9]
+; CHECK-NEXT:    add z17.d, z4.d, z17.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB44_43
+; CHECK-NEXT:    b .LBB44_44
+; CHECK-NEXT:  .LBB44_40: // %cond.load52
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v18.s }[2], [x9]
+; CHECK-NEXT:    add z20.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #19, .LBB44_38
+; CHECK-NEXT:  .LBB44_41: // %cond.load55
+; CHECK-NEXT:    mov z19.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v18.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #20, .LBB44_39
+; CHECK-NEXT:  .LBB44_42:
+; CHECK-NEXT:    // implicit-def: $q19
+; CHECK-NEXT:    add z17.d, z4.d, z17.d
+; CHECK-NEXT:    tbz w8, #21, .LBB44_44
+; CHECK-NEXT:  .LBB44_43: // %cond.load61
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.s }[1], [x9]
+; CHECK-NEXT:  .LBB44_44: // %else62
+; CHECK-NEXT:    tbnz w8, #22, .LBB44_48
+; CHECK-NEXT:  // %bb.45: // %else65
+; CHECK-NEXT:    add z20.d, z4.d, z16.d
+; CHECK-NEXT:    tbnz w8, #23, .LBB44_49
+; CHECK-NEXT:  .LBB44_46: // %else68
+; CHECK-NEXT:    tbz w8, #24, .LBB44_50
+; CHECK-NEXT:  .LBB44_47: // %cond.load70
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v16.s }[0], [x9]
+; CHECK-NEXT:    add z17.d, z4.d, z6.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB44_51
+; CHECK-NEXT:    b .LBB44_52
+; CHECK-NEXT:  .LBB44_48: // %cond.load64
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v19.s }[2], [x9]
+; CHECK-NEXT:    add z20.d, z4.d, z16.d
+; CHECK-NEXT:    tbz w8, #23, .LBB44_46
+; CHECK-NEXT:  .LBB44_49: // %cond.load67
+; CHECK-NEXT:    mov z16.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v19.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #24, .LBB44_47
+; CHECK-NEXT:  .LBB44_50:
+; CHECK-NEXT:    // implicit-def: $q16
+; CHECK-NEXT:    add z17.d, z4.d, z6.d
+; CHECK-NEXT:    tbz w8, #25, .LBB44_52
+; CHECK-NEXT:  .LBB44_51: // %cond.load73
+; CHECK-NEXT:    mov z6.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v16.s }[1], [x9]
+; CHECK-NEXT:  .LBB44_52: // %else74
+; CHECK-NEXT:    tbnz w8, #26, .LBB44_56
+; CHECK-NEXT:  // %bb.53: // %else77
+; CHECK-NEXT:    add z6.d, z4.d, z3.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB44_57
+; CHECK-NEXT:  .LBB44_54: // %else80
+; CHECK-NEXT:    tbz w8, #28, .LBB44_58
+; CHECK-NEXT:  .LBB44_55: // %cond.load82
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v3.s }[0], [x9]
+; CHECK-NEXT:    add z1.d, z4.d, z1.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB44_59
+; CHECK-NEXT:    b .LBB44_60
+; CHECK-NEXT:  .LBB44_56: // %cond.load76
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v16.s }[2], [x9]
+; CHECK-NEXT:    add z6.d, z4.d, z3.d
+; CHECK-NEXT:    tbz w8, #27, .LBB44_54
+; CHECK-NEXT:  .LBB44_57: // %cond.load79
+; CHECK-NEXT:    mov z3.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v16.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #28, .LBB44_55
+; CHECK-NEXT:  .LBB44_58:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    add z1.d, z4.d, z1.d
+; CHECK-NEXT:    tbz w8, #29, .LBB44_60
+; CHECK-NEXT:  .LBB44_59: // %cond.load85
+; CHECK-NEXT:    mov z4.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:  .LBB44_60: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB44_64
+; CHECK-NEXT:  // %bb.61: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB44_63
+; CHECK-NEXT:  .LBB44_62: // %cond.load91
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v3.s }[3], [x8]
+; CHECK-NEXT:  .LBB44_63: // %else92
+; CHECK-NEXT:    ldp d9, d8, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    stp q0, q2, [x0]
+; CHECK-NEXT:    ldr d10, [sp, #16] // 8-byte Folded Reload
+; CHECK-NEXT:    stp q5, q7, [x0, #32]
+; CHECK-NEXT:    stp q18, q19, [x0, #64]
+; CHECK-NEXT:    stp q16, q3, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #48
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB44_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB44_62
+; CHECK-NEXT:    b .LBB44_63
   %cvals = load <32 x float>, ptr %a
   %idxs = load <32 x i64>, ptr %b
   %byte_ptrs = getelementptr i8, ptr %base, <32 x i64> %idxs
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
   %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
   store <32 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_vec_plus_reg(ptr %a, ptr %b, i64 %off) #0 {
 ; CHECK-LABEL: masked_gather_vec_plus_reg:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [x2, z1.d]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #48
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    str d10, [sp, #16] // 8-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -32
+; CHECK-NEXT:    adrp x8, .LCPI45_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q2, q1, [x0, #64]
+; CHECK-NEXT:    ldr q26, [x8, :lo12:.LCPI45_0]
+; CHECK-NEXT:    ldp q30, q29, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z26.s
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z26.s
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z3.s, z2.s[1]
+; CHECK-NEXT:    fmov w8, s3
+; CHECK-NEXT:    mov z6.s, z2.s[2]
+; CHECK-NEXT:    mov z3.s, z2.s[3]
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z31.s, z1.s[2]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s6
+; CHECK-NEXT:    mov z3.s, z1.s[1]
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    mov z8.s, z1.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z30.s, z26.s
+; CHECK-NEXT:    bfi w9, w8, #18, #1
+; CHECK-NEXT:    fmov w8, s3
+; CHECK-NEXT:    bfi w9, w10, #19, #1
+; CHECK-NEXT:    fmov w10, s31
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    mov z30.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    mov z31.s, z30.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z9.s, z30.s[3]
+; CHECK-NEXT:    ldp q28, q27, [x0]
+; CHECK-NEXT:    orr w8, w9, w10, lsl #22
+; CHECK-NEXT:    fmov w9, s8
+; CHECK-NEXT:    fmov w10, s30
+; CHECK-NEXT:    mov z8.s, z30.s[2]
+; CHECK-NEXT:    fmov w11, s9
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z29.s, z26.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z29.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z28.s, z26.s
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s31
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s8
+; CHECK-NEXT:    mov z8.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z27.s, z26.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z9.s, z8.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z10.s, z8.s[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s8
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s9
+; CHECK-NEXT:    mov z9.s, z8.s[2]
+; CHECK-NEXT:    fmov w12, s10
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z27.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s9
+; CHECK-NEXT:    fmov w13, s27
+; CHECK-NEXT:    mov z30.s, z29.s[1]
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    mov z8.s, z27.s[2]
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z31.s, z29.s[2]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z28.s, z29.s[3]
+; CHECK-NEXT:    ldp q4, q0, [x0, #32]
+; CHECK-NEXT:    mov z29.s, z27.s[1]
+; CHECK-NEXT:    fmov w9, s30
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    fmov w13, s8
+; CHECK-NEXT:    fmov w12, s29
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z9.s, z27.s[3]
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z26.s
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    orr w9, w10, w13, lsl #6
+; CHECK-NEXT:    fmov w10, s9
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    mov z27.s, z4.s[1]
+; CHECK-NEXT:    fmov w12, s27
+; CHECK-NEXT:    mov z29.s, z4.s[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z30.s, z4.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z26.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w12, s0
+; CHECK-NEXT:    mov z4.s, z0.s[1]
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s30
+; CHECK-NEXT:    mov z26.s, z0.s[2]
+; CHECK-NEXT:    mov z27.s, z0.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w12, s31
+; CHECK-NEXT:    ldp q25, q24, [x1]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    ldp q2, q23, [x1, #32]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s28
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    ldp q5, q22, [x1, #64]
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q7, q21, [x1, #96]
+; CHECK-NEXT:    ldp q18, q20, [x1, #128]
+; CHECK-NEXT:    ldp q19, q17, [x1, #160]
+; CHECK-NEXT:    ldp q16, q6, [x1, #192]
+; CHECK-NEXT:    ldp q3, q1, [x1, #224]
+; CHECK-NEXT:    stp x2, x2, [sp]
+; CHECK-NEXT:    ldr q4, [sp]
+; CHECK-NEXT:    add z26.d, z25.d, z4.d
+; CHECK-NEXT:    tbz w8, #0, .LBB45_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    add z25.d, z24.d, z4.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB45_3
+; CHECK-NEXT:    b .LBB45_4
+; CHECK-NEXT:  .LBB45_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:    add z25.d, z24.d, z4.d
+; CHECK-NEXT:    tbz w8, #1, .LBB45_4
+; CHECK-NEXT:  .LBB45_3: // %cond.load1
+; CHECK-NEXT:    mov z24.d, z26.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB45_4: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB45_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    add z24.d, z2.d, z4.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB45_9
+; CHECK-NEXT:  .LBB45_6: // %else8
+; CHECK-NEXT:    tbz w8, #4, .LBB45_10
+; CHECK-NEXT:  .LBB45_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    add z23.d, z23.d, z4.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB45_11
+; CHECK-NEXT:    b .LBB45_12
+; CHECK-NEXT:  .LBB45_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    add z24.d, z2.d, z4.d
+; CHECK-NEXT:    tbz w8, #3, .LBB45_6
+; CHECK-NEXT:  .LBB45_9: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #4, .LBB45_7
+; CHECK-NEXT:  .LBB45_10:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:    add z23.d, z23.d, z4.d
+; CHECK-NEXT:    tbz w8, #5, .LBB45_12
+; CHECK-NEXT:  .LBB45_11: // %cond.load13
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB45_12: // %else14
+; CHECK-NEXT:    tbnz w8, #6, .LBB45_16
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    add z24.d, z5.d, z4.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB45_17
+; CHECK-NEXT:  .LBB45_14: // %else20
+; CHECK-NEXT:    tbz w8, #8, .LBB45_18
+; CHECK-NEXT:  .LBB45_15: // %cond.load22
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v5.s }[0], [x9]
+; CHECK-NEXT:    add z23.d, z22.d, z4.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB45_19
+; CHECK-NEXT:    b .LBB45_20
+; CHECK-NEXT:  .LBB45_16: // %cond.load16
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    add z24.d, z5.d, z4.d
+; CHECK-NEXT:    tbz w8, #7, .LBB45_14
+; CHECK-NEXT:  .LBB45_17: // %cond.load19
+; CHECK-NEXT:    mov z5.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #8, .LBB45_15
+; CHECK-NEXT:  .LBB45_18:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:    add z23.d, z22.d, z4.d
+; CHECK-NEXT:    tbz w8, #9, .LBB45_20
+; CHECK-NEXT:  .LBB45_19: // %cond.load25
+; CHECK-NEXT:    mov z22.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v5.s }[1], [x9]
+; CHECK-NEXT:  .LBB45_20: // %else26
+; CHECK-NEXT:    tbnz w8, #10, .LBB45_24
+; CHECK-NEXT:  // %bb.21: // %else29
+; CHECK-NEXT:    add z22.d, z7.d, z4.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB45_25
+; CHECK-NEXT:  .LBB45_22: // %else32
+; CHECK-NEXT:    tbz w8, #12, .LBB45_26
+; CHECK-NEXT:  .LBB45_23: // %cond.load34
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    add z21.d, z21.d, z4.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB45_27
+; CHECK-NEXT:    b .LBB45_28
+; CHECK-NEXT:  .LBB45_24: // %cond.load28
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v5.s }[2], [x9]
+; CHECK-NEXT:    add z22.d, z7.d, z4.d
+; CHECK-NEXT:    tbz w8, #11, .LBB45_22
+; CHECK-NEXT:  .LBB45_25: // %cond.load31
+; CHECK-NEXT:    mov z7.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v5.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #12, .LBB45_23
+; CHECK-NEXT:  .LBB45_26:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:    add z21.d, z21.d, z4.d
+; CHECK-NEXT:    tbz w8, #13, .LBB45_28
+; CHECK-NEXT:  .LBB45_27: // %cond.load37
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:  .LBB45_28: // %else38
+; CHECK-NEXT:    tbnz w8, #14, .LBB45_32
+; CHECK-NEXT:  // %bb.29: // %else41
+; CHECK-NEXT:    add z22.d, z18.d, z4.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB45_33
+; CHECK-NEXT:  .LBB45_30: // %else44
+; CHECK-NEXT:    tbz w8, #16, .LBB45_34
+; CHECK-NEXT:  .LBB45_31: // %cond.load46
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v18.s }[0], [x9]
+; CHECK-NEXT:    add z21.d, z20.d, z4.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB45_35
+; CHECK-NEXT:    b .LBB45_36
+; CHECK-NEXT:  .LBB45_32: // %cond.load40
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:    add z22.d, z18.d, z4.d
+; CHECK-NEXT:    tbz w8, #15, .LBB45_30
+; CHECK-NEXT:  .LBB45_33: // %cond.load43
+; CHECK-NEXT:    mov z18.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v7.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #16, .LBB45_31
+; CHECK-NEXT:  .LBB45_34:
+; CHECK-NEXT:    // implicit-def: $q18
+; CHECK-NEXT:    add z21.d, z20.d, z4.d
+; CHECK-NEXT:    tbz w8, #17, .LBB45_36
+; CHECK-NEXT:  .LBB45_35: // %cond.load49
+; CHECK-NEXT:    mov z20.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v18.s }[1], [x9]
+; CHECK-NEXT:  .LBB45_36: // %else50
+; CHECK-NEXT:    tbnz w8, #18, .LBB45_40
+; CHECK-NEXT:  // %bb.37: // %else53
+; CHECK-NEXT:    add z20.d, z19.d, z4.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB45_41
+; CHECK-NEXT:  .LBB45_38: // %else56
+; CHECK-NEXT:    tbz w8, #20, .LBB45_42
+; CHECK-NEXT:  .LBB45_39: // %cond.load58
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.s }[0], [x9]
+; CHECK-NEXT:    add z17.d, z17.d, z4.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB45_43
+; CHECK-NEXT:    b .LBB45_44
+; CHECK-NEXT:  .LBB45_40: // %cond.load52
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v18.s }[2], [x9]
+; CHECK-NEXT:    add z20.d, z19.d, z4.d
+; CHECK-NEXT:    tbz w8, #19, .LBB45_38
+; CHECK-NEXT:  .LBB45_41: // %cond.load55
+; CHECK-NEXT:    mov z19.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v18.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #20, .LBB45_39
+; CHECK-NEXT:  .LBB45_42:
+; CHECK-NEXT:    // implicit-def: $q19
+; CHECK-NEXT:    add z17.d, z17.d, z4.d
+; CHECK-NEXT:    tbz w8, #21, .LBB45_44
+; CHECK-NEXT:  .LBB45_43: // %cond.load61
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v19.s }[1], [x9]
+; CHECK-NEXT:  .LBB45_44: // %else62
+; CHECK-NEXT:    tbnz w8, #22, .LBB45_48
+; CHECK-NEXT:  // %bb.45: // %else65
+; CHECK-NEXT:    add z20.d, z16.d, z4.d
+; CHECK-NEXT:    tbnz w8, #23, .LBB45_49
+; CHECK-NEXT:  .LBB45_46: // %else68
+; CHECK-NEXT:    tbz w8, #24, .LBB45_50
+; CHECK-NEXT:  .LBB45_47: // %cond.load70
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v16.s }[0], [x9]
+; CHECK-NEXT:    add z17.d, z6.d, z4.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB45_51
+; CHECK-NEXT:    b .LBB45_52
+; CHECK-NEXT:  .LBB45_48: // %cond.load64
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v19.s }[2], [x9]
+; CHECK-NEXT:    add z20.d, z16.d, z4.d
+; CHECK-NEXT:    tbz w8, #23, .LBB45_46
+; CHECK-NEXT:  .LBB45_49: // %cond.load67
+; CHECK-NEXT:    mov z16.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v19.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #24, .LBB45_47
+; CHECK-NEXT:  .LBB45_50:
+; CHECK-NEXT:    // implicit-def: $q16
+; CHECK-NEXT:    add z17.d, z6.d, z4.d
+; CHECK-NEXT:    tbz w8, #25, .LBB45_52
+; CHECK-NEXT:  .LBB45_51: // %cond.load73
+; CHECK-NEXT:    mov z6.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v16.s }[1], [x9]
+; CHECK-NEXT:  .LBB45_52: // %else74
+; CHECK-NEXT:    tbnz w8, #26, .LBB45_56
+; CHECK-NEXT:  // %bb.53: // %else77
+; CHECK-NEXT:    add z6.d, z3.d, z4.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB45_57
+; CHECK-NEXT:  .LBB45_54: // %else80
+; CHECK-NEXT:    tbz w8, #28, .LBB45_58
+; CHECK-NEXT:  .LBB45_55: // %cond.load82
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v3.s }[0], [x9]
+; CHECK-NEXT:    add z1.d, z1.d, z4.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB45_59
+; CHECK-NEXT:    b .LBB45_60
+; CHECK-NEXT:  .LBB45_56: // %cond.load76
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v16.s }[2], [x9]
+; CHECK-NEXT:    add z6.d, z3.d, z4.d
+; CHECK-NEXT:    tbz w8, #27, .LBB45_54
+; CHECK-NEXT:  .LBB45_57: // %cond.load79
+; CHECK-NEXT:    mov z3.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v16.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #28, .LBB45_55
+; CHECK-NEXT:  .LBB45_58:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:    add z1.d, z1.d, z4.d
+; CHECK-NEXT:    tbz w8, #29, .LBB45_60
+; CHECK-NEXT:  .LBB45_59: // %cond.load85
+; CHECK-NEXT:    mov z4.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:  .LBB45_60: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB45_64
+; CHECK-NEXT:  // %bb.61: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB45_63
+; CHECK-NEXT:  .LBB45_62: // %cond.load91
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v3.s }[3], [x8]
+; CHECK-NEXT:  .LBB45_63: // %else92
+; CHECK-NEXT:    ldp d9, d8, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    stp q0, q2, [x0]
+; CHECK-NEXT:    ldr d10, [sp, #16] // 8-byte Folded Reload
+; CHECK-NEXT:    stp q5, q7, [x0, #32]
+; CHECK-NEXT:    stp q18, q19, [x0, #64]
+; CHECK-NEXT:    stp q16, q3, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #48
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB45_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB45_62
+; CHECK-NEXT:    b .LBB45_63
   %cvals = load <32 x float>, ptr %a
   %bases = load <32 x ptr>, ptr %b
   %byte_ptrs = getelementptr i8, <32 x ptr> %bases, i64 %off
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
   %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
   store <32 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_vec_plus_imm(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_vec_plus_imm:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d, #4]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI46_1
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q6, q1, [x0, #64]
+; CHECK-NEXT:    adrp x12, .LCPI46_0
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI46_1]
+; CHECK-NEXT:    ldp q5, q4, [x0]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z6.s, z3.s
+; CHECK-NEXT:    mov z7.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z3.s
+; CHECK-NEXT:    fmov w8, s7
+; CHECK-NEXT:    mov z17.s, z7.s[1]
+; CHECK-NEXT:    mov z18.s, z7.s[2]
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z19.s, z7.s[3]
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s19
+; CHECK-NEXT:    ldp q16, q6, [x0, #96]
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z7.s, z1.s[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    mov z17.s, z1.s[2]
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    mov z18.s, z1.s[3]
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z16.s, z3.s
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z7.s, z1.s[1]
+; CHECK-NEXT:    mov z16.s, z1.s[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z17.s, z1.s[3]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #23
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z6.s, z3.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z6.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z5.s, z3.s
+; CHECK-NEXT:    mov z17.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z3.s
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    mov z18.s, z17.s[1]
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    mov z18.s, z17.s[2]
+; CHECK-NEXT:    mov z19.s, z17.s[3]
+; CHECK-NEXT:    ldr q1, [x12, :lo12:.LCPI46_0]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s19
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z7.s, z6.s[1]
+; CHECK-NEXT:    fmov w13, s4
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z17.s, z4.s[2]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z16.s, z6.s[2]
+; CHECK-NEXT:    ldp q2, q0, [x0, #32]
+; CHECK-NEXT:    mov z5.s, z6.s[3]
+; CHECK-NEXT:    mov z6.s, z4.s[1]
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    fmov w13, s17
+; CHECK-NEXT:    fmov w12, s6
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z18.s, z4.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z3.s
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    orr w9, w10, w13, lsl #6
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z3.s
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s4
+; CHECK-NEXT:    mov z6.s, z2.s[2]
+; CHECK-NEXT:    mov z7.s, z2.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov w12, s0
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    mov z3.s, z0.s[2]
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    mov z4.s, z0.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    ldp q0, q2, [x1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    add z3.d, z0.d, z1.d
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbz w8, #0, .LBB46_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ldr s0, [x9]
+; CHECK-NEXT:    b .LBB46_3
+; CHECK-NEXT:  .LBB46_2:
+; CHECK-NEXT:    // implicit-def: $q0
+; CHECK-NEXT:  .LBB46_3: // %else
+; CHECK-NEXT:    ldr q4, [x1, #32]
+; CHECK-NEXT:    add z2.d, z2.d, z1.d
+; CHECK-NEXT:    tbz w8, #1, .LBB46_5
+; CHECK-NEXT:  // %bb.4: // %cond.load1
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:  .LBB46_5: // %else2
+; CHECK-NEXT:    tbz w8, #2, .LBB46_7
+; CHECK-NEXT:  // %bb.6: // %cond.load4
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:  .LBB46_7: // %else5
+; CHECK-NEXT:    ldr q3, [x1, #48]
+; CHECK-NEXT:    add z4.d, z4.d, z1.d
+; CHECK-NEXT:    tbz w8, #3, .LBB46_9
+; CHECK-NEXT:  // %bb.8: // %cond.load7
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:  .LBB46_9: // %else8
+; CHECK-NEXT:    tbz w8, #4, .LBB46_11
+; CHECK-NEXT:  // %bb.10: // %cond.load10
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    b .LBB46_12
+; CHECK-NEXT:  .LBB46_11:
+; CHECK-NEXT:    // implicit-def: $q2
+; CHECK-NEXT:  .LBB46_12: // %else11
+; CHECK-NEXT:    ldr q5, [x1, #64]
+; CHECK-NEXT:    add z3.d, z3.d, z1.d
+; CHECK-NEXT:    tbz w8, #5, .LBB46_14
+; CHECK-NEXT:  // %bb.13: // %cond.load13
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB46_14: // %else14
+; CHECK-NEXT:    tbz w8, #6, .LBB46_16
+; CHECK-NEXT:  // %bb.15: // %cond.load16
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:  .LBB46_16: // %else17
+; CHECK-NEXT:    ldr q4, [x1, #80]
+; CHECK-NEXT:    add z5.d, z5.d, z1.d
+; CHECK-NEXT:    tbz w8, #7, .LBB46_18
+; CHECK-NEXT:  // %bb.17: // %cond.load19
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:  .LBB46_18: // %else20
+; CHECK-NEXT:    tbz w8, #8, .LBB46_20
+; CHECK-NEXT:  // %bb.19: // %cond.load22
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v3.s }[0], [x9]
+; CHECK-NEXT:    b .LBB46_21
+; CHECK-NEXT:  .LBB46_20:
+; CHECK-NEXT:    // implicit-def: $q3
+; CHECK-NEXT:  .LBB46_21: // %else23
+; CHECK-NEXT:    ldr q6, [x1, #96]
+; CHECK-NEXT:    add z4.d, z4.d, z1.d
+; CHECK-NEXT:    tbz w8, #9, .LBB46_23
+; CHECK-NEXT:  // %bb.22: // %cond.load25
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:  .LBB46_23: // %else26
+; CHECK-NEXT:    tbz w8, #10, .LBB46_25
+; CHECK-NEXT:  // %bb.24: // %cond.load28
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:  .LBB46_25: // %else29
+; CHECK-NEXT:    ldr q5, [x1, #112]
+; CHECK-NEXT:    add z6.d, z6.d, z1.d
+; CHECK-NEXT:    tbz w8, #11, .LBB46_27
+; CHECK-NEXT:  // %bb.26: // %cond.load31
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[3], [x9]
+; CHECK-NEXT:  .LBB46_27: // %else32
+; CHECK-NEXT:    tbz w8, #12, .LBB46_29
+; CHECK-NEXT:  // %bb.28: // %cond.load34
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v4.s }[0], [x9]
+; CHECK-NEXT:    b .LBB46_30
+; CHECK-NEXT:  .LBB46_29:
+; CHECK-NEXT:    // implicit-def: $q4
+; CHECK-NEXT:  .LBB46_30: // %else35
+; CHECK-NEXT:    ldr q7, [x1, #128]
+; CHECK-NEXT:    add z5.d, z5.d, z1.d
+; CHECK-NEXT:    tbz w8, #13, .LBB46_32
+; CHECK-NEXT:  // %bb.31: // %cond.load37
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v4.s }[1], [x9]
+; CHECK-NEXT:  .LBB46_32: // %else38
+; CHECK-NEXT:    tbz w8, #14, .LBB46_34
+; CHECK-NEXT:  // %bb.33: // %cond.load40
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[2], [x9]
+; CHECK-NEXT:  .LBB46_34: // %else41
+; CHECK-NEXT:    ldr q6, [x1, #144]
+; CHECK-NEXT:    add z7.d, z7.d, z1.d
+; CHECK-NEXT:    tbz w8, #15, .LBB46_36
+; CHECK-NEXT:  // %bb.35: // %cond.load43
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[3], [x9]
+; CHECK-NEXT:  .LBB46_36: // %else44
+; CHECK-NEXT:    tbz w8, #16, .LBB46_38
+; CHECK-NEXT:  // %bb.37: // %cond.load46
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v5.s }[0], [x9]
+; CHECK-NEXT:    b .LBB46_39
+; CHECK-NEXT:  .LBB46_38:
+; CHECK-NEXT:    // implicit-def: $q5
+; CHECK-NEXT:  .LBB46_39: // %else47
+; CHECK-NEXT:    ldr q16, [x1, #160]
+; CHECK-NEXT:    add z6.d, z6.d, z1.d
+; CHECK-NEXT:    tbz w8, #17, .LBB46_41
+; CHECK-NEXT:  // %bb.40: // %cond.load49
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v5.s }[1], [x9]
+; CHECK-NEXT:  .LBB46_41: // %else50
+; CHECK-NEXT:    tbz w8, #18, .LBB46_43
+; CHECK-NEXT:  // %bb.42: // %cond.load52
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[2], [x9]
+; CHECK-NEXT:  .LBB46_43: // %else53
+; CHECK-NEXT:    ldr q7, [x1, #176]
+; CHECK-NEXT:    add z16.d, z16.d, z1.d
+; CHECK-NEXT:    tbz w8, #19, .LBB46_45
+; CHECK-NEXT:  // %bb.44: // %cond.load55
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[3], [x9]
+; CHECK-NEXT:  .LBB46_45: // %else56
+; CHECK-NEXT:    tbz w8, #20, .LBB46_47
+; CHECK-NEXT:  // %bb.46: // %cond.load58
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v6.s }[0], [x9]
+; CHECK-NEXT:    b .LBB46_48
+; CHECK-NEXT:  .LBB46_47:
+; CHECK-NEXT:    // implicit-def: $q6
+; CHECK-NEXT:  .LBB46_48: // %else59
+; CHECK-NEXT:    ldr q17, [x1, #192]
+; CHECK-NEXT:    add z7.d, z7.d, z1.d
+; CHECK-NEXT:    tbz w8, #21, .LBB46_50
+; CHECK-NEXT:  // %bb.49: // %cond.load61
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v6.s }[1], [x9]
+; CHECK-NEXT:  .LBB46_50: // %else62
+; CHECK-NEXT:    tbz w8, #22, .LBB46_52
+; CHECK-NEXT:  // %bb.51: // %cond.load64
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[2], [x9]
+; CHECK-NEXT:  .LBB46_52: // %else65
+; CHECK-NEXT:    ldr q16, [x1, #208]
+; CHECK-NEXT:    add z18.d, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #23, .LBB46_54
+; CHECK-NEXT:  // %bb.53: // %cond.load67
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[3], [x9]
+; CHECK-NEXT:  .LBB46_54: // %else68
+; CHECK-NEXT:    tbz w8, #24, .LBB46_56
+; CHECK-NEXT:  // %bb.55: // %cond.load70
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    b .LBB46_57
+; CHECK-NEXT:  .LBB46_56:
+; CHECK-NEXT:    // implicit-def: $q7
+; CHECK-NEXT:  .LBB46_57: // %else71
+; CHECK-NEXT:    ldr q17, [x1, #224]
+; CHECK-NEXT:    add z16.d, z16.d, z1.d
+; CHECK-NEXT:    tbz w8, #25, .LBB46_59
+; CHECK-NEXT:  // %bb.58: // %cond.load73
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:  .LBB46_59: // %else74
+; CHECK-NEXT:    tbz w8, #26, .LBB46_61
+; CHECK-NEXT:  // %bb.60: // %cond.load76
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:  .LBB46_61: // %else77
+; CHECK-NEXT:    ldr q18, [x1, #240]
+; CHECK-NEXT:    add z17.d, z17.d, z1.d
+; CHECK-NEXT:    tbz w8, #27, .LBB46_63
+; CHECK-NEXT:  // %bb.62: // %cond.load79
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[3], [x9]
+; CHECK-NEXT:  .LBB46_63: // %else80
+; CHECK-NEXT:    tbz w8, #28, .LBB46_65
+; CHECK-NEXT:  // %bb.64: // %cond.load82
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v16.s }[0], [x9]
+; CHECK-NEXT:    add z1.d, z18.d, z1.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB46_66
+; CHECK-NEXT:    b .LBB46_67
+; CHECK-NEXT:  .LBB46_65:
+; CHECK-NEXT:    // implicit-def: $q16
+; CHECK-NEXT:    add z1.d, z18.d, z1.d
+; CHECK-NEXT:    tbz w8, #29, .LBB46_67
+; CHECK-NEXT:  .LBB46_66: // %cond.load85
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    ld1 { v16.s }[1], [x9]
+; CHECK-NEXT:  .LBB46_67: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB46_71
+; CHECK-NEXT:  // %bb.68: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB46_70
+; CHECK-NEXT:  .LBB46_69: // %cond.load91
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v16.s }[3], [x8]
+; CHECK-NEXT:  .LBB46_70: // %else92
+; CHECK-NEXT:    stp q0, q2, [x0]
+; CHECK-NEXT:    stp q3, q4, [x0, #32]
+; CHECK-NEXT:    stp q5, q6, [x0, #64]
+; CHECK-NEXT:    stp q7, q16, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB46_71: // %cond.load88
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v16.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB46_69
+; CHECK-NEXT:    b .LBB46_70
   %cvals = load <32 x float>, ptr %a
   %bases = load <32 x ptr>, ptr %b
   %byte_ptrs = getelementptr i8, <32 x ptr> %bases, i64 4
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
   %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x float> undef)
   store <32 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_passthru(ptr %a, ptr %b, ptr %c) #0 {
 ; CHECK-LABEL: masked_gather_passthru:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x2]
-; CHECK-NEXT:    punpklo p2.h, p1.b
-; CHECK-NEXT:    ld1w { z1.d }, p2/z, [z1.d]
-; CHECK-NEXT:    uzp1 z1.s, z1.s, z1.s
-; CHECK-NEXT:    mov z0.s, p1/m, z1.s
-; CHECK-NEXT:    st1w { z0.s }, p0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI47_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q1, q2, [x0, #64]
+; CHECK-NEXT:    ldr q0, [x8, :lo12:.LCPI47_0]
+; CHECK-NEXT:    ldp q4, q3, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z5.s, z1.s[1]
+; CHECK-NEXT:    mov z6.s, z1.s[2]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    mov z7.s, z1.s[3]
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z5.s, z2.s[2]
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    mov z1.s, z2.s[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z0.s
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    mov z5.s, z4.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    mov z6.s, z4.s[2]
+; CHECK-NEXT:    mov z7.s, z4.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z3.s, z0.s
+; CHECK-NEXT:    ldp q16, q4, [x0]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    mov z5.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z16.s, z0.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    mov z17.s, z16.s[1]
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z17.s, z16.s[2]
+; CHECK-NEXT:    mov z18.s, z16.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z0.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.s, z5.s[1]
+; CHECK-NEXT:    fmov w13, s4
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z16.s, z4.s[2]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z7.s, z5.s[2]
+; CHECK-NEXT:    ldp q2, q1, [x0, #32]
+; CHECK-NEXT:    mov z3.s, z5.s[3]
+; CHECK-NEXT:    mov z5.s, z4.s[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    fmov w13, s16
+; CHECK-NEXT:    fmov w12, s5
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z17.s, z4.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    orr w9, w10, w13, lsl #6
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w12, s4
+; CHECK-NEXT:    mov z5.s, z2.s[2]
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    mov z0.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov w12, s0
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z4.s, z0.s[3]
+; CHECK-NEXT:    ldr q0, [x2]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s2
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbnz w8, #0, .LBB47_41
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB47_42
+; CHECK-NEXT:  .LBB47_2: // %else2
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbnz w8, #2, .LBB47_43
+; CHECK-NEXT:  .LBB47_3: // %else5
+; CHECK-NEXT:    tbz w8, #3, .LBB47_5
+; CHECK-NEXT:  .LBB47_4: // %cond.load7
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x9]
+; CHECK-NEXT:  .LBB47_5: // %else8
+; CHECK-NEXT:    ldr q2, [x1, #32]
+; CHECK-NEXT:    ldr q1, [x2, #16]
+; CHECK-NEXT:    tbnz w8, #4, .LBB47_44
+; CHECK-NEXT:  // %bb.6: // %else11
+; CHECK-NEXT:    tbnz w8, #5, .LBB47_45
+; CHECK-NEXT:  .LBB47_7: // %else14
+; CHECK-NEXT:    ldr q2, [x1, #48]
+; CHECK-NEXT:    tbnz w8, #6, .LBB47_46
+; CHECK-NEXT:  .LBB47_8: // %else17
+; CHECK-NEXT:    tbz w8, #7, .LBB47_10
+; CHECK-NEXT:  .LBB47_9: // %cond.load19
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[3], [x9]
+; CHECK-NEXT:  .LBB47_10: // %else20
+; CHECK-NEXT:    ldr q3, [x1, #64]
+; CHECK-NEXT:    ldr q2, [x2, #32]
+; CHECK-NEXT:    tbnz w8, #8, .LBB47_47
+; CHECK-NEXT:  // %bb.11: // %else23
+; CHECK-NEXT:    tbnz w8, #9, .LBB47_48
+; CHECK-NEXT:  .LBB47_12: // %else26
+; CHECK-NEXT:    ldr q3, [x1, #80]
+; CHECK-NEXT:    tbnz w8, #10, .LBB47_49
+; CHECK-NEXT:  .LBB47_13: // %else29
+; CHECK-NEXT:    tbz w8, #11, .LBB47_15
+; CHECK-NEXT:  .LBB47_14: // %cond.load31
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:  .LBB47_15: // %else32
+; CHECK-NEXT:    ldr q4, [x1, #96]
+; CHECK-NEXT:    ldr q3, [x2, #48]
+; CHECK-NEXT:    tbnz w8, #12, .LBB47_50
+; CHECK-NEXT:  // %bb.16: // %else35
+; CHECK-NEXT:    tbnz w8, #13, .LBB47_51
+; CHECK-NEXT:  .LBB47_17: // %else38
+; CHECK-NEXT:    ldr q4, [x1, #112]
+; CHECK-NEXT:    tbnz w8, #14, .LBB47_52
+; CHECK-NEXT:  .LBB47_18: // %else41
+; CHECK-NEXT:    tbz w8, #15, .LBB47_20
+; CHECK-NEXT:  .LBB47_19: // %cond.load43
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[3], [x9]
+; CHECK-NEXT:  .LBB47_20: // %else44
+; CHECK-NEXT:    ldr q5, [x1, #128]
+; CHECK-NEXT:    ldr q4, [x2, #64]
+; CHECK-NEXT:    tbnz w8, #16, .LBB47_53
+; CHECK-NEXT:  // %bb.21: // %else47
+; CHECK-NEXT:    tbnz w8, #17, .LBB47_54
+; CHECK-NEXT:  .LBB47_22: // %else50
+; CHECK-NEXT:    ldr q5, [x1, #144]
+; CHECK-NEXT:    tbnz w8, #18, .LBB47_55
+; CHECK-NEXT:  .LBB47_23: // %else53
+; CHECK-NEXT:    tbz w8, #19, .LBB47_25
+; CHECK-NEXT:  .LBB47_24: // %cond.load55
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[3], [x9]
+; CHECK-NEXT:  .LBB47_25: // %else56
+; CHECK-NEXT:    ldr q6, [x1, #160]
+; CHECK-NEXT:    ldr q5, [x2, #80]
+; CHECK-NEXT:    tbnz w8, #20, .LBB47_56
+; CHECK-NEXT:  // %bb.26: // %else59
+; CHECK-NEXT:    tbnz w8, #21, .LBB47_57
+; CHECK-NEXT:  .LBB47_27: // %else62
+; CHECK-NEXT:    ldr q6, [x1, #176]
+; CHECK-NEXT:    tbnz w8, #22, .LBB47_58
+; CHECK-NEXT:  .LBB47_28: // %else65
+; CHECK-NEXT:    tbz w8, #23, .LBB47_30
+; CHECK-NEXT:  .LBB47_29: // %cond.load67
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[3], [x9]
+; CHECK-NEXT:  .LBB47_30: // %else68
+; CHECK-NEXT:    ldr q7, [x1, #192]
+; CHECK-NEXT:    ldr q6, [x2, #96]
+; CHECK-NEXT:    tbnz w8, #24, .LBB47_59
+; CHECK-NEXT:  // %bb.31: // %else71
+; CHECK-NEXT:    tbnz w8, #25, .LBB47_60
+; CHECK-NEXT:  .LBB47_32: // %else74
+; CHECK-NEXT:    ldr q7, [x1, #208]
+; CHECK-NEXT:    tbnz w8, #26, .LBB47_61
+; CHECK-NEXT:  .LBB47_33: // %else77
+; CHECK-NEXT:    tbz w8, #27, .LBB47_35
+; CHECK-NEXT:  .LBB47_34: // %cond.load79
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[3], [x9]
+; CHECK-NEXT:  .LBB47_35: // %else80
+; CHECK-NEXT:    ldr q16, [x1, #224]
+; CHECK-NEXT:    ldr q7, [x2, #112]
+; CHECK-NEXT:    tbnz w8, #28, .LBB47_62
+; CHECK-NEXT:  // %bb.36: // %else83
+; CHECK-NEXT:    tbnz w8, #29, .LBB47_63
+; CHECK-NEXT:  .LBB47_37: // %else86
+; CHECK-NEXT:    ldr q16, [x1, #240]
+; CHECK-NEXT:    tbnz w8, #30, .LBB47_64
+; CHECK-NEXT:  .LBB47_38: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB47_40
+; CHECK-NEXT:  .LBB47_39: // %cond.load91
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x8, d16
+; CHECK-NEXT:    ld1 { v7.s }[3], [x8]
+; CHECK-NEXT:  .LBB47_40: // %else92
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    stp q2, q3, [x0, #32]
+; CHECK-NEXT:    stp q4, q5, [x0, #64]
+; CHECK-NEXT:    stp q6, q7, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB47_41: // %cond.load
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB47_2
+; CHECK-NEXT:  .LBB47_42: // %cond.load1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:    ldr q1, [x1, #16]
+; CHECK-NEXT:    tbz w8, #2, .LBB47_3
+; CHECK-NEXT:  .LBB47_43: // %cond.load4
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #3, .LBB47_4
+; CHECK-NEXT:    b .LBB47_5
+; CHECK-NEXT:  .LBB47_44: // %cond.load10
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB47_7
+; CHECK-NEXT:  .LBB47_45: // %cond.load13
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[1], [x9]
+; CHECK-NEXT:    ldr q2, [x1, #48]
+; CHECK-NEXT:    tbz w8, #6, .LBB47_8
+; CHECK-NEXT:  .LBB47_46: // %cond.load16
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    ld1 { v1.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB47_9
+; CHECK-NEXT:    b .LBB47_10
+; CHECK-NEXT:  .LBB47_47: // %cond.load22
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB47_12
+; CHECK-NEXT:  .LBB47_48: // %cond.load25
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:    ldr q3, [x1, #80]
+; CHECK-NEXT:    tbz w8, #10, .LBB47_13
+; CHECK-NEXT:  .LBB47_49: // %cond.load28
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #11, .LBB47_14
+; CHECK-NEXT:    b .LBB47_15
+; CHECK-NEXT:  .LBB47_50: // %cond.load34
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB47_17
+; CHECK-NEXT:  .LBB47_51: // %cond.load37
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:    ldr q4, [x1, #112]
+; CHECK-NEXT:    tbz w8, #14, .LBB47_18
+; CHECK-NEXT:  .LBB47_52: // %cond.load40
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #15, .LBB47_19
+; CHECK-NEXT:    b .LBB47_20
+; CHECK-NEXT:  .LBB47_53: // %cond.load46
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #17, .LBB47_22
+; CHECK-NEXT:  .LBB47_54: // %cond.load49
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[1], [x9]
+; CHECK-NEXT:    ldr q5, [x1, #144]
+; CHECK-NEXT:    tbz w8, #18, .LBB47_23
+; CHECK-NEXT:  .LBB47_55: // %cond.load52
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v4.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #19, .LBB47_24
+; CHECK-NEXT:    b .LBB47_25
+; CHECK-NEXT:  .LBB47_56: // %cond.load58
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #21, .LBB47_27
+; CHECK-NEXT:  .LBB47_57: // %cond.load61
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[1], [x9]
+; CHECK-NEXT:    ldr q6, [x1, #176]
+; CHECK-NEXT:    tbz w8, #22, .LBB47_28
+; CHECK-NEXT:  .LBB47_58: // %cond.load64
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v5.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #23, .LBB47_29
+; CHECK-NEXT:    b .LBB47_30
+; CHECK-NEXT:  .LBB47_59: // %cond.load70
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #25, .LBB47_32
+; CHECK-NEXT:  .LBB47_60: // %cond.load73
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[1], [x9]
+; CHECK-NEXT:    ldr q7, [x1, #208]
+; CHECK-NEXT:    tbz w8, #26, .LBB47_33
+; CHECK-NEXT:  .LBB47_61: // %cond.load76
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v6.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #27, .LBB47_34
+; CHECK-NEXT:    b .LBB47_35
+; CHECK-NEXT:  .LBB47_62: // %cond.load82
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #29, .LBB47_37
+; CHECK-NEXT:  .LBB47_63: // %cond.load85
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:    ldr q16, [x1, #240]
+; CHECK-NEXT:    tbz w8, #30, .LBB47_38
+; CHECK-NEXT:  .LBB47_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB47_39
+; CHECK-NEXT:    b .LBB47_40
   %cvals = load <32 x float>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %passthru = load <32 x float>, ptr %c
   %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
   %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x float> %passthru)
   store <32 x float> %vals, ptr %a
   ret void
 }
 
 define void @masked_gather_passthru_0(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_gather_passthru_0:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    ld1w { z0.d }, p0/z, [z1.d]
-; CHECK-NEXT:    st1w { z0.d }, p1, [x0]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #32
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    adrp x9, .LCPI48_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q1, q2, [x0, #64]
+; CHECK-NEXT:    ldr q0, [x9, :lo12:.LCPI48_0]
+; CHECK-NEXT:    ldp q5, q4, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z0.s
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z3.s, z1.s[1]
+; CHECK-NEXT:    mov z6.s, z1.s[2]
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    mov z7.s, z1.s[3]
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w10, #17, #1
+; CHECK-NEXT:    mov z3.s, z2.s[2]
+; CHECK-NEXT:    bfi w8, w11, #18, #1
+; CHECK-NEXT:    mov z1.s, z2.s[1]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    bfi w8, w12, #19, #1
+; CHECK-NEXT:    fmov w12, s3
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z5.s, z0.s
+; CHECK-NEXT:    bfi w8, w10, #20, #1
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    bfi w8, w11, #21, #1
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    mov z5.s, z1.s[1]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #22
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    mov z6.s, z1.s[2]
+; CHECK-NEXT:    mov z7.s, z1.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z0.s
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #23
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    orr w8, w8, w11, lsl #24
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    ldp q17, q16, [x0]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z4.s, z1.s[1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z5.s, z1.s[2]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #26
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z17.s, z0.s
+; CHECK-NEXT:    mov z6.s, z1.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldp q3, q2, [x0, #32]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    mov z5.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w11, lsl #29
+; CHECK-NEXT:    mov z7.s, z5.s[1]
+; CHECK-NEXT:    fmov w12, s5
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    mov z7.s, z5.s[2]
+; CHECK-NEXT:    mov z17.s, z5.s[3]
+; CHECK-NEXT:    fmov w13, s7
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z16.s, z0.s
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #30
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    mov z7.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w12, w11, #1, #1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    bfi w12, w13, #2, #1
+; CHECK-NEXT:    mov z16.s, z7.s[1]
+; CHECK-NEXT:    mov z17.s, z7.s[2]
+; CHECK-NEXT:    mov z18.s, z7.s[3]
+; CHECK-NEXT:    fmov w13, s16
+; CHECK-NEXT:    bfi w12, w10, #3, #1
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    bfi w12, w11, #4, #1
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z3.s, z0.s
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w12, w13, #5, #1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w13, s3
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z7.s, z3.s[1]
+; CHECK-NEXT:    orr w10, w12, w10, lsl #6
+; CHECK-NEXT:    mov z16.s, z3.s[2]
+; CHECK-NEXT:    and w12, w13, #0x1
+; CHECK-NEXT:    orr w10, w10, w11, lsl #7
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    mov z17.s, z3.s[3]
+; CHECK-NEXT:    orr w10, w10, w12, lsl #8
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z2.s, z0.s
+; CHECK-NEXT:    fmov w13, s6
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z2.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z3.s, z2.s[1]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #9
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    orr w10, w10, w12, lsl #10
+; CHECK-NEXT:    fmov w12, s2
+; CHECK-NEXT:    mov z6.s, z2.s[2]
+; CHECK-NEXT:    mov z7.s, z2.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w13, lsl #31
+; CHECK-NEXT:    ldp q4, q1, [x1, #224]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #11
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    fmov w12, s3
+; CHECK-NEXT:    orr w10, w10, w11, lsl #12
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w10, w10, w12, lsl #13
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    ldp q19, q5, [x1, #192]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #14
+; CHECK-NEXT:    orr w10, w10, w12, lsl #15
+; CHECK-NEXT:    and w10, w10, #0xffff
+; CHECK-NEXT:    orr w8, w10, w8
+; CHECK-NEXT:    ldp q21, q20, [x1, #160]
+; CHECK-NEXT:    ldp q23, q22, [x1, #128]
+; CHECK-NEXT:    ldp q25, q24, [x1, #96]
+; CHECK-NEXT:    ldp q27, q26, [x1, #64]
+; CHECK-NEXT:    ldp q29, q28, [x1, #32]
+; CHECK-NEXT:    ldp q7, q6, [x1]
+; CHECK-NEXT:    tbz w8, #0, .LBB48_2
+; CHECK-NEXT:  // %bb.1: // %cond.load
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    ldr q3, [x9, :lo12:.LCPI48_0]
+; CHECK-NEXT:    ldr s0, [x10]
+; CHECK-NEXT:    str wzr, [sp, #12]
+; CHECK-NEXT:    stp wzr, wzr, [sp, #4]
+; CHECK-NEXT:    str s0, [sp]
+; CHECK-NEXT:    ldr q2, [sp]
+; CHECK-NEXT:    mov v0.16b, v3.16b
+; CHECK-NEXT:    tbnz w8, #1, .LBB48_3
+; CHECK-NEXT:    b .LBB48_4
+; CHECK-NEXT:  .LBB48_2:
+; CHECK-NEXT:    mov v2.16b, v0.16b
+; CHECK-NEXT:    mov v3.16b, v0.16b
+; CHECK-NEXT:    tbz w8, #1, .LBB48_4
+; CHECK-NEXT:  .LBB48_3: // %cond.load1
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    ld1 { v2.s }[1], [x9]
+; CHECK-NEXT:  .LBB48_4: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB48_8
+; CHECK-NEXT:  // %bb.5: // %else5
+; CHECK-NEXT:    tbnz w8, #3, .LBB48_9
+; CHECK-NEXT:  .LBB48_6: // %else8
+; CHECK-NEXT:    tbz w8, #4, .LBB48_10
+; CHECK-NEXT:  .LBB48_7: // %cond.load10
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    mov v30.16b, v3.16b
+; CHECK-NEXT:    mov v6.16b, v3.16b
+; CHECK-NEXT:    ld1 { v30.s }[0], [x9]
+; CHECK-NEXT:    mov v7.16b, v3.16b
+; CHECK-NEXT:    mov v16.16b, v3.16b
+; CHECK-NEXT:    mov v17.16b, v3.16b
+; CHECK-NEXT:    mov v18.16b, v3.16b
+; CHECK-NEXT:    mov v3.16b, v30.16b
+; CHECK-NEXT:    tbnz w8, #5, .LBB48_11
+; CHECK-NEXT:    b .LBB48_12
+; CHECK-NEXT:  .LBB48_8: // %cond.load4
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v2.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB48_6
+; CHECK-NEXT:  .LBB48_9: // %cond.load7
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    ld1 { v2.s }[3], [x9]
+; CHECK-NEXT:    tbnz w8, #4, .LBB48_7
+; CHECK-NEXT:  .LBB48_10:
+; CHECK-NEXT:    mov v6.16b, v3.16b
+; CHECK-NEXT:    mov v7.16b, v3.16b
+; CHECK-NEXT:    mov v16.16b, v3.16b
+; CHECK-NEXT:    mov v17.16b, v3.16b
+; CHECK-NEXT:    mov v18.16b, v3.16b
+; CHECK-NEXT:    tbz w8, #5, .LBB48_12
+; CHECK-NEXT:  .LBB48_11: // %cond.load13
+; CHECK-NEXT:    mov z29.d, z29.d[1]
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    ld1 { v3.s }[1], [x9]
+; CHECK-NEXT:  .LBB48_12: // %else14
+; CHECK-NEXT:    tbnz w8, #6, .LBB48_40
+; CHECK-NEXT:  // %bb.13: // %else17
+; CHECK-NEXT:    tbnz w8, #7, .LBB48_41
+; CHECK-NEXT:  .LBB48_14: // %else20
+; CHECK-NEXT:    tbnz w8, #8, .LBB48_42
+; CHECK-NEXT:  .LBB48_15: // %else23
+; CHECK-NEXT:    tbnz w8, #9, .LBB48_43
+; CHECK-NEXT:  .LBB48_16: // %else26
+; CHECK-NEXT:    tbnz w8, #10, .LBB48_44
+; CHECK-NEXT:  .LBB48_17: // %else29
+; CHECK-NEXT:    tbnz w8, #11, .LBB48_45
+; CHECK-NEXT:  .LBB48_18: // %else32
+; CHECK-NEXT:    tbnz w8, #12, .LBB48_46
+; CHECK-NEXT:  .LBB48_19: // %else35
+; CHECK-NEXT:    tbnz w8, #13, .LBB48_47
+; CHECK-NEXT:  .LBB48_20: // %else38
+; CHECK-NEXT:    tbnz w8, #14, .LBB48_48
+; CHECK-NEXT:  .LBB48_21: // %else41
+; CHECK-NEXT:    tbnz w8, #15, .LBB48_49
+; CHECK-NEXT:  .LBB48_22: // %else44
+; CHECK-NEXT:    tbnz w8, #16, .LBB48_50
+; CHECK-NEXT:  .LBB48_23: // %else47
+; CHECK-NEXT:    tbnz w8, #17, .LBB48_51
+; CHECK-NEXT:  .LBB48_24: // %else50
+; CHECK-NEXT:    tbnz w8, #18, .LBB48_52
+; CHECK-NEXT:  .LBB48_25: // %else53
+; CHECK-NEXT:    tbnz w8, #19, .LBB48_53
+; CHECK-NEXT:  .LBB48_26: // %else56
+; CHECK-NEXT:    tbnz w8, #20, .LBB48_54
+; CHECK-NEXT:  .LBB48_27: // %else59
+; CHECK-NEXT:    tbnz w8, #21, .LBB48_55
+; CHECK-NEXT:  .LBB48_28: // %else62
+; CHECK-NEXT:    tbnz w8, #22, .LBB48_56
+; CHECK-NEXT:  .LBB48_29: // %else65
+; CHECK-NEXT:    tbnz w8, #23, .LBB48_57
+; CHECK-NEXT:  .LBB48_30: // %else68
+; CHECK-NEXT:    tbnz w8, #24, .LBB48_58
+; CHECK-NEXT:  .LBB48_31: // %else71
+; CHECK-NEXT:    tbnz w8, #25, .LBB48_59
+; CHECK-NEXT:  .LBB48_32: // %else74
+; CHECK-NEXT:    tbnz w8, #26, .LBB48_60
+; CHECK-NEXT:  .LBB48_33: // %else77
+; CHECK-NEXT:    tbnz w8, #27, .LBB48_61
+; CHECK-NEXT:  .LBB48_34: // %else80
+; CHECK-NEXT:    tbnz w8, #28, .LBB48_62
+; CHECK-NEXT:  .LBB48_35: // %else83
+; CHECK-NEXT:    tbnz w8, #29, .LBB48_63
+; CHECK-NEXT:  .LBB48_36: // %else86
+; CHECK-NEXT:    tbnz w8, #30, .LBB48_64
+; CHECK-NEXT:  .LBB48_37: // %else89
+; CHECK-NEXT:    tbz w8, #31, .LBB48_39
+; CHECK-NEXT:  .LBB48_38: // %cond.load91
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    ld1 { v0.s }[3], [x8]
+; CHECK-NEXT:  .LBB48_39: // %else92
+; CHECK-NEXT:    stp q2, q3, [x0]
+; CHECK-NEXT:    stp q6, q7, [x0, #32]
+; CHECK-NEXT:    stp q16, q17, [x0, #64]
+; CHECK-NEXT:    stp q18, q0, [x0, #96]
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB48_40: // %cond.load16
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    ld1 { v3.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB48_14
+; CHECK-NEXT:  .LBB48_41: // %cond.load19
+; CHECK-NEXT:    mov z28.d, z28.d[1]
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    ld1 { v3.s }[3], [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB48_15
+; CHECK-NEXT:  .LBB48_42: // %cond.load22
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    ld1 { v6.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB48_16
+; CHECK-NEXT:  .LBB48_43: // %cond.load25
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    ld1 { v6.s }[1], [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB48_17
+; CHECK-NEXT:  .LBB48_44: // %cond.load28
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    ld1 { v6.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB48_18
+; CHECK-NEXT:  .LBB48_45: // %cond.load31
+; CHECK-NEXT:    mov z26.d, z26.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    ld1 { v6.s }[3], [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB48_19
+; CHECK-NEXT:  .LBB48_46: // %cond.load34
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v7.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB48_20
+; CHECK-NEXT:  .LBB48_47: // %cond.load37
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    ld1 { v7.s }[1], [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB48_21
+; CHECK-NEXT:  .LBB48_48: // %cond.load40
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v7.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB48_22
+; CHECK-NEXT:  .LBB48_49: // %cond.load43
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    ld1 { v7.s }[3], [x9]
+; CHECK-NEXT:    tbz w8, #16, .LBB48_23
+; CHECK-NEXT:  .LBB48_50: // %cond.load46
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v16.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #17, .LBB48_24
+; CHECK-NEXT:  .LBB48_51: // %cond.load49
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    ld1 { v16.s }[1], [x9]
+; CHECK-NEXT:    tbz w8, #18, .LBB48_25
+; CHECK-NEXT:  .LBB48_52: // %cond.load52
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v16.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB48_26
+; CHECK-NEXT:  .LBB48_53: // %cond.load55
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    ld1 { v16.s }[3], [x9]
+; CHECK-NEXT:    tbz w8, #20, .LBB48_27
+; CHECK-NEXT:  .LBB48_54: // %cond.load58
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v17.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #21, .LBB48_28
+; CHECK-NEXT:  .LBB48_55: // %cond.load61
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    ld1 { v17.s }[1], [x9]
+; CHECK-NEXT:    tbz w8, #22, .LBB48_29
+; CHECK-NEXT:  .LBB48_56: // %cond.load64
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v17.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB48_30
+; CHECK-NEXT:  .LBB48_57: // %cond.load67
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    ld1 { v17.s }[3], [x9]
+; CHECK-NEXT:    tbz w8, #24, .LBB48_31
+; CHECK-NEXT:  .LBB48_58: // %cond.load70
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v18.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #25, .LBB48_32
+; CHECK-NEXT:  .LBB48_59: // %cond.load73
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    ld1 { v18.s }[1], [x9]
+; CHECK-NEXT:    tbz w8, #26, .LBB48_33
+; CHECK-NEXT:  .LBB48_60: // %cond.load76
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v18.s }[2], [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB48_34
+; CHECK-NEXT:  .LBB48_61: // %cond.load79
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    ld1 { v18.s }[3], [x9]
+; CHECK-NEXT:    tbz w8, #28, .LBB48_35
+; CHECK-NEXT:  .LBB48_62: // %cond.load82
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v0.s }[0], [x9]
+; CHECK-NEXT:    tbz w8, #29, .LBB48_36
+; CHECK-NEXT:  .LBB48_63: // %cond.load85
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    ld1 { v0.s }[1], [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB48_37
+; CHECK-NEXT:  .LBB48_64: // %cond.load88
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    ld1 { v0.s }[2], [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB48_38
+; CHECK-NEXT:    b .LBB48_39
   %cvals = load <32 x float>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = fcmp oeq <32 x float> %cvals, zeroinitializer
   %vals = call <32 x float> @llvm.masked.gather.v32f32(<32 x ptr> %ptrs, i32 8, <32 x i1> %mask, <32 x float> zeroinitializer)
   store <32 x float> %vals, ptr %a
   ret void
 }
 

llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-masked-scatter.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -force-streaming-compatible-sve < %s | FileCheck %s
 
 target triple = "aarch64-unknown-linux-gnu"
 
 ;
 ; ST1B
 ;
 
 define void @masked_scatter_v2i8(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v2i8:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    ldrb w9, [x0, #1]
+; CHECK-NEXT:    adrp x8, .LCPI0_0
+; CHECK-NEXT:    ptrue p0.s, vl2
+; CHECK-NEXT:    str w9, [sp, #4]
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI0_0]
 ; CHECK-NEXT:    ldrb w8, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl2
-; CHECK-NEXT:    ldr q2, [x1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    ldrb w8, [x0, #1]
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    cmeq v1.2s, v0.2s, #0
-; CHECK-NEXT:    ushll v0.2d, v0.2s, #0
-; CHECK-NEXT:    sshll v1.2d, v1.2s, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z1.d, #0
-; CHECK-NEXT:    st1b { z0.d }, p0, [z2.d]
+; CHECK-NEXT:    str w8, [sp]
+; CHECK-NEXT:    ldr d0, [sp]
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.s, z1.s[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbnz w9, #0, .LBB0_3
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB0_4
+; CHECK-NEXT:  .LBB0_2: // %else2
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB0_3: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB0_2
+; CHECK-NEXT:  .LBB0_4: // %cond.store1
+; CHECK-NEXT:    mov z0.s, z0.s[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strb w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <2 x i8>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = icmp eq <2 x i8> %vals, zeroinitializer
   call void @llvm.masked.scatter.v2i8(<2 x i8> %vals, <2 x ptr> %ptrs, i32 8, <2 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v4i8(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v4i8:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI1_0
 ; CHECK-NEXT:    ldr s0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    ushll v0.8h, v0.8b, #0
-; CHECK-NEXT:    cmeq v2.4h, v0.4h, #0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    sunpklo z2.s, z2.h
-; CHECK-NEXT:    sunpklo z2.d, z2.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z2.d, #0
-; CHECK-NEXT:    st1b { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    ptrue p0.h, vl4
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI1_0]
+; CHECK-NEXT:    uunpklo z0.h, z0.b
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z1.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.h, z1.h[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z2.h, z1.h[2]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z1.h, z1.h[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    ldp q2, q1, [x1]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB1_5
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB1_6
+; CHECK-NEXT:  .LBB1_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB1_7
+; CHECK-NEXT:  .LBB1_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB1_8
+; CHECK-NEXT:  .LBB1_4: // %else6
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB1_5: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB1_2
+; CHECK-NEXT:  .LBB1_6: // %cond.store1
+; CHECK-NEXT:    mov z3.h, z0.h[1]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB1_3
+; CHECK-NEXT:  .LBB1_7: // %cond.store3
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB1_4
+; CHECK-NEXT:  .LBB1_8: // %cond.store5
+; CHECK-NEXT:    mov z0.h, z0.h[3]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strb w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <4 x i8>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = icmp eq <4 x i8> %vals, zeroinitializer
   call void @llvm.masked.scatter.v4i8(<4 x i8> %vals, <4 x ptr> %ptrs, i32 8, <4 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v8i8(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_scatter_v8i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI2_0
+; CHECK-NEXT:    ldr d0, [x0]
+; CHECK-NEXT:    ptrue p0.b, vl8
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI2_0]
+; CHECK-NEXT:    cmpeq p0.b, p0/z, z0.b, z1.b
+; CHECK-NEXT:    mov z1.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.b, z1.b[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    mov z3.b, z1.b[2]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z4.b, z1.b[3]
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    mov z5.b, z1.b[4]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s5
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    mov z6.b, z1.b[5]
+; CHECK-NEXT:    mov z2.b, z1.b[6]
+; CHECK-NEXT:    bfi w9, w11, #3, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    bfi w9, w8, #4, #1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z1.b, z1.b[7]
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    ldp q2, q1, [x1, #32]
+; CHECK-NEXT:    orr w8, w9, w8, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    ldp q4, q3, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB2_9
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB2_10
+; CHECK-NEXT:  .LBB2_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB2_11
+; CHECK-NEXT:  .LBB2_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB2_12
+; CHECK-NEXT:  .LBB2_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB2_13
+; CHECK-NEXT:  .LBB2_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB2_14
+; CHECK-NEXT:  .LBB2_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB2_15
+; CHECK-NEXT:  .LBB2_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB2_16
+; CHECK-NEXT:  .LBB2_8: // %else14
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB2_9: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB2_2
+; CHECK-NEXT:  .LBB2_10: // %cond.store1
+; CHECK-NEXT:    mov z5.b, z0.b[1]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB2_3
+; CHECK-NEXT:  .LBB2_11: // %cond.store3
+; CHECK-NEXT:    mov z4.b, z0.b[2]
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB2_4
+; CHECK-NEXT:  .LBB2_12: // %cond.store5
+; CHECK-NEXT:    mov z4.b, z0.b[3]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB2_5
+; CHECK-NEXT:  .LBB2_13: // %cond.store7
+; CHECK-NEXT:    mov z3.b, z0.b[4]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB2_6
+; CHECK-NEXT:  .LBB2_14: // %cond.store9
+; CHECK-NEXT:    mov z3.b, z0.b[5]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB2_7
+; CHECK-NEXT:  .LBB2_15: // %cond.store11
+; CHECK-NEXT:    mov z2.b, z0.b[6]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB2_8
+; CHECK-NEXT:  .LBB2_16: // %cond.store13
+; CHECK-NEXT:    mov z0.b, z0.b[7]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strb w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %vals = load <8 x i8>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = icmp eq <8 x i8> %vals, zeroinitializer
   call void @llvm.masked.scatter.v8i8(<8 x i8> %vals, <8 x ptr> %ptrs, i32 8, <8 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v16i8(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v16i8:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI3_0
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmeq v2.16b, v0.16b, #0
-; CHECK-NEXT:    uunpklo z0.h, z0.b
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    sunpklo z2.h, z2.b
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    sunpklo z2.s, z2.h
-; CHECK-NEXT:    sunpklo z2.d, z2.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z2.d, #0
-; CHECK-NEXT:    st1b { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    ptrue p0.b, vl16
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI3_0]
+; CHECK-NEXT:    cmpeq p0.b, p0/z, z0.b, z1.b
+; CHECK-NEXT:    mov z1.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.b, z1.b[1]
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    mov z3.b, z1.b[2]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z4.b, z1.b[3]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    bfi w10, w8, #1, #1
+; CHECK-NEXT:    mov z5.b, z1.b[4]
+; CHECK-NEXT:    mov z7.b, z1.b[6]
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    mov z6.b, z1.b[5]
+; CHECK-NEXT:    fmov w8, s5
+; CHECK-NEXT:    bfi w10, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    mov z16.b, z1.b[7]
+; CHECK-NEXT:    bfi w10, w8, #4, #1
+; CHECK-NEXT:    mov z17.b, z1.b[8]
+; CHECK-NEXT:    and w8, w11, #0x1
+; CHECK-NEXT:    mov z18.b, z1.b[9]
+; CHECK-NEXT:    bfi w10, w9, #5, #1
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    orr w8, w10, w8, lsl #6
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    mov z19.b, z1.b[10]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z2.b, z1.b[11]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z3.b, z1.b[12]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    orr w8, w8, w9, lsl #9
+; CHECK-NEXT:    mov z20.b, z1.b[13]
+; CHECK-NEXT:    mov z21.b, z1.b[14]
+; CHECK-NEXT:    mov z5.b, z1.b[15]
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    ldp q2, q1, [x1, #96]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #11
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    ldp q4, q3, [x1, #64]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    ldp q6, q5, [x1, #32]
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    ldp q16, q7, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB3_17
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB3_18
+; CHECK-NEXT:  .LBB3_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB3_19
+; CHECK-NEXT:  .LBB3_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB3_20
+; CHECK-NEXT:  .LBB3_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB3_21
+; CHECK-NEXT:  .LBB3_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB3_22
+; CHECK-NEXT:  .LBB3_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB3_23
+; CHECK-NEXT:  .LBB3_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB3_24
+; CHECK-NEXT:  .LBB3_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB3_25
+; CHECK-NEXT:  .LBB3_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB3_26
+; CHECK-NEXT:  .LBB3_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB3_27
+; CHECK-NEXT:  .LBB3_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB3_28
+; CHECK-NEXT:  .LBB3_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB3_29
+; CHECK-NEXT:  .LBB3_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB3_30
+; CHECK-NEXT:  .LBB3_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB3_31
+; CHECK-NEXT:  .LBB3_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB3_32
+; CHECK-NEXT:  .LBB3_16: // %else30
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB3_17: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB3_2
+; CHECK-NEXT:  .LBB3_18: // %cond.store1
+; CHECK-NEXT:    mov z17.b, z0.b[1]
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB3_3
+; CHECK-NEXT:  .LBB3_19: // %cond.store3
+; CHECK-NEXT:    mov z16.b, z0.b[2]
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB3_4
+; CHECK-NEXT:  .LBB3_20: // %cond.store5
+; CHECK-NEXT:    mov z16.b, z0.b[3]
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB3_5
+; CHECK-NEXT:  .LBB3_21: // %cond.store7
+; CHECK-NEXT:    mov z7.b, z0.b[4]
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB3_6
+; CHECK-NEXT:  .LBB3_22: // %cond.store9
+; CHECK-NEXT:    mov z7.b, z0.b[5]
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB3_7
+; CHECK-NEXT:  .LBB3_23: // %cond.store11
+; CHECK-NEXT:    mov z6.b, z0.b[6]
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB3_8
+; CHECK-NEXT:  .LBB3_24: // %cond.store13
+; CHECK-NEXT:    mov z6.b, z0.b[7]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #8, .LBB3_9
+; CHECK-NEXT:  .LBB3_25: // %cond.store15
+; CHECK-NEXT:    mov z5.b, z0.b[8]
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #9, .LBB3_10
+; CHECK-NEXT:  .LBB3_26: // %cond.store17
+; CHECK-NEXT:    mov z5.b, z0.b[9]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #10, .LBB3_11
+; CHECK-NEXT:  .LBB3_27: // %cond.store19
+; CHECK-NEXT:    mov z4.b, z0.b[10]
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #11, .LBB3_12
+; CHECK-NEXT:  .LBB3_28: // %cond.store21
+; CHECK-NEXT:    mov z4.b, z0.b[11]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #12, .LBB3_13
+; CHECK-NEXT:  .LBB3_29: // %cond.store23
+; CHECK-NEXT:    mov z3.b, z0.b[12]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #13, .LBB3_14
+; CHECK-NEXT:  .LBB3_30: // %cond.store25
+; CHECK-NEXT:    mov z3.b, z0.b[13]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #14, .LBB3_15
+; CHECK-NEXT:  .LBB3_31: // %cond.store27
+; CHECK-NEXT:    mov z2.b, z0.b[14]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #15, .LBB3_16
+; CHECK-NEXT:  .LBB3_32: // %cond.store29
+; CHECK-NEXT:    mov z0.b, z0.b[15]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strb w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <16 x i8>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = icmp eq <16 x i8> %vals, zeroinitializer
   call void @llvm.masked.scatter.v16i8(<16 x i8> %vals, <16 x ptr> %ptrs, i32 8, <16 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v32i8(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v32i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.b, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1b { z0.b }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.b, p0/z, z0.b, #0
-; CHECK-NEXT:    uunpklo z0.h, z0.b
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1b { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI4_0
+; CHECK-NEXT:    ptrue p0.b, vl16
+; CHECK-NEXT:    ldp q3, q0, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI4_0]
+; CHECK-NEXT:    cmpeq p1.b, p0/z, z0.b, z1.b
+; CHECK-NEXT:    cmpeq p0.b, p0/z, z3.b, z1.b
+; CHECK-NEXT:    mov z2.b, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z4.b, z2.b[1]
+; CHECK-NEXT:    mov z5.b, z2.b[2]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    mov z6.b, z2.b[3]
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z7.b, z2.b[4]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    mov z17.b, z2.b[6]
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    mov z16.b, z2.b[5]
+; CHECK-NEXT:    fmov w8, s7
+; CHECK-NEXT:    bfi w9, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    mov z18.b, z2.b[7]
+; CHECK-NEXT:    mov z19.b, z2.b[8]
+; CHECK-NEXT:    bfi w9, w8, #20, #1
+; CHECK-NEXT:    mov z20.b, z2.b[9]
+; CHECK-NEXT:    and w8, w11, #0x1
+; CHECK-NEXT:    bfi w9, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    fmov w11, s20
+; CHECK-NEXT:    orr w8, w9, w8, lsl #22
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    mov z21.b, z2.b[10]
+; CHECK-NEXT:    mov z4.b, z2.b[11]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z22.b, z2.b[12]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z5.b, z2.b[13]
+; CHECK-NEXT:    mov z6.b, z2.b[14]
+; CHECK-NEXT:    mov z2.b, z2.b[15]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    mov z4.b, z1.b[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z7.b, z1.b[6]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z16.b, z1.b[7]
+; CHECK-NEXT:    mov z17.b, z1.b[8]
+; CHECK-NEXT:    mov z18.b, z1.b[9]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    fmov w11, s1
+; CHECK-NEXT:    mov z5.b, z1.b[4]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z6.b, z1.b[5]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z19.b, z1.b[10]
+; CHECK-NEXT:    mov z20.b, z1.b[11]
+; CHECK-NEXT:    mov z21.b, z1.b[12]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    orr w8, w8, w10, lsl #30
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    mov z2.b, z1.b[2]
+; CHECK-NEXT:    mov z4.b, z1.b[3]
+; CHECK-NEXT:    fmov w12, s2
+; CHECK-NEXT:    orr w8, w8, w9, lsl #31
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    bfi w9, w10, #1, #1
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    bfi w9, w12, #2, #1
+; CHECK-NEXT:    fmov w12, s6
+; CHECK-NEXT:    mov z22.b, z1.b[13]
+; CHECK-NEXT:    mov z23.b, z1.b[14]
+; CHECK-NEXT:    bfi w9, w10, #3, #1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    bfi w9, w11, #4, #1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    bfi w9, w12, #5, #1
+; CHECK-NEXT:    mov z24.b, z1.b[15]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #6
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    orr w9, w9, w11, lsl #7
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    ldp q2, q1, [x1, #224]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    orr w9, w9, w11, lsl #9
+; CHECK-NEXT:    fmov w11, s20
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    fmov w10, s21
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    ldp q5, q4, [x1, #192]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldp q7, q6, [x1, #160]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q17, q16, [x1, #128]
+; CHECK-NEXT:    ldp q19, q18, [x1, #96]
+; CHECK-NEXT:    ldp q21, q20, [x1, #64]
+; CHECK-NEXT:    ldp q23, q22, [x1, #32]
+; CHECK-NEXT:    ldp q25, q24, [x1]
+; CHECK-NEXT:    tbnz w8, #0, .LBB4_33
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB4_34
+; CHECK-NEXT:  .LBB4_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB4_35
+; CHECK-NEXT:  .LBB4_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB4_36
+; CHECK-NEXT:  .LBB4_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB4_37
+; CHECK-NEXT:  .LBB4_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB4_38
+; CHECK-NEXT:  .LBB4_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB4_39
+; CHECK-NEXT:  .LBB4_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB4_40
+; CHECK-NEXT:  .LBB4_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB4_41
+; CHECK-NEXT:  .LBB4_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB4_42
+; CHECK-NEXT:  .LBB4_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB4_43
+; CHECK-NEXT:  .LBB4_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB4_44
+; CHECK-NEXT:  .LBB4_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB4_45
+; CHECK-NEXT:  .LBB4_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB4_46
+; CHECK-NEXT:  .LBB4_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB4_47
+; CHECK-NEXT:  .LBB4_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB4_48
+; CHECK-NEXT:  .LBB4_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB4_49
+; CHECK-NEXT:  .LBB4_17: // %else32
+; CHECK-NEXT:    tbnz w8, #17, .LBB4_50
+; CHECK-NEXT:  .LBB4_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB4_51
+; CHECK-NEXT:  .LBB4_19: // %else36
+; CHECK-NEXT:    tbnz w8, #19, .LBB4_52
+; CHECK-NEXT:  .LBB4_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB4_53
+; CHECK-NEXT:  .LBB4_21: // %else40
+; CHECK-NEXT:    tbnz w8, #21, .LBB4_54
+; CHECK-NEXT:  .LBB4_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB4_55
+; CHECK-NEXT:  .LBB4_23: // %else44
+; CHECK-NEXT:    tbnz w8, #23, .LBB4_56
+; CHECK-NEXT:  .LBB4_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB4_57
+; CHECK-NEXT:  .LBB4_25: // %else48
+; CHECK-NEXT:    tbnz w8, #25, .LBB4_58
+; CHECK-NEXT:  .LBB4_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB4_59
+; CHECK-NEXT:  .LBB4_27: // %else52
+; CHECK-NEXT:    tbnz w8, #27, .LBB4_60
+; CHECK-NEXT:  .LBB4_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB4_61
+; CHECK-NEXT:  .LBB4_29: // %else56
+; CHECK-NEXT:    tbnz w8, #29, .LBB4_62
+; CHECK-NEXT:  .LBB4_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB4_63
+; CHECK-NEXT:  .LBB4_31: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB4_64
+; CHECK-NEXT:  .LBB4_32: // %else62
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB4_33: // %cond.store
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d25
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB4_2
+; CHECK-NEXT:  .LBB4_34: // %cond.store1
+; CHECK-NEXT:    mov z26.b, z3.b[1]
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    fmov w9, s26
+; CHECK-NEXT:    fmov x10, d25
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB4_3
+; CHECK-NEXT:  .LBB4_35: // %cond.store3
+; CHECK-NEXT:    mov z25.b, z3.b[2]
+; CHECK-NEXT:    fmov x10, d24
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB4_4
+; CHECK-NEXT:  .LBB4_36: // %cond.store5
+; CHECK-NEXT:    mov z25.b, z3.b[3]
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    fmov x10, d24
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB4_5
+; CHECK-NEXT:  .LBB4_37: // %cond.store7
+; CHECK-NEXT:    mov z24.b, z3.b[4]
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    fmov w9, s24
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB4_6
+; CHECK-NEXT:  .LBB4_38: // %cond.store9
+; CHECK-NEXT:    mov z24.b, z3.b[5]
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov w9, s24
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB4_7
+; CHECK-NEXT:  .LBB4_39: // %cond.store11
+; CHECK-NEXT:    mov z23.b, z3.b[6]
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB4_8
+; CHECK-NEXT:  .LBB4_40: // %cond.store13
+; CHECK-NEXT:    mov z23.b, z3.b[7]
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #8, .LBB4_9
+; CHECK-NEXT:  .LBB4_41: // %cond.store15
+; CHECK-NEXT:    mov z22.b, z3.b[8]
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    fmov w9, s22
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #9, .LBB4_10
+; CHECK-NEXT:  .LBB4_42: // %cond.store17
+; CHECK-NEXT:    mov z22.b, z3.b[9]
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov w9, s22
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #10, .LBB4_11
+; CHECK-NEXT:  .LBB4_43: // %cond.store19
+; CHECK-NEXT:    mov z21.b, z3.b[10]
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #11, .LBB4_12
+; CHECK-NEXT:  .LBB4_44: // %cond.store21
+; CHECK-NEXT:    mov z21.b, z3.b[11]
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #12, .LBB4_13
+; CHECK-NEXT:  .LBB4_45: // %cond.store23
+; CHECK-NEXT:    mov z20.b, z3.b[12]
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    fmov w9, s20
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #13, .LBB4_14
+; CHECK-NEXT:  .LBB4_46: // %cond.store25
+; CHECK-NEXT:    mov z20.b, z3.b[13]
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov w9, s20
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #14, .LBB4_15
+; CHECK-NEXT:  .LBB4_47: // %cond.store27
+; CHECK-NEXT:    mov z19.b, z3.b[14]
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    fmov w9, s19
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #15, .LBB4_16
+; CHECK-NEXT:  .LBB4_48: // %cond.store29
+; CHECK-NEXT:    mov z3.b, z3.b[15]
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #16, .LBB4_17
+; CHECK-NEXT:  .LBB4_49: // %cond.store31
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #17, .LBB4_18
+; CHECK-NEXT:  .LBB4_50: // %cond.store33
+; CHECK-NEXT:    mov z3.b, z0.b[1]
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #18, .LBB4_19
+; CHECK-NEXT:  .LBB4_51: // %cond.store35
+; CHECK-NEXT:    mov z3.b, z0.b[2]
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #19, .LBB4_20
+; CHECK-NEXT:  .LBB4_52: // %cond.store37
+; CHECK-NEXT:    mov z3.b, z0.b[3]
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #20, .LBB4_21
+; CHECK-NEXT:  .LBB4_53: // %cond.store39
+; CHECK-NEXT:    mov z3.b, z0.b[4]
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #21, .LBB4_22
+; CHECK-NEXT:  .LBB4_54: // %cond.store41
+; CHECK-NEXT:    mov z3.b, z0.b[5]
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #22, .LBB4_23
+; CHECK-NEXT:  .LBB4_55: // %cond.store43
+; CHECK-NEXT:    mov z3.b, z0.b[6]
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #23, .LBB4_24
+; CHECK-NEXT:  .LBB4_56: // %cond.store45
+; CHECK-NEXT:    mov z3.b, z0.b[7]
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #24, .LBB4_25
+; CHECK-NEXT:  .LBB4_57: // %cond.store47
+; CHECK-NEXT:    mov z3.b, z0.b[8]
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #25, .LBB4_26
+; CHECK-NEXT:  .LBB4_58: // %cond.store49
+; CHECK-NEXT:    mov z3.b, z0.b[9]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #26, .LBB4_27
+; CHECK-NEXT:  .LBB4_59: // %cond.store51
+; CHECK-NEXT:    mov z3.b, z0.b[10]
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #27, .LBB4_28
+; CHECK-NEXT:  .LBB4_60: // %cond.store53
+; CHECK-NEXT:    mov z3.b, z0.b[11]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #28, .LBB4_29
+; CHECK-NEXT:  .LBB4_61: // %cond.store55
+; CHECK-NEXT:    mov z3.b, z0.b[12]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #29, .LBB4_30
+; CHECK-NEXT:  .LBB4_62: // %cond.store57
+; CHECK-NEXT:    mov z3.b, z0.b[13]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #30, .LBB4_31
+; CHECK-NEXT:  .LBB4_63: // %cond.store59
+; CHECK-NEXT:    mov z2.b, z0.b[14]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strb w9, [x10]
+; CHECK-NEXT:    tbz w8, #31, .LBB4_32
+; CHECK-NEXT:  .LBB4_64: // %cond.store61
+; CHECK-NEXT:    mov z0.b, z0.b[15]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strb w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <32 x i8>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = icmp eq <32 x i8> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32i8(<32 x i8> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 ;
 ; ST1H
 ;
 
 define void @masked_scatter_v2i16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v2i16:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    ldrh w9, [x0, #2]
+; CHECK-NEXT:    adrp x8, .LCPI5_0
+; CHECK-NEXT:    ptrue p0.s, vl2
+; CHECK-NEXT:    str w9, [sp, #4]
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI5_0]
 ; CHECK-NEXT:    ldrh w8, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl2
-; CHECK-NEXT:    ldr q2, [x1]
-; CHECK-NEXT:    fmov s0, w8
-; CHECK-NEXT:    ldrh w8, [x0, #2]
-; CHECK-NEXT:    mov v0.s[1], w8
-; CHECK-NEXT:    cmeq v1.2s, v0.2s, #0
-; CHECK-NEXT:    ushll v0.2d, v0.2s, #0
-; CHECK-NEXT:    sshll v1.2d, v1.2s, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z1.d, #0
-; CHECK-NEXT:    st1h { z0.d }, p0, [z2.d]
+; CHECK-NEXT:    str w8, [sp]
+; CHECK-NEXT:    ldr d0, [sp]
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.s, z1.s[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbnz w9, #0, .LBB5_3
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB5_4
+; CHECK-NEXT:  .LBB5_2: // %else2
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB5_3: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB5_2
+; CHECK-NEXT:  .LBB5_4: // %cond.store1
+; CHECK-NEXT:    mov z0.s, z0.s[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strh w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <2 x i16>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = icmp eq <2 x i16> %vals, zeroinitializer
   call void @llvm.masked.scatter.v2i16(<2 x i16> %vals, <2 x ptr> %ptrs, i32 8, <2 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v4i16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v4i16:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI6_0
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmeq v2.4h, v0.4h, #0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    sunpklo z2.s, z2.h
-; CHECK-NEXT:    sunpklo z2.d, z2.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z2.d, #0
-; CHECK-NEXT:    st1h { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    ptrue p0.h, vl4
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI6_0]
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z1.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.h, z1.h[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z2.h, z1.h[2]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z1.h, z1.h[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    ldp q2, q1, [x1]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB6_5
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB6_6
+; CHECK-NEXT:  .LBB6_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB6_7
+; CHECK-NEXT:  .LBB6_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB6_8
+; CHECK-NEXT:  .LBB6_4: // %else6
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB6_5: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB6_2
+; CHECK-NEXT:  .LBB6_6: // %cond.store1
+; CHECK-NEXT:    mov z3.h, z0.h[1]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB6_3
+; CHECK-NEXT:  .LBB6_7: // %cond.store3
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB6_4
+; CHECK-NEXT:  .LBB6_8: // %cond.store5
+; CHECK-NEXT:    mov z0.h, z0.h[3]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strh w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <4 x i16>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = icmp eq <4 x i16> %vals, zeroinitializer
   call void @llvm.masked.scatter.v4i16(<4 x i16> %vals, <4 x ptr> %ptrs, i32 8, <4 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v8i16(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_scatter_v8i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI7_0
+; CHECK-NEXT:    ldr q0, [x0]
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI7_0]
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z1.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.h, z1.h[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    mov z3.h, z1.h[2]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z4.h, z1.h[3]
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    mov z5.h, z1.h[4]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s5
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    mov z6.h, z1.h[5]
+; CHECK-NEXT:    mov z2.h, z1.h[6]
+; CHECK-NEXT:    bfi w9, w11, #3, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    bfi w9, w8, #4, #1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z1.h, z1.h[7]
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    ldp q2, q1, [x1, #32]
+; CHECK-NEXT:    orr w8, w9, w8, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    ldp q4, q3, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB7_9
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB7_10
+; CHECK-NEXT:  .LBB7_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB7_11
+; CHECK-NEXT:  .LBB7_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB7_12
+; CHECK-NEXT:  .LBB7_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB7_13
+; CHECK-NEXT:  .LBB7_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB7_14
+; CHECK-NEXT:  .LBB7_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB7_15
+; CHECK-NEXT:  .LBB7_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB7_16
+; CHECK-NEXT:  .LBB7_8: // %else14
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB7_9: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB7_2
+; CHECK-NEXT:  .LBB7_10: // %cond.store1
+; CHECK-NEXT:    mov z5.h, z0.h[1]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB7_3
+; CHECK-NEXT:  .LBB7_11: // %cond.store3
+; CHECK-NEXT:    mov z4.h, z0.h[2]
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB7_4
+; CHECK-NEXT:  .LBB7_12: // %cond.store5
+; CHECK-NEXT:    mov z4.h, z0.h[3]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB7_5
+; CHECK-NEXT:  .LBB7_13: // %cond.store7
+; CHECK-NEXT:    mov z3.h, z0.h[4]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB7_6
+; CHECK-NEXT:  .LBB7_14: // %cond.store9
+; CHECK-NEXT:    mov z3.h, z0.h[5]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB7_7
+; CHECK-NEXT:  .LBB7_15: // %cond.store11
+; CHECK-NEXT:    mov z2.h, z0.h[6]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB7_8
+; CHECK-NEXT:  .LBB7_16: // %cond.store13
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strh w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %vals = load <8 x i16>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = icmp eq <8 x i16> %vals, zeroinitializer
   call void @llvm.masked.scatter.v8i16(<8 x i16> %vals, <8 x ptr> %ptrs, i32 8, <8 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v16i16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v16i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl16
-; CHECK-NEXT:    ptrue p1.d, vl16
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, #0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1h { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI8_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q2, q0, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI8_0]
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z2.h, z1.h
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z3.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z4.h, z3.h[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z5.h, z3.h[2]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    mov z6.h, z3.h[3]
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    mov z7.h, z3.h[4]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z16.h, z3.h[5]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s7
+; CHECK-NEXT:    mov z17.h, z3.h[6]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    mov z18.h, z3.h[7]
+; CHECK-NEXT:    bfi w9, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    bfi w9, w8, #4, #1
+; CHECK-NEXT:    fmov w8, s18
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z3.h, z1.h[1]
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z4.h, z1.h[2]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #6
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    orr w8, w9, w8, lsl #7
+; CHECK-NEXT:    orr w8, w8, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    mov z5.h, z1.h[3]
+; CHECK-NEXT:    mov z6.h, z1.h[4]
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z7.h, z1.h[5]
+; CHECK-NEXT:    mov z16.h, z1.h[6]
+; CHECK-NEXT:    mov z17.h, z1.h[7]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #9
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    ldp q3, q1, [x1, #96]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    ldp q5, q4, [x1, #64]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    ldp q7, q6, [x1, #32]
+; CHECK-NEXT:    ldp q17, q16, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB8_17
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB8_18
+; CHECK-NEXT:  .LBB8_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB8_19
+; CHECK-NEXT:  .LBB8_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB8_20
+; CHECK-NEXT:  .LBB8_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB8_21
+; CHECK-NEXT:  .LBB8_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB8_22
+; CHECK-NEXT:  .LBB8_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB8_23
+; CHECK-NEXT:  .LBB8_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB8_24
+; CHECK-NEXT:  .LBB8_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB8_25
+; CHECK-NEXT:  .LBB8_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB8_26
+; CHECK-NEXT:  .LBB8_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB8_27
+; CHECK-NEXT:  .LBB8_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB8_28
+; CHECK-NEXT:  .LBB8_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB8_29
+; CHECK-NEXT:  .LBB8_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB8_30
+; CHECK-NEXT:  .LBB8_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB8_31
+; CHECK-NEXT:  .LBB8_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB8_32
+; CHECK-NEXT:  .LBB8_16: // %else30
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB8_17: // %cond.store
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB8_2
+; CHECK-NEXT:  .LBB8_18: // %cond.store1
+; CHECK-NEXT:    mov z18.h, z2.h[1]
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB8_3
+; CHECK-NEXT:  .LBB8_19: // %cond.store3
+; CHECK-NEXT:    mov z17.h, z2.h[2]
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB8_4
+; CHECK-NEXT:  .LBB8_20: // %cond.store5
+; CHECK-NEXT:    mov z17.h, z2.h[3]
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB8_5
+; CHECK-NEXT:  .LBB8_21: // %cond.store7
+; CHECK-NEXT:    mov z16.h, z2.h[4]
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB8_6
+; CHECK-NEXT:  .LBB8_22: // %cond.store9
+; CHECK-NEXT:    mov z16.h, z2.h[5]
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB8_7
+; CHECK-NEXT:  .LBB8_23: // %cond.store11
+; CHECK-NEXT:    mov z7.h, z2.h[6]
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB8_8
+; CHECK-NEXT:  .LBB8_24: // %cond.store13
+; CHECK-NEXT:    mov z2.h, z2.h[7]
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #8, .LBB8_9
+; CHECK-NEXT:  .LBB8_25: // %cond.store15
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #9, .LBB8_10
+; CHECK-NEXT:  .LBB8_26: // %cond.store17
+; CHECK-NEXT:    mov z2.h, z0.h[1]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #10, .LBB8_11
+; CHECK-NEXT:  .LBB8_27: // %cond.store19
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #11, .LBB8_12
+; CHECK-NEXT:  .LBB8_28: // %cond.store21
+; CHECK-NEXT:    mov z2.h, z0.h[3]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #12, .LBB8_13
+; CHECK-NEXT:  .LBB8_29: // %cond.store23
+; CHECK-NEXT:    mov z2.h, z0.h[4]
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #13, .LBB8_14
+; CHECK-NEXT:  .LBB8_30: // %cond.store25
+; CHECK-NEXT:    mov z2.h, z0.h[5]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #14, .LBB8_15
+; CHECK-NEXT:  .LBB8_31: // %cond.store27
+; CHECK-NEXT:    mov z2.h, z0.h[6]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #15, .LBB8_16
+; CHECK-NEXT:  .LBB8_32: // %cond.store29
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strh w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <16 x i16>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = icmp eq <16 x i16> %vals, zeroinitializer
   call void @llvm.masked.scatter.v16i16(<16 x i16> %vals, <16 x ptr> %ptrs, i32 8, <16 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v32i16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v32i16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.h, p0/z, z0.h, #0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1h { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI9_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q1, q0, [x0, #32]
+; CHECK-NEXT:    ldr q7, [x8, :lo12:.LCPI9_0]
+; CHECK-NEXT:    ldp q17, q6, [x0]
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z1.h, z7.h
+; CHECK-NEXT:    mov z2.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z0.h, z7.h
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z3.h, z2.h[1]
+; CHECK-NEXT:    mov z4.h, z2.h[2]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z5.h, z2.h[3]
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z16.h, z2.h[4]
+; CHECK-NEXT:    mov z19.h, z2.h[6]
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z18.h, z2.h[5]
+; CHECK-NEXT:    mov z20.h, z2.h[7]
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s20
+; CHECK-NEXT:    bfi w8, w12, #20, #1
+; CHECK-NEXT:    mov z2.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    bfi w8, w9, #21, #1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z3.h, z2.h[1]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #22
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z4.h, z2.h[2]
+; CHECK-NEXT:    mov z5.h, z2.h[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.h, z2.h[4]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    mov z18.h, z2.h[5]
+; CHECK-NEXT:    mov z19.h, z2.h[6]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    cmpeq p1.h, p0/z, z17.h, z7.h
+; CHECK-NEXT:    mov z20.h, z2.h[7]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.h, z4.h[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    mov z16.h, z4.h[3]
+; CHECK-NEXT:    mov z18.h, z4.h[4]
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    mov z21.h, z4.h[6]
+; CHECK-NEXT:    mov z19.h, z4.h[5]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.h, z4.h[2]
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z22.h, z4.h[7]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #30
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    cmpeq p0.h, p0/z, z6.h, z7.h
+; CHECK-NEXT:    mov z7.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.h, z7.h[1]
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    bfi w10, w11, #2, #1
+; CHECK-NEXT:    fmov w11, s20
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    fmov w12, s19
+; CHECK-NEXT:    bfi w10, w9, #4, #1
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    mov z18.h, z7.h[2]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z19.h, z7.h[3]
+; CHECK-NEXT:    mov z20.h, z7.h[4]
+; CHECK-NEXT:    mov z21.h, z7.h[5]
+; CHECK-NEXT:    orr w9, w10, w9, lsl #6
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    mov z22.h, z7.h[6]
+; CHECK-NEXT:    mov z23.h, z7.h[7]
+; CHECK-NEXT:    ldp q3, q2, [x1, #224]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    orr w9, w9, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    ldp q5, q4, [x1, #192]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #9
+; CHECK-NEXT:    fmov w11, s19
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s21
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    ldp q16, q7, [x1, #160]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    ldp q19, q18, [x1, #128]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q21, q20, [x1, #96]
+; CHECK-NEXT:    ldp q23, q22, [x1, #64]
+; CHECK-NEXT:    ldp q25, q24, [x1, #32]
+; CHECK-NEXT:    ldp q27, q26, [x1]
+; CHECK-NEXT:    tbnz w8, #0, .LBB9_33
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB9_34
+; CHECK-NEXT:  .LBB9_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB9_35
+; CHECK-NEXT:  .LBB9_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB9_36
+; CHECK-NEXT:  .LBB9_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB9_37
+; CHECK-NEXT:  .LBB9_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB9_38
+; CHECK-NEXT:  .LBB9_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB9_39
+; CHECK-NEXT:  .LBB9_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB9_40
+; CHECK-NEXT:  .LBB9_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB9_41
+; CHECK-NEXT:  .LBB9_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB9_42
+; CHECK-NEXT:  .LBB9_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB9_43
+; CHECK-NEXT:  .LBB9_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB9_44
+; CHECK-NEXT:  .LBB9_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB9_45
+; CHECK-NEXT:  .LBB9_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB9_46
+; CHECK-NEXT:  .LBB9_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB9_47
+; CHECK-NEXT:  .LBB9_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB9_48
+; CHECK-NEXT:  .LBB9_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB9_49
+; CHECK-NEXT:  .LBB9_17: // %else32
+; CHECK-NEXT:    tbnz w8, #17, .LBB9_50
+; CHECK-NEXT:  .LBB9_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB9_51
+; CHECK-NEXT:  .LBB9_19: // %else36
+; CHECK-NEXT:    tbnz w8, #19, .LBB9_52
+; CHECK-NEXT:  .LBB9_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB9_53
+; CHECK-NEXT:  .LBB9_21: // %else40
+; CHECK-NEXT:    tbnz w8, #21, .LBB9_54
+; CHECK-NEXT:  .LBB9_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB9_55
+; CHECK-NEXT:  .LBB9_23: // %else44
+; CHECK-NEXT:    tbnz w8, #23, .LBB9_56
+; CHECK-NEXT:  .LBB9_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB9_57
+; CHECK-NEXT:  .LBB9_25: // %else48
+; CHECK-NEXT:    tbnz w8, #25, .LBB9_58
+; CHECK-NEXT:  .LBB9_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB9_59
+; CHECK-NEXT:  .LBB9_27: // %else52
+; CHECK-NEXT:    tbnz w8, #27, .LBB9_60
+; CHECK-NEXT:  .LBB9_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB9_61
+; CHECK-NEXT:  .LBB9_29: // %else56
+; CHECK-NEXT:    tbnz w8, #29, .LBB9_62
+; CHECK-NEXT:  .LBB9_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB9_63
+; CHECK-NEXT:  .LBB9_31: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB9_64
+; CHECK-NEXT:  .LBB9_32: // %else62
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB9_33: // %cond.store
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    fmov x10, d27
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB9_2
+; CHECK-NEXT:  .LBB9_34: // %cond.store1
+; CHECK-NEXT:    mov z28.h, z17.h[1]
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    fmov w9, s28
+; CHECK-NEXT:    fmov x10, d27
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB9_3
+; CHECK-NEXT:  .LBB9_35: // %cond.store3
+; CHECK-NEXT:    mov z27.h, z17.h[2]
+; CHECK-NEXT:    fmov x10, d26
+; CHECK-NEXT:    fmov w9, s27
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB9_4
+; CHECK-NEXT:  .LBB9_36: // %cond.store5
+; CHECK-NEXT:    mov z27.h, z17.h[3]
+; CHECK-NEXT:    mov z26.d, z26.d[1]
+; CHECK-NEXT:    fmov w9, s27
+; CHECK-NEXT:    fmov x10, d26
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB9_5
+; CHECK-NEXT:  .LBB9_37: // %cond.store7
+; CHECK-NEXT:    mov z26.h, z17.h[4]
+; CHECK-NEXT:    fmov x10, d25
+; CHECK-NEXT:    fmov w9, s26
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB9_6
+; CHECK-NEXT:  .LBB9_38: // %cond.store9
+; CHECK-NEXT:    mov z26.h, z17.h[5]
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    fmov w9, s26
+; CHECK-NEXT:    fmov x10, d25
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB9_7
+; CHECK-NEXT:  .LBB9_39: // %cond.store11
+; CHECK-NEXT:    mov z25.h, z17.h[6]
+; CHECK-NEXT:    fmov x10, d24
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB9_8
+; CHECK-NEXT:  .LBB9_40: // %cond.store13
+; CHECK-NEXT:    mov z17.h, z17.h[7]
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    fmov x10, d24
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #8, .LBB9_9
+; CHECK-NEXT:  .LBB9_41: // %cond.store15
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #9, .LBB9_10
+; CHECK-NEXT:  .LBB9_42: // %cond.store17
+; CHECK-NEXT:    mov z17.h, z6.h[1]
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #10, .LBB9_11
+; CHECK-NEXT:  .LBB9_43: // %cond.store19
+; CHECK-NEXT:    mov z17.h, z6.h[2]
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #11, .LBB9_12
+; CHECK-NEXT:  .LBB9_44: // %cond.store21
+; CHECK-NEXT:    mov z17.h, z6.h[3]
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #12, .LBB9_13
+; CHECK-NEXT:  .LBB9_45: // %cond.store23
+; CHECK-NEXT:    mov z17.h, z6.h[4]
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #13, .LBB9_14
+; CHECK-NEXT:  .LBB9_46: // %cond.store25
+; CHECK-NEXT:    mov z17.h, z6.h[5]
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #14, .LBB9_15
+; CHECK-NEXT:  .LBB9_47: // %cond.store27
+; CHECK-NEXT:    mov z17.h, z6.h[6]
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #15, .LBB9_16
+; CHECK-NEXT:  .LBB9_48: // %cond.store29
+; CHECK-NEXT:    mov z6.h, z6.h[7]
+; CHECK-NEXT:    mov z17.d, z20.d[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #16, .LBB9_17
+; CHECK-NEXT:  .LBB9_49: // %cond.store31
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #17, .LBB9_18
+; CHECK-NEXT:  .LBB9_50: // %cond.store33
+; CHECK-NEXT:    mov z6.h, z1.h[1]
+; CHECK-NEXT:    mov z17.d, z19.d[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #18, .LBB9_19
+; CHECK-NEXT:  .LBB9_51: // %cond.store35
+; CHECK-NEXT:    mov z6.h, z1.h[2]
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #19, .LBB9_20
+; CHECK-NEXT:  .LBB9_52: // %cond.store37
+; CHECK-NEXT:    mov z6.h, z1.h[3]
+; CHECK-NEXT:    mov z17.d, z18.d[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #20, .LBB9_21
+; CHECK-NEXT:  .LBB9_53: // %cond.store39
+; CHECK-NEXT:    mov z6.h, z1.h[4]
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #21, .LBB9_22
+; CHECK-NEXT:  .LBB9_54: // %cond.store41
+; CHECK-NEXT:    mov z6.h, z1.h[5]
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #22, .LBB9_23
+; CHECK-NEXT:  .LBB9_55: // %cond.store43
+; CHECK-NEXT:    mov z6.h, z1.h[6]
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #23, .LBB9_24
+; CHECK-NEXT:  .LBB9_56: // %cond.store45
+; CHECK-NEXT:    mov z1.h, z1.h[7]
+; CHECK-NEXT:    mov z6.d, z7.d[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #24, .LBB9_25
+; CHECK-NEXT:  .LBB9_57: // %cond.store47
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #25, .LBB9_26
+; CHECK-NEXT:  .LBB9_58: // %cond.store49
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #26, .LBB9_27
+; CHECK-NEXT:  .LBB9_59: // %cond.store51
+; CHECK-NEXT:    mov z1.h, z0.h[2]
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #27, .LBB9_28
+; CHECK-NEXT:  .LBB9_60: // %cond.store53
+; CHECK-NEXT:    mov z1.h, z0.h[3]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #28, .LBB9_29
+; CHECK-NEXT:  .LBB9_61: // %cond.store55
+; CHECK-NEXT:    mov z1.h, z0.h[4]
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #29, .LBB9_30
+; CHECK-NEXT:  .LBB9_62: // %cond.store57
+; CHECK-NEXT:    mov z1.h, z0.h[5]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #30, .LBB9_31
+; CHECK-NEXT:  .LBB9_63: // %cond.store59
+; CHECK-NEXT:    mov z1.h, z0.h[6]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    strh w9, [x10]
+; CHECK-NEXT:    tbz w8, #31, .LBB9_32
+; CHECK-NEXT:  .LBB9_64: // %cond.store61
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    strh w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <32 x i16>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = icmp eq <32 x i16> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32i16(<32 x i16> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 ;
 ; ST1W
 ;
 
 define void @masked_scatter_v2i32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v2i32:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI10_0
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl2
-; CHECK-NEXT:    ldr q2, [x1]
-; CHECK-NEXT:    cmeq v1.2s, v0.2s, #0
-; CHECK-NEXT:    ushll v0.2d, v0.2s, #0
-; CHECK-NEXT:    sshll v1.2d, v1.2s, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z1.d, #0
-; CHECK-NEXT:    st1w { z0.d }, p0, [z2.d]
+; CHECK-NEXT:    ptrue p0.s, vl2
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI10_0]
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.s, z1.s[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbnz w9, #0, .LBB10_3
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB10_4
+; CHECK-NEXT:  .LBB10_2: // %else2
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB10_3: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB10_2
+; CHECK-NEXT:  .LBB10_4: // %cond.store1
+; CHECK-NEXT:    mov z0.s, z0.s[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <2 x i32>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = icmp eq <2 x i32> %vals, zeroinitializer
   call void @llvm.masked.scatter.v2i32(<2 x i32> %vals, <2 x ptr> %ptrs, i32 8, <2 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v4i32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v4i32:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI11_0
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmeq v2.4s, v0.4s, #0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    sunpklo z2.d, z2.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z2.d, #0
-; CHECK-NEXT:    st1w { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI11_0]
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.s, z1.s[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z2.s, z1.s[2]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    ldp q2, q1, [x1]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB11_5
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB11_6
+; CHECK-NEXT:  .LBB11_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB11_7
+; CHECK-NEXT:  .LBB11_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB11_8
+; CHECK-NEXT:  .LBB11_4: // %else6
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB11_5: // %cond.store
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB11_2
+; CHECK-NEXT:  .LBB11_6: // %cond.store1
+; CHECK-NEXT:    mov z3.s, z0.s[1]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB11_3
+; CHECK-NEXT:  .LBB11_7: // %cond.store3
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB11_4
+; CHECK-NEXT:  .LBB11_8: // %cond.store5
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <4 x i32>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = icmp eq <4 x i32> %vals, zeroinitializer
   call void @llvm.masked.scatter.v4i32(<4 x i32> %vals, <4 x ptr> %ptrs, i32 8, <4 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v8i32(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_scatter_v8i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI12_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q1, q0, [x0]
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI12_0]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z1.s, z2.s
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z2.s
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z3.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z5.s, z2.s[2]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    mov z2.s, z3.s[1]
+; CHECK-NEXT:    mov z4.s, z3.s[2]
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    mov z2.s, z3.s[3]
+; CHECK-NEXT:    ldp q5, q4, [x1]
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    ldp q3, q2, [x1, #32]
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbnz w9, #0, .LBB12_9
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB12_10
+; CHECK-NEXT:  .LBB12_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB12_11
+; CHECK-NEXT:  .LBB12_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB12_12
+; CHECK-NEXT:  .LBB12_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB12_13
+; CHECK-NEXT:  .LBB12_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB12_14
+; CHECK-NEXT:  .LBB12_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB12_15
+; CHECK-NEXT:  .LBB12_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB12_16
+; CHECK-NEXT:  .LBB12_8: // %else14
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB12_9: // %cond.store
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB12_2
+; CHECK-NEXT:  .LBB12_10: // %cond.store1
+; CHECK-NEXT:    mov z6.s, z1.s[1]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB12_3
+; CHECK-NEXT:  .LBB12_11: // %cond.store3
+; CHECK-NEXT:    mov z5.s, z1.s[2]
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB12_4
+; CHECK-NEXT:  .LBB12_12: // %cond.store5
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB12_5
+; CHECK-NEXT:  .LBB12_13: // %cond.store7
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB12_6
+; CHECK-NEXT:  .LBB12_14: // %cond.store9
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB12_7
+; CHECK-NEXT:  .LBB12_15: // %cond.store11
+; CHECK-NEXT:    mov z1.s, z0.s[2]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB12_8
+; CHECK-NEXT:  .LBB12_16: // %cond.store13
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %vals = load <8 x i32>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = icmp eq <8 x i32> %vals, zeroinitializer
   call void @llvm.masked.scatter.v8i32(<8 x i32> %vals, <8 x ptr> %ptrs, i32 8, <8 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v16i32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v16i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl16
-; CHECK-NEXT:    ptrue p1.d, vl16
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, #0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1w { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI13_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q6, q4, [x0]
+; CHECK-NEXT:    ldr q5, [x8, :lo12:.LCPI13_0]
+; CHECK-NEXT:    ldp q2, q0, [x0, #32]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z6.s, z5.s
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z4.s, z5.s
+; CHECK-NEXT:    mov z7.s, z1.s[1]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z16.s, z1.s[2]
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    mov z17.s, z1.s[3]
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    mov z7.s, z3.s[2]
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    mov z1.s, z3.s[1]
+; CHECK-NEXT:    mov z16.s, z3.s[3]
+; CHECK-NEXT:    fmov w12, s3
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    bfi w8, w12, #4, #1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z2.s, z5.s
+; CHECK-NEXT:    bfi w8, w9, #5, #1
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #6
+; CHECK-NEXT:    mov z3.s, z1.s[1]
+; CHECK-NEXT:    mov z7.s, z1.s[2]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    mov z16.s, z1.s[3]
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, z5.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z5.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z7.s, z5.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #9
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    mov z16.s, z5.s[2]
+; CHECK-NEXT:    mov z17.s, z5.s[3]
+; CHECK-NEXT:    ldp q3, q1, [x1, #96]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    ldp q7, q5, [x1, #64]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    ldp q17, q16, [x1, #32]
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    ldp q19, q18, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB13_17
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB13_18
+; CHECK-NEXT:  .LBB13_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB13_19
+; CHECK-NEXT:  .LBB13_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB13_20
+; CHECK-NEXT:  .LBB13_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB13_21
+; CHECK-NEXT:  .LBB13_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB13_22
+; CHECK-NEXT:  .LBB13_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB13_23
+; CHECK-NEXT:  .LBB13_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB13_24
+; CHECK-NEXT:  .LBB13_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB13_25
+; CHECK-NEXT:  .LBB13_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB13_26
+; CHECK-NEXT:  .LBB13_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB13_27
+; CHECK-NEXT:  .LBB13_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB13_28
+; CHECK-NEXT:  .LBB13_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB13_29
+; CHECK-NEXT:  .LBB13_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB13_30
+; CHECK-NEXT:  .LBB13_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB13_31
+; CHECK-NEXT:  .LBB13_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB13_32
+; CHECK-NEXT:  .LBB13_16: // %else30
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB13_17: // %cond.store
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB13_2
+; CHECK-NEXT:  .LBB13_18: // %cond.store1
+; CHECK-NEXT:    mov z20.s, z6.s[1]
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov w9, s20
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB13_3
+; CHECK-NEXT:  .LBB13_19: // %cond.store3
+; CHECK-NEXT:    mov z19.s, z6.s[2]
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    fmov w9, s19
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB13_4
+; CHECK-NEXT:  .LBB13_20: // %cond.store5
+; CHECK-NEXT:    mov z6.s, z6.s[3]
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB13_5
+; CHECK-NEXT:  .LBB13_21: // %cond.store7
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB13_6
+; CHECK-NEXT:  .LBB13_22: // %cond.store9
+; CHECK-NEXT:    mov z6.s, z4.s[1]
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB13_7
+; CHECK-NEXT:  .LBB13_23: // %cond.store11
+; CHECK-NEXT:    mov z6.s, z4.s[2]
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB13_8
+; CHECK-NEXT:  .LBB13_24: // %cond.store13
+; CHECK-NEXT:    mov z4.s, z4.s[3]
+; CHECK-NEXT:    mov z6.d, z16.d[1]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #8, .LBB13_9
+; CHECK-NEXT:  .LBB13_25: // %cond.store15
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #9, .LBB13_10
+; CHECK-NEXT:  .LBB13_26: // %cond.store17
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    mov z6.d, z7.d[1]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #10, .LBB13_11
+; CHECK-NEXT:  .LBB13_27: // %cond.store19
+; CHECK-NEXT:    mov z4.s, z2.s[2]
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #11, .LBB13_12
+; CHECK-NEXT:  .LBB13_28: // %cond.store21
+; CHECK-NEXT:    mov z2.s, z2.s[3]
+; CHECK-NEXT:    mov z4.d, z5.d[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #12, .LBB13_13
+; CHECK-NEXT:  .LBB13_29: // %cond.store23
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #13, .LBB13_14
+; CHECK-NEXT:  .LBB13_30: // %cond.store25
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #14, .LBB13_15
+; CHECK-NEXT:  .LBB13_31: // %cond.store27
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #15, .LBB13_16
+; CHECK-NEXT:  .LBB13_32: // %cond.store29
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str w8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <16 x i32>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = icmp eq <16 x i32> %vals, zeroinitializer
   call void @llvm.masked.scatter.v16i32(<16 x i32> %vals, <16 x ptr> %ptrs, i32 8, <16 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v32i32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v32i32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    cmpeq p0.s, p0/z, z0.s, #0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1w { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    str d8, [sp, #-16]! // 8-byte Folded Spill
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    .cfi_offset b8, -16
+; CHECK-NEXT:    adrp x8, .LCPI14_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q5, q3, [x0, #64]
+; CHECK-NEXT:    ldr q22, [x8, :lo12:.LCPI14_0]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z5.s, z22.s
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z3.s, z22.s
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z6.s, z2.s[1]
+; CHECK-NEXT:    mov z7.s, z2.s[2]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    mov z16.s, z2.s[3]
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z6.s, z4.s[2]
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    mov z2.s, z4.s[1]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z7.s, z4.s[3]
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z1.s, z22.s
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    mov z6.s, z2.s[2]
+; CHECK-NEXT:    mov z7.s, z2.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z0.s, z22.s
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    ldp q25, q23, [x0]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z6.s, z2.s[2]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z25.s, z22.s
+; CHECK-NEXT:    mov z19.s, z2.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldp q20, q18, [x0, #32]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    mov z6.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    mov z7.s, z6.s[1]
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    mov z7.s, z6.s[2]
+; CHECK-NEXT:    mov z16.s, z6.s[3]
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z23.s, z22.s
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #30
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    mov z16.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w11, w10, #1, #1
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    bfi w11, w12, #2, #1
+; CHECK-NEXT:    mov z17.s, z16.s[1]
+; CHECK-NEXT:    mov z21.s, z16.s[2]
+; CHECK-NEXT:    mov z24.s, z16.s[3]
+; CHECK-NEXT:    fmov w12, s17
+; CHECK-NEXT:    bfi w11, w9, #3, #1
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    bfi w11, w10, #4, #1
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    cmpeq p1.s, p0/z, z20.s, z22.s
+; CHECK-NEXT:    mov z21.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w11, w12, #5, #1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w12, s21
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z24.s, z21.s[1]
+; CHECK-NEXT:    orr w9, w11, w9, lsl #6
+; CHECK-NEXT:    mov z26.s, z21.s[2]
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    mov z27.s, z21.s[3]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z18.s, z22.s
+; CHECK-NEXT:    fmov w12, s19
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z22.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z24.s, z22.s[1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    mov z26.s, z22.s[2]
+; CHECK-NEXT:    mov z27.s, z22.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #31
+; CHECK-NEXT:    ldp q4, q2, [x1, #224]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #11
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    ldp q7, q6, [x1, #192]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q17, q16, [x1, #160]
+; CHECK-NEXT:    ldp q21, q19, [x1, #128]
+; CHECK-NEXT:    ldp q24, q22, [x1, #96]
+; CHECK-NEXT:    ldp q27, q26, [x1, #64]
+; CHECK-NEXT:    ldp q29, q28, [x1, #32]
+; CHECK-NEXT:    ldp q31, q30, [x1]
+; CHECK-NEXT:    tbnz w8, #0, .LBB14_33
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB14_34
+; CHECK-NEXT:  .LBB14_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB14_35
+; CHECK-NEXT:  .LBB14_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB14_36
+; CHECK-NEXT:  .LBB14_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB14_37
+; CHECK-NEXT:  .LBB14_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB14_38
+; CHECK-NEXT:  .LBB14_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB14_39
+; CHECK-NEXT:  .LBB14_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB14_40
+; CHECK-NEXT:  .LBB14_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB14_41
+; CHECK-NEXT:  .LBB14_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB14_42
+; CHECK-NEXT:  .LBB14_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB14_43
+; CHECK-NEXT:  .LBB14_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB14_44
+; CHECK-NEXT:  .LBB14_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB14_45
+; CHECK-NEXT:  .LBB14_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB14_46
+; CHECK-NEXT:  .LBB14_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB14_47
+; CHECK-NEXT:  .LBB14_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB14_48
+; CHECK-NEXT:  .LBB14_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB14_49
+; CHECK-NEXT:  .LBB14_17: // %else32
+; CHECK-NEXT:    tbnz w8, #17, .LBB14_50
+; CHECK-NEXT:  .LBB14_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB14_51
+; CHECK-NEXT:  .LBB14_19: // %else36
+; CHECK-NEXT:    tbnz w8, #19, .LBB14_52
+; CHECK-NEXT:  .LBB14_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB14_53
+; CHECK-NEXT:  .LBB14_21: // %else40
+; CHECK-NEXT:    tbnz w8, #21, .LBB14_54
+; CHECK-NEXT:  .LBB14_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB14_55
+; CHECK-NEXT:  .LBB14_23: // %else44
+; CHECK-NEXT:    tbnz w8, #23, .LBB14_56
+; CHECK-NEXT:  .LBB14_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB14_57
+; CHECK-NEXT:  .LBB14_25: // %else48
+; CHECK-NEXT:    tbnz w8, #25, .LBB14_58
+; CHECK-NEXT:  .LBB14_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB14_59
+; CHECK-NEXT:  .LBB14_27: // %else52
+; CHECK-NEXT:    tbnz w8, #27, .LBB14_60
+; CHECK-NEXT:  .LBB14_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB14_61
+; CHECK-NEXT:  .LBB14_29: // %else56
+; CHECK-NEXT:    tbnz w8, #29, .LBB14_62
+; CHECK-NEXT:  .LBB14_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB14_63
+; CHECK-NEXT:  .LBB14_31: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB14_64
+; CHECK-NEXT:  .LBB14_32: // %else62
+; CHECK-NEXT:    ldr d8, [sp], #16 // 8-byte Folded Reload
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB14_33: // %cond.store
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    fmov x10, d31
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB14_2
+; CHECK-NEXT:  .LBB14_34: // %cond.store1
+; CHECK-NEXT:    mov z8.s, z25.s[1]
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    fmov w9, s8
+; CHECK-NEXT:    fmov x10, d31
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB14_3
+; CHECK-NEXT:  .LBB14_35: // %cond.store3
+; CHECK-NEXT:    mov z31.s, z25.s[2]
+; CHECK-NEXT:    fmov x10, d30
+; CHECK-NEXT:    fmov w9, s31
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB14_4
+; CHECK-NEXT:  .LBB14_36: // %cond.store5
+; CHECK-NEXT:    mov z25.s, z25.s[3]
+; CHECK-NEXT:    mov z30.d, z30.d[1]
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    fmov x10, d30
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB14_5
+; CHECK-NEXT:  .LBB14_37: // %cond.store7
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    fmov x10, d29
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB14_6
+; CHECK-NEXT:  .LBB14_38: // %cond.store9
+; CHECK-NEXT:    mov z25.s, z23.s[1]
+; CHECK-NEXT:    mov z29.d, z29.d[1]
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    fmov x10, d29
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB14_7
+; CHECK-NEXT:  .LBB14_39: // %cond.store11
+; CHECK-NEXT:    mov z25.s, z23.s[2]
+; CHECK-NEXT:    fmov x10, d28
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB14_8
+; CHECK-NEXT:  .LBB14_40: // %cond.store13
+; CHECK-NEXT:    mov z23.s, z23.s[3]
+; CHECK-NEXT:    mov z25.d, z28.d[1]
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    fmov x10, d25
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #8, .LBB14_9
+; CHECK-NEXT:  .LBB14_41: // %cond.store15
+; CHECK-NEXT:    fmov w9, s20
+; CHECK-NEXT:    fmov x10, d27
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #9, .LBB14_10
+; CHECK-NEXT:  .LBB14_42: // %cond.store17
+; CHECK-NEXT:    mov z23.s, z20.s[1]
+; CHECK-NEXT:    mov z25.d, z27.d[1]
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    fmov x10, d25
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #10, .LBB14_11
+; CHECK-NEXT:  .LBB14_43: // %cond.store19
+; CHECK-NEXT:    mov z23.s, z20.s[2]
+; CHECK-NEXT:    fmov x10, d26
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #11, .LBB14_12
+; CHECK-NEXT:  .LBB14_44: // %cond.store21
+; CHECK-NEXT:    mov z20.s, z20.s[3]
+; CHECK-NEXT:    mov z23.d, z26.d[1]
+; CHECK-NEXT:    fmov w9, s20
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #12, .LBB14_13
+; CHECK-NEXT:  .LBB14_45: // %cond.store23
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    fmov x10, d24
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #13, .LBB14_14
+; CHECK-NEXT:  .LBB14_46: // %cond.store25
+; CHECK-NEXT:    mov z20.s, z18.s[1]
+; CHECK-NEXT:    mov z23.d, z24.d[1]
+; CHECK-NEXT:    fmov w9, s20
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #14, .LBB14_15
+; CHECK-NEXT:  .LBB14_47: // %cond.store27
+; CHECK-NEXT:    mov z20.s, z18.s[2]
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    fmov w9, s20
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #15, .LBB14_16
+; CHECK-NEXT:  .LBB14_48: // %cond.store29
+; CHECK-NEXT:    mov z18.s, z18.s[3]
+; CHECK-NEXT:    mov z20.d, z22.d[1]
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #16, .LBB14_17
+; CHECK-NEXT:  .LBB14_49: // %cond.store31
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #17, .LBB14_18
+; CHECK-NEXT:  .LBB14_50: // %cond.store33
+; CHECK-NEXT:    mov z18.s, z5.s[1]
+; CHECK-NEXT:    mov z20.d, z21.d[1]
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #18, .LBB14_19
+; CHECK-NEXT:  .LBB14_51: // %cond.store35
+; CHECK-NEXT:    mov z18.s, z5.s[2]
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #19, .LBB14_20
+; CHECK-NEXT:  .LBB14_52: // %cond.store37
+; CHECK-NEXT:    mov z5.s, z5.s[3]
+; CHECK-NEXT:    mov z18.d, z19.d[1]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #20, .LBB14_21
+; CHECK-NEXT:  .LBB14_53: // %cond.store39
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #21, .LBB14_22
+; CHECK-NEXT:  .LBB14_54: // %cond.store41
+; CHECK-NEXT:    mov z5.s, z3.s[1]
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #22, .LBB14_23
+; CHECK-NEXT:  .LBB14_55: // %cond.store43
+; CHECK-NEXT:    mov z5.s, z3.s[2]
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #23, .LBB14_24
+; CHECK-NEXT:  .LBB14_56: // %cond.store45
+; CHECK-NEXT:    mov z3.s, z3.s[3]
+; CHECK-NEXT:    mov z5.d, z16.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #24, .LBB14_25
+; CHECK-NEXT:  .LBB14_57: // %cond.store47
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #25, .LBB14_26
+; CHECK-NEXT:  .LBB14_58: // %cond.store49
+; CHECK-NEXT:    mov z3.s, z1.s[1]
+; CHECK-NEXT:    mov z5.d, z7.d[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #26, .LBB14_27
+; CHECK-NEXT:  .LBB14_59: // %cond.store51
+; CHECK-NEXT:    mov z3.s, z1.s[2]
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #27, .LBB14_28
+; CHECK-NEXT:  .LBB14_60: // %cond.store53
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    mov z3.d, z6.d[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #28, .LBB14_29
+; CHECK-NEXT:  .LBB14_61: // %cond.store55
+; CHECK-NEXT:    fmov w9, s0
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #29, .LBB14_30
+; CHECK-NEXT:  .LBB14_62: // %cond.store57
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    mov z3.d, z4.d[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #30, .LBB14_31
+; CHECK-NEXT:  .LBB14_63: // %cond.store59
+; CHECK-NEXT:    mov z1.s, z0.s[2]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    str w9, [x10]
+; CHECK-NEXT:    tbz w8, #31, .LBB14_32
+; CHECK-NEXT:  .LBB14_64: // %cond.store61
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str w8, [x9]
+; CHECK-NEXT:    ldr d8, [sp], #16 // 8-byte Folded Reload
 ; CHECK-NEXT:    ret
   %vals = load <32 x i32>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = icmp eq <32 x i32> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32i32(<32 x i32> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 ;
 ; ST1D
 ;
 
 ; Scalarize 1 x i64 scatters
 define void @masked_scatter_v1i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v1i64:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    ldr d0, [x0]
 ; CHECK-NEXT:    fmov x8, d0
 ; CHECK-NEXT:    cbnz x8, .LBB15_2
 ; CHECK-NEXT:  // %bb.1: // %cond.store
 ; CHECK-NEXT:    ldr d1, [x1]
-; CHECK-NEXT:    fmov x8, d1
-; CHECK-NEXT:    str d0, [x8]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str x8, [x9]
 ; CHECK-NEXT:  .LBB15_2: // %else
 ; CHECK-NEXT:    ret
   %vals = load <1 x i64>, ptr %a
   %ptrs = load <1 x ptr>, ptr %b
   %mask = icmp eq <1 x i64> %vals, zeroinitializer
   call void @llvm.masked.scatter.v1i64(<1 x i64> %vals, <1 x ptr> %ptrs, i32 8, <1 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v2i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v2i64:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI16_0
 ; CHECK-NEXT:    ldr q0, [x0]
 ; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI16_0]
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    mov z1.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fmov x8, d2
 ; CHECK-NEXT:    ldr q1, [x1]
-; CHECK-NEXT:    cmeq v2.2d, v0.2d, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z2.d, #0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbnz w9, #0, .LBB16_3
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB16_4
+; CHECK-NEXT:  .LBB16_2: // %else2
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB16_3: // %cond.store
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB16_2
+; CHECK-NEXT:  .LBB16_4: // %cond.store1
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str x8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <2 x i64>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = icmp eq <2 x i64> %vals, zeroinitializer
   call void @llvm.masked.scatter.v2i64(<2 x i64> %vals, <2 x ptr> %ptrs, i32 8, <2 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v4i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v4i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI17_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q1, q0, [x0]
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI17_0]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z1.d, z2.d
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, z2.d
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z4.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ldp q3, q2, [x1]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB17_5
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB17_6
+; CHECK-NEXT:  .LBB17_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB17_7
+; CHECK-NEXT:  .LBB17_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB17_8
+; CHECK-NEXT:  .LBB17_4: // %else6
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB17_5: // %cond.store
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB17_2
+; CHECK-NEXT:  .LBB17_6: // %cond.store1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB17_3
+; CHECK-NEXT:  .LBB17_7: // %cond.store3
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB17_4
+; CHECK-NEXT:  .LBB17_8: // %cond.store5
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str x8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <4 x i64>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = icmp eq <4 x i64> %vals, zeroinitializer
   call void @llvm.masked.scatter.v4i64(<4 x i64> %vals, <4 x ptr> %ptrs, i32 8, <4 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v8i64(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_scatter_v8i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI18_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q4, q2, [x0]
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI18_0]
+; CHECK-NEXT:    ldp q1, q0, [x0, #32]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z4.d, z3.d
+; CHECK-NEXT:    cmpeq p2.d, p0/z, z2.d, z3.d
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.d, p2/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z7.d, z5.d[1]
+; CHECK-NEXT:    fmov x8, d5
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z1.d, z3.d
+; CHECK-NEXT:    mov z5.d, z6.d[1]
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, z3.d
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov x11, d5
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    mov z6.d, z3.d[1]
+; CHECK-NEXT:    mov z3.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z5.d, z3.d[1]
+; CHECK-NEXT:    ldp q7, q6, [x1]
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    ldp q5, q3, [x1, #32]
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbnz w9, #0, .LBB18_9
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB18_10
+; CHECK-NEXT:  .LBB18_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB18_11
+; CHECK-NEXT:  .LBB18_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB18_12
+; CHECK-NEXT:  .LBB18_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB18_13
+; CHECK-NEXT:  .LBB18_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB18_14
+; CHECK-NEXT:  .LBB18_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB18_15
+; CHECK-NEXT:  .LBB18_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB18_16
+; CHECK-NEXT:  .LBB18_8: // %else14
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB18_9: // %cond.store
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB18_2
+; CHECK-NEXT:  .LBB18_10: // %cond.store1
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB18_3
+; CHECK-NEXT:  .LBB18_11: // %cond.store3
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB18_4
+; CHECK-NEXT:  .LBB18_12: // %cond.store5
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    mov z4.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB18_5
+; CHECK-NEXT:  .LBB18_13: // %cond.store7
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB18_6
+; CHECK-NEXT:  .LBB18_14: // %cond.store9
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z2.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB18_7
+; CHECK-NEXT:  .LBB18_15: // %cond.store11
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB18_8
+; CHECK-NEXT:  .LBB18_16: // %cond.store13
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    mov z1.d, z3.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str x8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %vals = load <8 x i64>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = icmp eq <8 x i64> %vals, zeroinitializer
   call void @llvm.masked.scatter.v8i64(<8 x i64> %vals, <8 x ptr> %ptrs, i32 8, <8 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v16i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v16i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI19_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q18, q17, [x0]
+; CHECK-NEXT:    ldr q20, [x8, :lo12:.LCPI19_0]
+; CHECK-NEXT:    ldp q7, q6, [x0, #32]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z18.d, z20.d
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z17.d, z20.d
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    mov z2.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z7.d, z20.d
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x11, d2
+; CHECK-NEXT:    bfi w8, w10, #1, #1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    mov z2.d, z3.d[1]
+; CHECK-NEXT:    bfi w8, w9, #2, #1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z6.d, z20.d
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q5, q3, [x0, #64]
+; CHECK-NEXT:    bfi w8, w10, #4, #1
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    fmov x11, d4
+; CHECK-NEXT:    bfi w8, w9, #5, #1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z5.d, z20.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z3.d, z20.d
+; CHECK-NEXT:    orr w8, w8, w10, lsl #6
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x11, d2
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #9
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z1.d, z20.d
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, z20.d
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    mov z21.d, z2.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z20.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q4, q2, [x1, #96]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.d, z20.d[1]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #12
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldp q19, q16, [x1, #64]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #13
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    ldp q21, q20, [x1, #32]
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    ldp q23, q22, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB19_17
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB19_18
+; CHECK-NEXT:  .LBB19_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB19_19
+; CHECK-NEXT:  .LBB19_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB19_20
+; CHECK-NEXT:  .LBB19_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB19_21
+; CHECK-NEXT:  .LBB19_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB19_22
+; CHECK-NEXT:  .LBB19_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB19_23
+; CHECK-NEXT:  .LBB19_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB19_24
+; CHECK-NEXT:  .LBB19_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB19_25
+; CHECK-NEXT:  .LBB19_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB19_26
+; CHECK-NEXT:  .LBB19_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB19_27
+; CHECK-NEXT:  .LBB19_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB19_28
+; CHECK-NEXT:  .LBB19_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB19_29
+; CHECK-NEXT:  .LBB19_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB19_30
+; CHECK-NEXT:  .LBB19_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB19_31
+; CHECK-NEXT:  .LBB19_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB19_32
+; CHECK-NEXT:  .LBB19_16: // %else30
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB19_17: // %cond.store
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB19_2
+; CHECK-NEXT:  .LBB19_18: // %cond.store1
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB19_3
+; CHECK-NEXT:  .LBB19_19: // %cond.store3
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB19_4
+; CHECK-NEXT:  .LBB19_20: // %cond.store5
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    mov z18.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB19_5
+; CHECK-NEXT:  .LBB19_21: // %cond.store7
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB19_6
+; CHECK-NEXT:  .LBB19_22: // %cond.store9
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    mov z17.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB19_7
+; CHECK-NEXT:  .LBB19_23: // %cond.store11
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB19_8
+; CHECK-NEXT:  .LBB19_24: // %cond.store13
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    mov z7.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #8, .LBB19_9
+; CHECK-NEXT:  .LBB19_25: // %cond.store15
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    fmov x10, d19
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #9, .LBB19_10
+; CHECK-NEXT:  .LBB19_26: // %cond.store17
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z6.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #10, .LBB19_11
+; CHECK-NEXT:  .LBB19_27: // %cond.store19
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #11, .LBB19_12
+; CHECK-NEXT:  .LBB19_28: // %cond.store21
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    mov z5.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #12, .LBB19_13
+; CHECK-NEXT:  .LBB19_29: // %cond.store23
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #13, .LBB19_14
+; CHECK-NEXT:  .LBB19_30: // %cond.store25
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z3.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #14, .LBB19_15
+; CHECK-NEXT:  .LBB19_31: // %cond.store27
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #15, .LBB19_16
+; CHECK-NEXT:  .LBB19_32: // %cond.store29
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str x8, [x9]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <16 x i64>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = icmp eq <16 x i64> %vals, zeroinitializer
   call void @llvm.masked.scatter.v16i64(<16 x i64> %vals, <16 x ptr> %ptrs, i32 8, <16 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v32i64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v32i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    cmpeq p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z1.d]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #80
+; CHECK-NEXT:    .cfi_def_cfa_offset 80
+; CHECK-NEXT:    stp d15, d14, [sp, #16] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d13, d12, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d11, d10, [sp, #48] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #64] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -24
+; CHECK-NEXT:    .cfi_offset b11, -32
+; CHECK-NEXT:    .cfi_offset b12, -40
+; CHECK-NEXT:    .cfi_offset b13, -48
+; CHECK-NEXT:    .cfi_offset b14, -56
+; CHECK-NEXT:    .cfi_offset b15, -64
+; CHECK-NEXT:    adrp x8, .LCPI20_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q20, q7, [x0, #128]
+; CHECK-NEXT:    ldr q9, [x8, :lo12:.LCPI20_0]
+; CHECK-NEXT:    ldp q19, q16, [x0, #160]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z20.d, z9.d
+; CHECK-NEXT:    mov z0.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z7.d, z9.d
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    mov z1.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z3.d, z1.d[1]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z19.d, z9.d
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z16.d, z9.d
+; CHECK-NEXT:    mov z17.d, z5.d[1]
+; CHECK-NEXT:    ldp q6, q4, [x0, #192]
+; CHECK-NEXT:    bfi w8, w9, #18, #1
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w10, #19, #1
+; CHECK-NEXT:    fmov x11, d5
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    mov z22.d, z5.d[1]
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z6.d, z9.d
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    ldp q2, q0, [x0, #224]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z22.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z4.d, z9.d
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    mov z23.d, z22.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z22.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x11, d22
+; CHECK-NEXT:    mov z28.d, z22.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z2.d, z9.d
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    ldp q31, q30, [x0]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z31.d, z9.d
+; CHECK-NEXT:    fmov x10, d28
+; CHECK-NEXT:    mov z29.d, z28.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q27, q26, [x0, #32]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d29
+; CHECK-NEXT:    mov z29.d, z28.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z30.d, z9.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldp q25, q24, [x0, #64]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    fmov x10, d28
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z27.d, z9.d
+; CHECK-NEXT:    mov z29.d, z28.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov x11, d29
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z26.d, z9.d
+; CHECK-NEXT:    mov z29.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z8.d, z28.d[1]
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    fmov x12, d29
+; CHECK-NEXT:    bfi w10, w11, #3, #1
+; CHECK-NEXT:    fmov x11, d8
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z0.d, z9.d
+; CHECK-NEXT:    bfi w10, w9, #4, #1
+; CHECK-NEXT:    mov z28.d, z29.d[1]
+; CHECK-NEXT:    and w9, w12, #0x1
+; CHECK-NEXT:    mov z29.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w10, w11, #5, #1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z25.d, z9.d
+; CHECK-NEXT:    orr w9, w10, w9, lsl #6
+; CHECK-NEXT:    fmov x10, d28
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z24.d, z9.d
+; CHECK-NEXT:    fmov x11, d28
+; CHECK-NEXT:    mov z10.d, z28.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q21, q18, [x0, #96]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov x10, d10
+; CHECK-NEXT:    fmov x12, d29
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov x11, d28
+; CHECK-NEXT:    mov z8.d, z29.d[1]
+; CHECK-NEXT:    mov z10.d, z28.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    cmpeq p1.d, p0/z, z21.d, z9.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov x10, d8
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    fmov x11, d10
+; CHECK-NEXT:    mov z8.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    mov z11.d, z8.d[1]
+; CHECK-NEXT:    fmov x12, d8
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #31
+; CHECK-NEXT:    fmov x10, d11
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z18.d, z9.d
+; CHECK-NEXT:    mov z9.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov x11, d9
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w12, lsl #12
+; CHECK-NEXT:    mov z11.d, z9.d[1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov x11, d11
+; CHECK-NEXT:    ldp q3, q1, [x1, #224]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q17, q5, [x1, #192]
+; CHECK-NEXT:    ldp q23, q22, [x1, #160]
+; CHECK-NEXT:    ldp q29, q28, [x1, #128]
+; CHECK-NEXT:    ldp q10, q8, [x1, #96]
+; CHECK-NEXT:    ldp q11, q9, [x1, #64]
+; CHECK-NEXT:    ldp q13, q12, [x1, #32]
+; CHECK-NEXT:    ldp q15, q14, [x1]
+; CHECK-NEXT:    tbnz w8, #0, .LBB20_34
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB20_35
+; CHECK-NEXT:  .LBB20_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB20_36
+; CHECK-NEXT:  .LBB20_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB20_37
+; CHECK-NEXT:  .LBB20_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB20_38
+; CHECK-NEXT:  .LBB20_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB20_39
+; CHECK-NEXT:  .LBB20_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB20_40
+; CHECK-NEXT:  .LBB20_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB20_41
+; CHECK-NEXT:  .LBB20_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB20_42
+; CHECK-NEXT:  .LBB20_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB20_43
+; CHECK-NEXT:  .LBB20_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB20_44
+; CHECK-NEXT:  .LBB20_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB20_45
+; CHECK-NEXT:  .LBB20_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB20_46
+; CHECK-NEXT:  .LBB20_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB20_47
+; CHECK-NEXT:  .LBB20_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB20_48
+; CHECK-NEXT:  .LBB20_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB20_49
+; CHECK-NEXT:  .LBB20_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB20_50
+; CHECK-NEXT:  .LBB20_17: // %else32
+; CHECK-NEXT:    tbnz w8, #17, .LBB20_51
+; CHECK-NEXT:  .LBB20_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB20_52
+; CHECK-NEXT:  .LBB20_19: // %else36
+; CHECK-NEXT:    tbnz w8, #19, .LBB20_53
+; CHECK-NEXT:  .LBB20_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB20_54
+; CHECK-NEXT:  .LBB20_21: // %else40
+; CHECK-NEXT:    tbnz w8, #21, .LBB20_55
+; CHECK-NEXT:  .LBB20_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB20_56
+; CHECK-NEXT:  .LBB20_23: // %else44
+; CHECK-NEXT:    tbnz w8, #23, .LBB20_57
+; CHECK-NEXT:  .LBB20_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB20_58
+; CHECK-NEXT:  .LBB20_25: // %else48
+; CHECK-NEXT:    tbnz w8, #25, .LBB20_59
+; CHECK-NEXT:  .LBB20_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB20_60
+; CHECK-NEXT:  .LBB20_27: // %else52
+; CHECK-NEXT:    tbnz w8, #27, .LBB20_61
+; CHECK-NEXT:  .LBB20_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB20_62
+; CHECK-NEXT:  .LBB20_29: // %else56
+; CHECK-NEXT:    tbnz w8, #29, .LBB20_63
+; CHECK-NEXT:  .LBB20_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB20_64
+; CHECK-NEXT:  .LBB20_31: // %else60
+; CHECK-NEXT:    tbz w8, #31, .LBB20_33
+; CHECK-NEXT:  .LBB20_32: // %cond.store61
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str x8, [x9]
+; CHECK-NEXT:  .LBB20_33: // %else62
+; CHECK-NEXT:    ldp d9, d8, [sp, #64] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d11, d10, [sp, #48] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d13, d12, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d15, d14, [sp, #16] // 16-byte Folded Reload
+; CHECK-NEXT:    add sp, sp, #80
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB20_34: // %cond.store
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    fmov x10, d15
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #1, .LBB20_2
+; CHECK-NEXT:  .LBB20_35: // %cond.store1
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    mov z15.d, z15.d[1]
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    fmov x10, d15
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #2, .LBB20_3
+; CHECK-NEXT:  .LBB20_36: // %cond.store3
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    fmov x10, d14
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #3, .LBB20_4
+; CHECK-NEXT:  .LBB20_37: // %cond.store5
+; CHECK-NEXT:    mov z30.d, z30.d[1]
+; CHECK-NEXT:    mov z31.d, z14.d[1]
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    fmov x10, d31
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #4, .LBB20_5
+; CHECK-NEXT:  .LBB20_38: // %cond.store7
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    fmov x10, d13
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #5, .LBB20_6
+; CHECK-NEXT:  .LBB20_39: // %cond.store9
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    mov z30.d, z13.d[1]
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    fmov x10, d30
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #6, .LBB20_7
+; CHECK-NEXT:  .LBB20_40: // %cond.store11
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    fmov x10, d12
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #7, .LBB20_8
+; CHECK-NEXT:  .LBB20_41: // %cond.store13
+; CHECK-NEXT:    mov z26.d, z26.d[1]
+; CHECK-NEXT:    mov z27.d, z12.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    fmov x10, d27
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #8, .LBB20_9
+; CHECK-NEXT:  .LBB20_42: // %cond.store15
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    fmov x10, d11
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #9, .LBB20_10
+; CHECK-NEXT:  .LBB20_43: // %cond.store17
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    mov z26.d, z11.d[1]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    fmov x10, d26
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #10, .LBB20_11
+; CHECK-NEXT:  .LBB20_44: // %cond.store19
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    fmov x10, d9
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #11, .LBB20_12
+; CHECK-NEXT:  .LBB20_45: // %cond.store21
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    mov z25.d, z9.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    fmov x10, d25
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #12, .LBB20_13
+; CHECK-NEXT:  .LBB20_46: // %cond.store23
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    fmov x10, d10
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #13, .LBB20_14
+; CHECK-NEXT:  .LBB20_47: // %cond.store25
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    mov z24.d, z10.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    fmov x10, d24
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #14, .LBB20_15
+; CHECK-NEXT:  .LBB20_48: // %cond.store27
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    fmov x10, d8
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #15, .LBB20_16
+; CHECK-NEXT:  .LBB20_49: // %cond.store29
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    mov z21.d, z8.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #16, .LBB20_17
+; CHECK-NEXT:  .LBB20_50: // %cond.store31
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    fmov x10, d29
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #17, .LBB20_18
+; CHECK-NEXT:  .LBB20_51: // %cond.store33
+; CHECK-NEXT:    mov z18.d, z20.d[1]
+; CHECK-NEXT:    mov z20.d, z29.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #18, .LBB20_19
+; CHECK-NEXT:  .LBB20_52: // %cond.store35
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    fmov x10, d28
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #19, .LBB20_20
+; CHECK-NEXT:  .LBB20_53: // %cond.store37
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    mov z18.d, z28.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #20, .LBB20_21
+; CHECK-NEXT:  .LBB20_54: // %cond.store39
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #21, .LBB20_22
+; CHECK-NEXT:  .LBB20_55: // %cond.store41
+; CHECK-NEXT:    mov z7.d, z19.d[1]
+; CHECK-NEXT:    mov z18.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    fmov x10, d18
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #22, .LBB20_23
+; CHECK-NEXT:  .LBB20_56: // %cond.store43
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #23, .LBB20_24
+; CHECK-NEXT:  .LBB20_57: // %cond.store45
+; CHECK-NEXT:    mov z7.d, z16.d[1]
+; CHECK-NEXT:    mov z16.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    fmov x10, d16
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #24, .LBB20_25
+; CHECK-NEXT:  .LBB20_58: // %cond.store47
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #25, .LBB20_26
+; CHECK-NEXT:  .LBB20_59: // %cond.store49
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    mov z7.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    fmov x10, d7
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #26, .LBB20_27
+; CHECK-NEXT:  .LBB20_60: // %cond.store51
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #27, .LBB20_28
+; CHECK-NEXT:  .LBB20_61: // %cond.store53
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #28, .LBB20_29
+; CHECK-NEXT:  .LBB20_62: // %cond.store55
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #29, .LBB20_30
+; CHECK-NEXT:  .LBB20_63: // %cond.store57
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbz w8, #30, .LBB20_31
+; CHECK-NEXT:  .LBB20_64: // %cond.store59
+; CHECK-NEXT:    fmov x9, d0
+; CHECK-NEXT:    fmov x10, d1
+; CHECK-NEXT:    str x9, [x10]
+; CHECK-NEXT:    tbnz w8, #31, .LBB20_32
+; CHECK-NEXT:    b .LBB20_33
   %vals = load <32 x i64>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = icmp eq <32 x i64> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32i64(<32 x i64> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 ;
 ; ST1H (float)
 ;
 
 define void @masked_scatter_v2f16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v2f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr s1, [x0]
-; CHECK-NEXT:    movi d0, #0000000000000000
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    fcmeq v2.4h, v1.4h, #0.0
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI21_0
+; CHECK-NEXT:    ldr s0, [x0]
+; CHECK-NEXT:    ptrue p0.h, vl4
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI21_0]
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z1.h, p0/z, #-1 // =0xffffffffffffffff
 ; CHECK-NEXT:    uunpklo z1.s, z1.h
-; CHECK-NEXT:    sshll v2.4s, v2.4h, #0
+; CHECK-NEXT:    mov z2.s, z1.s[1]
+; CHECK-NEXT:    fmov w9, s1
 ; CHECK-NEXT:    fmov w8, s2
-; CHECK-NEXT:    mov w9, v2.s[1]
-; CHECK-NEXT:    ldr q2, [x1]
-; CHECK-NEXT:    mov v0.h[0], w8
-; CHECK-NEXT:    mov v0.h[1], w9
-; CHECK-NEXT:    shl v0.4h, v0.4h, #15
-; CHECK-NEXT:    cmlt v0.4h, v0.4h, #0
-; CHECK-NEXT:    sunpklo z0.s, z0.h
-; CHECK-NEXT:    sunpklo z0.d, z0.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z0.d, #0
-; CHECK-NEXT:    uunpklo z0.d, z1.s
-; CHECK-NEXT:    st1h { z0.d }, p0, [z2.d]
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbnz w9, #0, .LBB21_3
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB21_4
+; CHECK-NEXT:  .LBB21_2: // %else2
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB21_3: // %cond.store
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB21_2
+; CHECK-NEXT:  .LBB21_4: // %cond.store1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <2 x half>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = fcmp oeq <2 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v2f16(<2 x half> %vals, <2 x ptr> %ptrs, i32 8, <2 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v4f16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v4f16:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI22_0
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq v2.4h, v0.4h, #0.0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    sunpklo z2.s, z2.h
-; CHECK-NEXT:    sunpklo z2.d, z2.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z2.d, #0
-; CHECK-NEXT:    st1h { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    ptrue p0.h, vl4
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI22_0]
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z1.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.h, z1.h[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z2.h, z1.h[2]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z1.h, z1.h[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    ldp q2, q1, [x1]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB22_5
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB22_6
+; CHECK-NEXT:  .LBB22_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB22_7
+; CHECK-NEXT:  .LBB22_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB22_8
+; CHECK-NEXT:  .LBB22_4: // %else6
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB22_5: // %cond.store
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB22_2
+; CHECK-NEXT:  .LBB22_6: // %cond.store1
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[1]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB22_3
+; CHECK-NEXT:  .LBB22_7: // %cond.store3
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB22_4
+; CHECK-NEXT:  .LBB22_8: // %cond.store5
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <4 x half>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = fcmp oeq <4 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v4f16(<4 x half> %vals, <4 x ptr> %ptrs, i32 8, <4 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v8f16(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_scatter_v8f16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI23_0
+; CHECK-NEXT:    ldr q0, [x0]
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI23_0]
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z1.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.h, z1.h[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    mov z3.h, z1.h[2]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z4.h, z1.h[3]
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    mov z5.h, z1.h[4]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s5
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    mov z6.h, z1.h[5]
+; CHECK-NEXT:    mov z2.h, z1.h[6]
+; CHECK-NEXT:    bfi w9, w11, #3, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    bfi w9, w8, #4, #1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z1.h, z1.h[7]
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    ldp q2, q1, [x1, #32]
+; CHECK-NEXT:    orr w8, w9, w8, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    ldp q4, q3, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB23_9
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB23_10
+; CHECK-NEXT:  .LBB23_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB23_11
+; CHECK-NEXT:  .LBB23_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB23_12
+; CHECK-NEXT:  .LBB23_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB23_13
+; CHECK-NEXT:  .LBB23_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB23_14
+; CHECK-NEXT:  .LBB23_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB23_15
+; CHECK-NEXT:  .LBB23_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB23_16
+; CHECK-NEXT:  .LBB23_8: // %else14
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB23_9: // %cond.store
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB23_2
+; CHECK-NEXT:  .LBB23_10: // %cond.store1
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z4.h, z0.h[1]
+; CHECK-NEXT:    str h4, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB23_3
+; CHECK-NEXT:  .LBB23_11: // %cond.store3
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z4.h, z0.h[2]
+; CHECK-NEXT:    str h4, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB23_4
+; CHECK-NEXT:  .LBB23_12: // %cond.store5
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z3.h, z0.h[3]
+; CHECK-NEXT:    str h3, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB23_5
+; CHECK-NEXT:  .LBB23_13: // %cond.store7
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z3.h, z0.h[4]
+; CHECK-NEXT:    str h3, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB23_6
+; CHECK-NEXT:  .LBB23_14: // %cond.store9
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[5]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB23_7
+; CHECK-NEXT:  .LBB23_15: // %cond.store11
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.h, z0.h[6]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB23_8
+; CHECK-NEXT:  .LBB23_16: // %cond.store13
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %vals = load <8 x half>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = fcmp oeq <8 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v8f16(<8 x half> %vals, <8 x ptr> %ptrs, i32 8, <8 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v16f16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v16f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl16
-; CHECK-NEXT:    ptrue p1.d, vl16
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1h { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI24_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q2, q0, [x0]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI24_0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z2.h, z1.h
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, z1.h
+; CHECK-NEXT:    mov z3.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z1.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z4.h, z3.h[1]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z5.h, z3.h[2]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    mov z6.h, z3.h[3]
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    mov z7.h, z3.h[4]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    mov z16.h, z3.h[5]
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s7
+; CHECK-NEXT:    mov z17.h, z3.h[6]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    mov z18.h, z3.h[7]
+; CHECK-NEXT:    bfi w9, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    bfi w9, w8, #4, #1
+; CHECK-NEXT:    fmov w8, s18
+; CHECK-NEXT:    bfi w9, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z3.h, z1.h[1]
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z4.h, z1.h[2]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #6
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    orr w8, w9, w8, lsl #7
+; CHECK-NEXT:    orr w8, w8, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    mov z5.h, z1.h[3]
+; CHECK-NEXT:    mov z6.h, z1.h[4]
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z7.h, z1.h[5]
+; CHECK-NEXT:    mov z16.h, z1.h[6]
+; CHECK-NEXT:    mov z17.h, z1.h[7]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #9
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    ldp q3, q1, [x1, #96]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    ldp q5, q4, [x1, #64]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    ldp q7, q6, [x1, #32]
+; CHECK-NEXT:    ldp q17, q16, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB24_17
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB24_18
+; CHECK-NEXT:  .LBB24_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB24_19
+; CHECK-NEXT:  .LBB24_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB24_20
+; CHECK-NEXT:  .LBB24_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB24_21
+; CHECK-NEXT:  .LBB24_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB24_22
+; CHECK-NEXT:  .LBB24_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB24_23
+; CHECK-NEXT:  .LBB24_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB24_24
+; CHECK-NEXT:  .LBB24_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB24_25
+; CHECK-NEXT:  .LBB24_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB24_26
+; CHECK-NEXT:  .LBB24_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB24_27
+; CHECK-NEXT:  .LBB24_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB24_28
+; CHECK-NEXT:  .LBB24_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB24_29
+; CHECK-NEXT:  .LBB24_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB24_30
+; CHECK-NEXT:  .LBB24_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB24_31
+; CHECK-NEXT:  .LBB24_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB24_32
+; CHECK-NEXT:  .LBB24_16: // %else30
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB24_17: // %cond.store
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB24_2
+; CHECK-NEXT:  .LBB24_18: // %cond.store1
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.h, z2.h[1]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB24_3
+; CHECK-NEXT:  .LBB24_19: // %cond.store3
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z17.h, z2.h[2]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB24_4
+; CHECK-NEXT:  .LBB24_20: // %cond.store5
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.h, z2.h[3]
+; CHECK-NEXT:    str h16, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB24_5
+; CHECK-NEXT:  .LBB24_21: // %cond.store7
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z16.h, z2.h[4]
+; CHECK-NEXT:    str h16, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB24_6
+; CHECK-NEXT:  .LBB24_22: // %cond.store9
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z7.h, z2.h[5]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB24_7
+; CHECK-NEXT:  .LBB24_23: // %cond.store11
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z7.h, z2.h[6]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB24_8
+; CHECK-NEXT:  .LBB24_24: // %cond.store13
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    mov z2.h, z2.h[7]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB24_9
+; CHECK-NEXT:  .LBB24_25: // %cond.store15
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB24_10
+; CHECK-NEXT:  .LBB24_26: // %cond.store17
+; CHECK-NEXT:    mov z2.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[1]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB24_11
+; CHECK-NEXT:  .LBB24_27: // %cond.store19
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z2.h, z0.h[2]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB24_12
+; CHECK-NEXT:  .LBB24_28: // %cond.store21
+; CHECK-NEXT:    mov z2.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[3]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB24_13
+; CHECK-NEXT:  .LBB24_29: // %cond.store23
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z2.h, z0.h[4]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB24_14
+; CHECK-NEXT:  .LBB24_30: // %cond.store25
+; CHECK-NEXT:    mov z2.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[5]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB24_15
+; CHECK-NEXT:  .LBB24_31: // %cond.store27
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.h, z0.h[6]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB24_16
+; CHECK-NEXT:  .LBB24_32: // %cond.store29
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <16 x half>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = fcmp oeq <16 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v16f16(<16 x half> %vals, <16 x ptr> %ptrs, i32 8, <16 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v32f16(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v32f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1h { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI25_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q1, q0, [x0, #32]
+; CHECK-NEXT:    ldr q7, [x8, :lo12:.LCPI25_0]
+; CHECK-NEXT:    ldp q17, q6, [x0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z1.h, z7.h
+; CHECK-NEXT:    mov z2.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z0.h, z7.h
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z3.h, z2.h[1]
+; CHECK-NEXT:    mov z4.h, z2.h[2]
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z5.h, z2.h[3]
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z16.h, z2.h[4]
+; CHECK-NEXT:    mov z19.h, z2.h[6]
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z18.h, z2.h[5]
+; CHECK-NEXT:    mov z20.h, z2.h[7]
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s20
+; CHECK-NEXT:    bfi w8, w12, #20, #1
+; CHECK-NEXT:    mov z2.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    bfi w8, w9, #21, #1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z3.h, z2.h[1]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #22
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    mov z4.h, z2.h[2]
+; CHECK-NEXT:    mov z5.h, z2.h[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.h, z2.h[4]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    mov z18.h, z2.h[5]
+; CHECK-NEXT:    mov z19.h, z2.h[6]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z17.h, z7.h
+; CHECK-NEXT:    mov z20.h, z2.h[7]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.h, z4.h[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    mov z16.h, z4.h[3]
+; CHECK-NEXT:    mov z18.h, z4.h[4]
+; CHECK-NEXT:    fmov w12, s16
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s19
+; CHECK-NEXT:    mov z21.h, z4.h[6]
+; CHECK-NEXT:    mov z19.h, z4.h[5]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.h, z4.h[2]
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z22.h, z4.h[7]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #30
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z6.h, z7.h
+; CHECK-NEXT:    mov z7.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z16.h, z7.h[1]
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    bfi w10, w11, #2, #1
+; CHECK-NEXT:    fmov w11, s20
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    fmov w12, s19
+; CHECK-NEXT:    bfi w10, w9, #4, #1
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    bfi w10, w12, #5, #1
+; CHECK-NEXT:    mov z18.h, z7.h[2]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z19.h, z7.h[3]
+; CHECK-NEXT:    mov z20.h, z7.h[4]
+; CHECK-NEXT:    mov z21.h, z7.h[5]
+; CHECK-NEXT:    orr w9, w10, w9, lsl #6
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    mov z22.h, z7.h[6]
+; CHECK-NEXT:    mov z23.h, z7.h[7]
+; CHECK-NEXT:    ldp q3, q2, [x1, #224]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    orr w9, w9, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s18
+; CHECK-NEXT:    ldp q5, q4, [x1, #192]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #9
+; CHECK-NEXT:    fmov w11, s19
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s21
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    ldp q16, q7, [x1, #160]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    ldp q19, q18, [x1, #128]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q21, q20, [x1, #96]
+; CHECK-NEXT:    ldp q23, q22, [x1, #64]
+; CHECK-NEXT:    ldp q25, q24, [x1, #32]
+; CHECK-NEXT:    ldp q27, q26, [x1]
+; CHECK-NEXT:    tbnz w8, #0, .LBB25_33
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB25_34
+; CHECK-NEXT:  .LBB25_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB25_35
+; CHECK-NEXT:  .LBB25_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB25_36
+; CHECK-NEXT:  .LBB25_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB25_37
+; CHECK-NEXT:  .LBB25_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB25_38
+; CHECK-NEXT:  .LBB25_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB25_39
+; CHECK-NEXT:  .LBB25_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB25_40
+; CHECK-NEXT:  .LBB25_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB25_41
+; CHECK-NEXT:  .LBB25_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB25_42
+; CHECK-NEXT:  .LBB25_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB25_43
+; CHECK-NEXT:  .LBB25_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB25_44
+; CHECK-NEXT:  .LBB25_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB25_45
+; CHECK-NEXT:  .LBB25_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB25_46
+; CHECK-NEXT:  .LBB25_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB25_47
+; CHECK-NEXT:  .LBB25_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB25_48
+; CHECK-NEXT:  .LBB25_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB25_49
+; CHECK-NEXT:  .LBB25_17: // %else32
+; CHECK-NEXT:    tbnz w8, #17, .LBB25_50
+; CHECK-NEXT:  .LBB25_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB25_51
+; CHECK-NEXT:  .LBB25_19: // %else36
+; CHECK-NEXT:    tbnz w8, #19, .LBB25_52
+; CHECK-NEXT:  .LBB25_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB25_53
+; CHECK-NEXT:  .LBB25_21: // %else40
+; CHECK-NEXT:    tbnz w8, #21, .LBB25_54
+; CHECK-NEXT:  .LBB25_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB25_55
+; CHECK-NEXT:  .LBB25_23: // %else44
+; CHECK-NEXT:    tbnz w8, #23, .LBB25_56
+; CHECK-NEXT:  .LBB25_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB25_57
+; CHECK-NEXT:  .LBB25_25: // %else48
+; CHECK-NEXT:    tbnz w8, #25, .LBB25_58
+; CHECK-NEXT:  .LBB25_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB25_59
+; CHECK-NEXT:  .LBB25_27: // %else52
+; CHECK-NEXT:    tbnz w8, #27, .LBB25_60
+; CHECK-NEXT:  .LBB25_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB25_61
+; CHECK-NEXT:  .LBB25_29: // %else56
+; CHECK-NEXT:    tbnz w8, #29, .LBB25_62
+; CHECK-NEXT:  .LBB25_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB25_63
+; CHECK-NEXT:  .LBB25_31: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB25_64
+; CHECK-NEXT:  .LBB25_32: // %else62
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB25_33: // %cond.store
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB25_2
+; CHECK-NEXT:  .LBB25_34: // %cond.store1
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    mov z27.h, z17.h[1]
+; CHECK-NEXT:    str h27, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB25_3
+; CHECK-NEXT:  .LBB25_35: // %cond.store3
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    mov z27.h, z17.h[2]
+; CHECK-NEXT:    str h27, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB25_4
+; CHECK-NEXT:  .LBB25_36: // %cond.store5
+; CHECK-NEXT:    mov z26.d, z26.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    mov z26.h, z17.h[3]
+; CHECK-NEXT:    str h26, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB25_5
+; CHECK-NEXT:  .LBB25_37: // %cond.store7
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    mov z26.h, z17.h[4]
+; CHECK-NEXT:    str h26, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB25_6
+; CHECK-NEXT:  .LBB25_38: // %cond.store9
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    mov z25.h, z17.h[5]
+; CHECK-NEXT:    str h25, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB25_7
+; CHECK-NEXT:  .LBB25_39: // %cond.store11
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    mov z25.h, z17.h[6]
+; CHECK-NEXT:    str h25, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB25_8
+; CHECK-NEXT:  .LBB25_40: // %cond.store13
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    mov z17.h, z17.h[7]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB25_9
+; CHECK-NEXT:  .LBB25_41: // %cond.store15
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB25_10
+; CHECK-NEXT:  .LBB25_42: // %cond.store17
+; CHECK-NEXT:    mov z17.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.h, z6.h[1]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB25_11
+; CHECK-NEXT:  .LBB25_43: // %cond.store19
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z17.h, z6.h[2]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB25_12
+; CHECK-NEXT:  .LBB25_44: // %cond.store21
+; CHECK-NEXT:    mov z17.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.h, z6.h[3]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB25_13
+; CHECK-NEXT:  .LBB25_45: // %cond.store23
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z17.h, z6.h[4]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB25_14
+; CHECK-NEXT:  .LBB25_46: // %cond.store25
+; CHECK-NEXT:    mov z17.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.h, z6.h[5]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB25_15
+; CHECK-NEXT:  .LBB25_47: // %cond.store27
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z17.h, z6.h[6]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB25_16
+; CHECK-NEXT:  .LBB25_48: // %cond.store29
+; CHECK-NEXT:    mov z17.d, z20.d[1]
+; CHECK-NEXT:    mov z6.h, z6.h[7]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    tbz w8, #16, .LBB25_17
+; CHECK-NEXT:  .LBB25_49: // %cond.store31
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #17, .LBB25_18
+; CHECK-NEXT:  .LBB25_50: // %cond.store33
+; CHECK-NEXT:    mov z6.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.h, z1.h[1]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    tbz w8, #18, .LBB25_19
+; CHECK-NEXT:  .LBB25_51: // %cond.store35
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z6.h, z1.h[2]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB25_20
+; CHECK-NEXT:  .LBB25_52: // %cond.store37
+; CHECK-NEXT:    mov z6.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.h, z1.h[3]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    tbz w8, #20, .LBB25_21
+; CHECK-NEXT:  .LBB25_53: // %cond.store39
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z6.h, z1.h[4]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    tbz w8, #21, .LBB25_22
+; CHECK-NEXT:  .LBB25_54: // %cond.store41
+; CHECK-NEXT:    mov z6.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.h, z1.h[5]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    tbz w8, #22, .LBB25_23
+; CHECK-NEXT:  .LBB25_55: // %cond.store43
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z6.h, z1.h[6]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB25_24
+; CHECK-NEXT:  .LBB25_56: // %cond.store45
+; CHECK-NEXT:    mov z6.d, z7.d[1]
+; CHECK-NEXT:    mov z1.h, z1.h[7]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #24, .LBB25_25
+; CHECK-NEXT:  .LBB25_57: // %cond.store47
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:    tbz w8, #25, .LBB25_26
+; CHECK-NEXT:  .LBB25_58: // %cond.store49
+; CHECK-NEXT:    mov z1.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.h, z0.h[1]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #26, .LBB25_27
+; CHECK-NEXT:  .LBB25_59: // %cond.store51
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z1.h, z0.h[2]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB25_28
+; CHECK-NEXT:  .LBB25_60: // %cond.store53
+; CHECK-NEXT:    mov z1.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.h, z0.h[3]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #28, .LBB25_29
+; CHECK-NEXT:  .LBB25_61: // %cond.store55
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z1.h, z0.h[4]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #29, .LBB25_30
+; CHECK-NEXT:  .LBB25_62: // %cond.store57
+; CHECK-NEXT:    mov z1.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.h, z0.h[5]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB25_31
+; CHECK-NEXT:  .LBB25_63: // %cond.store59
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z1.h, z0.h[6]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #31, .LBB25_32
+; CHECK-NEXT:  .LBB25_64: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <32 x half>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = fcmp oeq <32 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 ;
 ; ST1W (float)
 ;
 
 define void @masked_scatter_v2f32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v2f32:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI26_0
 ; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl2
-; CHECK-NEXT:    ldr q2, [x1]
-; CHECK-NEXT:    fcmeq v1.2s, v0.2s, #0.0
-; CHECK-NEXT:    ushll v0.2d, v0.2s, #0
-; CHECK-NEXT:    sshll v1.2d, v1.2s, #0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z1.d, #0
-; CHECK-NEXT:    st1w { z0.d }, p0, [z2.d]
+; CHECK-NEXT:    ptrue p0.s, vl2
+; CHECK-NEXT:    ldr d1, [x8, :lo12:.LCPI26_0]
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.s, z1.s[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbnz w9, #0, .LBB26_3
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB26_4
+; CHECK-NEXT:  .LBB26_2: // %else2
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB26_3: // %cond.store
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB26_2
+; CHECK-NEXT:  .LBB26_4: // %cond.store1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <2 x float>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = fcmp oeq <2 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v2f32(<2 x float> %vals, <2 x ptr> %ptrs, i32 8, <2 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v4f32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v4f32:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI27_0
 ; CHECK-NEXT:    ldr q0, [x0]
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq v2.4s, v0.4s, #0.0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    sunpklo z2.d, z2.s
-; CHECK-NEXT:    cmpne p0.d, p0/z, z2.d, #0
-; CHECK-NEXT:    st1w { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI27_0]
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z1.s
+; CHECK-NEXT:    mov z1.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.s, z1.s[1]
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z2.s, z1.s[2]
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    ldp q2, q1, [x1]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB27_5
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB27_6
+; CHECK-NEXT:  .LBB27_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB27_7
+; CHECK-NEXT:  .LBB27_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB27_8
+; CHECK-NEXT:  .LBB27_4: // %else6
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB27_5: // %cond.store
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB27_2
+; CHECK-NEXT:  .LBB27_6: // %cond.store1
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB27_3
+; CHECK-NEXT:  .LBB27_7: // %cond.store3
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB27_4
+; CHECK-NEXT:  .LBB27_8: // %cond.store5
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <4 x float>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = fcmp oeq <4 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v4f32(<4 x float> %vals, <4 x ptr> %ptrs, i32 8, <4 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v8f32(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_scatter_v8f32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI28_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q1, q0, [x0]
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI28_0]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z2.s
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z2.s
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z3.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z5.s, z2.s[2]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    mov z6.s, z2.s[3]
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    mov z2.s, z3.s[1]
+; CHECK-NEXT:    mov z4.s, z3.s[2]
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    mov z2.s, z3.s[3]
+; CHECK-NEXT:    ldp q5, q4, [x1]
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    ldp q3, q2, [x1, #32]
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbnz w9, #0, .LBB28_9
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB28_10
+; CHECK-NEXT:  .LBB28_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB28_11
+; CHECK-NEXT:  .LBB28_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB28_12
+; CHECK-NEXT:  .LBB28_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB28_13
+; CHECK-NEXT:  .LBB28_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB28_14
+; CHECK-NEXT:  .LBB28_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB28_15
+; CHECK-NEXT:  .LBB28_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB28_16
+; CHECK-NEXT:  .LBB28_8: // %else14
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB28_9: // %cond.store
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB28_2
+; CHECK-NEXT:  .LBB28_10: // %cond.store1
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z5.s, z1.s[1]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB28_3
+; CHECK-NEXT:  .LBB28_11: // %cond.store3
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z5.s, z1.s[2]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB28_4
+; CHECK-NEXT:  .LBB28_12: // %cond.store5
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB28_5
+; CHECK-NEXT:  .LBB28_13: // %cond.store7
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB28_6
+; CHECK-NEXT:  .LBB28_14: // %cond.store9
+; CHECK-NEXT:    mov z1.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB28_7
+; CHECK-NEXT:  .LBB28_15: // %cond.store11
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z1.s, z0.s[2]
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB28_8
+; CHECK-NEXT:  .LBB28_16: // %cond.store13
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %vals = load <8 x float>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = fcmp oeq <8 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v8f32(<8 x float> %vals, <8 x ptr> %ptrs, i32 8, <8 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v16f32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v16f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl16
-; CHECK-NEXT:    ptrue p1.d, vl16
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1w { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI29_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q6, q4, [x0]
+; CHECK-NEXT:    ldr q5, [x8, :lo12:.LCPI29_0]
+; CHECK-NEXT:    ldp q2, q0, [x0, #32]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z6.s, z5.s
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z5.s
+; CHECK-NEXT:    mov z7.s, z1.s[1]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov z16.s, z1.s[2]
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    mov z17.s, z1.s[3]
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    mov z7.s, z3.s[2]
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    mov z1.s, z3.s[1]
+; CHECK-NEXT:    mov z16.s, z3.s[3]
+; CHECK-NEXT:    fmov w12, s3
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    bfi w8, w12, #4, #1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z5.s
+; CHECK-NEXT:    bfi w8, w9, #5, #1
+; CHECK-NEXT:    mov z1.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #6
+; CHECK-NEXT:    mov z3.s, z1.s[1]
+; CHECK-NEXT:    mov z7.s, z1.s[2]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    fmov w9, s1
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    mov z16.s, z1.s[3]
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, z5.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z5.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z7.s, z5.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #9
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    mov z16.s, z5.s[2]
+; CHECK-NEXT:    mov z17.s, z5.s[3]
+; CHECK-NEXT:    ldp q3, q1, [x1, #96]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    orr w8, w8, w9, lsl #12
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    ldp q7, q5, [x1, #64]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    ldp q17, q16, [x1, #32]
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    ldp q19, q18, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB29_17
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB29_18
+; CHECK-NEXT:  .LBB29_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB29_19
+; CHECK-NEXT:  .LBB29_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB29_20
+; CHECK-NEXT:  .LBB29_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB29_21
+; CHECK-NEXT:  .LBB29_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB29_22
+; CHECK-NEXT:  .LBB29_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB29_23
+; CHECK-NEXT:  .LBB29_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB29_24
+; CHECK-NEXT:  .LBB29_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB29_25
+; CHECK-NEXT:  .LBB29_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB29_26
+; CHECK-NEXT:  .LBB29_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB29_27
+; CHECK-NEXT:  .LBB29_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB29_28
+; CHECK-NEXT:  .LBB29_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB29_29
+; CHECK-NEXT:  .LBB29_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB29_30
+; CHECK-NEXT:  .LBB29_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB29_31
+; CHECK-NEXT:  .LBB29_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB29_32
+; CHECK-NEXT:  .LBB29_16: // %else30
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB29_17: // %cond.store
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB29_2
+; CHECK-NEXT:  .LBB29_18: // %cond.store1
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z19.s, z6.s[1]
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB29_3
+; CHECK-NEXT:  .LBB29_19: // %cond.store3
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z19.s, z6.s[2]
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB29_4
+; CHECK-NEXT:  .LBB29_20: // %cond.store5
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    mov z6.s, z6.s[3]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB29_5
+; CHECK-NEXT:  .LBB29_21: // %cond.store7
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str s4, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB29_6
+; CHECK-NEXT:  .LBB29_22: // %cond.store9
+; CHECK-NEXT:    mov z6.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.s, z4.s[1]
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB29_7
+; CHECK-NEXT:  .LBB29_23: // %cond.store11
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z6.s, z4.s[2]
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB29_8
+; CHECK-NEXT:  .LBB29_24: // %cond.store13
+; CHECK-NEXT:    mov z6.d, z16.d[1]
+; CHECK-NEXT:    mov z4.s, z4.s[3]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str s4, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB29_9
+; CHECK-NEXT:  .LBB29_25: // %cond.store15
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB29_10
+; CHECK-NEXT:  .LBB29_26: // %cond.store17
+; CHECK-NEXT:    mov z4.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    str s4, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB29_11
+; CHECK-NEXT:  .LBB29_27: // %cond.store19
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z4.s, z2.s[2]
+; CHECK-NEXT:    str s4, [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB29_12
+; CHECK-NEXT:  .LBB29_28: // %cond.store21
+; CHECK-NEXT:    mov z4.d, z5.d[1]
+; CHECK-NEXT:    mov z2.s, z2.s[3]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB29_13
+; CHECK-NEXT:  .LBB29_29: // %cond.store23
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB29_14
+; CHECK-NEXT:  .LBB29_30: // %cond.store25
+; CHECK-NEXT:    mov z2.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB29_15
+; CHECK-NEXT:  .LBB29_31: // %cond.store27
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB29_16
+; CHECK-NEXT:  .LBB29_32: // %cond.store29
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <16 x float>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = fcmp oeq <16 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v16f32(<16 x float> %vals, <16 x ptr> %ptrs, i32 8, <16 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v32f32(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v32f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1w { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI30_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q5, q3, [x0, #64]
+; CHECK-NEXT:    ldr q23, [x8, :lo12:.LCPI30_0]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z5.s, z23.s
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z3.s, z23.s
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov z6.s, z2.s[1]
+; CHECK-NEXT:    mov z7.s, z2.s[2]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    mov z16.s, z2.s[3]
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s16
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z6.s, z4.s[2]
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    mov z2.s, z4.s[1]
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z7.s, z4.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z23.s
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s2
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    mov z6.s, z2.s[2]
+; CHECK-NEXT:    mov z7.s, z2.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, z23.s
+; CHECK-NEXT:    mov z2.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    ldp q25, q22, [x0]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.s, z2.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z6.s, z2.s[2]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s2
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z25.s, z23.s
+; CHECK-NEXT:    mov z20.s, z2.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldp q19, q17, [x0, #32]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    mov z6.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    mov z7.s, z6.s[1]
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    mov z7.s, z6.s[2]
+; CHECK-NEXT:    mov z16.s, z6.s[3]
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z22.s, z23.s
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #30
+; CHECK-NEXT:    fmov w9, s16
+; CHECK-NEXT:    mov z16.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w11, w10, #1, #1
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    bfi w11, w12, #2, #1
+; CHECK-NEXT:    mov z18.s, z16.s[1]
+; CHECK-NEXT:    mov z21.s, z16.s[2]
+; CHECK-NEXT:    mov z24.s, z16.s[3]
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    bfi w11, w9, #3, #1
+; CHECK-NEXT:    fmov w9, s21
+; CHECK-NEXT:    bfi w11, w10, #4, #1
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z19.s, z23.s
+; CHECK-NEXT:    mov z21.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w11, w12, #5, #1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w12, s21
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z24.s, z21.s[1]
+; CHECK-NEXT:    orr w9, w11, w9, lsl #6
+; CHECK-NEXT:    mov z26.s, z21.s[2]
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    mov z27.s, z21.s[3]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z17.s, z23.s
+; CHECK-NEXT:    fmov w12, s20
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z23.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z24.s, z23.s[1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    mov z26.s, z23.s[2]
+; CHECK-NEXT:    mov z27.s, z23.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #31
+; CHECK-NEXT:    ldp q4, q2, [x1, #224]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #11
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    ldp q7, q6, [x1, #192]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q18, q16, [x1, #160]
+; CHECK-NEXT:    ldp q21, q20, [x1, #128]
+; CHECK-NEXT:    ldp q24, q23, [x1, #96]
+; CHECK-NEXT:    ldp q27, q26, [x1, #64]
+; CHECK-NEXT:    ldp q29, q28, [x1, #32]
+; CHECK-NEXT:    ldp q31, q30, [x1]
+; CHECK-NEXT:    tbnz w8, #0, .LBB30_33
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB30_34
+; CHECK-NEXT:  .LBB30_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB30_35
+; CHECK-NEXT:  .LBB30_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB30_36
+; CHECK-NEXT:  .LBB30_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB30_37
+; CHECK-NEXT:  .LBB30_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB30_38
+; CHECK-NEXT:  .LBB30_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB30_39
+; CHECK-NEXT:  .LBB30_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB30_40
+; CHECK-NEXT:  .LBB30_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB30_41
+; CHECK-NEXT:  .LBB30_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB30_42
+; CHECK-NEXT:  .LBB30_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB30_43
+; CHECK-NEXT:  .LBB30_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB30_44
+; CHECK-NEXT:  .LBB30_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB30_45
+; CHECK-NEXT:  .LBB30_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB30_46
+; CHECK-NEXT:  .LBB30_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB30_47
+; CHECK-NEXT:  .LBB30_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB30_48
+; CHECK-NEXT:  .LBB30_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB30_49
+; CHECK-NEXT:  .LBB30_17: // %else32
+; CHECK-NEXT:    tbnz w8, #17, .LBB30_50
+; CHECK-NEXT:  .LBB30_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB30_51
+; CHECK-NEXT:  .LBB30_19: // %else36
+; CHECK-NEXT:    tbnz w8, #19, .LBB30_52
+; CHECK-NEXT:  .LBB30_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB30_53
+; CHECK-NEXT:  .LBB30_21: // %else40
+; CHECK-NEXT:    tbnz w8, #21, .LBB30_54
+; CHECK-NEXT:  .LBB30_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB30_55
+; CHECK-NEXT:  .LBB30_23: // %else44
+; CHECK-NEXT:    tbnz w8, #23, .LBB30_56
+; CHECK-NEXT:  .LBB30_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB30_57
+; CHECK-NEXT:  .LBB30_25: // %else48
+; CHECK-NEXT:    tbnz w8, #25, .LBB30_58
+; CHECK-NEXT:  .LBB30_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB30_59
+; CHECK-NEXT:  .LBB30_27: // %else52
+; CHECK-NEXT:    tbnz w8, #27, .LBB30_60
+; CHECK-NEXT:  .LBB30_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB30_61
+; CHECK-NEXT:  .LBB30_29: // %else56
+; CHECK-NEXT:    tbnz w8, #29, .LBB30_62
+; CHECK-NEXT:  .LBB30_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB30_63
+; CHECK-NEXT:  .LBB30_31: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB30_64
+; CHECK-NEXT:  .LBB30_32: // %else62
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB30_33: // %cond.store
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    str s25, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB30_2
+; CHECK-NEXT:  .LBB30_34: // %cond.store1
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    mov z31.s, z25.s[1]
+; CHECK-NEXT:    str s31, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB30_3
+; CHECK-NEXT:  .LBB30_35: // %cond.store3
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    mov z31.s, z25.s[2]
+; CHECK-NEXT:    str s31, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB30_4
+; CHECK-NEXT:  .LBB30_36: // %cond.store5
+; CHECK-NEXT:    mov z30.d, z30.d[1]
+; CHECK-NEXT:    mov z25.s, z25.s[3]
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    str s25, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB30_5
+; CHECK-NEXT:  .LBB30_37: // %cond.store7
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    str s22, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB30_6
+; CHECK-NEXT:  .LBB30_38: // %cond.store9
+; CHECK-NEXT:    mov z25.d, z29.d[1]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    mov z25.s, z22.s[1]
+; CHECK-NEXT:    str s25, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB30_7
+; CHECK-NEXT:  .LBB30_39: // %cond.store11
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z25.s, z22.s[2]
+; CHECK-NEXT:    str s25, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB30_8
+; CHECK-NEXT:  .LBB30_40: // %cond.store13
+; CHECK-NEXT:    mov z25.d, z28.d[1]
+; CHECK-NEXT:    mov z22.s, z22.s[3]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str s22, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB30_9
+; CHECK-NEXT:  .LBB30_41: // %cond.store15
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB30_10
+; CHECK-NEXT:  .LBB30_42: // %cond.store17
+; CHECK-NEXT:    mov z22.d, z27.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z22.s, z19.s[1]
+; CHECK-NEXT:    str s22, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB30_11
+; CHECK-NEXT:  .LBB30_43: // %cond.store19
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    mov z22.s, z19.s[2]
+; CHECK-NEXT:    str s22, [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB30_12
+; CHECK-NEXT:  .LBB30_44: // %cond.store21
+; CHECK-NEXT:    mov z22.d, z26.d[1]
+; CHECK-NEXT:    mov z19.s, z19.s[3]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB30_13
+; CHECK-NEXT:  .LBB30_45: // %cond.store23
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    str s17, [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB30_14
+; CHECK-NEXT:  .LBB30_46: // %cond.store25
+; CHECK-NEXT:    mov z19.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z19.s, z17.s[1]
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB30_15
+; CHECK-NEXT:  .LBB30_47: // %cond.store27
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    mov z19.s, z17.s[2]
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB30_16
+; CHECK-NEXT:  .LBB30_48: // %cond.store29
+; CHECK-NEXT:    mov z19.d, z23.d[1]
+; CHECK-NEXT:    mov z17.s, z17.s[3]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str s17, [x9]
+; CHECK-NEXT:    tbz w8, #16, .LBB30_17
+; CHECK-NEXT:  .LBB30_49: // %cond.store31
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #17, .LBB30_18
+; CHECK-NEXT:  .LBB30_50: // %cond.store33
+; CHECK-NEXT:    mov z17.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.s, z5.s[1]
+; CHECK-NEXT:    str s17, [x9]
+; CHECK-NEXT:    tbz w8, #18, .LBB30_19
+; CHECK-NEXT:  .LBB30_51: // %cond.store35
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z17.s, z5.s[2]
+; CHECK-NEXT:    str s17, [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB30_20
+; CHECK-NEXT:  .LBB30_52: // %cond.store37
+; CHECK-NEXT:    mov z17.d, z20.d[1]
+; CHECK-NEXT:    mov z5.s, z5.s[3]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #20, .LBB30_21
+; CHECK-NEXT:  .LBB30_53: // %cond.store39
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    str s3, [x9]
+; CHECK-NEXT:    tbz w8, #21, .LBB30_22
+; CHECK-NEXT:  .LBB30_54: // %cond.store41
+; CHECK-NEXT:    mov z5.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z5.s, z3.s[1]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #22, .LBB30_23
+; CHECK-NEXT:  .LBB30_55: // %cond.store43
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z5.s, z3.s[2]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB30_24
+; CHECK-NEXT:  .LBB30_56: // %cond.store45
+; CHECK-NEXT:    mov z5.d, z16.d[1]
+; CHECK-NEXT:    mov z3.s, z3.s[3]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str s3, [x9]
+; CHECK-NEXT:    tbz w8, #24, .LBB30_25
+; CHECK-NEXT:  .LBB30_57: // %cond.store47
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #25, .LBB30_26
+; CHECK-NEXT:  .LBB30_58: // %cond.store49
+; CHECK-NEXT:    mov z3.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z3.s, z1.s[1]
+; CHECK-NEXT:    str s3, [x9]
+; CHECK-NEXT:    tbz w8, #26, .LBB30_27
+; CHECK-NEXT:  .LBB30_59: // %cond.store51
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z3.s, z1.s[2]
+; CHECK-NEXT:    str s3, [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB30_28
+; CHECK-NEXT:  .LBB30_60: // %cond.store53
+; CHECK-NEXT:    mov z3.d, z6.d[1]
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #28, .LBB30_29
+; CHECK-NEXT:  .LBB30_61: // %cond.store55
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    tbz w8, #29, .LBB30_30
+; CHECK-NEXT:  .LBB30_62: // %cond.store57
+; CHECK-NEXT:    mov z1.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.s, z0.s[1]
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB30_31
+; CHECK-NEXT:  .LBB30_63: // %cond.store59
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z1.s, z0.s[2]
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #31, .LBB30_32
+; CHECK-NEXT:  .LBB30_64: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <32 x float>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = fcmp oeq <32 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
; CHECK-NEXT:    ret
   %mask = fcmp oeq <1 x double> %vals, zeroinitializer
   call void @llvm.masked.scatter.v1f64(<1 x double> %vals, <1 x ptr> %ptrs, i32 8, <1 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v2f64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v2f64:
 ; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI32_0
 ; CHECK-NEXT:    ldr q0, [x0]
 ; CHECK-NEXT:    ptrue p0.d, vl2
-; CHECK-NEXT:    ldr q2, [x1]
-; CHECK-NEXT:    fcmeq v1.2d, v0.2d, #0.0
-; CHECK-NEXT:    cmpne p0.d, p0/z, z1.d, #0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z2.d]
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI32_0]
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, z1.d
+; CHECK-NEXT:    mov z1.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ldr q1, [x1]
+; CHECK-NEXT:    bfi w9, w8, #1, #31
+; CHECK-NEXT:    and w8, w9, #0x3
+; CHECK-NEXT:    tbnz w9, #0, .LBB32_3
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB32_4
+; CHECK-NEXT:  .LBB32_2: // %else2
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB32_3: // %cond.store
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str d0, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB32_2
+; CHECK-NEXT:  .LBB32_4: // %cond.store1
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str d0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <2 x double>, ptr %a
   %ptrs = load <2 x ptr>, ptr %b
   %mask = fcmp oeq <2 x double> %vals, zeroinitializer
   call void @llvm.masked.scatter.v2f64(<2 x double> %vals, <2 x ptr> %ptrs, i32 8, <2 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v4f64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v4f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl4
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI33_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q1, q0, [x0]
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI33_0]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z2.d
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, z2.d
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z2.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z4.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    fmov x8, d4
+; CHECK-NEXT:    fmov x10, d2
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    bfi w9, w8, #1, #1
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    ldp q3, q2, [x1]
+; CHECK-NEXT:    bfi w9, w10, #2, #1
+; CHECK-NEXT:    orr w9, w9, w8, lsl #3
+; CHECK-NEXT:    and w8, w9, #0xf
+; CHECK-NEXT:    tbnz w9, #0, .LBB33_5
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB33_6
+; CHECK-NEXT:  .LBB33_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB33_7
+; CHECK-NEXT:  .LBB33_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB33_8
+; CHECK-NEXT:  .LBB33_4: // %else6
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB33_5: // %cond.store
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB33_2
+; CHECK-NEXT:  .LBB33_6: // %cond.store1
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB33_3
+; CHECK-NEXT:  .LBB33_7: // %cond.store3
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str d0, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB33_4
+; CHECK-NEXT:  .LBB33_8: // %cond.store5
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str d0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <4 x double>, ptr %a
   %ptrs = load <4 x ptr>, ptr %b
   %mask = fcmp oeq <4 x double> %vals, zeroinitializer
   call void @llvm.masked.scatter.v4f64(<4 x double> %vals, <4 x ptr> %ptrs, i32 8, <4 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v8f64(ptr %a, ptr %b) #0 {
+; CHECK-LABEL: masked_scatter_v8f64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI34_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q4, q2, [x0]
+; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI34_0]
+; CHECK-NEXT:    ldp q1, q0, [x0, #32]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z4.d, z3.d
+; CHECK-NEXT:    fcmeq p2.d, p0/z, z2.d, z3.d
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.d, p2/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z7.d, z5.d[1]
+; CHECK-NEXT:    fmov x8, d5
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z3.d
+; CHECK-NEXT:    mov z5.d, z6.d[1]
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, z3.d
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov x11, d5
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    mov z6.d, z3.d[1]
+; CHECK-NEXT:    mov z3.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z5.d, z3.d[1]
+; CHECK-NEXT:    ldp q7, q6, [x1]
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov x10, d5
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    ldp q5, q3, [x1, #32]
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbnz w9, #0, .LBB34_9
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB34_10
+; CHECK-NEXT:  .LBB34_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB34_11
+; CHECK-NEXT:  .LBB34_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB34_12
+; CHECK-NEXT:  .LBB34_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB34_13
+; CHECK-NEXT:  .LBB34_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB34_14
+; CHECK-NEXT:  .LBB34_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB34_15
+; CHECK-NEXT:  .LBB34_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB34_16
+; CHECK-NEXT:  .LBB34_8: // %else14
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB34_9: // %cond.store
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str d4, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB34_2
+; CHECK-NEXT:  .LBB34_10: // %cond.store1
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str d4, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB34_3
+; CHECK-NEXT:  .LBB34_11: // %cond.store3
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB34_4
+; CHECK-NEXT:  .LBB34_12: // %cond.store5
+; CHECK-NEXT:    mov z4.d, z6.d[1]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB34_5
+; CHECK-NEXT:  .LBB34_13: // %cond.store7
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB34_6
+; CHECK-NEXT:  .LBB34_14: // %cond.store9
+; CHECK-NEXT:    mov z2.d, z5.d[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB34_7
+; CHECK-NEXT:  .LBB34_15: // %cond.store11
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d0, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB34_8
+; CHECK-NEXT:  .LBB34_16: // %cond.store13
+; CHECK-NEXT:    mov z1.d, z3.d[1]
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str d0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
   %vals = load <8 x double>, ptr %a
   %ptrs = load <8 x ptr>, ptr %b
   %mask = fcmp oeq <8 x double> %vals, zeroinitializer
   call void @llvm.masked.scatter.v8f64(<8 x double> %vals, <8 x ptr> %ptrs, i32 8, <8 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v16f64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v16f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl16
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z1.d]
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    adrp x8, .LCPI35_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q18, q17, [x0]
+; CHECK-NEXT:    ldr q20, [x8, :lo12:.LCPI35_0]
+; CHECK-NEXT:    ldp q7, q5, [x0, #32]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z18.d, z20.d
+; CHECK-NEXT:    mov z2.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z17.d, z20.d
+; CHECK-NEXT:    mov z4.d, z2.d[1]
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    mov z2.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z7.d, z20.d
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x11, d2
+; CHECK-NEXT:    bfi w8, w10, #1, #1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    bfi w8, w9, #2, #1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z5.d, z20.d
+; CHECK-NEXT:    mov z6.d, z3.d[1]
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w10, #4, #1
+; CHECK-NEXT:    ldp q4, q2, [x0, #64]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.d, z3.d[1]
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    fmov x11, d6
+; CHECK-NEXT:    bfi w8, w9, #5, #1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z4.d, z20.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z2.d, z20.d
+; CHECK-NEXT:    orr w8, w8, w10, lsl #6
+; CHECK-NEXT:    mov z6.d, z3.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #7
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x11, d3
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #8
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #9
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z20.d
+; CHECK-NEXT:    orr w8, w8, w9, lsl #10
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, z20.d
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    mov z21.d, z3.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z20.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q6, q3, [x1, #96]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #11
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.d, z20.d[1]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #12
+; CHECK-NEXT:    fmov x10, d20
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    ldp q19, q16, [x1, #64]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #13
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    orr w8, w8, w9, lsl #14
+; CHECK-NEXT:    orr w9, w8, w10, lsl #15
+; CHECK-NEXT:    ldp q21, q20, [x1, #32]
+; CHECK-NEXT:    and w8, w9, #0xffff
+; CHECK-NEXT:    ldp q23, q22, [x1]
+; CHECK-NEXT:    tbnz w9, #0, .LBB35_17
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB35_18
+; CHECK-NEXT:  .LBB35_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB35_19
+; CHECK-NEXT:  .LBB35_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB35_20
+; CHECK-NEXT:  .LBB35_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB35_21
+; CHECK-NEXT:  .LBB35_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB35_22
+; CHECK-NEXT:  .LBB35_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB35_23
+; CHECK-NEXT:  .LBB35_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB35_24
+; CHECK-NEXT:  .LBB35_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB35_25
+; CHECK-NEXT:  .LBB35_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB35_26
+; CHECK-NEXT:  .LBB35_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB35_27
+; CHECK-NEXT:  .LBB35_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB35_28
+; CHECK-NEXT:  .LBB35_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB35_29
+; CHECK-NEXT:  .LBB35_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB35_30
+; CHECK-NEXT:  .LBB35_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB35_31
+; CHECK-NEXT:  .LBB35_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB35_32
+; CHECK-NEXT:  .LBB35_16: // %else30
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB35_17: // %cond.store
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    str d18, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB35_2
+; CHECK-NEXT:  .LBB35_18: // %cond.store1
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    str d18, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB35_3
+; CHECK-NEXT:  .LBB35_19: // %cond.store3
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    str d17, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB35_4
+; CHECK-NEXT:  .LBB35_20: // %cond.store5
+; CHECK-NEXT:    mov z18.d, z22.d[1]
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    str d17, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB35_5
+; CHECK-NEXT:  .LBB35_21: // %cond.store7
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    str d7, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB35_6
+; CHECK-NEXT:  .LBB35_22: // %cond.store9
+; CHECK-NEXT:    mov z17.d, z21.d[1]
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str d7, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB35_7
+; CHECK-NEXT:  .LBB35_23: // %cond.store11
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str d5, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB35_8
+; CHECK-NEXT:  .LBB35_24: // %cond.store13
+; CHECK-NEXT:    mov z7.d, z20.d[1]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str d5, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB35_9
+; CHECK-NEXT:  .LBB35_25: // %cond.store15
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str d4, [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB35_10
+; CHECK-NEXT:  .LBB35_26: // %cond.store17
+; CHECK-NEXT:    mov z5.d, z19.d[1]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str d4, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB35_11
+; CHECK-NEXT:  .LBB35_27: // %cond.store19
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB35_12
+; CHECK-NEXT:  .LBB35_28: // %cond.store21
+; CHECK-NEXT:    mov z4.d, z16.d[1]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB35_13
+; CHECK-NEXT:  .LBB35_29: // %cond.store23
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB35_14
+; CHECK-NEXT:  .LBB35_30: // %cond.store25
+; CHECK-NEXT:    mov z2.d, z6.d[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB35_15
+; CHECK-NEXT:  .LBB35_31: // %cond.store27
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d0, [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB35_16
+; CHECK-NEXT:  .LBB35_32: // %cond.store29
+; CHECK-NEXT:    mov z1.d, z3.d[1]
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str d0, [x8]
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
   %vals = load <16 x double>, ptr %a
   %ptrs = load <16 x ptr>, ptr %b
   %mask = fcmp oeq <16 x double> %vals, zeroinitializer
   call void @llvm.masked.scatter.v16f64(<16 x double> %vals, <16 x ptr> %ptrs, i32 8, <16 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_v32f64(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_v32f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z1.d]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #80
+; CHECK-NEXT:    .cfi_def_cfa_offset 80
+; CHECK-NEXT:    stp d15, d14, [sp, #16] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d13, d12, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d11, d10, [sp, #48] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #64] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -24
+; CHECK-NEXT:    .cfi_offset b11, -32
+; CHECK-NEXT:    .cfi_offset b12, -40
+; CHECK-NEXT:    .cfi_offset b13, -48
+; CHECK-NEXT:    .cfi_offset b14, -56
+; CHECK-NEXT:    .cfi_offset b15, -64
+; CHECK-NEXT:    adrp x8, .LCPI36_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q20, q7, [x0, #128]
+; CHECK-NEXT:    ldr q9, [x8, :lo12:.LCPI36_0]
+; CHECK-NEXT:    ldp q19, q16, [x0, #160]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z20.d, z9.d
+; CHECK-NEXT:    mov z0.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z7.d, z9.d
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    mov z1.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z3.d, z1.d[1]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z19.d, z9.d
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov x10, d3
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z6.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z16.d, z9.d
+; CHECK-NEXT:    mov z17.d, z6.d[1]
+; CHECK-NEXT:    ldp q5, q4, [x0, #192]
+; CHECK-NEXT:    bfi w8, w9, #18, #1
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w10, #19, #1
+; CHECK-NEXT:    fmov x11, d6
+; CHECK-NEXT:    fmov x10, d17
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    mov z22.d, z6.d[1]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z5.d, z9.d
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    ldp q2, q0, [x0, #224]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z22.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z4.d, z9.d
+; CHECK-NEXT:    fmov x10, d22
+; CHECK-NEXT:    mov z23.d, z22.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z22.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x11, d22
+; CHECK-NEXT:    mov z28.d, z22.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d23
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z2.d, z9.d
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    ldp q31, q30, [x0]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z31.d, z9.d
+; CHECK-NEXT:    fmov x10, d28
+; CHECK-NEXT:    mov z29.d, z28.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q27, q26, [x0, #32]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d29
+; CHECK-NEXT:    mov z29.d, z28.d[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z30.d, z9.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldp q25, q24, [x0, #64]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    fmov x10, d28
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z27.d, z9.d
+; CHECK-NEXT:    mov z29.d, z28.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov x11, d29
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z26.d, z9.d
+; CHECK-NEXT:    mov z29.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z8.d, z28.d[1]
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    fmov x12, d29
+; CHECK-NEXT:    bfi w10, w11, #3, #1
+; CHECK-NEXT:    fmov x11, d8
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, z9.d
+; CHECK-NEXT:    bfi w10, w9, #4, #1
+; CHECK-NEXT:    mov z28.d, z29.d[1]
+; CHECK-NEXT:    and w9, w12, #0x1
+; CHECK-NEXT:    mov z29.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w10, w11, #5, #1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z25.d, z9.d
+; CHECK-NEXT:    orr w9, w10, w9, lsl #6
+; CHECK-NEXT:    fmov x10, d28
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z24.d, z9.d
+; CHECK-NEXT:    fmov x11, d28
+; CHECK-NEXT:    mov z10.d, z28.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z28.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q21, q18, [x0, #96]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov x10, d10
+; CHECK-NEXT:    fmov x12, d29
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov x11, d28
+; CHECK-NEXT:    mov z8.d, z29.d[1]
+; CHECK-NEXT:    mov z10.d, z28.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z21.d, z9.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov x10, d8
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    fmov x11, d10
+; CHECK-NEXT:    mov z8.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    mov z11.d, z8.d[1]
+; CHECK-NEXT:    fmov x12, d8
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #31
+; CHECK-NEXT:    fmov x10, d11
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z18.d, z9.d
+; CHECK-NEXT:    mov z9.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov x11, d9
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w12, lsl #12
+; CHECK-NEXT:    mov z11.d, z9.d[1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov x11, d11
+; CHECK-NEXT:    ldp q3, q1, [x1, #224]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q17, q6, [x1, #192]
+; CHECK-NEXT:    ldp q23, q22, [x1, #160]
+; CHECK-NEXT:    ldp q29, q28, [x1, #128]
+; CHECK-NEXT:    ldp q10, q8, [x1, #96]
+; CHECK-NEXT:    ldp q11, q9, [x1, #64]
+; CHECK-NEXT:    ldp q13, q12, [x1, #32]
+; CHECK-NEXT:    ldp q15, q14, [x1]
+; CHECK-NEXT:    tbnz w8, #0, .LBB36_34
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB36_35
+; CHECK-NEXT:  .LBB36_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB36_36
+; CHECK-NEXT:  .LBB36_3: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB36_37
+; CHECK-NEXT:  .LBB36_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB36_38
+; CHECK-NEXT:  .LBB36_5: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB36_39
+; CHECK-NEXT:  .LBB36_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB36_40
+; CHECK-NEXT:  .LBB36_7: // %else12
+; CHECK-NEXT:    tbnz w8, #7, .LBB36_41
+; CHECK-NEXT:  .LBB36_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB36_42
+; CHECK-NEXT:  .LBB36_9: // %else16
+; CHECK-NEXT:    tbnz w8, #9, .LBB36_43
+; CHECK-NEXT:  .LBB36_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB36_44
+; CHECK-NEXT:  .LBB36_11: // %else20
+; CHECK-NEXT:    tbnz w8, #11, .LBB36_45
+; CHECK-NEXT:  .LBB36_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB36_46
+; CHECK-NEXT:  .LBB36_13: // %else24
+; CHECK-NEXT:    tbnz w8, #13, .LBB36_47
+; CHECK-NEXT:  .LBB36_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB36_48
+; CHECK-NEXT:  .LBB36_15: // %else28
+; CHECK-NEXT:    tbnz w8, #15, .LBB36_49
+; CHECK-NEXT:  .LBB36_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB36_50
+; CHECK-NEXT:  .LBB36_17: // %else32
+; CHECK-NEXT:    tbnz w8, #17, .LBB36_51
+; CHECK-NEXT:  .LBB36_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB36_52
+; CHECK-NEXT:  .LBB36_19: // %else36
+; CHECK-NEXT:    tbnz w8, #19, .LBB36_53
+; CHECK-NEXT:  .LBB36_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB36_54
+; CHECK-NEXT:  .LBB36_21: // %else40
+; CHECK-NEXT:    tbnz w8, #21, .LBB36_55
+; CHECK-NEXT:  .LBB36_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB36_56
+; CHECK-NEXT:  .LBB36_23: // %else44
+; CHECK-NEXT:    tbnz w8, #23, .LBB36_57
+; CHECK-NEXT:  .LBB36_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB36_58
+; CHECK-NEXT:  .LBB36_25: // %else48
+; CHECK-NEXT:    tbnz w8, #25, .LBB36_59
+; CHECK-NEXT:  .LBB36_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB36_60
+; CHECK-NEXT:  .LBB36_27: // %else52
+; CHECK-NEXT:    tbnz w8, #27, .LBB36_61
+; CHECK-NEXT:  .LBB36_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB36_62
+; CHECK-NEXT:  .LBB36_29: // %else56
+; CHECK-NEXT:    tbnz w8, #29, .LBB36_63
+; CHECK-NEXT:  .LBB36_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB36_64
+; CHECK-NEXT:  .LBB36_31: // %else60
+; CHECK-NEXT:    tbz w8, #31, .LBB36_33
+; CHECK-NEXT:  .LBB36_32: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str d0, [x8]
+; CHECK-NEXT:  .LBB36_33: // %else62
+; CHECK-NEXT:    ldp d9, d8, [sp, #64] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d11, d10, [sp, #48] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d13, d12, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d15, d14, [sp, #16] // 16-byte Folded Reload
+; CHECK-NEXT:    add sp, sp, #80
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB36_34: // %cond.store
+; CHECK-NEXT:    fmov x9, d15
+; CHECK-NEXT:    str d31, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB36_2
+; CHECK-NEXT:  .LBB36_35: // %cond.store1
+; CHECK-NEXT:    mov z15.d, z15.d[1]
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    fmov x9, d15
+; CHECK-NEXT:    str d31, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB36_3
+; CHECK-NEXT:  .LBB36_36: // %cond.store3
+; CHECK-NEXT:    fmov x9, d14
+; CHECK-NEXT:    str d30, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB36_4
+; CHECK-NEXT:  .LBB36_37: // %cond.store5
+; CHECK-NEXT:    mov z31.d, z14.d[1]
+; CHECK-NEXT:    mov z30.d, z30.d[1]
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    str d30, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB36_5
+; CHECK-NEXT:  .LBB36_38: // %cond.store7
+; CHECK-NEXT:    fmov x9, d13
+; CHECK-NEXT:    str d27, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB36_6
+; CHECK-NEXT:  .LBB36_39: // %cond.store9
+; CHECK-NEXT:    mov z30.d, z13.d[1]
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    str d27, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB36_7
+; CHECK-NEXT:  .LBB36_40: // %cond.store11
+; CHECK-NEXT:    fmov x9, d12
+; CHECK-NEXT:    str d26, [x9]
+; CHECK-NEXT:    tbz w8, #7, .LBB36_8
+; CHECK-NEXT:  .LBB36_41: // %cond.store13
+; CHECK-NEXT:    mov z27.d, z12.d[1]
+; CHECK-NEXT:    mov z26.d, z26.d[1]
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str d26, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB36_9
+; CHECK-NEXT:  .LBB36_42: // %cond.store15
+; CHECK-NEXT:    fmov x9, d11
+; CHECK-NEXT:    str d25, [x9]
+; CHECK-NEXT:    tbz w8, #9, .LBB36_10
+; CHECK-NEXT:  .LBB36_43: // %cond.store17
+; CHECK-NEXT:    mov z26.d, z11.d[1]
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    str d25, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB36_11
+; CHECK-NEXT:  .LBB36_44: // %cond.store19
+; CHECK-NEXT:    fmov x9, d9
+; CHECK-NEXT:    str d24, [x9]
+; CHECK-NEXT:    tbz w8, #11, .LBB36_12
+; CHECK-NEXT:  .LBB36_45: // %cond.store21
+; CHECK-NEXT:    mov z25.d, z9.d[1]
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str d24, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB36_13
+; CHECK-NEXT:  .LBB36_46: // %cond.store23
+; CHECK-NEXT:    fmov x9, d10
+; CHECK-NEXT:    str d21, [x9]
+; CHECK-NEXT:    tbz w8, #13, .LBB36_14
+; CHECK-NEXT:  .LBB36_47: // %cond.store25
+; CHECK-NEXT:    mov z24.d, z10.d[1]
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    str d21, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB36_15
+; CHECK-NEXT:  .LBB36_48: // %cond.store27
+; CHECK-NEXT:    fmov x9, d8
+; CHECK-NEXT:    str d18, [x9]
+; CHECK-NEXT:    tbz w8, #15, .LBB36_16
+; CHECK-NEXT:  .LBB36_49: // %cond.store29
+; CHECK-NEXT:    mov z21.d, z8.d[1]
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    str d18, [x9]
+; CHECK-NEXT:    tbz w8, #16, .LBB36_17
+; CHECK-NEXT:  .LBB36_50: // %cond.store31
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    str d20, [x9]
+; CHECK-NEXT:    tbz w8, #17, .LBB36_18
+; CHECK-NEXT:  .LBB36_51: // %cond.store33
+; CHECK-NEXT:    mov z18.d, z29.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z18.d, z20.d[1]
+; CHECK-NEXT:    str d18, [x9]
+; CHECK-NEXT:    tbz w8, #18, .LBB36_19
+; CHECK-NEXT:  .LBB36_52: // %cond.store35
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    str d7, [x9]
+; CHECK-NEXT:    tbz w8, #19, .LBB36_20
+; CHECK-NEXT:  .LBB36_53: // %cond.store37
+; CHECK-NEXT:    mov z18.d, z28.d[1]
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    str d7, [x9]
+; CHECK-NEXT:    tbz w8, #20, .LBB36_21
+; CHECK-NEXT:  .LBB36_54: // %cond.store39
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    str d19, [x9]
+; CHECK-NEXT:    tbz w8, #21, .LBB36_22
+; CHECK-NEXT:  .LBB36_55: // %cond.store41
+; CHECK-NEXT:    mov z7.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z7.d, z19.d[1]
+; CHECK-NEXT:    str d7, [x9]
+; CHECK-NEXT:    tbz w8, #22, .LBB36_23
+; CHECK-NEXT:  .LBB36_56: // %cond.store43
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    str d16, [x9]
+; CHECK-NEXT:    tbz w8, #23, .LBB36_24
+; CHECK-NEXT:  .LBB36_57: // %cond.store45
+; CHECK-NEXT:    mov z7.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z7.d, z16.d[1]
+; CHECK-NEXT:    str d7, [x9]
+; CHECK-NEXT:    tbz w8, #24, .LBB36_25
+; CHECK-NEXT:  .LBB36_58: // %cond.store47
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str d5, [x9]
+; CHECK-NEXT:    tbz w8, #25, .LBB36_26
+; CHECK-NEXT:  .LBB36_59: // %cond.store49
+; CHECK-NEXT:    mov z7.d, z17.d[1]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str d5, [x9]
+; CHECK-NEXT:    tbz w8, #26, .LBB36_27
+; CHECK-NEXT:  .LBB36_60: // %cond.store51
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str d4, [x9]
+; CHECK-NEXT:    tbz w8, #27, .LBB36_28
+; CHECK-NEXT:  .LBB36_61: // %cond.store53
+; CHECK-NEXT:    mov z5.d, z6.d[1]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str d4, [x9]
+; CHECK-NEXT:    tbz w8, #28, .LBB36_29
+; CHECK-NEXT:  .LBB36_62: // %cond.store55
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:    tbz w8, #29, .LBB36_30
+; CHECK-NEXT:  .LBB36_63: // %cond.store57
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB36_31
+; CHECK-NEXT:  .LBB36_64: // %cond.store59
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    str d0, [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB36_32
+; CHECK-NEXT:    b .LBB36_33
   %vals = load <32 x double>, ptr %a
   %ptrs = load <32 x ptr>, ptr %b
   %mask = fcmp oeq <32 x double> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f64(<32 x double> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 ; The above tests test the types, the below tests check that the addressing
 ; modes still function
 
 define void @masked_scatter_32b_scaled_sext_f16(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_scatter_32b_scaled_sext_f16:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.s, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1h { z0.s }, p0, [x2, z1.s, sxtw #1]
+; CHECK-NEXT:    adrp x8, .LCPI37_1
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q2, q0, [x0, #32]
+; CHECK-NEXT:    adrp x12, .LCPI37_0
+; CHECK-NEXT:    ldr q6, [x8, :lo12:.LCPI37_1]
+; CHECK-NEXT:    ldp q18, q7, [x0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z2.h, z6.h
+; CHECK-NEXT:    mov z3.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z19.h, z3.h[1]
+; CHECK-NEXT:    mov z22.h, z3.h[2]
+; CHECK-NEXT:    fmov w8, s19
+; CHECK-NEXT:    mov z23.h, z3.h[3]
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z24.h, z3.h[4]
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s24
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    mov z26.h, z3.h[6]
+; CHECK-NEXT:    bfi w9, w11, #19, #1
+; CHECK-NEXT:    mov z25.h, z3.h[5]
+; CHECK-NEXT:    ldp q4, q1, [x1, #96]
+; CHECK-NEXT:    bfi w9, w8, #20, #1
+; CHECK-NEXT:    fmov w8, s26
+; CHECK-NEXT:    mov z27.h, z3.h[7]
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    ldp q16, q5, [x1, #64]
+; CHECK-NEXT:    ldp q19, q17, [x1, #32]
+; CHECK-NEXT:    ldp q21, q20, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    bfi w9, w10, #21, #1
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z0.h, z6.h
+; CHECK-NEXT:    orr w8, w9, w8, lsl #22
+; CHECK-NEXT:    fmov w9, s27
+; CHECK-NEXT:    mov z23.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z18.h, z6.h
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    mov z24.h, z23.h[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z25.h, z23.h[2]
+; CHECK-NEXT:    mov z26.h, z23.h[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z27.h, z23.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    mov z28.h, z23.h[5]
+; CHECK-NEXT:    mov z29.h, z23.h[6]
+; CHECK-NEXT:    mov z22.h, z23.h[7]
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    mov z23.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    ldr q3, [x12, :lo12:.LCPI37_0]
+; CHECK-NEXT:    mov z24.h, z23.h[1]
+; CHECK-NEXT:    fmov w12, s28
+; CHECK-NEXT:    fmov w14, s23
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w13, s24
+; CHECK-NEXT:    mov z24.h, z23.h[2]
+; CHECK-NEXT:    mov z25.h, z23.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z26.h, z23.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w12, w14, #0x1
+; CHECK-NEXT:    fmov w14, s25
+; CHECK-NEXT:    mov z27.h, z23.h[5]
+; CHECK-NEXT:    bfi w12, w13, #1, #1
+; CHECK-NEXT:    fmov w13, s26
+; CHECK-NEXT:    bfi w12, w11, #2, #1
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    mov z28.h, z23.h[6]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    bfi w12, w14, #3, #1
+; CHECK-NEXT:    mov z30.h, z23.h[7]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    bfi w12, w13, #4, #1
+; CHECK-NEXT:    fmov w10, s28
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z7.h, z6.h
+; CHECK-NEXT:    bfi w12, w11, #5, #1
+; CHECK-NEXT:    mov z6.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s30
+; CHECK-NEXT:    fmov w13, s6
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z23.h, z6.h[1]
+; CHECK-NEXT:    mov z24.h, z6.h[2]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w10, w12, w10, lsl #6
+; CHECK-NEXT:    and w12, w13, #0x1
+; CHECK-NEXT:    fmov w13, s23
+; CHECK-NEXT:    fmov w9, s29
+; CHECK-NEXT:    orr w10, w10, w11, lsl #7
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z25.h, z6.h[3]
+; CHECK-NEXT:    orr w10, w10, w12, lsl #8
+; CHECK-NEXT:    fmov w12, s25
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z26.h, z6.h[4]
+; CHECK-NEXT:    orr w10, w10, w13, lsl #9
+; CHECK-NEXT:    mov z27.h, z6.h[5]
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #30
+; CHECK-NEXT:    orr w9, w10, w11, lsl #10
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    mov z28.h, z6.h[6]
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    orr w9, w9, w12, lsl #11
+; CHECK-NEXT:    fmov w12, s28
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z29.h, z6.h[7]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    ldr q6, [sp]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    sunpklo z22.d, z21.s
+; CHECK-NEXT:    ext z21.b, z21.b, z21.b, #8
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    sunpklo z21.d, z21.s
+; CHECK-NEXT:    orr w8, w8, w10, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    movprfx z23, z22
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z3.d
+; CHECK-NEXT:    movprfx z22, z21
+; CHECK-NEXT:    lsl z22.d, p0/m, z22.d, z3.d
+; CHECK-NEXT:    add z21.d, z6.d, z23.d
+; CHECK-NEXT:    tbz w8, #0, .LBB37_2
+; CHECK-NEXT:  // %bb.1: // %cond.store
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB37_2: // %else
+; CHECK-NEXT:    sunpklo z23.d, z20.s
+; CHECK-NEXT:    add z22.d, z6.d, z22.d
+; CHECK-NEXT:    tbz w8, #1, .LBB37_4
+; CHECK-NEXT:  // %bb.3: // %cond.store1
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.h, z18.h[1]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:  .LBB37_4: // %else2
+; CHECK-NEXT:    ext z20.b, z20.b, z20.b, #8
+; CHECK-NEXT:    movprfx z21, z23
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z3.d
+; CHECK-NEXT:    tbz w8, #2, .LBB37_6
+; CHECK-NEXT:  // %bb.5: // %cond.store3
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z23.h, z18.h[2]
+; CHECK-NEXT:    str h23, [x9]
+; CHECK-NEXT:  .LBB37_6: // %else4
+; CHECK-NEXT:    sunpklo z20.d, z20.s
+; CHECK-NEXT:    add z21.d, z6.d, z21.d
+; CHECK-NEXT:    tbz w8, #3, .LBB37_8
+; CHECK-NEXT:  // %bb.7: // %cond.store5
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z22.h, z18.h[3]
+; CHECK-NEXT:    str h22, [x9]
+; CHECK-NEXT:  .LBB37_8: // %else6
+; CHECK-NEXT:    lsl z20.d, p0/m, z20.d, z3.d
+; CHECK-NEXT:    tbz w8, #4, .LBB37_10
+; CHECK-NEXT:  // %bb.9: // %cond.store7
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z22.h, z18.h[4]
+; CHECK-NEXT:    str h22, [x9]
+; CHECK-NEXT:  .LBB37_10: // %else8
+; CHECK-NEXT:    sunpklo z22.d, z19.s
+; CHECK-NEXT:    add z20.d, z6.d, z20.d
+; CHECK-NEXT:    tbz w8, #5, .LBB37_12
+; CHECK-NEXT:  // %bb.11: // %cond.store9
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.h, z18.h[5]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:  .LBB37_12: // %else10
+; CHECK-NEXT:    ext z19.b, z19.b, z19.b, #8
+; CHECK-NEXT:    lsl z22.d, p0/m, z22.d, z3.d
+; CHECK-NEXT:    tbz w8, #6, .LBB37_14
+; CHECK-NEXT:  // %bb.13: // %cond.store11
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z21.h, z18.h[6]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:  .LBB37_14: // %else12
+; CHECK-NEXT:    sunpklo z21.d, z19.s
+; CHECK-NEXT:    add z19.d, z6.d, z22.d
+; CHECK-NEXT:    tbz w8, #7, .LBB37_16
+; CHECK-NEXT:  // %bb.15: // %cond.store13
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    mov z18.h, z18.h[7]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB37_16: // %else14
+; CHECK-NEXT:    movprfx z18, z21
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #8, .LBB37_18
+; CHECK-NEXT:  // %bb.17: // %cond.store15
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB37_18: // %else16
+; CHECK-NEXT:    sunpklo z20.d, z17.s
+; CHECK-NEXT:    add z18.d, z6.d, z18.d
+; CHECK-NEXT:    tbz w8, #9, .LBB37_20
+; CHECK-NEXT:  // %bb.19: // %cond.store17
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z19.h, z7.h[1]
+; CHECK-NEXT:    str h19, [x9]
+; CHECK-NEXT:  .LBB37_20: // %else18
+; CHECK-NEXT:    ext z17.b, z17.b, z17.b, #8
+; CHECK-NEXT:    movprfx z19, z20
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #10, .LBB37_22
+; CHECK-NEXT:  // %bb.21: // %cond.store19
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z20.h, z7.h[2]
+; CHECK-NEXT:    str h20, [x9]
+; CHECK-NEXT:  .LBB37_22: // %else20
+; CHECK-NEXT:    sunpklo z17.d, z17.s
+; CHECK-NEXT:    add z19.d, z6.d, z19.d
+; CHECK-NEXT:    tbz w8, #11, .LBB37_24
+; CHECK-NEXT:  // %bb.23: // %cond.store21
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z18.h, z7.h[3]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB37_24: // %else22
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z3.d
+; CHECK-NEXT:    tbz w8, #12, .LBB37_26
+; CHECK-NEXT:  // %bb.25: // %cond.store23
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z18.h, z7.h[4]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB37_26: // %else24
+; CHECK-NEXT:    sunpklo z18.d, z16.s
+; CHECK-NEXT:    add z17.d, z6.d, z17.d
+; CHECK-NEXT:    tbz w8, #13, .LBB37_28
+; CHECK-NEXT:  // %bb.27: // %cond.store25
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z19.h, z7.h[5]
+; CHECK-NEXT:    str h19, [x9]
+; CHECK-NEXT:  .LBB37_28: // %else26
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    movprfx z19, z18
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #14, .LBB37_30
+; CHECK-NEXT:  // %bb.29: // %cond.store27
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z18.h, z7.h[6]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB37_30: // %else28
+; CHECK-NEXT:    sunpklo z18.d, z16.s
+; CHECK-NEXT:    add z16.d, z6.d, z19.d
+; CHECK-NEXT:    tbz w8, #15, .LBB37_32
+; CHECK-NEXT:  // %bb.31: // %cond.store29
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    mov z7.h, z7.h[7]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB37_32: // %else30
+; CHECK-NEXT:    movprfx z7, z18
+; CHECK-NEXT:    lsl z7.d, p0/m, z7.d, z3.d
+; CHECK-NEXT:    tbz w8, #16, .LBB37_34
+; CHECK-NEXT:  // %bb.33: // %cond.store31
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:  .LBB37_34: // %else32
+; CHECK-NEXT:    sunpklo z17.d, z5.s
+; CHECK-NEXT:    add z7.d, z6.d, z7.d
+; CHECK-NEXT:    tbz w8, #17, .LBB37_36
+; CHECK-NEXT:  // %bb.35: // %cond.store33
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.h, z2.h[1]
+; CHECK-NEXT:    str h16, [x9]
+; CHECK-NEXT:  .LBB37_36: // %else34
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    movprfx z16, z17
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z3.d
+; CHECK-NEXT:    tbz w8, #18, .LBB37_38
+; CHECK-NEXT:  // %bb.37: // %cond.store35
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z17.h, z2.h[2]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:  .LBB37_38: // %else36
+; CHECK-NEXT:    sunpklo z5.d, z5.s
+; CHECK-NEXT:    add z16.d, z6.d, z16.d
+; CHECK-NEXT:    tbz w8, #19, .LBB37_40
+; CHECK-NEXT:  // %bb.39: // %cond.store37
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z7.h, z2.h[3]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB37_40: // %else38
+; CHECK-NEXT:    lsl z5.d, p0/m, z5.d, z3.d
+; CHECK-NEXT:    tbz w8, #20, .LBB37_42
+; CHECK-NEXT:  // %bb.41: // %cond.store39
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z7.h, z2.h[4]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB37_42: // %else40
+; CHECK-NEXT:    sunpklo z7.d, z4.s
+; CHECK-NEXT:    add z5.d, z6.d, z5.d
+; CHECK-NEXT:    tbz w8, #21, .LBB37_44
+; CHECK-NEXT:  // %bb.43: // %cond.store41
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.h, z2.h[5]
+; CHECK-NEXT:    str h16, [x9]
+; CHECK-NEXT:  .LBB37_44: // %else42
+; CHECK-NEXT:    ext z4.b, z4.b, z4.b, #8
+; CHECK-NEXT:    movprfx z16, z7
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z3.d
+; CHECK-NEXT:    tbz w8, #22, .LBB37_46
+; CHECK-NEXT:  // %bb.45: // %cond.store43
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z7.h, z2.h[6]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB37_46: // %else44
+; CHECK-NEXT:    sunpklo z7.d, z4.s
+; CHECK-NEXT:    add z4.d, z6.d, z16.d
+; CHECK-NEXT:    tbz w8, #23, .LBB37_48
+; CHECK-NEXT:  // %bb.47: // %cond.store45
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z2.h, z2.h[7]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:  .LBB37_48: // %else46
+; CHECK-NEXT:    movprfx z2, z7
+; CHECK-NEXT:    lsl z2.d, p0/m, z2.d, z3.d
+; CHECK-NEXT:    tbz w8, #24, .LBB37_50
+; CHECK-NEXT:  // %bb.49: // %cond.store47
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:  .LBB37_50: // %else48
+; CHECK-NEXT:    sunpklo z5.d, z1.s
+; CHECK-NEXT:    add z2.d, z6.d, z2.d
+; CHECK-NEXT:    tbz w8, #25, .LBB37_52
+; CHECK-NEXT:  // %bb.51: // %cond.store49
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z4.h, z0.h[1]
+; CHECK-NEXT:    str h4, [x9]
+; CHECK-NEXT:  .LBB37_52: // %else50
+; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    lsl z5.d, p0/m, z5.d, z3.d
+; CHECK-NEXT:    tbz w8, #26, .LBB37_54
+; CHECK-NEXT:  // %bb.53: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z4.h, z0.h[2]
+; CHECK-NEXT:    str h4, [x9]
+; CHECK-NEXT:  .LBB37_54: // %else52
+; CHECK-NEXT:    sunpklo z4.d, z1.s
+; CHECK-NEXT:    add z1.d, z6.d, z5.d
+; CHECK-NEXT:    tbz w8, #27, .LBB37_56
+; CHECK-NEXT:  // %bb.55: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[3]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:  .LBB37_56: // %else54
+; CHECK-NEXT:    movprfx z2, z4
+; CHECK-NEXT:    lsl z2.d, p0/m, z2.d, z3.d
+; CHECK-NEXT:    tbnz w8, #28, .LBB37_61
+; CHECK-NEXT:  // %bb.57: // %else56
+; CHECK-NEXT:    add z2.d, z6.d, z2.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB37_62
+; CHECK-NEXT:  .LBB37_58: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB37_63
+; CHECK-NEXT:  .LBB37_59: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB37_64
+; CHECK-NEXT:  .LBB37_60: // %else62
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB37_61: // %cond.store55
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z3.h, z0.h[4]
+; CHECK-NEXT:    str h3, [x9]
+; CHECK-NEXT:    add z2.d, z6.d, z2.d
+; CHECK-NEXT:    tbz w8, #29, .LBB37_58
+; CHECK-NEXT:  .LBB37_62: // %cond.store57
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.h, z0.h[5]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB37_59
+; CHECK-NEXT:  .LBB37_63: // %cond.store59
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z1.h, z0.h[6]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #31, .LBB37_60
+; CHECK-NEXT:  .LBB37_64: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #32
 ; CHECK-NEXT:    ret
   %vals = load <32 x half>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = sext <32 x i32> %idxs to <32 x i64>
   %ptrs = getelementptr half, ptr %base, <32 x i64> %ext
   %mask = fcmp oeq <32 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_32b_scaled_sext_f32(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_scatter_32b_scaled_sext_f32:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    st1w { z0.s }, p0, [x2, z1.s, sxtw #2]
+; CHECK-NEXT:    adrp x8, .LCPI38_1
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q5, q2, [x0, #64]
+; CHECK-NEXT:    ldr q7, [x8, :lo12:.LCPI38_1]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z5.s, z7.s
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z7.s
+; CHECK-NEXT:    fmov w9, s4
+; CHECK-NEXT:    mov z6.s, z4.s[1]
+; CHECK-NEXT:    mov z20.s, z4.s[2]
+; CHECK-NEXT:    fmov w8, s6
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    mov z24.s, z4.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z26.s, z4.s[2]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s24
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    mov z24.s, z4.s[1]
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z27.s, z4.s[3]
+; CHECK-NEXT:    bfi w9, w8, #19, #1
+; CHECK-NEXT:    fmov w8, s26
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z7.s
+; CHECK-NEXT:    bfi w9, w10, #20, #1
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w9, w11, #21, #1
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    ldp q19, q17, [x0, #32]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w9, w8, lsl #22
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z26.s, z4.s[1]
+; CHECK-NEXT:    mov z27.s, z4.s[2]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #23
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    fmov w9, s26
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    mov z28.s, z4.s[3]
+; CHECK-NEXT:    ldp q23, q21, [x0]
+; CHECK-NEXT:    adrp x11, .LCPI38_0
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldp q6, q3, [x1, #96]
+; CHECK-NEXT:    ldp q18, q16, [x1, #64]
+; CHECK-NEXT:    ldp q22, q20, [x1, #32]
+; CHECK-NEXT:    ldp q25, q24, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, z7.s
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    fmov w9, s28
+; CHECK-NEXT:    mov z27.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z23.s, z7.s
+; CHECK-NEXT:    mov z28.s, z27.s[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z29.s, z27.s[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z26.s, z27.s[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    mov z27.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    mov z30.s, z27.s[1]
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    fmov w9, s30
+; CHECK-NEXT:    mov z30.s, z27.s[2]
+; CHECK-NEXT:    mov z31.s, z27.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z21.s, z7.s
+; CHECK-NEXT:    fmov w12, s31
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z27.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    fmov w9, s30
+; CHECK-NEXT:    ldr q4, [x11, :lo12:.LCPI38_0]
+; CHECK-NEXT:    fmov w11, s28
+; CHECK-NEXT:    mov z28.s, z27.s[1]
+; CHECK-NEXT:    fmov w13, s27
+; CHECK-NEXT:    bfi w10, w9, #2, #1
+; CHECK-NEXT:    fmov w9, s28
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z30.s, z27.s[2]
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    mov z31.s, z27.s[3]
+; CHECK-NEXT:    bfi w10, w9, #5, #1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z19.s, z7.s
+; CHECK-NEXT:    fmov w9, s30
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z27.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w12, s31
+; CHECK-NEXT:    fmov w13, s27
+; CHECK-NEXT:    orr w8, w8, w11, lsl #29
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z28.s, z27.s[1]
+; CHECK-NEXT:    orr w9, w10, w9, lsl #6
+; CHECK-NEXT:    and w10, w13, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w12, lsl #7
+; CHECK-NEXT:    mov z29.s, z27.s[2]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s28
+; CHECK-NEXT:    mov z30.s, z27.s[3]
+; CHECK-NEXT:    orr w8, w8, w11, lsl #30
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    fmov w12, s30
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z17.s, z7.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z7.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z27.s, z7.s[1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    mov z28.s, z7.s[2]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    fmov w12, s28
+; CHECK-NEXT:    mov z29.s, z7.s[3]
+; CHECK-NEXT:    ldr q7, [sp]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    sunpklo z26.d, z25.s
+; CHECK-NEXT:    ext z25.b, z25.b, z25.b, #8
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    sunpklo z25.d, z25.s
+; CHECK-NEXT:    orr w8, w8, w10, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    lsl z26.d, p0/m, z26.d, z4.d
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z4.d
+; CHECK-NEXT:    add z27.d, z7.d, z26.d
+; CHECK-NEXT:    tbz w8, #0, .LBB38_2
+; CHECK-NEXT:  // %bb.1: // %cond.store
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:  .LBB38_2: // %else
+; CHECK-NEXT:    sunpklo z26.d, z24.s
+; CHECK-NEXT:    add z25.d, z7.d, z25.d
+; CHECK-NEXT:    tbz w8, #1, .LBB38_4
+; CHECK-NEXT:  // %bb.3: // %cond.store1
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    mov z27.s, z23.s[1]
+; CHECK-NEXT:    str s27, [x9]
+; CHECK-NEXT:  .LBB38_4: // %else2
+; CHECK-NEXT:    ext z24.b, z24.b, z24.b, #8
+; CHECK-NEXT:    movprfx z27, z26
+; CHECK-NEXT:    lsl z27.d, p0/m, z27.d, z4.d
+; CHECK-NEXT:    tbz w8, #2, .LBB38_6
+; CHECK-NEXT:  // %bb.5: // %cond.store3
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    mov z26.s, z23.s[2]
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:  .LBB38_6: // %else4
+; CHECK-NEXT:    sunpklo z26.d, z24.s
+; CHECK-NEXT:    add z24.d, z7.d, z27.d
+; CHECK-NEXT:    tbz w8, #3, .LBB38_8
+; CHECK-NEXT:  // %bb.7: // %cond.store5
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    mov z23.s, z23.s[3]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:  .LBB38_8: // %else6
+; CHECK-NEXT:    movprfx z23, z26
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z4.d
+; CHECK-NEXT:    tbz w8, #4, .LBB38_10
+; CHECK-NEXT:  // %bb.9: // %cond.store7
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    str s21, [x9]
+; CHECK-NEXT:  .LBB38_10: // %else8
+; CHECK-NEXT:    sunpklo z25.d, z22.s
+; CHECK-NEXT:    add z23.d, z7.d, z23.d
+; CHECK-NEXT:    tbz w8, #5, .LBB38_12
+; CHECK-NEXT:  // %bb.11: // %cond.store9
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    mov z24.s, z21.s[1]
+; CHECK-NEXT:    str s24, [x9]
+; CHECK-NEXT:  .LBB38_12: // %else10
+; CHECK-NEXT:    ext z22.b, z22.b, z22.b, #8
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z4.d
+; CHECK-NEXT:    tbz w8, #6, .LBB38_14
+; CHECK-NEXT:  // %bb.13: // %cond.store11
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    mov z24.s, z21.s[2]
+; CHECK-NEXT:    str s24, [x9]
+; CHECK-NEXT:  .LBB38_14: // %else12
+; CHECK-NEXT:    sunpklo z24.d, z22.s
+; CHECK-NEXT:    add z22.d, z7.d, z25.d
+; CHECK-NEXT:    tbz w8, #7, .LBB38_16
+; CHECK-NEXT:  // %bb.15: // %cond.store13
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    mov z21.s, z21.s[3]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    str s21, [x9]
+; CHECK-NEXT:  .LBB38_16: // %else14
+; CHECK-NEXT:    movprfx z21, z24
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z4.d
+; CHECK-NEXT:    tbz w8, #8, .LBB38_18
+; CHECK-NEXT:  // %bb.17: // %cond.store15
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:  .LBB38_18: // %else16
+; CHECK-NEXT:    sunpklo z23.d, z20.s
+; CHECK-NEXT:    add z21.d, z7.d, z21.d
+; CHECK-NEXT:    tbz w8, #9, .LBB38_20
+; CHECK-NEXT:  // %bb.19: // %cond.store17
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z22.s, z19.s[1]
+; CHECK-NEXT:    str s22, [x9]
+; CHECK-NEXT:  .LBB38_20: // %else18
+; CHECK-NEXT:    ext z20.b, z20.b, z20.b, #8
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z4.d
+; CHECK-NEXT:    tbz w8, #10, .LBB38_22
+; CHECK-NEXT:  // %bb.21: // %cond.store19
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z22.s, z19.s[2]
+; CHECK-NEXT:    str s22, [x9]
+; CHECK-NEXT:  .LBB38_22: // %else20
+; CHECK-NEXT:    sunpklo z22.d, z20.s
+; CHECK-NEXT:    add z20.d, z7.d, z23.d
+; CHECK-NEXT:    tbz w8, #11, .LBB38_24
+; CHECK-NEXT:  // %bb.23: // %cond.store21
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    mov z19.s, z19.s[3]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:  .LBB38_24: // %else22
+; CHECK-NEXT:    movprfx z19, z22
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z4.d
+; CHECK-NEXT:    tbz w8, #12, .LBB38_26
+; CHECK-NEXT:  // %bb.25: // %cond.store23
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str s17, [x9]
+; CHECK-NEXT:  .LBB38_26: // %else24
+; CHECK-NEXT:    sunpklo z21.d, z18.s
+; CHECK-NEXT:    add z19.d, z7.d, z19.d
+; CHECK-NEXT:    tbz w8, #13, .LBB38_28
+; CHECK-NEXT:  // %bb.27: // %cond.store25
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z20.s, z17.s[1]
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:  .LBB38_28: // %else26
+; CHECK-NEXT:    ext z18.b, z18.b, z18.b, #8
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z4.d
+; CHECK-NEXT:    tbz w8, #14, .LBB38_30
+; CHECK-NEXT:  // %bb.29: // %cond.store27
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z20.s, z17.s[2]
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:  .LBB38_30: // %else28
+; CHECK-NEXT:    sunpklo z20.d, z18.s
+; CHECK-NEXT:    add z18.d, z7.d, z21.d
+; CHECK-NEXT:    tbz w8, #15, .LBB38_32
+; CHECK-NEXT:  // %bb.31: // %cond.store29
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    mov z17.s, z17.s[3]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str s17, [x9]
+; CHECK-NEXT:  .LBB38_32: // %else30
+; CHECK-NEXT:    movprfx z17, z20
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z4.d
+; CHECK-NEXT:    tbz w8, #16, .LBB38_34
+; CHECK-NEXT:  // %bb.33: // %cond.store31
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:  .LBB38_34: // %else32
+; CHECK-NEXT:    sunpklo z19.d, z16.s
+; CHECK-NEXT:    add z17.d, z7.d, z17.d
+; CHECK-NEXT:    tbz w8, #17, .LBB38_36
+; CHECK-NEXT:  // %bb.35: // %cond.store33
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z18.s, z5.s[1]
+; CHECK-NEXT:    str s18, [x9]
+; CHECK-NEXT:  .LBB38_36: // %else34
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z4.d
+; CHECK-NEXT:    tbz w8, #18, .LBB38_38
+; CHECK-NEXT:  // %bb.37: // %cond.store35
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z18.s, z5.s[2]
+; CHECK-NEXT:    str s18, [x9]
+; CHECK-NEXT:  .LBB38_38: // %else36
+; CHECK-NEXT:    sunpklo z18.d, z16.s
+; CHECK-NEXT:    add z16.d, z7.d, z19.d
+; CHECK-NEXT:    tbz w8, #19, .LBB38_40
+; CHECK-NEXT:  // %bb.39: // %cond.store37
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    mov z5.s, z5.s[3]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:  .LBB38_40: // %else38
+; CHECK-NEXT:    movprfx z5, z18
+; CHECK-NEXT:    lsl z5.d, p0/m, z5.d, z4.d
+; CHECK-NEXT:    tbz w8, #20, .LBB38_42
+; CHECK-NEXT:  // %bb.41: // %cond.store39
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:  .LBB38_42: // %else40
+; CHECK-NEXT:    sunpklo z17.d, z6.s
+; CHECK-NEXT:    add z5.d, z7.d, z5.d
+; CHECK-NEXT:    tbz w8, #21, .LBB38_44
+; CHECK-NEXT:  // %bb.43: // %cond.store41
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.s, z2.s[1]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:  .LBB38_44: // %else42
+; CHECK-NEXT:    ext z6.b, z6.b, z6.b, #8
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z4.d
+; CHECK-NEXT:    tbz w8, #22, .LBB38_46
+; CHECK-NEXT:  // %bb.45: // %cond.store43
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z16.s, z2.s[2]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:  .LBB38_46: // %else44
+; CHECK-NEXT:    sunpklo z16.d, z6.s
+; CHECK-NEXT:    add z6.d, z7.d, z17.d
+; CHECK-NEXT:    tbz w8, #23, .LBB38_48
+; CHECK-NEXT:  // %bb.47: // %cond.store45
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z2.s, z2.s[3]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:  .LBB38_48: // %else46
+; CHECK-NEXT:    movprfx z2, z16
+; CHECK-NEXT:    lsl z2.d, p0/m, z2.d, z4.d
+; CHECK-NEXT:    tbz w8, #24, .LBB38_50
+; CHECK-NEXT:  // %bb.49: // %cond.store47
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:  .LBB38_50: // %else48
+; CHECK-NEXT:    sunpklo z5.d, z3.s
+; CHECK-NEXT:    add z2.d, z7.d, z2.d
+; CHECK-NEXT:    tbz w8, #25, .LBB38_52
+; CHECK-NEXT:  // %bb.51: // %cond.store49
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.s, z1.s[1]
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:  .LBB38_52: // %else50
+; CHECK-NEXT:    ext z3.b, z3.b, z3.b, #8
+; CHECK-NEXT:    movprfx z6, z5
+; CHECK-NEXT:    lsl z6.d, p0/m, z6.d, z4.d
+; CHECK-NEXT:    tbz w8, #26, .LBB38_54
+; CHECK-NEXT:  // %bb.53: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z5.s, z1.s[2]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:  .LBB38_54: // %else52
+; CHECK-NEXT:    sunpklo z5.d, z3.s
+; CHECK-NEXT:    add z3.d, z7.d, z6.d
+; CHECK-NEXT:    tbz w8, #27, .LBB38_56
+; CHECK-NEXT:  // %bb.55: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:  .LBB38_56: // %else54
+; CHECK-NEXT:    movprfx z1, z5
+; CHECK-NEXT:    lsl z1.d, p0/m, z1.d, z4.d
+; CHECK-NEXT:    tbnz w8, #28, .LBB38_61
+; CHECK-NEXT:  // %bb.57: // %else56
+; CHECK-NEXT:    add z1.d, z7.d, z1.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB38_62
+; CHECK-NEXT:  .LBB38_58: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB38_63
+; CHECK-NEXT:  .LBB38_59: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB38_64
+; CHECK-NEXT:  .LBB38_60: // %else62
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB38_61: // %cond.store55
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    add z1.d, z7.d, z1.d
+; CHECK-NEXT:    tbz w8, #29, .LBB38_58
+; CHECK-NEXT:  .LBB38_62: // %cond.store57
+; CHECK-NEXT:    mov z2.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB38_59
+; CHECK-NEXT:  .LBB38_63: // %cond.store59
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #31, .LBB38_60
+; CHECK-NEXT:  .LBB38_64: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:    add sp, sp, #32
 ; CHECK-NEXT:    ret
   %vals = load <32 x float>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = sext <32 x i32> %idxs to <32 x i64>
   %ptrs = getelementptr float, ptr %base, <32 x i64> %ext
   %mask = fcmp oeq <32 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_32b_scaled_sext_f64(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_scatter_32b_scaled_sext_f64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl32
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    ld1sw { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    st1d { z0.d }, p0, [x2, z1.d, lsl #3]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #64
+; CHECK-NEXT:    .cfi_def_cfa_offset 64
+; CHECK-NEXT:    str d12, [sp, #16] // 8-byte Folded Spill
+; CHECK-NEXT:    stp d11, d10, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #48] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -24
+; CHECK-NEXT:    .cfi_offset b11, -32
+; CHECK-NEXT:    .cfi_offset b12, -48
+; CHECK-NEXT:    adrp x8, .LCPI39_1
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldp q19, q7, [x0, #128]
+; CHECK-NEXT:    ldr q22, [x8, :lo12:.LCPI39_1]
+; CHECK-NEXT:    ldp q16, q5, [x0, #160]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z19.d, z22.d
+; CHECK-NEXT:    mov z0.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z7.d, z22.d
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    mov z1.d, z0.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.d, z3.d[1]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z16.d, z22.d
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z3.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z5.d, z22.d
+; CHECK-NEXT:    ldp q4, q2, [x0, #192]
+; CHECK-NEXT:    bfi w8, w9, #18, #1
+; CHECK-NEXT:    mov z21.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.d, z3.d[1]
+; CHECK-NEXT:    fmov x11, d3
+; CHECK-NEXT:    bfi w8, w10, #19, #1
+; CHECK-NEXT:    fmov x10, d21
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.d, z21.d[1]
+; CHECK-NEXT:    bfi w8, w11, #20, #1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z4.d, z22.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    bfi w8, w9, #21, #1
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #22
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    mov z10.d, z6.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z2.d, z22.d
+; CHECK-NEXT:    ldp q1, q0, [x0, #224]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    adrp x9, .LCPI39_0
+; CHECK-NEXT:    mov z6.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov x10, d10
+; CHECK-NEXT:    mov z10.d, z6.d[1]
+; CHECK-NEXT:    fmov x11, d6
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z22.d
+; CHECK-NEXT:    ldp q31, q30, [x0]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    orr w8, w8, w11, lsl #26
+; CHECK-NEXT:    ldr q6, [x9, :lo12:.LCPI39_0]
+; CHECK-NEXT:    fmov x9, d10
+; CHECK-NEXT:    mov z10.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z31.d, z22.d
+; CHECK-NEXT:    fmov x10, d10
+; CHECK-NEXT:    mov z11.d, z10.d[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z10.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q28, q27, [x0, #32]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    mov z12.d, z10.d[1]
+; CHECK-NEXT:    fmov x11, d10
+; CHECK-NEXT:    orr w8, w8, w10, lsl #28
+; CHECK-NEXT:    fmov x10, d12
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z30.d, z22.d
+; CHECK-NEXT:    fmov x9, d11
+; CHECK-NEXT:    mov z10.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z11.d, z10.d[1]
+; CHECK-NEXT:    fmov x12, d10
+; CHECK-NEXT:    bfi w11, w10, #1, #1
+; CHECK-NEXT:    fmov x10, d11
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z28.d, z22.d
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z10.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z27.d, z22.d
+; CHECK-NEXT:    ldp q25, q24, [x0, #64]
+; CHECK-NEXT:    bfi w11, w12, #2, #1
+; CHECK-NEXT:    mov z11.d, z10.d[1]
+; CHECK-NEXT:    fmov x12, d10
+; CHECK-NEXT:    mov z10.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    bfi w11, w10, #3, #1
+; CHECK-NEXT:    fmov x9, d11
+; CHECK-NEXT:    fmov x10, d10
+; CHECK-NEXT:    bfi w11, w12, #4, #1
+; CHECK-NEXT:    mov z11.d, z10.d[1]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z25.d, z22.d
+; CHECK-NEXT:    bfi w11, w9, #5, #1
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov x10, d11
+; CHECK-NEXT:    mov z10.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov x12, d10
+; CHECK-NEXT:    mov z11.d, z10.d[1]
+; CHECK-NEXT:    orr w9, w11, w9, lsl #6
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z0.d, z22.d
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z10.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q20, q18, [x0, #96]
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov x10, d11
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z24.d, z22.d
+; CHECK-NEXT:    mov z11.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov x11, d11
+; CHECK-NEXT:    mov z12.d, z11.d[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z20.d, z22.d
+; CHECK-NEXT:    mov z11.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z18.d, z22.d
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov x11, d12
+; CHECK-NEXT:    fmov x12, d11
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    mov z12.d, z11.d[1]
+; CHECK-NEXT:    mov z22.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z11.d, z10.d[1]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    fmov x12, d22
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov x11, d10
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov x10, d12
+; CHECK-NEXT:    mov z10.d, z22.d[1]
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    ldp q9, q8, [x1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov x10, d10
+; CHECK-NEXT:    orr w9, w9, w12, lsl #14
+; CHECK-NEXT:    fmov x12, d11
+; CHECK-NEXT:    orr w8, w8, w11, lsl #30
+; CHECK-NEXT:    sunpklo z10.d, z9.s
+; CHECK-NEXT:    ldp q17, q3, [x1, #96]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w12, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    ext z9.b, z9.b, z9.b, #8
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    sunpklo z9.d, z9.s
+; CHECK-NEXT:    lsl z10.d, p0/m, z10.d, z6.d
+; CHECK-NEXT:    lsl z9.d, p0/m, z9.d, z6.d
+; CHECK-NEXT:    ldp q23, q21, [x1, #64]
+; CHECK-NEXT:    ldp q29, q26, [x1, #32]
+; CHECK-NEXT:    stp x2, x2, [sp]
+; CHECK-NEXT:    ldr q22, [sp]
+; CHECK-NEXT:    add z10.d, z22.d, z10.d
+; CHECK-NEXT:    tbz w8, #0, .LBB39_2
+; CHECK-NEXT:  // %bb.1: // %cond.store
+; CHECK-NEXT:    fmov x9, d10
+; CHECK-NEXT:    str d31, [x9]
+; CHECK-NEXT:  .LBB39_2: // %else
+; CHECK-NEXT:    sunpklo z11.d, z8.s
+; CHECK-NEXT:    add z9.d, z22.d, z9.d
+; CHECK-NEXT:    tbz w8, #1, .LBB39_4
+; CHECK-NEXT:  // %bb.3: // %cond.store1
+; CHECK-NEXT:    mov z10.d, z10.d[1]
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    fmov x9, d10
+; CHECK-NEXT:    str d31, [x9]
+; CHECK-NEXT:  .LBB39_4: // %else2
+; CHECK-NEXT:    ext z8.b, z8.b, z8.b, #8
+; CHECK-NEXT:    movprfx z31, z11
+; CHECK-NEXT:    lsl z31.d, p0/m, z31.d, z6.d
+; CHECK-NEXT:    tbz w8, #2, .LBB39_6
+; CHECK-NEXT:  // %bb.5: // %cond.store3
+; CHECK-NEXT:    fmov x9, d9
+; CHECK-NEXT:    str d30, [x9]
+; CHECK-NEXT:  .LBB39_6: // %else4
+; CHECK-NEXT:    sunpklo z8.d, z8.s
+; CHECK-NEXT:    add z31.d, z22.d, z31.d
+; CHECK-NEXT:    tbz w8, #3, .LBB39_8
+; CHECK-NEXT:  // %bb.7: // %cond.store5
+; CHECK-NEXT:    mov z9.d, z9.d[1]
+; CHECK-NEXT:    mov z30.d, z30.d[1]
+; CHECK-NEXT:    fmov x9, d9
+; CHECK-NEXT:    str d30, [x9]
+; CHECK-NEXT:  .LBB39_8: // %else6
+; CHECK-NEXT:    movprfx z30, z8
+; CHECK-NEXT:    lsl z30.d, p0/m, z30.d, z6.d
+; CHECK-NEXT:    tbz w8, #4, .LBB39_10
+; CHECK-NEXT:  // %bb.9: // %cond.store7
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    str d28, [x9]
+; CHECK-NEXT:  .LBB39_10: // %else8
+; CHECK-NEXT:    sunpklo z8.d, z29.s
+; CHECK-NEXT:    add z30.d, z22.d, z30.d
+; CHECK-NEXT:    tbz w8, #5, .LBB39_12
+; CHECK-NEXT:  // %bb.11: // %cond.store9
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    mov z28.d, z28.d[1]
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    str d28, [x9]
+; CHECK-NEXT:  .LBB39_12: // %else10
+; CHECK-NEXT:    ext z29.b, z29.b, z29.b, #8
+; CHECK-NEXT:    movprfx z28, z8
+; CHECK-NEXT:    lsl z28.d, p0/m, z28.d, z6.d
+; CHECK-NEXT:    tbz w8, #6, .LBB39_14
+; CHECK-NEXT:  // %bb.13: // %cond.store11
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    str d27, [x9]
+; CHECK-NEXT:  .LBB39_14: // %else12
+; CHECK-NEXT:    sunpklo z29.d, z29.s
+; CHECK-NEXT:    add z28.d, z22.d, z28.d
+; CHECK-NEXT:    tbz w8, #7, .LBB39_16
+; CHECK-NEXT:  // %bb.15: // %cond.store13
+; CHECK-NEXT:    mov z30.d, z30.d[1]
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    str d27, [x9]
+; CHECK-NEXT:  .LBB39_16: // %else14
+; CHECK-NEXT:    movprfx z27, z29
+; CHECK-NEXT:    lsl z27.d, p0/m, z27.d, z6.d
+; CHECK-NEXT:    tbz w8, #8, .LBB39_18
+; CHECK-NEXT:  // %bb.17: // %cond.store15
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    str d25, [x9]
+; CHECK-NEXT:  .LBB39_18: // %else16
+; CHECK-NEXT:    sunpklo z29.d, z26.s
+; CHECK-NEXT:    add z27.d, z22.d, z27.d
+; CHECK-NEXT:    tbz w8, #9, .LBB39_20
+; CHECK-NEXT:  // %bb.19: // %cond.store17
+; CHECK-NEXT:    mov z28.d, z28.d[1]
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    str d25, [x9]
+; CHECK-NEXT:  .LBB39_20: // %else18
+; CHECK-NEXT:    ext z26.b, z26.b, z26.b, #8
+; CHECK-NEXT:    movprfx z25, z29
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z6.d
+; CHECK-NEXT:    tbz w8, #10, .LBB39_22
+; CHECK-NEXT:  // %bb.21: // %cond.store19
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str d24, [x9]
+; CHECK-NEXT:  .LBB39_22: // %else20
+; CHECK-NEXT:    sunpklo z26.d, z26.s
+; CHECK-NEXT:    add z25.d, z22.d, z25.d
+; CHECK-NEXT:    tbz w8, #11, .LBB39_24
+; CHECK-NEXT:  // %bb.23: // %cond.store21
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str d24, [x9]
+; CHECK-NEXT:  .LBB39_24: // %else22
+; CHECK-NEXT:    movprfx z24, z26
+; CHECK-NEXT:    lsl z24.d, p0/m, z24.d, z6.d
+; CHECK-NEXT:    tbz w8, #12, .LBB39_26
+; CHECK-NEXT:  // %bb.25: // %cond.store23
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str d20, [x9]
+; CHECK-NEXT:  .LBB39_26: // %else24
+; CHECK-NEXT:    sunpklo z26.d, z23.s
+; CHECK-NEXT:    add z24.d, z22.d, z24.d
+; CHECK-NEXT:    tbz w8, #13, .LBB39_28
+; CHECK-NEXT:  // %bb.27: // %cond.store25
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str d20, [x9]
+; CHECK-NEXT:  .LBB39_28: // %else26
+; CHECK-NEXT:    ext z23.b, z23.b, z23.b, #8
+; CHECK-NEXT:    movprfx z20, z26
+; CHECK-NEXT:    lsl z20.d, p0/m, z20.d, z6.d
+; CHECK-NEXT:    tbz w8, #14, .LBB39_30
+; CHECK-NEXT:  // %bb.29: // %cond.store27
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    str d18, [x9]
+; CHECK-NEXT:  .LBB39_30: // %else28
+; CHECK-NEXT:    sunpklo z23.d, z23.s
+; CHECK-NEXT:    add z20.d, z22.d, z20.d
+; CHECK-NEXT:    tbz w8, #15, .LBB39_32
+; CHECK-NEXT:  // %bb.31: // %cond.store29
+; CHECK-NEXT:    mov z24.d, z24.d[1]
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    str d18, [x9]
+; CHECK-NEXT:  .LBB39_32: // %else30
+; CHECK-NEXT:    movprfx z18, z23
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z6.d
+; CHECK-NEXT:    tbz w8, #16, .LBB39_34
+; CHECK-NEXT:  // %bb.33: // %cond.store31
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str d19, [x9]
+; CHECK-NEXT:  .LBB39_34: // %else32
+; CHECK-NEXT:    sunpklo z23.d, z21.s
+; CHECK-NEXT:    add z18.d, z22.d, z18.d
+; CHECK-NEXT:    tbz w8, #17, .LBB39_36
+; CHECK-NEXT:  // %bb.35: // %cond.store33
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str d19, [x9]
+; CHECK-NEXT:  .LBB39_36: // %else34
+; CHECK-NEXT:    ext z21.b, z21.b, z21.b, #8
+; CHECK-NEXT:    movprfx z19, z23
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z6.d
+; CHECK-NEXT:    tbz w8, #18, .LBB39_38
+; CHECK-NEXT:  // %bb.37: // %cond.store35
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    str d7, [x9]
+; CHECK-NEXT:  .LBB39_38: // %else36
+; CHECK-NEXT:    sunpklo z20.d, z21.s
+; CHECK-NEXT:    add z19.d, z22.d, z19.d
+; CHECK-NEXT:    tbz w8, #19, .LBB39_40
+; CHECK-NEXT:  // %bb.39: // %cond.store37
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    str d7, [x9]
+; CHECK-NEXT:  .LBB39_40: // %else38
+; CHECK-NEXT:    movprfx z7, z20
+; CHECK-NEXT:    lsl z7.d, p0/m, z7.d, z6.d
+; CHECK-NEXT:    tbz w8, #20, .LBB39_42
+; CHECK-NEXT:  // %bb.41: // %cond.store39
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str d16, [x9]
+; CHECK-NEXT:  .LBB39_42: // %else40
+; CHECK-NEXT:    sunpklo z18.d, z17.s
+; CHECK-NEXT:    add z7.d, z22.d, z7.d
+; CHECK-NEXT:    tbz w8, #21, .LBB39_44
+; CHECK-NEXT:  // %bb.43: // %cond.store41
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str d16, [x9]
+; CHECK-NEXT:  .LBB39_44: // %else42
+; CHECK-NEXT:    ext z17.b, z17.b, z17.b, #8
+; CHECK-NEXT:    movprfx z16, z18
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z6.d
+; CHECK-NEXT:    tbz w8, #22, .LBB39_46
+; CHECK-NEXT:  // %bb.45: // %cond.store43
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str d5, [x9]
+; CHECK-NEXT:  .LBB39_46: // %else44
+; CHECK-NEXT:    sunpklo z17.d, z17.s
+; CHECK-NEXT:    add z16.d, z22.d, z16.d
+; CHECK-NEXT:    tbz w8, #23, .LBB39_48
+; CHECK-NEXT:  // %bb.47: // %cond.store45
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str d5, [x9]
+; CHECK-NEXT:  .LBB39_48: // %else46
+; CHECK-NEXT:    movprfx z5, z17
+; CHECK-NEXT:    lsl z5.d, p0/m, z5.d, z6.d
+; CHECK-NEXT:    tbz w8, #24, .LBB39_50
+; CHECK-NEXT:  // %bb.49: // %cond.store47
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str d4, [x9]
+; CHECK-NEXT:  .LBB39_50: // %else48
+; CHECK-NEXT:    sunpklo z7.d, z3.s
+; CHECK-NEXT:    add z5.d, z22.d, z5.d
+; CHECK-NEXT:    tbz w8, #25, .LBB39_52
+; CHECK-NEXT:  // %bb.51: // %cond.store49
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str d4, [x9]
+; CHECK-NEXT:  .LBB39_52: // %else50
+; CHECK-NEXT:    ext z3.b, z3.b, z3.b, #8
+; CHECK-NEXT:    lsl z7.d, p0/m, z7.d, z6.d
+; CHECK-NEXT:    tbz w8, #26, .LBB39_54
+; CHECK-NEXT:  // %bb.53: // %cond.store51
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:  .LBB39_54: // %else52
+; CHECK-NEXT:    sunpklo z4.d, z3.s
+; CHECK-NEXT:    add z3.d, z22.d, z7.d
+; CHECK-NEXT:    tbz w8, #27, .LBB39_56
+; CHECK-NEXT:  // %bb.55: // %cond.store53
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:  .LBB39_56: // %else54
+; CHECK-NEXT:    movprfx z2, z4
+; CHECK-NEXT:    lsl z2.d, p0/m, z2.d, z6.d
+; CHECK-NEXT:    tbnz w8, #28, .LBB39_62
+; CHECK-NEXT:  // %bb.57: // %else56
+; CHECK-NEXT:    add z2.d, z22.d, z2.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB39_63
+; CHECK-NEXT:  .LBB39_58: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB39_64
+; CHECK-NEXT:  .LBB39_59: // %else60
+; CHECK-NEXT:    tbz w8, #31, .LBB39_61
+; CHECK-NEXT:  .LBB39_60: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str d0, [x8]
+; CHECK-NEXT:  .LBB39_61: // %else62
+; CHECK-NEXT:    ldp d9, d8, [sp, #48] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d11, d10, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    ldr d12, [sp, #16] // 8-byte Folded Reload
+; CHECK-NEXT:    add sp, sp, #64
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB39_62: // %cond.store55
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    add z2.d, z22.d, z2.d
+; CHECK-NEXT:    tbz w8, #29, .LBB39_58
+; CHECK-NEXT:  .LBB39_63: // %cond.store57
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB39_59
+; CHECK-NEXT:  .LBB39_64: // %cond.store59
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str d0, [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB39_60
+; CHECK-NEXT:    b .LBB39_61
   %vals = load <32 x double>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = sext <32 x i32> %idxs to <32 x i64>
   %ptrs = getelementptr double, ptr %base, <32 x i64> %ext
   %mask = fcmp oeq <32 x double> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f64(<32 x double> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_32b_scaled_zext(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_scatter_32b_scaled_zext:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.s, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1h { z0.s }, p0, [x2, z1.s, uxtw #1]
+; CHECK-NEXT:    adrp x8, .LCPI40_1
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q2, q0, [x0, #32]
+; CHECK-NEXT:    adrp x12, .LCPI40_0
+; CHECK-NEXT:    ldr q6, [x8, :lo12:.LCPI40_1]
+; CHECK-NEXT:    ldp q18, q7, [x0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z2.h, z6.h
+; CHECK-NEXT:    mov z3.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z19.h, z3.h[1]
+; CHECK-NEXT:    mov z22.h, z3.h[2]
+; CHECK-NEXT:    fmov w8, s19
+; CHECK-NEXT:    mov z23.h, z3.h[3]
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z24.h, z3.h[4]
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    fmov w11, s23
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s24
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    mov z26.h, z3.h[6]
+; CHECK-NEXT:    bfi w9, w11, #19, #1
+; CHECK-NEXT:    mov z25.h, z3.h[5]
+; CHECK-NEXT:    ldp q4, q1, [x1, #96]
+; CHECK-NEXT:    bfi w9, w8, #20, #1
+; CHECK-NEXT:    fmov w8, s26
+; CHECK-NEXT:    mov z27.h, z3.h[7]
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    ldp q16, q5, [x1, #64]
+; CHECK-NEXT:    ldp q19, q17, [x1, #32]
+; CHECK-NEXT:    ldp q21, q20, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    bfi w9, w10, #21, #1
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z0.h, z6.h
+; CHECK-NEXT:    orr w8, w9, w8, lsl #22
+; CHECK-NEXT:    fmov w9, s27
+; CHECK-NEXT:    mov z23.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z18.h, z6.h
+; CHECK-NEXT:    fmov w10, s23
+; CHECK-NEXT:    mov z24.h, z23.h[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z25.h, z23.h[2]
+; CHECK-NEXT:    mov z26.h, z23.h[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    mov z27.h, z23.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    mov z28.h, z23.h[5]
+; CHECK-NEXT:    mov z29.h, z23.h[6]
+; CHECK-NEXT:    mov z22.h, z23.h[7]
+; CHECK-NEXT:    and w9, w10, #0x1
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    mov z23.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    ldr q3, [x12, :lo12:.LCPI40_0]
+; CHECK-NEXT:    mov z24.h, z23.h[1]
+; CHECK-NEXT:    fmov w12, s28
+; CHECK-NEXT:    fmov w14, s23
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w13, s24
+; CHECK-NEXT:    mov z24.h, z23.h[2]
+; CHECK-NEXT:    mov z25.h, z23.h[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z26.h, z23.h[4]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    and w12, w14, #0x1
+; CHECK-NEXT:    fmov w14, s25
+; CHECK-NEXT:    mov z27.h, z23.h[5]
+; CHECK-NEXT:    bfi w12, w13, #1, #1
+; CHECK-NEXT:    fmov w13, s26
+; CHECK-NEXT:    bfi w12, w11, #2, #1
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    mov z28.h, z23.h[6]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    bfi w12, w14, #3, #1
+; CHECK-NEXT:    mov z30.h, z23.h[7]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    bfi w12, w13, #4, #1
+; CHECK-NEXT:    fmov w10, s28
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z7.h, z6.h
+; CHECK-NEXT:    bfi w12, w11, #5, #1
+; CHECK-NEXT:    mov z6.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s30
+; CHECK-NEXT:    fmov w13, s6
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z23.h, z6.h[1]
+; CHECK-NEXT:    mov z24.h, z6.h[2]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w10, w12, w10, lsl #6
+; CHECK-NEXT:    and w12, w13, #0x1
+; CHECK-NEXT:    fmov w13, s23
+; CHECK-NEXT:    fmov w9, s29
+; CHECK-NEXT:    orr w10, w10, w11, lsl #7
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z25.h, z6.h[3]
+; CHECK-NEXT:    orr w10, w10, w12, lsl #8
+; CHECK-NEXT:    fmov w12, s25
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z26.h, z6.h[4]
+; CHECK-NEXT:    orr w10, w10, w13, lsl #9
+; CHECK-NEXT:    mov z27.h, z6.h[5]
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #30
+; CHECK-NEXT:    orr w9, w10, w11, lsl #10
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    mov z28.h, z6.h[6]
+; CHECK-NEXT:    fmov w11, s27
+; CHECK-NEXT:    orr w9, w9, w12, lsl #11
+; CHECK-NEXT:    fmov w12, s28
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z29.h, z6.h[7]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    ldr q6, [sp]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s29
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    fmov w10, s22
+; CHECK-NEXT:    uunpklo z22.d, z21.s
+; CHECK-NEXT:    ext z21.b, z21.b, z21.b, #8
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    uunpklo z21.d, z21.s
+; CHECK-NEXT:    orr w8, w8, w10, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    movprfx z23, z22
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z3.d
+; CHECK-NEXT:    movprfx z22, z21
+; CHECK-NEXT:    lsl z22.d, p0/m, z22.d, z3.d
+; CHECK-NEXT:    add z21.d, z6.d, z23.d
+; CHECK-NEXT:    tbz w8, #0, .LBB40_2
+; CHECK-NEXT:  // %bb.1: // %cond.store
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB40_2: // %else
+; CHECK-NEXT:    uunpklo z23.d, z20.s
+; CHECK-NEXT:    add z22.d, z6.d, z22.d
+; CHECK-NEXT:    tbz w8, #1, .LBB40_4
+; CHECK-NEXT:  // %bb.3: // %cond.store1
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.h, z18.h[1]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:  .LBB40_4: // %else2
+; CHECK-NEXT:    ext z20.b, z20.b, z20.b, #8
+; CHECK-NEXT:    movprfx z21, z23
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z3.d
+; CHECK-NEXT:    tbz w8, #2, .LBB40_6
+; CHECK-NEXT:  // %bb.5: // %cond.store3
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z23.h, z18.h[2]
+; CHECK-NEXT:    str h23, [x9]
+; CHECK-NEXT:  .LBB40_6: // %else4
+; CHECK-NEXT:    uunpklo z20.d, z20.s
+; CHECK-NEXT:    add z21.d, z6.d, z21.d
+; CHECK-NEXT:    tbz w8, #3, .LBB40_8
+; CHECK-NEXT:  // %bb.7: // %cond.store5
+; CHECK-NEXT:    mov z22.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    mov z22.h, z18.h[3]
+; CHECK-NEXT:    str h22, [x9]
+; CHECK-NEXT:  .LBB40_8: // %else6
+; CHECK-NEXT:    lsl z20.d, p0/m, z20.d, z3.d
+; CHECK-NEXT:    tbz w8, #4, .LBB40_10
+; CHECK-NEXT:  // %bb.9: // %cond.store7
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z22.h, z18.h[4]
+; CHECK-NEXT:    str h22, [x9]
+; CHECK-NEXT:  .LBB40_10: // %else8
+; CHECK-NEXT:    uunpklo z22.d, z19.s
+; CHECK-NEXT:    add z20.d, z6.d, z20.d
+; CHECK-NEXT:    tbz w8, #5, .LBB40_12
+; CHECK-NEXT:  // %bb.11: // %cond.store9
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.h, z18.h[5]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:  .LBB40_12: // %else10
+; CHECK-NEXT:    ext z19.b, z19.b, z19.b, #8
+; CHECK-NEXT:    lsl z22.d, p0/m, z22.d, z3.d
+; CHECK-NEXT:    tbz w8, #6, .LBB40_14
+; CHECK-NEXT:  // %bb.13: // %cond.store11
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z21.h, z18.h[6]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:  .LBB40_14: // %else12
+; CHECK-NEXT:    uunpklo z21.d, z19.s
+; CHECK-NEXT:    add z19.d, z6.d, z22.d
+; CHECK-NEXT:    tbz w8, #7, .LBB40_16
+; CHECK-NEXT:  // %bb.15: // %cond.store13
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    mov z18.h, z18.h[7]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB40_16: // %else14
+; CHECK-NEXT:    movprfx z18, z21
+; CHECK-NEXT:    lsl z18.d, p0/m, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #8, .LBB40_18
+; CHECK-NEXT:  // %bb.17: // %cond.store15
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB40_18: // %else16
+; CHECK-NEXT:    uunpklo z20.d, z17.s
+; CHECK-NEXT:    add z18.d, z6.d, z18.d
+; CHECK-NEXT:    tbz w8, #9, .LBB40_20
+; CHECK-NEXT:  // %bb.19: // %cond.store17
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z19.h, z7.h[1]
+; CHECK-NEXT:    str h19, [x9]
+; CHECK-NEXT:  .LBB40_20: // %else18
+; CHECK-NEXT:    ext z17.b, z17.b, z17.b, #8
+; CHECK-NEXT:    movprfx z19, z20
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #10, .LBB40_22
+; CHECK-NEXT:  // %bb.21: // %cond.store19
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z20.h, z7.h[2]
+; CHECK-NEXT:    str h20, [x9]
+; CHECK-NEXT:  .LBB40_22: // %else20
+; CHECK-NEXT:    uunpklo z17.d, z17.s
+; CHECK-NEXT:    add z19.d, z6.d, z19.d
+; CHECK-NEXT:    tbz w8, #11, .LBB40_24
+; CHECK-NEXT:  // %bb.23: // %cond.store21
+; CHECK-NEXT:    mov z18.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z18.h, z7.h[3]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB40_24: // %else22
+; CHECK-NEXT:    lsl z17.d, p0/m, z17.d, z3.d
+; CHECK-NEXT:    tbz w8, #12, .LBB40_26
+; CHECK-NEXT:  // %bb.25: // %cond.store23
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z18.h, z7.h[4]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB40_26: // %else24
+; CHECK-NEXT:    uunpklo z18.d, z16.s
+; CHECK-NEXT:    add z17.d, z6.d, z17.d
+; CHECK-NEXT:    tbz w8, #13, .LBB40_28
+; CHECK-NEXT:  // %bb.27: // %cond.store25
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z19.h, z7.h[5]
+; CHECK-NEXT:    str h19, [x9]
+; CHECK-NEXT:  .LBB40_28: // %else26
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    movprfx z19, z18
+; CHECK-NEXT:    lsl z19.d, p0/m, z19.d, z3.d
+; CHECK-NEXT:    tbz w8, #14, .LBB40_30
+; CHECK-NEXT:  // %bb.29: // %cond.store27
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z18.h, z7.h[6]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB40_30: // %else28
+; CHECK-NEXT:    uunpklo z18.d, z16.s
+; CHECK-NEXT:    add z16.d, z6.d, z19.d
+; CHECK-NEXT:    tbz w8, #15, .LBB40_32
+; CHECK-NEXT:  // %bb.31: // %cond.store29
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    mov z7.h, z7.h[7]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB40_32: // %else30
+; CHECK-NEXT:    movprfx z7, z18
+; CHECK-NEXT:    lsl z7.d, p0/m, z7.d, z3.d
+; CHECK-NEXT:    tbz w8, #16, .LBB40_34
+; CHECK-NEXT:  // %bb.33: // %cond.store31
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:  .LBB40_34: // %else32
+; CHECK-NEXT:    uunpklo z17.d, z5.s
+; CHECK-NEXT:    add z7.d, z6.d, z7.d
+; CHECK-NEXT:    tbz w8, #17, .LBB40_36
+; CHECK-NEXT:  // %bb.35: // %cond.store33
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.h, z2.h[1]
+; CHECK-NEXT:    str h16, [x9]
+; CHECK-NEXT:  .LBB40_36: // %else34
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    movprfx z16, z17
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z3.d
+; CHECK-NEXT:    tbz w8, #18, .LBB40_38
+; CHECK-NEXT:  // %bb.37: // %cond.store35
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z17.h, z2.h[2]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:  .LBB40_38: // %else36
+; CHECK-NEXT:    uunpklo z5.d, z5.s
+; CHECK-NEXT:    add z16.d, z6.d, z16.d
+; CHECK-NEXT:    tbz w8, #19, .LBB40_40
+; CHECK-NEXT:  // %bb.39: // %cond.store37
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z7.h, z2.h[3]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB40_40: // %else38
+; CHECK-NEXT:    lsl z5.d, p0/m, z5.d, z3.d
+; CHECK-NEXT:    tbz w8, #20, .LBB40_42
+; CHECK-NEXT:  // %bb.41: // %cond.store39
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z7.h, z2.h[4]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB40_42: // %else40
+; CHECK-NEXT:    uunpklo z7.d, z4.s
+; CHECK-NEXT:    add z5.d, z6.d, z5.d
+; CHECK-NEXT:    tbz w8, #21, .LBB40_44
+; CHECK-NEXT:  // %bb.43: // %cond.store41
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.h, z2.h[5]
+; CHECK-NEXT:    str h16, [x9]
+; CHECK-NEXT:  .LBB40_44: // %else42
+; CHECK-NEXT:    ext z4.b, z4.b, z4.b, #8
+; CHECK-NEXT:    movprfx z16, z7
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z3.d
+; CHECK-NEXT:    tbz w8, #22, .LBB40_46
+; CHECK-NEXT:  // %bb.45: // %cond.store43
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z7.h, z2.h[6]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB40_46: // %else44
+; CHECK-NEXT:    uunpklo z7.d, z4.s
+; CHECK-NEXT:    add z4.d, z6.d, z16.d
+; CHECK-NEXT:    tbz w8, #23, .LBB40_48
+; CHECK-NEXT:  // %bb.47: // %cond.store45
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z2.h, z2.h[7]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:  .LBB40_48: // %else46
+; CHECK-NEXT:    movprfx z2, z7
+; CHECK-NEXT:    lsl z2.d, p0/m, z2.d, z3.d
+; CHECK-NEXT:    tbz w8, #24, .LBB40_50
+; CHECK-NEXT:  // %bb.49: // %cond.store47
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:  .LBB40_50: // %else48
+; CHECK-NEXT:    uunpklo z5.d, z1.s
+; CHECK-NEXT:    add z2.d, z6.d, z2.d
+; CHECK-NEXT:    tbz w8, #25, .LBB40_52
+; CHECK-NEXT:  // %bb.51: // %cond.store49
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z4.h, z0.h[1]
+; CHECK-NEXT:    str h4, [x9]
+; CHECK-NEXT:  .LBB40_52: // %else50
+; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    lsl z5.d, p0/m, z5.d, z3.d
+; CHECK-NEXT:    tbz w8, #26, .LBB40_54
+; CHECK-NEXT:  // %bb.53: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z4.h, z0.h[2]
+; CHECK-NEXT:    str h4, [x9]
+; CHECK-NEXT:  .LBB40_54: // %else52
+; CHECK-NEXT:    uunpklo z4.d, z1.s
+; CHECK-NEXT:    add z1.d, z6.d, z5.d
+; CHECK-NEXT:    tbz w8, #27, .LBB40_56
+; CHECK-NEXT:  // %bb.55: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[3]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:  .LBB40_56: // %else54
+; CHECK-NEXT:    movprfx z2, z4
+; CHECK-NEXT:    lsl z2.d, p0/m, z2.d, z3.d
+; CHECK-NEXT:    tbnz w8, #28, .LBB40_61
+; CHECK-NEXT:  // %bb.57: // %else56
+; CHECK-NEXT:    add z2.d, z6.d, z2.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB40_62
+; CHECK-NEXT:  .LBB40_58: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB40_63
+; CHECK-NEXT:  .LBB40_59: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB40_64
+; CHECK-NEXT:  .LBB40_60: // %else62
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB40_61: // %cond.store55
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z3.h, z0.h[4]
+; CHECK-NEXT:    str h3, [x9]
+; CHECK-NEXT:    add z2.d, z6.d, z2.d
+; CHECK-NEXT:    tbz w8, #29, .LBB40_58
+; CHECK-NEXT:  .LBB40_62: // %cond.store57
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z1.h, z0.h[5]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB40_59
+; CHECK-NEXT:  .LBB40_63: // %cond.store59
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z1.h, z0.h[6]
+; CHECK-NEXT:    str h1, [x9]
+; CHECK-NEXT:    tbz w8, #31, .LBB40_60
+; CHECK-NEXT:  .LBB40_64: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z2.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #32
 ; CHECK-NEXT:    ret
   %vals = load <32 x half>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = zext <32 x i32> %idxs to <32 x i64>
   %ptrs = getelementptr half, ptr %base, <32 x i64> %ext
   %mask = fcmp oeq <32 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_32b_unscaled_sext(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_scatter_32b_unscaled_sext:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.s, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1h { z0.s }, p0, [x2, z1.s, sxtw]
+; CHECK-NEXT:    adrp x8, .LCPI41_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q2, q0, [x0, #32]
+; CHECK-NEXT:    ldr q4, [x8, :lo12:.LCPI41_0]
+; CHECK-NEXT:    ldp q17, q6, [x0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z2.h, z4.h
+; CHECK-NEXT:    mov z5.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z0.h, z4.h
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    mov z7.h, z5.h[1]
+; CHECK-NEXT:    fmov w8, s7
+; CHECK-NEXT:    mov z16.h, z5.h[2]
+; CHECK-NEXT:    mov z18.h, z5.h[3]
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z19.h, z5.h[4]
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    fmov w11, s19
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s18
+; CHECK-NEXT:    mov z21.h, z5.h[6]
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    mov z20.h, z5.h[5]
+; CHECK-NEXT:    mov z22.h, z5.h[7]
+; CHECK-NEXT:    bfi w9, w8, #19, #1
+; CHECK-NEXT:    fmov w10, s21
+; CHECK-NEXT:    fmov w8, s20
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    mov z19.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z20.h, z19.h[1]
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    mov z23.h, z19.h[2]
+; CHECK-NEXT:    and w8, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #22
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    mov z24.h, z19.h[3]
+; CHECK-NEXT:    orr w8, w9, w8, lsl #23
+; CHECK-NEXT:    fmov w9, s19
+; CHECK-NEXT:    ldp q3, q1, [x1, #96]
+; CHECK-NEXT:    mov z25.h, z19.h[4]
+; CHECK-NEXT:    mov z26.h, z19.h[5]
+; CHECK-NEXT:    mov z22.h, z19.h[6]
+; CHECK-NEXT:    mov z21.h, z19.h[7]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    ldp q7, q5, [x1, #64]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldp q18, q16, [x1, #32]
+; CHECK-NEXT:    ldp q20, q19, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z17.h, z4.h
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    mov z23.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z24.h, z23.h[1]
+; CHECK-NEXT:    fmov w12, s23
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z24.h, z23.h[2]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z25.h, z23.h[3]
+; CHECK-NEXT:    fmov w13, s24
+; CHECK-NEXT:    mov z26.h, z23.h[4]
+; CHECK-NEXT:    mov z28.h, z23.h[6]
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    fmov w14, s25
+; CHECK-NEXT:    mov z27.h, z23.h[5]
+; CHECK-NEXT:    bfi w12, w11, #1, #1
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    fmov w9, s28
+; CHECK-NEXT:    mov z29.h, z23.h[7]
+; CHECK-NEXT:    bfi w12, w13, #2, #1
+; CHECK-NEXT:    fmov w13, s27
+; CHECK-NEXT:    bfi w12, w14, #3, #1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z6.h, z4.h
+; CHECK-NEXT:    fmov w10, s29
+; CHECK-NEXT:    bfi w12, w11, #4, #1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w12, w13, #5, #1
+; CHECK-NEXT:    mov z23.h, z4.h[1]
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w12, w9, lsl #6
+; CHECK-NEXT:    fmov w12, s23
+; CHECK-NEXT:    mov z24.h, z4.h[2]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    mov z25.h, z4.h[3]
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z26.h, z4.h[4]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    orr w9, w9, w12, lsl #9
+; CHECK-NEXT:    fmov w12, s26
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z27.h, z4.h[5]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z28.h, z4.h[6]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    orr w9, w9, w11, lsl #12
+; CHECK-NEXT:    fmov w11, s28
+; CHECK-NEXT:    fmov w12, s22
+; CHECK-NEXT:    mov z29.h, z4.h[7]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldr q4, [sp]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    sunpklo z22.d, z20.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s29
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s21
+; CHECK-NEXT:    ext z20.b, z20.b, z20.b, #8
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    sunpklo z21.d, z20.s
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    add z20.d, z4.d, z22.d
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbnz w8, #0, .LBB41_40
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    add z21.d, z4.d, z21.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB41_41
+; CHECK-NEXT:  .LBB41_2: // %else2
+; CHECK-NEXT:    sunpklo z20.d, z19.s
+; CHECK-NEXT:    tbz w8, #2, .LBB41_4
+; CHECK-NEXT:  .LBB41_3: // %cond.store3
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z22.h, z17.h[2]
+; CHECK-NEXT:    str h22, [x9]
+; CHECK-NEXT:  .LBB41_4: // %else4
+; CHECK-NEXT:    ext z19.b, z19.b, z19.b, #8
+; CHECK-NEXT:    add z20.d, z4.d, z20.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB41_42
+; CHECK-NEXT:  // %bb.5: // %else6
+; CHECK-NEXT:    sunpklo z19.d, z19.s
+; CHECK-NEXT:    tbnz w8, #4, .LBB41_43
+; CHECK-NEXT:  .LBB41_6: // %else8
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB41_44
+; CHECK-NEXT:  .LBB41_7: // %else10
+; CHECK-NEXT:    sunpklo z20.d, z18.s
+; CHECK-NEXT:    tbz w8, #6, .LBB41_9
+; CHECK-NEXT:  .LBB41_8: // %cond.store11
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z21.h, z17.h[6]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:  .LBB41_9: // %else12
+; CHECK-NEXT:    ext z18.b, z18.b, z18.b, #8
+; CHECK-NEXT:    add z20.d, z4.d, z20.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB41_45
+; CHECK-NEXT:  // %bb.10: // %else14
+; CHECK-NEXT:    sunpklo z17.d, z18.s
+; CHECK-NEXT:    tbnz w8, #8, .LBB41_46
+; CHECK-NEXT:  .LBB41_11: // %else16
+; CHECK-NEXT:    add z17.d, z4.d, z17.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB41_47
+; CHECK-NEXT:  .LBB41_12: // %else18
+; CHECK-NEXT:    sunpklo z18.d, z16.s
+; CHECK-NEXT:    tbz w8, #10, .LBB41_14
+; CHECK-NEXT:  .LBB41_13: // %cond.store19
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z19.h, z6.h[2]
+; CHECK-NEXT:    str h19, [x9]
+; CHECK-NEXT:  .LBB41_14: // %else20
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB41_48
+; CHECK-NEXT:  // %bb.15: // %else22
+; CHECK-NEXT:    sunpklo z16.d, z16.s
+; CHECK-NEXT:    tbnz w8, #12, .LBB41_49
+; CHECK-NEXT:  .LBB41_16: // %else24
+; CHECK-NEXT:    add z16.d, z4.d, z16.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB41_50
+; CHECK-NEXT:  .LBB41_17: // %else26
+; CHECK-NEXT:    sunpklo z17.d, z7.s
+; CHECK-NEXT:    tbz w8, #14, .LBB41_19
+; CHECK-NEXT:  .LBB41_18: // %cond.store27
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z18.h, z6.h[6]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB41_19: // %else28
+; CHECK-NEXT:    ext z7.b, z7.b, z7.b, #8
+; CHECK-NEXT:    add z17.d, z4.d, z17.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB41_51
+; CHECK-NEXT:  // %bb.20: // %else30
+; CHECK-NEXT:    sunpklo z6.d, z7.s
+; CHECK-NEXT:    tbnz w8, #16, .LBB41_52
+; CHECK-NEXT:  .LBB41_21: // %else32
+; CHECK-NEXT:    add z6.d, z4.d, z6.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB41_53
+; CHECK-NEXT:  .LBB41_22: // %else34
+; CHECK-NEXT:    sunpklo z7.d, z5.s
+; CHECK-NEXT:    tbz w8, #18, .LBB41_24
+; CHECK-NEXT:  .LBB41_23: // %cond.store35
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z16.h, z2.h[2]
+; CHECK-NEXT:    str h16, [x9]
+; CHECK-NEXT:  .LBB41_24: // %else36
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    add z7.d, z4.d, z7.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB41_54
+; CHECK-NEXT:  // %bb.25: // %else38
+; CHECK-NEXT:    sunpklo z5.d, z5.s
+; CHECK-NEXT:    tbnz w8, #20, .LBB41_55
+; CHECK-NEXT:  .LBB41_26: // %else40
+; CHECK-NEXT:    add z5.d, z4.d, z5.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB41_56
+; CHECK-NEXT:  .LBB41_27: // %else42
+; CHECK-NEXT:    sunpklo z6.d, z3.s
+; CHECK-NEXT:    tbz w8, #22, .LBB41_29
+; CHECK-NEXT:  .LBB41_28: // %cond.store43
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z7.h, z2.h[6]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB41_29: // %else44
+; CHECK-NEXT:    ext z3.b, z3.b, z3.b, #8
+; CHECK-NEXT:    add z6.d, z4.d, z6.d
+; CHECK-NEXT:    tbnz w8, #23, .LBB41_57
+; CHECK-NEXT:  // %bb.30: // %else46
+; CHECK-NEXT:    sunpklo z2.d, z3.s
+; CHECK-NEXT:    tbnz w8, #24, .LBB41_58
+; CHECK-NEXT:  .LBB41_31: // %else48
+; CHECK-NEXT:    add z2.d, z4.d, z2.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB41_59
+; CHECK-NEXT:  .LBB41_32: // %else50
+; CHECK-NEXT:    sunpklo z3.d, z1.s
+; CHECK-NEXT:    tbz w8, #26, .LBB41_34
+; CHECK-NEXT:  .LBB41_33: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z5.h, z0.h[2]
+; CHECK-NEXT:    str h5, [x9]
+; CHECK-NEXT:  .LBB41_34: // %else52
+; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    add z3.d, z4.d, z3.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB41_60
+; CHECK-NEXT:  // %bb.35: // %else54
+; CHECK-NEXT:    sunpklo z1.d, z1.s
+; CHECK-NEXT:    tbnz w8, #28, .LBB41_61
+; CHECK-NEXT:  .LBB41_36: // %else56
+; CHECK-NEXT:    add z1.d, z4.d, z1.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB41_62
+; CHECK-NEXT:  .LBB41_37: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB41_63
+; CHECK-NEXT:  .LBB41_38: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB41_64
+; CHECK-NEXT:  .LBB41_39: // %else62
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB41_40: // %cond.store
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    add z21.d, z4.d, z21.d
+; CHECK-NEXT:    tbz w8, #1, .LBB41_2
+; CHECK-NEXT:  .LBB41_41: // %cond.store1
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z20.h, z17.h[1]
+; CHECK-NEXT:    str h20, [x9]
+; CHECK-NEXT:    sunpklo z20.d, z19.s
+; CHECK-NEXT:    tbnz w8, #2, .LBB41_3
+; CHECK-NEXT:    b .LBB41_4
+; CHECK-NEXT:  .LBB41_42: // %cond.store5
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.h, z17.h[3]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:    sunpklo z19.d, z19.s
+; CHECK-NEXT:    tbz w8, #4, .LBB41_6
+; CHECK-NEXT:  .LBB41_43: // %cond.store7
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z21.h, z17.h[4]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #5, .LBB41_7
+; CHECK-NEXT:  .LBB41_44: // %cond.store9
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z20.h, z17.h[5]
+; CHECK-NEXT:    str h20, [x9]
+; CHECK-NEXT:    sunpklo z20.d, z18.s
+; CHECK-NEXT:    tbnz w8, #6, .LBB41_8
+; CHECK-NEXT:    b .LBB41_9
+; CHECK-NEXT:  .LBB41_45: // %cond.store13
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    mov z17.h, z17.h[7]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    sunpklo z17.d, z18.s
+; CHECK-NEXT:    tbz w8, #8, .LBB41_11
+; CHECK-NEXT:  .LBB41_46: // %cond.store15
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    add z17.d, z4.d, z17.d
+; CHECK-NEXT:    tbz w8, #9, .LBB41_12
+; CHECK-NEXT:  .LBB41_47: // %cond.store17
+; CHECK-NEXT:    mov z18.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z18.h, z6.h[1]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:    sunpklo z18.d, z16.s
+; CHECK-NEXT:    tbnz w8, #10, .LBB41_13
+; CHECK-NEXT:    b .LBB41_14
+; CHECK-NEXT:  .LBB41_48: // %cond.store21
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.h, z6.h[3]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    sunpklo z16.d, z16.s
+; CHECK-NEXT:    tbz w8, #12, .LBB41_16
+; CHECK-NEXT:  .LBB41_49: // %cond.store23
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z17.h, z6.h[4]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    add z16.d, z4.d, z16.d
+; CHECK-NEXT:    tbz w8, #13, .LBB41_17
+; CHECK-NEXT:  .LBB41_50: // %cond.store25
+; CHECK-NEXT:    mov z17.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.h, z6.h[5]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    sunpklo z17.d, z7.s
+; CHECK-NEXT:    tbnz w8, #14, .LBB41_18
+; CHECK-NEXT:    b .LBB41_19
+; CHECK-NEXT:  .LBB41_51: // %cond.store29
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    mov z6.h, z6.h[7]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    sunpklo z6.d, z7.s
+; CHECK-NEXT:    tbz w8, #16, .LBB41_21
+; CHECK-NEXT:  .LBB41_52: // %cond.store31
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    add z6.d, z4.d, z6.d
+; CHECK-NEXT:    tbz w8, #17, .LBB41_22
+; CHECK-NEXT:  .LBB41_53: // %cond.store33
+; CHECK-NEXT:    mov z7.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z7.h, z2.h[1]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:    sunpklo z7.d, z5.s
+; CHECK-NEXT:    tbnz w8, #18, .LBB41_23
+; CHECK-NEXT:    b .LBB41_24
+; CHECK-NEXT:  .LBB41_54: // %cond.store37
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.h, z2.h[3]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    sunpklo z5.d, z5.s
+; CHECK-NEXT:    tbz w8, #20, .LBB41_26
+; CHECK-NEXT:  .LBB41_55: // %cond.store39
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z6.h, z2.h[4]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    add z5.d, z4.d, z5.d
+; CHECK-NEXT:    tbz w8, #21, .LBB41_27
+; CHECK-NEXT:  .LBB41_56: // %cond.store41
+; CHECK-NEXT:    mov z6.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.h, z2.h[5]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    sunpklo z6.d, z3.s
+; CHECK-NEXT:    tbnz w8, #22, .LBB41_28
+; CHECK-NEXT:    b .LBB41_29
+; CHECK-NEXT:  .LBB41_57: // %cond.store45
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z2.h, z2.h[7]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    sunpklo z2.d, z3.s
+; CHECK-NEXT:    tbz w8, #24, .LBB41_31
+; CHECK-NEXT:  .LBB41_58: // %cond.store47
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:    add z2.d, z4.d, z2.d
+; CHECK-NEXT:    tbz w8, #25, .LBB41_32
+; CHECK-NEXT:  .LBB41_59: // %cond.store49
+; CHECK-NEXT:    mov z3.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z3.h, z0.h[1]
+; CHECK-NEXT:    str h3, [x9]
+; CHECK-NEXT:    sunpklo z3.d, z1.s
+; CHECK-NEXT:    tbnz w8, #26, .LBB41_33
+; CHECK-NEXT:    b .LBB41_34
+; CHECK-NEXT:  .LBB41_60: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[3]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    sunpklo z1.d, z1.s
+; CHECK-NEXT:    tbz w8, #28, .LBB41_36
+; CHECK-NEXT:  .LBB41_61: // %cond.store55
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z2.h, z0.h[4]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    add z1.d, z4.d, z1.d
+; CHECK-NEXT:    tbz w8, #29, .LBB41_37
+; CHECK-NEXT:  .LBB41_62: // %cond.store57
+; CHECK-NEXT:    mov z2.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[5]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB41_38
+; CHECK-NEXT:  .LBB41_63: // %cond.store59
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.h, z0.h[6]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #31, .LBB41_39
+; CHECK-NEXT:  .LBB41_64: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #32
 ; CHECK-NEXT:    ret
   %vals = load <32 x half>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = sext <32 x i32> %idxs to <32 x i64>
   %byte_ptrs = getelementptr i8, ptr %base, <32 x i64> %ext
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_32b_unscaled_zext(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_scatter_32b_unscaled_zext:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.h, vl32
-; CHECK-NEXT:    ptrue p1.s, vl32
-; CHECK-NEXT:    ld1h { z0.h }, p0/z, [x0]
-; CHECK-NEXT:    ld1w { z1.s }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.h, p0/z, z0.h, #0.0
-; CHECK-NEXT:    uunpklo z0.s, z0.h
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1h { z0.s }, p0, [x2, z1.s, uxtw]
+; CHECK-NEXT:    adrp x8, .LCPI42_0
+; CHECK-NEXT:    ptrue p0.h, vl8
+; CHECK-NEXT:    ldp q2, q0, [x0, #32]
+; CHECK-NEXT:    ldr q4, [x8, :lo12:.LCPI42_0]
+; CHECK-NEXT:    ldp q17, q6, [x0]
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z2.h, z4.h
+; CHECK-NEXT:    mov z5.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z0.h, z4.h
+; CHECK-NEXT:    fmov w9, s5
+; CHECK-NEXT:    mov z7.h, z5.h[1]
+; CHECK-NEXT:    fmov w8, s7
+; CHECK-NEXT:    mov z16.h, z5.h[2]
+; CHECK-NEXT:    mov z18.h, z5.h[3]
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z19.h, z5.h[4]
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    fmov w11, s19
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s18
+; CHECK-NEXT:    mov z21.h, z5.h[6]
+; CHECK-NEXT:    bfi w9, w10, #18, #1
+; CHECK-NEXT:    mov z20.h, z5.h[5]
+; CHECK-NEXT:    mov z22.h, z5.h[7]
+; CHECK-NEXT:    bfi w9, w8, #19, #1
+; CHECK-NEXT:    fmov w10, s21
+; CHECK-NEXT:    fmov w8, s20
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    fmov w11, s22
+; CHECK-NEXT:    mov z19.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z20.h, z19.h[1]
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    mov z23.h, z19.h[2]
+; CHECK-NEXT:    and w8, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #22
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    mov z24.h, z19.h[3]
+; CHECK-NEXT:    orr w8, w9, w8, lsl #23
+; CHECK-NEXT:    fmov w9, s19
+; CHECK-NEXT:    ldp q3, q1, [x1, #96]
+; CHECK-NEXT:    mov z25.h, z19.h[4]
+; CHECK-NEXT:    mov z26.h, z19.h[5]
+; CHECK-NEXT:    mov z22.h, z19.h[6]
+; CHECK-NEXT:    mov z21.h, z19.h[7]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #24
+; CHECK-NEXT:    fmov w9, s23
+; CHECK-NEXT:    ldp q7, q5, [x1, #64]
+; CHECK-NEXT:    orr w8, w8, w10, lsl #25
+; CHECK-NEXT:    fmov w10, s24
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldp q18, q16, [x1, #32]
+; CHECK-NEXT:    ldp q20, q19, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp, #-32]!
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    fcmeq p1.h, p0/z, z17.h, z4.h
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov w9, s25
+; CHECK-NEXT:    mov z23.h, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z24.h, z23.h[1]
+; CHECK-NEXT:    fmov w12, s23
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    fmov w10, s26
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    mov z24.h, z23.h[2]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z25.h, z23.h[3]
+; CHECK-NEXT:    fmov w13, s24
+; CHECK-NEXT:    mov z26.h, z23.h[4]
+; CHECK-NEXT:    mov z28.h, z23.h[6]
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    fmov w14, s25
+; CHECK-NEXT:    mov z27.h, z23.h[5]
+; CHECK-NEXT:    bfi w12, w11, #1, #1
+; CHECK-NEXT:    fmov w11, s26
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #28
+; CHECK-NEXT:    fmov w9, s28
+; CHECK-NEXT:    mov z29.h, z23.h[7]
+; CHECK-NEXT:    bfi w12, w13, #2, #1
+; CHECK-NEXT:    fmov w13, s27
+; CHECK-NEXT:    bfi w12, w14, #3, #1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    fcmeq p0.h, p0/z, z6.h, z4.h
+; CHECK-NEXT:    fmov w10, s29
+; CHECK-NEXT:    bfi w12, w11, #4, #1
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z4.h, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w12, w13, #5, #1
+; CHECK-NEXT:    mov z23.h, z4.h[1]
+; CHECK-NEXT:    fmov w11, s4
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w12, w9, lsl #6
+; CHECK-NEXT:    fmov w12, s23
+; CHECK-NEXT:    mov z24.h, z4.h[2]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    mov z25.h, z4.h[3]
+; CHECK-NEXT:    fmov w11, s24
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z26.h, z4.h[4]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #8
+; CHECK-NEXT:    fmov w10, s25
+; CHECK-NEXT:    orr w9, w9, w12, lsl #9
+; CHECK-NEXT:    fmov w12, s26
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z27.h, z4.h[5]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z28.h, z4.h[6]
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #11
+; CHECK-NEXT:    fmov w10, s27
+; CHECK-NEXT:    orr w9, w9, w11, lsl #12
+; CHECK-NEXT:    fmov w11, s28
+; CHECK-NEXT:    fmov w12, s22
+; CHECK-NEXT:    mov z29.h, z4.h[7]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    ldr q4, [sp]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    uunpklo z22.d, z20.s
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s29
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s21
+; CHECK-NEXT:    ext z20.b, z20.b, z20.b, #8
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    uunpklo z21.d, z20.s
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    add z20.d, z4.d, z22.d
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    tbnz w8, #0, .LBB42_40
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    add z21.d, z4.d, z21.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB42_41
+; CHECK-NEXT:  .LBB42_2: // %else2
+; CHECK-NEXT:    uunpklo z20.d, z19.s
+; CHECK-NEXT:    tbz w8, #2, .LBB42_4
+; CHECK-NEXT:  .LBB42_3: // %cond.store3
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z22.h, z17.h[2]
+; CHECK-NEXT:    str h22, [x9]
+; CHECK-NEXT:  .LBB42_4: // %else4
+; CHECK-NEXT:    ext z19.b, z19.b, z19.b, #8
+; CHECK-NEXT:    add z20.d, z4.d, z20.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB42_42
+; CHECK-NEXT:  // %bb.5: // %else6
+; CHECK-NEXT:    uunpklo z19.d, z19.s
+; CHECK-NEXT:    tbnz w8, #4, .LBB42_43
+; CHECK-NEXT:  .LBB42_6: // %else8
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB42_44
+; CHECK-NEXT:  .LBB42_7: // %else10
+; CHECK-NEXT:    uunpklo z20.d, z18.s
+; CHECK-NEXT:    tbz w8, #6, .LBB42_9
+; CHECK-NEXT:  .LBB42_8: // %cond.store11
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z21.h, z17.h[6]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:  .LBB42_9: // %else12
+; CHECK-NEXT:    ext z18.b, z18.b, z18.b, #8
+; CHECK-NEXT:    add z20.d, z4.d, z20.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB42_45
+; CHECK-NEXT:  // %bb.10: // %else14
+; CHECK-NEXT:    uunpklo z17.d, z18.s
+; CHECK-NEXT:    tbnz w8, #8, .LBB42_46
+; CHECK-NEXT:  .LBB42_11: // %else16
+; CHECK-NEXT:    add z17.d, z4.d, z17.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB42_47
+; CHECK-NEXT:  .LBB42_12: // %else18
+; CHECK-NEXT:    uunpklo z18.d, z16.s
+; CHECK-NEXT:    tbz w8, #10, .LBB42_14
+; CHECK-NEXT:  .LBB42_13: // %cond.store19
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z19.h, z6.h[2]
+; CHECK-NEXT:    str h19, [x9]
+; CHECK-NEXT:  .LBB42_14: // %else20
+; CHECK-NEXT:    ext z16.b, z16.b, z16.b, #8
+; CHECK-NEXT:    add z18.d, z4.d, z18.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB42_48
+; CHECK-NEXT:  // %bb.15: // %else22
+; CHECK-NEXT:    uunpklo z16.d, z16.s
+; CHECK-NEXT:    tbnz w8, #12, .LBB42_49
+; CHECK-NEXT:  .LBB42_16: // %else24
+; CHECK-NEXT:    add z16.d, z4.d, z16.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB42_50
+; CHECK-NEXT:  .LBB42_17: // %else26
+; CHECK-NEXT:    uunpklo z17.d, z7.s
+; CHECK-NEXT:    tbz w8, #14, .LBB42_19
+; CHECK-NEXT:  .LBB42_18: // %cond.store27
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z18.h, z6.h[6]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:  .LBB42_19: // %else28
+; CHECK-NEXT:    ext z7.b, z7.b, z7.b, #8
+; CHECK-NEXT:    add z17.d, z4.d, z17.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB42_51
+; CHECK-NEXT:  // %bb.20: // %else30
+; CHECK-NEXT:    uunpklo z6.d, z7.s
+; CHECK-NEXT:    tbnz w8, #16, .LBB42_52
+; CHECK-NEXT:  .LBB42_21: // %else32
+; CHECK-NEXT:    add z6.d, z4.d, z6.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB42_53
+; CHECK-NEXT:  .LBB42_22: // %else34
+; CHECK-NEXT:    uunpklo z7.d, z5.s
+; CHECK-NEXT:    tbz w8, #18, .LBB42_24
+; CHECK-NEXT:  .LBB42_23: // %cond.store35
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z16.h, z2.h[2]
+; CHECK-NEXT:    str h16, [x9]
+; CHECK-NEXT:  .LBB42_24: // %else36
+; CHECK-NEXT:    ext z5.b, z5.b, z5.b, #8
+; CHECK-NEXT:    add z7.d, z4.d, z7.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB42_54
+; CHECK-NEXT:  // %bb.25: // %else38
+; CHECK-NEXT:    uunpklo z5.d, z5.s
+; CHECK-NEXT:    tbnz w8, #20, .LBB42_55
+; CHECK-NEXT:  .LBB42_26: // %else40
+; CHECK-NEXT:    add z5.d, z4.d, z5.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB42_56
+; CHECK-NEXT:  .LBB42_27: // %else42
+; CHECK-NEXT:    uunpklo z6.d, z3.s
+; CHECK-NEXT:    tbz w8, #22, .LBB42_29
+; CHECK-NEXT:  .LBB42_28: // %cond.store43
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z7.h, z2.h[6]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:  .LBB42_29: // %else44
+; CHECK-NEXT:    ext z3.b, z3.b, z3.b, #8
+; CHECK-NEXT:    add z6.d, z4.d, z6.d
+; CHECK-NEXT:    tbnz w8, #23, .LBB42_57
+; CHECK-NEXT:  // %bb.30: // %else46
+; CHECK-NEXT:    uunpklo z2.d, z3.s
+; CHECK-NEXT:    tbnz w8, #24, .LBB42_58
+; CHECK-NEXT:  .LBB42_31: // %else48
+; CHECK-NEXT:    add z2.d, z4.d, z2.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB42_59
+; CHECK-NEXT:  .LBB42_32: // %else50
+; CHECK-NEXT:    uunpklo z3.d, z1.s
+; CHECK-NEXT:    tbz w8, #26, .LBB42_34
+; CHECK-NEXT:  .LBB42_33: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z5.h, z0.h[2]
+; CHECK-NEXT:    str h5, [x9]
+; CHECK-NEXT:  .LBB42_34: // %else52
+; CHECK-NEXT:    ext z1.b, z1.b, z1.b, #8
+; CHECK-NEXT:    add z3.d, z4.d, z3.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB42_60
+; CHECK-NEXT:  // %bb.35: // %else54
+; CHECK-NEXT:    uunpklo z1.d, z1.s
+; CHECK-NEXT:    tbnz w8, #28, .LBB42_61
+; CHECK-NEXT:  .LBB42_36: // %else56
+; CHECK-NEXT:    add z1.d, z4.d, z1.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB42_62
+; CHECK-NEXT:  .LBB42_37: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB42_63
+; CHECK-NEXT:  .LBB42_38: // %else60
+; CHECK-NEXT:    tbnz w8, #31, .LBB42_64
+; CHECK-NEXT:  .LBB42_39: // %else62
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB42_40: // %cond.store
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    add z21.d, z4.d, z21.d
+; CHECK-NEXT:    tbz w8, #1, .LBB42_2
+; CHECK-NEXT:  .LBB42_41: // %cond.store1
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z20.h, z17.h[1]
+; CHECK-NEXT:    str h20, [x9]
+; CHECK-NEXT:    uunpklo z20.d, z19.s
+; CHECK-NEXT:    tbnz w8, #2, .LBB42_3
+; CHECK-NEXT:    b .LBB42_4
+; CHECK-NEXT:  .LBB42_42: // %cond.store5
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.h, z17.h[3]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:    uunpklo z19.d, z19.s
+; CHECK-NEXT:    tbz w8, #4, .LBB42_6
+; CHECK-NEXT:  .LBB42_43: // %cond.store7
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z21.h, z17.h[4]
+; CHECK-NEXT:    str h21, [x9]
+; CHECK-NEXT:    add z19.d, z4.d, z19.d
+; CHECK-NEXT:    tbz w8, #5, .LBB42_7
+; CHECK-NEXT:  .LBB42_44: // %cond.store9
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z20.h, z17.h[5]
+; CHECK-NEXT:    str h20, [x9]
+; CHECK-NEXT:    uunpklo z20.d, z18.s
+; CHECK-NEXT:    tbnz w8, #6, .LBB42_8
+; CHECK-NEXT:    b .LBB42_9
+; CHECK-NEXT:  .LBB42_45: // %cond.store13
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    mov z17.h, z17.h[7]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    uunpklo z17.d, z18.s
+; CHECK-NEXT:    tbz w8, #8, .LBB42_11
+; CHECK-NEXT:  .LBB42_46: // %cond.store15
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    add z17.d, z4.d, z17.d
+; CHECK-NEXT:    tbz w8, #9, .LBB42_12
+; CHECK-NEXT:  .LBB42_47: // %cond.store17
+; CHECK-NEXT:    mov z18.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z18.h, z6.h[1]
+; CHECK-NEXT:    str h18, [x9]
+; CHECK-NEXT:    uunpklo z18.d, z16.s
+; CHECK-NEXT:    tbnz w8, #10, .LBB42_13
+; CHECK-NEXT:    b .LBB42_14
+; CHECK-NEXT:  .LBB42_48: // %cond.store21
+; CHECK-NEXT:    mov z17.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.h, z6.h[3]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    uunpklo z16.d, z16.s
+; CHECK-NEXT:    tbz w8, #12, .LBB42_16
+; CHECK-NEXT:  .LBB42_49: // %cond.store23
+; CHECK-NEXT:    fmov x9, d18
+; CHECK-NEXT:    mov z17.h, z6.h[4]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    add z16.d, z4.d, z16.d
+; CHECK-NEXT:    tbz w8, #13, .LBB42_17
+; CHECK-NEXT:  .LBB42_50: // %cond.store25
+; CHECK-NEXT:    mov z17.d, z18.d[1]
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    mov z17.h, z6.h[5]
+; CHECK-NEXT:    str h17, [x9]
+; CHECK-NEXT:    uunpklo z17.d, z7.s
+; CHECK-NEXT:    tbnz w8, #14, .LBB42_18
+; CHECK-NEXT:    b .LBB42_19
+; CHECK-NEXT:  .LBB42_51: // %cond.store29
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    mov z6.h, z6.h[7]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    uunpklo z6.d, z7.s
+; CHECK-NEXT:    tbz w8, #16, .LBB42_21
+; CHECK-NEXT:  .LBB42_52: // %cond.store31
+; CHECK-NEXT:    fmov x9, d17
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    add z6.d, z4.d, z6.d
+; CHECK-NEXT:    tbz w8, #17, .LBB42_22
+; CHECK-NEXT:  .LBB42_53: // %cond.store33
+; CHECK-NEXT:    mov z7.d, z17.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z7.h, z2.h[1]
+; CHECK-NEXT:    str h7, [x9]
+; CHECK-NEXT:    uunpklo z7.d, z5.s
+; CHECK-NEXT:    tbnz w8, #18, .LBB42_23
+; CHECK-NEXT:    b .LBB42_24
+; CHECK-NEXT:  .LBB42_54: // %cond.store37
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.h, z2.h[3]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    uunpklo z5.d, z5.s
+; CHECK-NEXT:    tbz w8, #20, .LBB42_26
+; CHECK-NEXT:  .LBB42_55: // %cond.store39
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z6.h, z2.h[4]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    add z5.d, z4.d, z5.d
+; CHECK-NEXT:    tbz w8, #21, .LBB42_27
+; CHECK-NEXT:  .LBB42_56: // %cond.store41
+; CHECK-NEXT:    mov z6.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.h, z2.h[5]
+; CHECK-NEXT:    str h6, [x9]
+; CHECK-NEXT:    uunpklo z6.d, z3.s
+; CHECK-NEXT:    tbnz w8, #22, .LBB42_28
+; CHECK-NEXT:    b .LBB42_29
+; CHECK-NEXT:  .LBB42_57: // %cond.store45
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z2.h, z2.h[7]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    uunpklo z2.d, z3.s
+; CHECK-NEXT:    tbz w8, #24, .LBB42_31
+; CHECK-NEXT:  .LBB42_58: // %cond.store47
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str h0, [x9]
+; CHECK-NEXT:    add z2.d, z4.d, z2.d
+; CHECK-NEXT:    tbz w8, #25, .LBB42_32
+; CHECK-NEXT:  .LBB42_59: // %cond.store49
+; CHECK-NEXT:    mov z3.d, z6.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z3.h, z0.h[1]
+; CHECK-NEXT:    str h3, [x9]
+; CHECK-NEXT:    uunpklo z3.d, z1.s
+; CHECK-NEXT:    tbnz w8, #26, .LBB42_33
+; CHECK-NEXT:    b .LBB42_34
+; CHECK-NEXT:  .LBB42_60: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[3]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    uunpklo z1.d, z1.s
+; CHECK-NEXT:    tbz w8, #28, .LBB42_36
+; CHECK-NEXT:  .LBB42_61: // %cond.store55
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    mov z2.h, z0.h[4]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    add z1.d, z4.d, z1.d
+; CHECK-NEXT:    tbz w8, #29, .LBB42_37
+; CHECK-NEXT:  .LBB42_62: // %cond.store57
+; CHECK-NEXT:    mov z2.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.h, z0.h[5]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB42_38
+; CHECK-NEXT:  .LBB42_63: // %cond.store59
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.h, z0.h[6]
+; CHECK-NEXT:    str h2, [x9]
+; CHECK-NEXT:    tbz w8, #31, .LBB42_39
+; CHECK-NEXT:  .LBB42_64: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.h, z0.h[7]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str h0, [x8]
+; CHECK-NEXT:    add sp, sp, #32
 ; CHECK-NEXT:    ret
   %vals = load <32 x half>, ptr %a
   %idxs = load <32 x i32>, ptr %b
   %ext = zext <32 x i32> %idxs to <32 x i64>
   %byte_ptrs = getelementptr i8, ptr %base, <32 x i64> %ext
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x half> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f16(<32 x half> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_64b_scaled(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_scatter_64b_scaled:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1w { z0.d }, p0, [x2, z1.d, lsl #2]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #96
+; CHECK-NEXT:    .cfi_def_cfa_offset 96
+; CHECK-NEXT:    stp d15, d14, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d13, d12, [sp, #48] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d11, d10, [sp, #64] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #80] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -24
+; CHECK-NEXT:    .cfi_offset b11, -32
+; CHECK-NEXT:    .cfi_offset b12, -40
+; CHECK-NEXT:    .cfi_offset b13, -48
+; CHECK-NEXT:    .cfi_offset b14, -56
+; CHECK-NEXT:    .cfi_offset b15, -64
+; CHECK-NEXT:    adrp x8, .LCPI43_1
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q6, q2, [x0, #64]
+; CHECK-NEXT:    ldr q19, [x8, :lo12:.LCPI43_1]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z6.s, z19.s
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z19.s
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z4.s, z3.s[1]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    mov z5.s, z3.s[2]
+; CHECK-NEXT:    mov z4.s, z3.s[3]
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z4.s, z3.s[1]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s5
+; CHECK-NEXT:    mov z5.s, z3.s[2]
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z10.s, z3.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z19.s
+; CHECK-NEXT:    bfi w9, w8, #18, #1
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    bfi w9, w10, #19, #1
+; CHECK-NEXT:    fmov w10, s5
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    mov z4.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    mov z11.s, z4.s[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z12.s, z4.s[3]
+; CHECK-NEXT:    ldp q29, q26, [x0]
+; CHECK-NEXT:    orr w8, w9, w10, lsl #22
+; CHECK-NEXT:    fmov w9, s10
+; CHECK-NEXT:    mov z10.s, z4.s[1]
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    fmov w11, s10
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, z19.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p2.s, p0/z, z29.s, z19.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z10.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s12
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    fmov w9, s11
+; CHECK-NEXT:    adrp x11, .LCPI43_0
+; CHECK-NEXT:    mov z11.s, p2/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z13.s, z11.s[1]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z26.s, z19.s
+; CHECK-NEXT:    ldr q4, [x11, :lo12:.LCPI43_0]
+; CHECK-NEXT:    mov z14.s, z11.s[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #26
+; CHECK-NEXT:    fmov w9, s13
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    fmov w10, s11
+; CHECK-NEXT:    mov z13.s, z11.s[2]
+; CHECK-NEXT:    mov z11.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s13
+; CHECK-NEXT:    fmov w12, s14
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z14.s, z11.s[2]
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    mov z13.s, z11.s[1]
+; CHECK-NEXT:    fmov w13, s11
+; CHECK-NEXT:    bfi w10, w11, #2, #1
+; CHECK-NEXT:    fmov w11, s14
+; CHECK-NEXT:    fmov w9, s13
+; CHECK-NEXT:    ldp q21, q16, [x0, #32]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z15.s, z11.s[3]
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    bfi w10, w9, #5, #1
+; CHECK-NEXT:    mov z12.s, z10.s[1]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #6
+; CHECK-NEXT:    fmov w11, s15
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z21.s, z19.s
+; CHECK-NEXT:    mov z11.s, z10.s[2]
+; CHECK-NEXT:    mov z13.s, z10.s[3]
+; CHECK-NEXT:    fmov w12, s10
+; CHECK-NEXT:    mov z10.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w13, s10
+; CHECK-NEXT:    fmov w9, s12
+; CHECK-NEXT:    mov z12.s, z10.s[1]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #7
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    mov z14.s, z10.s[2]
+; CHECK-NEXT:    and w11, w13, #0x1
+; CHECK-NEXT:    fmov w13, s12
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #28
+; CHECK-NEXT:    orr w10, w10, w11, lsl #8
+; CHECK-NEXT:    mov z15.s, z10.s[3]
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z16.s, z19.s
+; CHECK-NEXT:    fmov w11, s15
+; CHECK-NEXT:    orr w9, w10, w13, lsl #9
+; CHECK-NEXT:    fmov w10, s14
+; CHECK-NEXT:    mov z19.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    fmov w12, s19
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z10.s, z19.s[1]
+; CHECK-NEXT:    mov z12.s, z19.s[2]
+; CHECK-NEXT:    mov z14.s, z19.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s12
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s10
+; CHECK-NEXT:    fmov w12, s11
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s14
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s13
+; CHECK-NEXT:    ldp q31, q30, [x1, #32]
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q28, q27, [x1, #64]
+; CHECK-NEXT:    ldp q25, q24, [x1, #96]
+; CHECK-NEXT:    ldp q23, q22, [x1, #128]
+; CHECK-NEXT:    ldp q20, q18, [x1, #160]
+; CHECK-NEXT:    ldp q17, q7, [x1, #192]
+; CHECK-NEXT:    ldp q5, q3, [x1, #224]
+; CHECK-NEXT:    ldp q8, q9, [x1]
+; CHECK-NEXT:    stp x2, x2, [sp]
+; CHECK-NEXT:    ldr q19, [sp]
+; CHECK-NEXT:    movprfx z10, z8
+; CHECK-NEXT:    lsl z10.d, p0/m, z10.d, z4.d
+; CHECK-NEXT:    movprfx z8, z9
+; CHECK-NEXT:    lsl z8.d, p0/m, z8.d, z4.d
+; CHECK-NEXT:    add z9.d, z19.d, z10.d
+; CHECK-NEXT:    tbnz w8, #0, .LBB43_44
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    add z8.d, z19.d, z8.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB43_45
+; CHECK-NEXT:  .LBB43_2: // %else2
+; CHECK-NEXT:    lsl z31.d, p0/m, z31.d, z4.d
+; CHECK-NEXT:    tbnz w8, #2, .LBB43_46
+; CHECK-NEXT:  .LBB43_3: // %else4
+; CHECK-NEXT:    add z31.d, z19.d, z31.d
+; CHECK-NEXT:    tbz w8, #3, .LBB43_5
+; CHECK-NEXT:  .LBB43_4: // %cond.store5
+; CHECK-NEXT:    mov z8.d, z8.d[1]
+; CHECK-NEXT:    mov z29.s, z29.s[3]
+; CHECK-NEXT:    fmov x9, d8
+; CHECK-NEXT:    str s29, [x9]
+; CHECK-NEXT:  .LBB43_5: // %else6
+; CHECK-NEXT:    movprfx z29, z30
+; CHECK-NEXT:    lsl z29.d, p0/m, z29.d, z4.d
+; CHECK-NEXT:    tbnz w8, #4, .LBB43_47
+; CHECK-NEXT:  // %bb.6: // %else8
+; CHECK-NEXT:    add z29.d, z19.d, z29.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB43_48
+; CHECK-NEXT:  .LBB43_7: // %else10
+; CHECK-NEXT:    lsl z28.d, p0/m, z28.d, z4.d
+; CHECK-NEXT:    tbnz w8, #6, .LBB43_49
+; CHECK-NEXT:  .LBB43_8: // %else12
+; CHECK-NEXT:    add z28.d, z19.d, z28.d
+; CHECK-NEXT:    tbz w8, #7, .LBB43_10
+; CHECK-NEXT:  .LBB43_9: // %cond.store13
+; CHECK-NEXT:    mov z29.d, z29.d[1]
+; CHECK-NEXT:    mov z26.s, z26.s[3]
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:  .LBB43_10: // %else14
+; CHECK-NEXT:    movprfx z26, z27
+; CHECK-NEXT:    lsl z26.d, p0/m, z26.d, z4.d
+; CHECK-NEXT:    tbnz w8, #8, .LBB43_50
+; CHECK-NEXT:  // %bb.11: // %else16
+; CHECK-NEXT:    add z26.d, z19.d, z26.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB43_51
+; CHECK-NEXT:  .LBB43_12: // %else18
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z4.d
+; CHECK-NEXT:    tbnz w8, #10, .LBB43_52
+; CHECK-NEXT:  .LBB43_13: // %else20
+; CHECK-NEXT:    add z25.d, z19.d, z25.d
+; CHECK-NEXT:    tbz w8, #11, .LBB43_15
+; CHECK-NEXT:  .LBB43_14: // %cond.store21
+; CHECK-NEXT:    mov z26.d, z26.d[1]
+; CHECK-NEXT:    mov z21.s, z21.s[3]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    str s21, [x9]
+; CHECK-NEXT:  .LBB43_15: // %else22
+; CHECK-NEXT:    movprfx z21, z24
+; CHECK-NEXT:    lsl z21.d, p0/m, z21.d, z4.d
+; CHECK-NEXT:    tbnz w8, #12, .LBB43_53
+; CHECK-NEXT:  // %bb.16: // %else24
+; CHECK-NEXT:    add z21.d, z19.d, z21.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB43_54
+; CHECK-NEXT:  .LBB43_17: // %else26
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z4.d
+; CHECK-NEXT:    tbnz w8, #14, .LBB43_55
+; CHECK-NEXT:  .LBB43_18: // %else28
+; CHECK-NEXT:    add z23.d, z19.d, z23.d
+; CHECK-NEXT:    tbz w8, #15, .LBB43_20
+; CHECK-NEXT:  .LBB43_19: // %cond.store29
+; CHECK-NEXT:    mov z21.d, z21.d[1]
+; CHECK-NEXT:    mov z16.s, z16.s[3]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:  .LBB43_20: // %else30
+; CHECK-NEXT:    movprfx z16, z22
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z4.d
+; CHECK-NEXT:    tbnz w8, #16, .LBB43_56
+; CHECK-NEXT:  // %bb.21: // %else32
+; CHECK-NEXT:    add z16.d, z19.d, z16.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB43_57
+; CHECK-NEXT:  .LBB43_22: // %else34
+; CHECK-NEXT:    lsl z20.d, p0/m, z20.d, z4.d
+; CHECK-NEXT:    tbnz w8, #18, .LBB43_58
+; CHECK-NEXT:  .LBB43_23: // %else36
+; CHECK-NEXT:    add z20.d, z19.d, z20.d
+; CHECK-NEXT:    tbz w8, #19, .LBB43_25
+; CHECK-NEXT:  .LBB43_24: // %cond.store37
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    mov z6.s, z6.s[3]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:  .LBB43_25: // %else38
+; CHECK-NEXT:    movprfx z6, z18
+; CHECK-NEXT:    lsl z6.d, p0/m, z6.d, z4.d
+; CHECK-NEXT:    tbz w8, #20, .LBB43_27
+; CHECK-NEXT:  // %bb.26: // %cond.store39
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:  .LBB43_27: // %else40
+; CHECK-NEXT:    add z6.d, z19.d, z6.d
+; CHECK-NEXT:    tbz w8, #21, .LBB43_29
+; CHECK-NEXT:  // %bb.28: // %cond.store41
+; CHECK-NEXT:    mov z16.d, z20.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.s, z2.s[1]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:  .LBB43_29: // %else42
+; CHECK-NEXT:    movprfx z16, z17
+; CHECK-NEXT:    lsl z16.d, p0/m, z16.d, z4.d
+; CHECK-NEXT:    tbz w8, #22, .LBB43_31
+; CHECK-NEXT:  // %bb.30: // %cond.store43
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z17.s, z2.s[2]
+; CHECK-NEXT:    str s17, [x9]
+; CHECK-NEXT:  .LBB43_31: // %else44
+; CHECK-NEXT:    add z16.d, z19.d, z16.d
+; CHECK-NEXT:    tbz w8, #23, .LBB43_33
+; CHECK-NEXT:  // %bb.32: // %cond.store45
+; CHECK-NEXT:    mov z6.d, z6.d[1]
+; CHECK-NEXT:    mov z2.s, z2.s[3]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:  .LBB43_33: // %else46
+; CHECK-NEXT:    movprfx z2, z7
+; CHECK-NEXT:    lsl z2.d, p0/m, z2.d, z4.d
+; CHECK-NEXT:    tbnz w8, #24, .LBB43_59
+; CHECK-NEXT:  // %bb.34: // %else48
+; CHECK-NEXT:    add z2.d, z19.d, z2.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB43_60
+; CHECK-NEXT:  .LBB43_35: // %else50
+; CHECK-NEXT:    lsl z5.d, p0/m, z5.d, z4.d
+; CHECK-NEXT:    tbnz w8, #26, .LBB43_61
+; CHECK-NEXT:  .LBB43_36: // %else52
+; CHECK-NEXT:    add z5.d, z19.d, z5.d
+; CHECK-NEXT:    tbz w8, #27, .LBB43_38
+; CHECK-NEXT:  .LBB43_37: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:  .LBB43_38: // %else54
+; CHECK-NEXT:    movprfx z1, z3
+; CHECK-NEXT:    lsl z1.d, p0/m, z1.d, z4.d
+; CHECK-NEXT:    tbnz w8, #28, .LBB43_62
+; CHECK-NEXT:  // %bb.39: // %else56
+; CHECK-NEXT:    add z1.d, z19.d, z1.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB43_63
+; CHECK-NEXT:  .LBB43_40: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB43_64
+; CHECK-NEXT:  .LBB43_41: // %else60
+; CHECK-NEXT:    tbz w8, #31, .LBB43_43
+; CHECK-NEXT:  .LBB43_42: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:  .LBB43_43: // %else62
+; CHECK-NEXT:    ldp d9, d8, [sp, #80] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d11, d10, [sp, #64] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d13, d12, [sp, #48] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d15, d14, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    add sp, sp, #96
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB43_44: // %cond.store
+; CHECK-NEXT:    fmov x9, d9
+; CHECK-NEXT:    str s29, [x9]
+; CHECK-NEXT:    add z8.d, z19.d, z8.d
+; CHECK-NEXT:    tbz w8, #1, .LBB43_2
+; CHECK-NEXT:  .LBB43_45: // %cond.store1
+; CHECK-NEXT:    mov z9.d, z9.d[1]
+; CHECK-NEXT:    fmov x9, d9
+; CHECK-NEXT:    mov z9.s, z29.s[1]
+; CHECK-NEXT:    str s9, [x9]
+; CHECK-NEXT:    lsl z31.d, p0/m, z31.d, z4.d
+; CHECK-NEXT:    tbz w8, #2, .LBB43_3
+; CHECK-NEXT:  .LBB43_46: // %cond.store3
+; CHECK-NEXT:    fmov x9, d8
+; CHECK-NEXT:    mov z9.s, z29.s[2]
+; CHECK-NEXT:    str s9, [x9]
+; CHECK-NEXT:    add z31.d, z19.d, z31.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB43_4
+; CHECK-NEXT:    b .LBB43_5
+; CHECK-NEXT:  .LBB43_47: // %cond.store7
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:    add z29.d, z19.d, z29.d
+; CHECK-NEXT:    tbz w8, #5, .LBB43_7
+; CHECK-NEXT:  .LBB43_48: // %cond.store9
+; CHECK-NEXT:    mov z30.d, z31.d[1]
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    mov z30.s, z26.s[1]
+; CHECK-NEXT:    str s30, [x9]
+; CHECK-NEXT:    lsl z28.d, p0/m, z28.d, z4.d
+; CHECK-NEXT:    tbz w8, #6, .LBB43_8
+; CHECK-NEXT:  .LBB43_49: // %cond.store11
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    mov z30.s, z26.s[2]
+; CHECK-NEXT:    str s30, [x9]
+; CHECK-NEXT:    add z28.d, z19.d, z28.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB43_9
+; CHECK-NEXT:    b .LBB43_10
+; CHECK-NEXT:  .LBB43_50: // %cond.store15
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    str s21, [x9]
+; CHECK-NEXT:    add z26.d, z19.d, z26.d
+; CHECK-NEXT:    tbz w8, #9, .LBB43_12
+; CHECK-NEXT:  .LBB43_51: // %cond.store17
+; CHECK-NEXT:    mov z27.d, z28.d[1]
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    mov z27.s, z21.s[1]
+; CHECK-NEXT:    str s27, [x9]
+; CHECK-NEXT:    lsl z25.d, p0/m, z25.d, z4.d
+; CHECK-NEXT:    tbz w8, #10, .LBB43_13
+; CHECK-NEXT:  .LBB43_52: // %cond.store19
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    mov z27.s, z21.s[2]
+; CHECK-NEXT:    str s27, [x9]
+; CHECK-NEXT:    add z25.d, z19.d, z25.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB43_14
+; CHECK-NEXT:    b .LBB43_15
+; CHECK-NEXT:  .LBB43_53: // %cond.store23
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    add z21.d, z19.d, z21.d
+; CHECK-NEXT:    tbz w8, #13, .LBB43_17
+; CHECK-NEXT:  .LBB43_54: // %cond.store25
+; CHECK-NEXT:    mov z24.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    mov z24.s, z16.s[1]
+; CHECK-NEXT:    str s24, [x9]
+; CHECK-NEXT:    lsl z23.d, p0/m, z23.d, z4.d
+; CHECK-NEXT:    tbz w8, #14, .LBB43_18
+; CHECK-NEXT:  .LBB43_55: // %cond.store27
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z24.s, z16.s[2]
+; CHECK-NEXT:    str s24, [x9]
+; CHECK-NEXT:    add z23.d, z19.d, z23.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB43_19
+; CHECK-NEXT:    b .LBB43_20
+; CHECK-NEXT:  .LBB43_56: // %cond.store31
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:    add z16.d, z19.d, z16.d
+; CHECK-NEXT:    tbz w8, #17, .LBB43_22
+; CHECK-NEXT:  .LBB43_57: // %cond.store33
+; CHECK-NEXT:    mov z21.d, z23.d[1]
+; CHECK-NEXT:    fmov x9, d21
+; CHECK-NEXT:    mov z21.s, z6.s[1]
+; CHECK-NEXT:    str s21, [x9]
+; CHECK-NEXT:    lsl z20.d, p0/m, z20.d, z4.d
+; CHECK-NEXT:    tbz w8, #18, .LBB43_23
+; CHECK-NEXT:  .LBB43_58: // %cond.store35
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z21.s, z6.s[2]
+; CHECK-NEXT:    str s21, [x9]
+; CHECK-NEXT:    add z20.d, z19.d, z20.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB43_24
+; CHECK-NEXT:    b .LBB43_25
+; CHECK-NEXT:  .LBB43_59: // %cond.store47
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    add z2.d, z19.d, z2.d
+; CHECK-NEXT:    tbz w8, #25, .LBB43_35
+; CHECK-NEXT:  .LBB43_60: // %cond.store49
+; CHECK-NEXT:    mov z6.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    mov z6.s, z1.s[1]
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:    lsl z5.d, p0/m, z5.d, z4.d
+; CHECK-NEXT:    tbz w8, #26, .LBB43_36
+; CHECK-NEXT:  .LBB43_61: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z6.s, z1.s[2]
+; CHECK-NEXT:    str s6, [x9]
+; CHECK-NEXT:    add z5.d, z19.d, z5.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB43_37
+; CHECK-NEXT:    b .LBB43_38
+; CHECK-NEXT:  .LBB43_62: // %cond.store55
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    add z1.d, z19.d, z1.d
+; CHECK-NEXT:    tbz w8, #29, .LBB43_40
+; CHECK-NEXT:  .LBB43_63: // %cond.store57
+; CHECK-NEXT:    mov z2.d, z5.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB43_41
+; CHECK-NEXT:  .LBB43_64: // %cond.store59
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB43_42
+; CHECK-NEXT:    b .LBB43_43
   %vals = load <32 x float>, ptr %a
   %idxs = load <32 x i64>, ptr %b
   %ptrs = getelementptr float, ptr %base, <32 x i64> %idxs
   %mask = fcmp oeq <32 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_64b_unscaled(ptr %a, ptr %b, ptr %base) #0 {
 ; CHECK-LABEL: masked_scatter_64b_unscaled:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1w { z0.d }, p0, [x2, z1.d]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #80
+; CHECK-NEXT:    .cfi_def_cfa_offset 80
+; CHECK-NEXT:    str d14, [sp, #16] // 8-byte Folded Spill
+; CHECK-NEXT:    stp d13, d12, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d11, d10, [sp, #48] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #64] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -24
+; CHECK-NEXT:    .cfi_offset b11, -32
+; CHECK-NEXT:    .cfi_offset b12, -40
+; CHECK-NEXT:    .cfi_offset b13, -48
+; CHECK-NEXT:    .cfi_offset b14, -64
+; CHECK-NEXT:    adrp x8, .LCPI44_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q5, q2, [x0, #64]
+; CHECK-NEXT:    ldr q18, [x8, :lo12:.LCPI44_0]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z5.s, z18.s
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z18.s
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z4.s, z3.s[1]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    mov z6.s, z3.s[2]
+; CHECK-NEXT:    mov z4.s, z3.s[3]
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z4.s, z3.s[1]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s6
+; CHECK-NEXT:    mov z6.s, z3.s[2]
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z9.s, z3.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z18.s
+; CHECK-NEXT:    bfi w9, w8, #18, #1
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    bfi w9, w10, #19, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    mov z10.s, z3.s[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z11.s, z3.s[3]
+; CHECK-NEXT:    ldp q28, q25, [x0]
+; CHECK-NEXT:    orr w8, w9, w10, lsl #22
+; CHECK-NEXT:    fmov w9, s9
+; CHECK-NEXT:    mov z9.s, z3.s[1]
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, z18.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s9
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s10
+; CHECK-NEXT:    mov z9.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z28.s, z18.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z10.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z12.s, z10.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    fmov w9, s11
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s10
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z25.s, z18.s
+; CHECK-NEXT:    mov z13.s, z10.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w12, s13
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z11.s, z9.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s12
+; CHECK-NEXT:    mov z12.s, z10.s[2]
+; CHECK-NEXT:    mov z10.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s12
+; CHECK-NEXT:    mov z13.s, z10.s[2]
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    mov z12.s, z10.s[1]
+; CHECK-NEXT:    fmov w13, s10
+; CHECK-NEXT:    fmov w9, s12
+; CHECK-NEXT:    bfi w10, w11, #2, #1
+; CHECK-NEXT:    fmov w11, s13
+; CHECK-NEXT:    ldp q20, q16, [x0, #32]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z14.s, z10.s[3]
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    bfi w10, w9, #5, #1
+; CHECK-NEXT:    mov z10.s, z9.s[2]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #6
+; CHECK-NEXT:    fmov w11, s14
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z20.s, z18.s
+; CHECK-NEXT:    mov z12.s, z9.s[3]
+; CHECK-NEXT:    fmov w12, s9
+; CHECK-NEXT:    mov z9.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w13, s9
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w9, s11
+; CHECK-NEXT:    mov z11.s, z9.s[1]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #7
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w13, #0x1
+; CHECK-NEXT:    fmov w13, s11
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z13.s, z9.s[2]
+; CHECK-NEXT:    orr w8, w8, w12, lsl #28
+; CHECK-NEXT:    orr w10, w10, w11, lsl #8
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    mov z14.s, z9.s[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z16.s, z18.s
+; CHECK-NEXT:    orr w9, w10, w13, lsl #9
+; CHECK-NEXT:    fmov w10, s13
+; CHECK-NEXT:    mov z18.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s14
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    mov z9.s, z18.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z11.s, z18.s[2]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z13.s, z18.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s11
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s9
+; CHECK-NEXT:    fmov w12, s10
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s13
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s12
+; CHECK-NEXT:    ldp q8, q31, [x1]
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q30, q29, [x1, #32]
+; CHECK-NEXT:    ldp q27, q26, [x1, #64]
+; CHECK-NEXT:    ldp q24, q23, [x1, #96]
+; CHECK-NEXT:    ldp q22, q21, [x1, #128]
+; CHECK-NEXT:    ldp q19, q17, [x1, #160]
+; CHECK-NEXT:    ldp q7, q6, [x1, #192]
+; CHECK-NEXT:    ldp q4, q3, [x1, #224]
+; CHECK-NEXT:    stp x2, x2, [sp]
+; CHECK-NEXT:    ldr q18, [sp]
+; CHECK-NEXT:    add z8.d, z18.d, z8.d
+; CHECK-NEXT:    tbnz w8, #0, .LBB44_34
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    add z31.d, z18.d, z31.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB44_35
+; CHECK-NEXT:  .LBB44_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB44_36
+; CHECK-NEXT:  .LBB44_3: // %else4
+; CHECK-NEXT:    add z30.d, z18.d, z30.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB44_37
+; CHECK-NEXT:  .LBB44_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB44_38
+; CHECK-NEXT:  .LBB44_5: // %else8
+; CHECK-NEXT:    add z28.d, z18.d, z29.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB44_39
+; CHECK-NEXT:  .LBB44_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB44_40
+; CHECK-NEXT:  .LBB44_7: // %else12
+; CHECK-NEXT:    add z27.d, z18.d, z27.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB44_41
+; CHECK-NEXT:  .LBB44_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB44_42
+; CHECK-NEXT:  .LBB44_9: // %else16
+; CHECK-NEXT:    add z25.d, z18.d, z26.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB44_43
+; CHECK-NEXT:  .LBB44_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB44_44
+; CHECK-NEXT:  .LBB44_11: // %else20
+; CHECK-NEXT:    add z24.d, z18.d, z24.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB44_45
+; CHECK-NEXT:  .LBB44_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB44_46
+; CHECK-NEXT:  .LBB44_13: // %else24
+; CHECK-NEXT:    add z20.d, z18.d, z23.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB44_47
+; CHECK-NEXT:  .LBB44_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB44_48
+; CHECK-NEXT:  .LBB44_15: // %else28
+; CHECK-NEXT:    add z22.d, z18.d, z22.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB44_49
+; CHECK-NEXT:  .LBB44_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB44_50
+; CHECK-NEXT:  .LBB44_17: // %else32
+; CHECK-NEXT:    add z16.d, z18.d, z21.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB44_51
+; CHECK-NEXT:  .LBB44_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB44_52
+; CHECK-NEXT:  .LBB44_19: // %else36
+; CHECK-NEXT:    add z19.d, z18.d, z19.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB44_53
+; CHECK-NEXT:  .LBB44_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB44_54
+; CHECK-NEXT:  .LBB44_21: // %else40
+; CHECK-NEXT:    add z5.d, z18.d, z17.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB44_55
+; CHECK-NEXT:  .LBB44_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB44_56
+; CHECK-NEXT:  .LBB44_23: // %else44
+; CHECK-NEXT:    add z7.d, z18.d, z7.d
+; CHECK-NEXT:    tbnz w8, #23, .LBB44_57
+; CHECK-NEXT:  .LBB44_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB44_58
+; CHECK-NEXT:  .LBB44_25: // %else48
+; CHECK-NEXT:    add z2.d, z18.d, z6.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB44_59
+; CHECK-NEXT:  .LBB44_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB44_60
+; CHECK-NEXT:  .LBB44_27: // %else52
+; CHECK-NEXT:    add z4.d, z18.d, z4.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB44_61
+; CHECK-NEXT:  .LBB44_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB44_62
+; CHECK-NEXT:  .LBB44_29: // %else56
+; CHECK-NEXT:    add z1.d, z18.d, z3.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB44_63
+; CHECK-NEXT:  .LBB44_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB44_64
+; CHECK-NEXT:  .LBB44_31: // %else60
+; CHECK-NEXT:    tbz w8, #31, .LBB44_33
+; CHECK-NEXT:  .LBB44_32: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:  .LBB44_33: // %else62
+; CHECK-NEXT:    ldp d9, d8, [sp, #64] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d11, d10, [sp, #48] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d13, d12, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    ldr d14, [sp, #16] // 8-byte Folded Reload
+; CHECK-NEXT:    add sp, sp, #80
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB44_34: // %cond.store
+; CHECK-NEXT:    fmov x9, d8
+; CHECK-NEXT:    str s28, [x9]
+; CHECK-NEXT:    add z31.d, z18.d, z31.d
+; CHECK-NEXT:    tbz w8, #1, .LBB44_2
+; CHECK-NEXT:  .LBB44_35: // %cond.store1
+; CHECK-NEXT:    mov z8.d, z8.d[1]
+; CHECK-NEXT:    fmov x9, d8
+; CHECK-NEXT:    mov z8.s, z28.s[1]
+; CHECK-NEXT:    str s8, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB44_3
+; CHECK-NEXT:  .LBB44_36: // %cond.store3
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    mov z8.s, z28.s[2]
+; CHECK-NEXT:    str s8, [x9]
+; CHECK-NEXT:    add z30.d, z18.d, z30.d
+; CHECK-NEXT:    tbz w8, #3, .LBB44_4
+; CHECK-NEXT:  .LBB44_37: // %cond.store5
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    mov z28.s, z28.s[3]
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    str s28, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB44_5
+; CHECK-NEXT:  .LBB44_38: // %cond.store7
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    str s25, [x9]
+; CHECK-NEXT:    add z28.d, z18.d, z29.d
+; CHECK-NEXT:    tbz w8, #5, .LBB44_6
+; CHECK-NEXT:  .LBB44_39: // %cond.store9
+; CHECK-NEXT:    mov z29.d, z30.d[1]
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    mov z29.s, z25.s[1]
+; CHECK-NEXT:    str s29, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB44_7
+; CHECK-NEXT:  .LBB44_40: // %cond.store11
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z29.s, z25.s[2]
+; CHECK-NEXT:    str s29, [x9]
+; CHECK-NEXT:    add z27.d, z18.d, z27.d
+; CHECK-NEXT:    tbz w8, #7, .LBB44_8
+; CHECK-NEXT:  .LBB44_41: // %cond.store13
+; CHECK-NEXT:    mov z28.d, z28.d[1]
+; CHECK-NEXT:    mov z25.s, z25.s[3]
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    str s25, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB44_9
+; CHECK-NEXT:  .LBB44_42: // %cond.store15
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:    add z25.d, z18.d, z26.d
+; CHECK-NEXT:    tbz w8, #9, .LBB44_10
+; CHECK-NEXT:  .LBB44_43: // %cond.store17
+; CHECK-NEXT:    mov z26.d, z27.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    mov z26.s, z20.s[1]
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB44_11
+; CHECK-NEXT:  .LBB44_44: // %cond.store19
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    mov z26.s, z20.s[2]
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:    add z24.d, z18.d, z24.d
+; CHECK-NEXT:    tbz w8, #11, .LBB44_12
+; CHECK-NEXT:  .LBB44_45: // %cond.store21
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    mov z20.s, z20.s[3]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB44_13
+; CHECK-NEXT:  .LBB44_46: // %cond.store23
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    add z20.d, z18.d, z23.d
+; CHECK-NEXT:    tbz w8, #13, .LBB44_14
+; CHECK-NEXT:  .LBB44_47: // %cond.store25
+; CHECK-NEXT:    mov z23.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    mov z23.s, z16.s[1]
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB44_15
+; CHECK-NEXT:  .LBB44_48: // %cond.store27
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z23.s, z16.s[2]
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:    add z22.d, z18.d, z22.d
+; CHECK-NEXT:    tbz w8, #15, .LBB44_16
+; CHECK-NEXT:  .LBB44_49: // %cond.store29
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    mov z16.s, z16.s[3]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    tbz w8, #16, .LBB44_17
+; CHECK-NEXT:  .LBB44_50: // %cond.store31
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    add z16.d, z18.d, z21.d
+; CHECK-NEXT:    tbz w8, #17, .LBB44_18
+; CHECK-NEXT:  .LBB44_51: // %cond.store33
+; CHECK-NEXT:    mov z20.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z20.s, z5.s[1]
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:    tbz w8, #18, .LBB44_19
+; CHECK-NEXT:  .LBB44_52: // %cond.store35
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z20.s, z5.s[2]
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:    add z19.d, z18.d, z19.d
+; CHECK-NEXT:    tbz w8, #19, .LBB44_20
+; CHECK-NEXT:  .LBB44_53: // %cond.store37
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    mov z5.s, z5.s[3]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #20, .LBB44_21
+; CHECK-NEXT:  .LBB44_54: // %cond.store39
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    add z5.d, z18.d, z17.d
+; CHECK-NEXT:    tbz w8, #21, .LBB44_22
+; CHECK-NEXT:  .LBB44_55: // %cond.store41
+; CHECK-NEXT:    mov z16.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.s, z2.s[1]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    tbz w8, #22, .LBB44_23
+; CHECK-NEXT:  .LBB44_56: // %cond.store43
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z16.s, z2.s[2]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    add z7.d, z18.d, z7.d
+; CHECK-NEXT:    tbz w8, #23, .LBB44_24
+; CHECK-NEXT:  .LBB44_57: // %cond.store45
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z2.s, z2.s[3]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #24, .LBB44_25
+; CHECK-NEXT:  .LBB44_58: // %cond.store47
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    add z2.d, z18.d, z6.d
+; CHECK-NEXT:    tbz w8, #25, .LBB44_26
+; CHECK-NEXT:  .LBB44_59: // %cond.store49
+; CHECK-NEXT:    mov z5.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z5.s, z1.s[1]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #26, .LBB44_27
+; CHECK-NEXT:  .LBB44_60: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z5.s, z1.s[2]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    add z4.d, z18.d, z4.d
+; CHECK-NEXT:    tbz w8, #27, .LBB44_28
+; CHECK-NEXT:  .LBB44_61: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #28, .LBB44_29
+; CHECK-NEXT:  .LBB44_62: // %cond.store55
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    add z1.d, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #29, .LBB44_30
+; CHECK-NEXT:  .LBB44_63: // %cond.store57
+; CHECK-NEXT:    mov z2.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB44_31
+; CHECK-NEXT:  .LBB44_64: // %cond.store59
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB44_32
+; CHECK-NEXT:    b .LBB44_33
   %vals = load <32 x float>, ptr %a
   %idxs = load <32 x i64>, ptr %b
   %byte_ptrs = getelementptr i8, ptr %base, <32 x i64> %idxs
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_vec_plus_reg(ptr %a, ptr %b, i64 %off) #0 {
 ; CHECK-LABEL: masked_scatter_vec_plus_reg:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1w { z0.d }, p0, [x2, z1.d]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #80
+; CHECK-NEXT:    .cfi_def_cfa_offset 80
+; CHECK-NEXT:    str d14, [sp, #16] // 8-byte Folded Spill
+; CHECK-NEXT:    stp d13, d12, [sp, #32] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d11, d10, [sp, #48] // 16-byte Folded Spill
+; CHECK-NEXT:    stp d9, d8, [sp, #64] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    .cfi_offset b10, -24
+; CHECK-NEXT:    .cfi_offset b11, -32
+; CHECK-NEXT:    .cfi_offset b12, -40
+; CHECK-NEXT:    .cfi_offset b13, -48
+; CHECK-NEXT:    .cfi_offset b14, -64
+; CHECK-NEXT:    adrp x8, .LCPI45_0
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q5, q2, [x0, #64]
+; CHECK-NEXT:    ldr q18, [x8, :lo12:.LCPI45_0]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z5.s, z18.s
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z18.s
+; CHECK-NEXT:    fmov w9, s3
+; CHECK-NEXT:    mov z4.s, z3.s[1]
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    mov z6.s, z3.s[2]
+; CHECK-NEXT:    mov z4.s, z3.s[3]
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s4
+; CHECK-NEXT:    lsl w9, w9, #16
+; CHECK-NEXT:    mov z4.s, z3.s[1]
+; CHECK-NEXT:    bfi w9, w8, #17, #1
+; CHECK-NEXT:    fmov w8, s6
+; CHECK-NEXT:    mov z6.s, z3.s[2]
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    mov z9.s, z3.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z18.s
+; CHECK-NEXT:    bfi w9, w8, #18, #1
+; CHECK-NEXT:    fmov w8, s4
+; CHECK-NEXT:    bfi w9, w10, #19, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    bfi w9, w11, #20, #1
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w9, w8, #21, #1
+; CHECK-NEXT:    mov z10.s, z3.s[2]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z11.s, z3.s[3]
+; CHECK-NEXT:    ldp q28, q25, [x0]
+; CHECK-NEXT:    orr w8, w9, w10, lsl #22
+; CHECK-NEXT:    fmov w9, s9
+; CHECK-NEXT:    mov z9.s, z3.s[1]
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, z18.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s9
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s10
+; CHECK-NEXT:    mov z9.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z28.s, z18.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z10.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z12.s, z10.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    fmov w9, s11
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s10
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z25.s, z18.s
+; CHECK-NEXT:    mov z13.s, z10.s[3]
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w12, s13
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z11.s, z9.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #27
+; CHECK-NEXT:    fmov w9, s12
+; CHECK-NEXT:    mov z12.s, z10.s[2]
+; CHECK-NEXT:    mov z10.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s12
+; CHECK-NEXT:    mov z13.s, z10.s[2]
+; CHECK-NEXT:    bfi w10, w9, #1, #1
+; CHECK-NEXT:    mov z12.s, z10.s[1]
+; CHECK-NEXT:    fmov w13, s10
+; CHECK-NEXT:    fmov w9, s12
+; CHECK-NEXT:    bfi w10, w11, #2, #1
+; CHECK-NEXT:    fmov w11, s13
+; CHECK-NEXT:    ldp q20, q16, [x0, #32]
+; CHECK-NEXT:    bfi w10, w12, #3, #1
+; CHECK-NEXT:    mov z14.s, z10.s[3]
+; CHECK-NEXT:    bfi w10, w13, #4, #1
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    bfi w10, w9, #5, #1
+; CHECK-NEXT:    mov z10.s, z9.s[2]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #6
+; CHECK-NEXT:    fmov w11, s14
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z20.s, z18.s
+; CHECK-NEXT:    mov z12.s, z9.s[3]
+; CHECK-NEXT:    fmov w12, s9
+; CHECK-NEXT:    mov z9.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w13, s9
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    fmov w9, s11
+; CHECK-NEXT:    mov z11.s, z9.s[1]
+; CHECK-NEXT:    orr w10, w10, w11, lsl #7
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    and w11, w13, #0x1
+; CHECK-NEXT:    fmov w13, s11
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z13.s, z9.s[2]
+; CHECK-NEXT:    orr w8, w8, w12, lsl #28
+; CHECK-NEXT:    orr w10, w10, w11, lsl #8
+; CHECK-NEXT:    and w13, w13, #0x1
+; CHECK-NEXT:    mov z14.s, z9.s[3]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #29
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z16.s, z18.s
+; CHECK-NEXT:    orr w9, w10, w13, lsl #9
+; CHECK-NEXT:    fmov w10, s13
+; CHECK-NEXT:    mov z18.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fmov w11, s14
+; CHECK-NEXT:    fmov w12, s18
+; CHECK-NEXT:    mov z9.s, z18.s[1]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z11.s, z18.s[2]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    mov z13.s, z18.s[3]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #10
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #11
+; CHECK-NEXT:    fmov w11, s11
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s9
+; CHECK-NEXT:    fmov w12, s10
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    and w12, w12, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #13
+; CHECK-NEXT:    fmov w10, s13
+; CHECK-NEXT:    orr w9, w9, w11, lsl #14
+; CHECK-NEXT:    fmov w11, s12
+; CHECK-NEXT:    ldp q8, q31, [x1]
+; CHECK-NEXT:    orr w8, w8, w12, lsl #30
+; CHECK-NEXT:    orr w9, w9, w10, lsl #15
+; CHECK-NEXT:    orr w8, w8, w11, lsl #31
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q30, q29, [x1, #32]
+; CHECK-NEXT:    ldp q27, q26, [x1, #64]
+; CHECK-NEXT:    ldp q24, q23, [x1, #96]
+; CHECK-NEXT:    ldp q22, q21, [x1, #128]
+; CHECK-NEXT:    ldp q19, q17, [x1, #160]
+; CHECK-NEXT:    ldp q7, q6, [x1, #192]
+; CHECK-NEXT:    ldp q4, q3, [x1, #224]
+; CHECK-NEXT:    stp x2, x2, [sp]
+; CHECK-NEXT:    ldr q18, [sp]
+; CHECK-NEXT:    add z8.d, z8.d, z18.d
+; CHECK-NEXT:    tbnz w8, #0, .LBB45_34
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    add z31.d, z31.d, z18.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB45_35
+; CHECK-NEXT:  .LBB45_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB45_36
+; CHECK-NEXT:  .LBB45_3: // %else4
+; CHECK-NEXT:    add z30.d, z30.d, z18.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB45_37
+; CHECK-NEXT:  .LBB45_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB45_38
+; CHECK-NEXT:  .LBB45_5: // %else8
+; CHECK-NEXT:    add z28.d, z29.d, z18.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB45_39
+; CHECK-NEXT:  .LBB45_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB45_40
+; CHECK-NEXT:  .LBB45_7: // %else12
+; CHECK-NEXT:    add z27.d, z27.d, z18.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB45_41
+; CHECK-NEXT:  .LBB45_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB45_42
+; CHECK-NEXT:  .LBB45_9: // %else16
+; CHECK-NEXT:    add z25.d, z26.d, z18.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB45_43
+; CHECK-NEXT:  .LBB45_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB45_44
+; CHECK-NEXT:  .LBB45_11: // %else20
+; CHECK-NEXT:    add z24.d, z24.d, z18.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB45_45
+; CHECK-NEXT:  .LBB45_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB45_46
+; CHECK-NEXT:  .LBB45_13: // %else24
+; CHECK-NEXT:    add z20.d, z23.d, z18.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB45_47
+; CHECK-NEXT:  .LBB45_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB45_48
+; CHECK-NEXT:  .LBB45_15: // %else28
+; CHECK-NEXT:    add z22.d, z22.d, z18.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB45_49
+; CHECK-NEXT:  .LBB45_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB45_50
+; CHECK-NEXT:  .LBB45_17: // %else32
+; CHECK-NEXT:    add z16.d, z21.d, z18.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB45_51
+; CHECK-NEXT:  .LBB45_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB45_52
+; CHECK-NEXT:  .LBB45_19: // %else36
+; CHECK-NEXT:    add z19.d, z19.d, z18.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB45_53
+; CHECK-NEXT:  .LBB45_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB45_54
+; CHECK-NEXT:  .LBB45_21: // %else40
+; CHECK-NEXT:    add z5.d, z17.d, z18.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB45_55
+; CHECK-NEXT:  .LBB45_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB45_56
+; CHECK-NEXT:  .LBB45_23: // %else44
+; CHECK-NEXT:    add z7.d, z7.d, z18.d
+; CHECK-NEXT:    tbnz w8, #23, .LBB45_57
+; CHECK-NEXT:  .LBB45_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB45_58
+; CHECK-NEXT:  .LBB45_25: // %else48
+; CHECK-NEXT:    add z2.d, z6.d, z18.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB45_59
+; CHECK-NEXT:  .LBB45_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB45_60
+; CHECK-NEXT:  .LBB45_27: // %else52
+; CHECK-NEXT:    add z4.d, z4.d, z18.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB45_61
+; CHECK-NEXT:  .LBB45_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB45_62
+; CHECK-NEXT:  .LBB45_29: // %else56
+; CHECK-NEXT:    add z1.d, z3.d, z18.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB45_63
+; CHECK-NEXT:  .LBB45_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB45_64
+; CHECK-NEXT:  .LBB45_31: // %else60
+; CHECK-NEXT:    tbz w8, #31, .LBB45_33
+; CHECK-NEXT:  .LBB45_32: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:  .LBB45_33: // %else62
+; CHECK-NEXT:    ldp d9, d8, [sp, #64] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d11, d10, [sp, #48] // 16-byte Folded Reload
+; CHECK-NEXT:    ldp d13, d12, [sp, #32] // 16-byte Folded Reload
+; CHECK-NEXT:    ldr d14, [sp, #16] // 8-byte Folded Reload
+; CHECK-NEXT:    add sp, sp, #80
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB45_34: // %cond.store
+; CHECK-NEXT:    fmov x9, d8
+; CHECK-NEXT:    str s28, [x9]
+; CHECK-NEXT:    add z31.d, z31.d, z18.d
+; CHECK-NEXT:    tbz w8, #1, .LBB45_2
+; CHECK-NEXT:  .LBB45_35: // %cond.store1
+; CHECK-NEXT:    mov z8.d, z8.d[1]
+; CHECK-NEXT:    fmov x9, d8
+; CHECK-NEXT:    mov z8.s, z28.s[1]
+; CHECK-NEXT:    str s8, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB45_3
+; CHECK-NEXT:  .LBB45_36: // %cond.store3
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    mov z8.s, z28.s[2]
+; CHECK-NEXT:    str s8, [x9]
+; CHECK-NEXT:    add z30.d, z30.d, z18.d
+; CHECK-NEXT:    tbz w8, #3, .LBB45_4
+; CHECK-NEXT:  .LBB45_37: // %cond.store5
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    mov z28.s, z28.s[3]
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    str s28, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB45_5
+; CHECK-NEXT:  .LBB45_38: // %cond.store7
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    str s25, [x9]
+; CHECK-NEXT:    add z28.d, z29.d, z18.d
+; CHECK-NEXT:    tbz w8, #5, .LBB45_6
+; CHECK-NEXT:  .LBB45_39: // %cond.store9
+; CHECK-NEXT:    mov z29.d, z30.d[1]
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    mov z29.s, z25.s[1]
+; CHECK-NEXT:    str s29, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB45_7
+; CHECK-NEXT:  .LBB45_40: // %cond.store11
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    mov z29.s, z25.s[2]
+; CHECK-NEXT:    str s29, [x9]
+; CHECK-NEXT:    add z27.d, z27.d, z18.d
+; CHECK-NEXT:    tbz w8, #7, .LBB45_8
+; CHECK-NEXT:  .LBB45_41: // %cond.store13
+; CHECK-NEXT:    mov z28.d, z28.d[1]
+; CHECK-NEXT:    mov z25.s, z25.s[3]
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    str s25, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB45_9
+; CHECK-NEXT:  .LBB45_42: // %cond.store15
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:    add z25.d, z26.d, z18.d
+; CHECK-NEXT:    tbz w8, #9, .LBB45_10
+; CHECK-NEXT:  .LBB45_43: // %cond.store17
+; CHECK-NEXT:    mov z26.d, z27.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    mov z26.s, z20.s[1]
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB45_11
+; CHECK-NEXT:  .LBB45_44: // %cond.store19
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    mov z26.s, z20.s[2]
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:    add z24.d, z24.d, z18.d
+; CHECK-NEXT:    tbz w8, #11, .LBB45_12
+; CHECK-NEXT:  .LBB45_45: // %cond.store21
+; CHECK-NEXT:    mov z25.d, z25.d[1]
+; CHECK-NEXT:    mov z20.s, z20.s[3]
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB45_13
+; CHECK-NEXT:  .LBB45_46: // %cond.store23
+; CHECK-NEXT:    fmov x9, d24
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    add z20.d, z23.d, z18.d
+; CHECK-NEXT:    tbz w8, #13, .LBB45_14
+; CHECK-NEXT:  .LBB45_47: // %cond.store25
+; CHECK-NEXT:    mov z23.d, z24.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    mov z23.s, z16.s[1]
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB45_15
+; CHECK-NEXT:  .LBB45_48: // %cond.store27
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z23.s, z16.s[2]
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:    add z22.d, z22.d, z18.d
+; CHECK-NEXT:    tbz w8, #15, .LBB45_16
+; CHECK-NEXT:  .LBB45_49: // %cond.store29
+; CHECK-NEXT:    mov z20.d, z20.d[1]
+; CHECK-NEXT:    mov z16.s, z16.s[3]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    tbz w8, #16, .LBB45_17
+; CHECK-NEXT:  .LBB45_50: // %cond.store31
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    add z16.d, z21.d, z18.d
+; CHECK-NEXT:    tbz w8, #17, .LBB45_18
+; CHECK-NEXT:  .LBB45_51: // %cond.store33
+; CHECK-NEXT:    mov z20.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d20
+; CHECK-NEXT:    mov z20.s, z5.s[1]
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:    tbz w8, #18, .LBB45_19
+; CHECK-NEXT:  .LBB45_52: // %cond.store35
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z20.s, z5.s[2]
+; CHECK-NEXT:    str s20, [x9]
+; CHECK-NEXT:    add z19.d, z19.d, z18.d
+; CHECK-NEXT:    tbz w8, #19, .LBB45_20
+; CHECK-NEXT:  .LBB45_53: // %cond.store37
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    mov z5.s, z5.s[3]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #20, .LBB45_21
+; CHECK-NEXT:  .LBB45_54: // %cond.store39
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    add z5.d, z17.d, z18.d
+; CHECK-NEXT:    tbz w8, #21, .LBB45_22
+; CHECK-NEXT:  .LBB45_55: // %cond.store41
+; CHECK-NEXT:    mov z16.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.s, z2.s[1]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    tbz w8, #22, .LBB45_23
+; CHECK-NEXT:  .LBB45_56: // %cond.store43
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z16.s, z2.s[2]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    add z7.d, z7.d, z18.d
+; CHECK-NEXT:    tbz w8, #23, .LBB45_24
+; CHECK-NEXT:  .LBB45_57: // %cond.store45
+; CHECK-NEXT:    mov z5.d, z5.d[1]
+; CHECK-NEXT:    mov z2.s, z2.s[3]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #24, .LBB45_25
+; CHECK-NEXT:  .LBB45_58: // %cond.store47
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    add z2.d, z6.d, z18.d
+; CHECK-NEXT:    tbz w8, #25, .LBB45_26
+; CHECK-NEXT:  .LBB45_59: // %cond.store49
+; CHECK-NEXT:    mov z5.d, z7.d[1]
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    mov z5.s, z1.s[1]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    tbz w8, #26, .LBB45_27
+; CHECK-NEXT:  .LBB45_60: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z5.s, z1.s[2]
+; CHECK-NEXT:    str s5, [x9]
+; CHECK-NEXT:    add z4.d, z4.d, z18.d
+; CHECK-NEXT:    tbz w8, #27, .LBB45_28
+; CHECK-NEXT:  .LBB45_61: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #28, .LBB45_29
+; CHECK-NEXT:  .LBB45_62: // %cond.store55
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    add z1.d, z3.d, z18.d
+; CHECK-NEXT:    tbz w8, #29, .LBB45_30
+; CHECK-NEXT:  .LBB45_63: // %cond.store57
+; CHECK-NEXT:    mov z2.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB45_31
+; CHECK-NEXT:  .LBB45_64: // %cond.store59
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB45_32
+; CHECK-NEXT:    b .LBB45_33
   %vals = load <32 x float>, ptr %a
   %bases = load <32 x ptr>, ptr %b
   %byte_ptrs = getelementptr i8, <32 x ptr> %bases, i64 %off
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 define void @masked_scatter_vec_plus_imm(ptr %a, ptr %b) #0 {
 ; CHECK-LABEL: masked_scatter_vec_plus_imm:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.s, vl32
-; CHECK-NEXT:    ptrue p1.d, vl32
-; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0]
-; CHECK-NEXT:    ld1d { z1.d }, p1/z, [x1]
-; CHECK-NEXT:    fcmeq p0.s, p0/z, z0.s, #0.0
-; CHECK-NEXT:    uunpklo z0.d, z0.s
-; CHECK-NEXT:    punpklo p0.h, p0.b
-; CHECK-NEXT:    st1w { z0.d }, p0, [z1.d, #4]
-; CHECK-NEXT:    ret
+; CHECK-NEXT:    sub sp, sp, #32
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    stp d9, d8, [sp, #16] // 16-byte Folded Spill
+; CHECK-NEXT:    .cfi_offset b8, -8
+; CHECK-NEXT:    .cfi_offset b9, -16
+; CHECK-NEXT:    adrp x8, .LCPI46_1
+; CHECK-NEXT:    ptrue p0.s, vl4
+; CHECK-NEXT:    ldp q4, q2, [x0, #64]
+; CHECK-NEXT:    ldr q5, [x8, :lo12:.LCPI46_1]
+; CHECK-NEXT:    ldp q1, q0, [x0, #96]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z4.s, z5.s
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z2.s, z5.s
+; CHECK-NEXT:    fmov w8, s3
+; CHECK-NEXT:    mov z7.s, z3.s[1]
+; CHECK-NEXT:    mov z16.s, z3.s[2]
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    mov z17.s, z3.s[3]
+; CHECK-NEXT:    fmov w10, s16
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    lsl w8, w8, #16
+; CHECK-NEXT:    mov z6.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w8, w9, #17, #1
+; CHECK-NEXT:    mov z7.s, z6.s[2]
+; CHECK-NEXT:    bfi w8, w10, #18, #1
+; CHECK-NEXT:    mov z3.s, z6.s[1]
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    bfi w8, w11, #19, #1
+; CHECK-NEXT:    fmov w11, s7
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    mov z17.s, z6.s[3]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z1.s, z5.s
+; CHECK-NEXT:    bfi w8, w9, #20, #1
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w11, #0x1
+; CHECK-NEXT:    bfi w8, w10, #21, #1
+; CHECK-NEXT:    fmov w10, s3
+; CHECK-NEXT:    mov z6.s, z3.s[1]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #22
+; CHECK-NEXT:    fmov w9, s17
+; CHECK-NEXT:    mov z7.s, z3.s[2]
+; CHECK-NEXT:    mov z17.s, z3.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z0.s, z5.s
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    mov z3.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    ldp q27, q23, [x0]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #23
+; CHECK-NEXT:    fmov w9, s6
+; CHECK-NEXT:    orr w8, w8, w10, lsl #24
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    mov z6.s, z3.s[1]
+; CHECK-NEXT:    fmov w11, s3
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w12, s6
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z7.s, z3.s[2]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #25
+; CHECK-NEXT:    adrp x9, .LCPI46_0
+; CHECK-NEXT:    orr w8, w8, w10, lsl #26
+; CHECK-NEXT:    fmov w10, s17
+; CHECK-NEXT:    mov z17.s, z3.s[3]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    ldr q3, [x9, :lo12:.LCPI46_0]
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z27.s, z5.s
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    fmov w9, s7
+; CHECK-NEXT:    mov z6.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z23.s, z5.s
+; CHECK-NEXT:    orr w8, w8, w10, lsl #27
+; CHECK-NEXT:    and w10, w12, #0x1
+; CHECK-NEXT:    orr w8, w8, w11, lsl #28
+; CHECK-NEXT:    mov z7.s, z6.s[1]
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    orr w8, w8, w10, lsl #29
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    mov z7.s, z6.s[2]
+; CHECK-NEXT:    mov z18.s, z6.s[3]
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    ldp q19, q16, [x0, #32]
+; CHECK-NEXT:    orr w8, w8, w9, lsl #30
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    mov z6.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    bfi w11, w10, #1, #1
+; CHECK-NEXT:    fmov w10, s6
+; CHECK-NEXT:    bfi w11, w12, #2, #1
+; CHECK-NEXT:    mov z7.s, z6.s[1]
+; CHECK-NEXT:    mov z18.s, z6.s[2]
+; CHECK-NEXT:    mov z20.s, z6.s[3]
+; CHECK-NEXT:    fmov w12, s7
+; CHECK-NEXT:    bfi w11, w9, #3, #1
+; CHECK-NEXT:    fmov w9, s18
+; CHECK-NEXT:    bfi w11, w10, #4, #1
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    fcmeq p1.s, p0/z, z19.s, z5.s
+; CHECK-NEXT:    bfi w11, w12, #5, #1
+; CHECK-NEXT:    mov z6.s, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    fmov w12, s6
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z7.s, z6.s[1]
+; CHECK-NEXT:    orr w9, w11, w9, lsl #6
+; CHECK-NEXT:    orr w9, w9, w10, lsl #7
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    and w11, w12, #0x1
+; CHECK-NEXT:    mov z18.s, z6.s[2]
+; CHECK-NEXT:    mov z20.s, z6.s[3]
+; CHECK-NEXT:    fcmeq p0.s, p0/z, z16.s, z5.s
+; CHECK-NEXT:    orr w9, w9, w11, lsl #8
+; CHECK-NEXT:    fmov w11, s18
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    mov z5.s, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.s, z5.s[1]
+; CHECK-NEXT:    mov z7.s, z5.s[2]
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    orr w9, w9, w10, lsl #9
+; CHECK-NEXT:    fmov w10, s20
+; CHECK-NEXT:    fmov w12, s17
+; CHECK-NEXT:    orr w9, w9, w11, lsl #10
+; CHECK-NEXT:    fmov w11, s5
+; CHECK-NEXT:    mov z17.s, z5.s[3]
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w8, w8, w12, lsl #31
+; CHECK-NEXT:    ldp q31, q30, [x1]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #11
+; CHECK-NEXT:    and w10, w11, #0x1
+; CHECK-NEXT:    fmov w11, s6
+; CHECK-NEXT:    orr w9, w9, w10, lsl #12
+; CHECK-NEXT:    fmov w10, s7
+; CHECK-NEXT:    add z9.d, z31.d, z3.d
+; CHECK-NEXT:    and w11, w11, #0x1
+; CHECK-NEXT:    and w10, w10, #0x1
+; CHECK-NEXT:    orr w9, w9, w11, lsl #13
+; CHECK-NEXT:    fmov w11, s17
+; CHECK-NEXT:    ldp q8, q29, [x1, #32]
+; CHECK-NEXT:    orr w9, w9, w10, lsl #14
+; CHECK-NEXT:    orr w9, w9, w11, lsl #15
+; CHECK-NEXT:    and w9, w9, #0xffff
+; CHECK-NEXT:    orr w8, w9, w8
+; CHECK-NEXT:    ldp q28, q26, [x1, #64]
+; CHECK-NEXT:    ldp q25, q24, [x1, #96]
+; CHECK-NEXT:    ldp q22, q21, [x1, #128]
+; CHECK-NEXT:    ldp q20, q18, [x1, #160]
+; CHECK-NEXT:    ldp q17, q7, [x1, #192]
+; CHECK-NEXT:    ldp q6, q5, [x1, #224]
+; CHECK-NEXT:    tbnz w8, #0, .LBB46_34
+; CHECK-NEXT:  // %bb.1: // %else
+; CHECK-NEXT:    add z31.d, z30.d, z3.d
+; CHECK-NEXT:    tbnz w8, #1, .LBB46_35
+; CHECK-NEXT:  .LBB46_2: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB46_36
+; CHECK-NEXT:  .LBB46_3: // %else4
+; CHECK-NEXT:    add z30.d, z8.d, z3.d
+; CHECK-NEXT:    tbnz w8, #3, .LBB46_37
+; CHECK-NEXT:  .LBB46_4: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB46_38
+; CHECK-NEXT:  .LBB46_5: // %else8
+; CHECK-NEXT:    add z27.d, z29.d, z3.d
+; CHECK-NEXT:    tbnz w8, #5, .LBB46_39
+; CHECK-NEXT:  .LBB46_6: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB46_40
+; CHECK-NEXT:  .LBB46_7: // %else12
+; CHECK-NEXT:    add z28.d, z28.d, z3.d
+; CHECK-NEXT:    tbnz w8, #7, .LBB46_41
+; CHECK-NEXT:  .LBB46_8: // %else14
+; CHECK-NEXT:    tbnz w8, #8, .LBB46_42
+; CHECK-NEXT:  .LBB46_9: // %else16
+; CHECK-NEXT:    add z23.d, z26.d, z3.d
+; CHECK-NEXT:    tbnz w8, #9, .LBB46_43
+; CHECK-NEXT:  .LBB46_10: // %else18
+; CHECK-NEXT:    tbnz w8, #10, .LBB46_44
+; CHECK-NEXT:  .LBB46_11: // %else20
+; CHECK-NEXT:    add z25.d, z25.d, z3.d
+; CHECK-NEXT:    tbnz w8, #11, .LBB46_45
+; CHECK-NEXT:  .LBB46_12: // %else22
+; CHECK-NEXT:    tbnz w8, #12, .LBB46_46
+; CHECK-NEXT:  .LBB46_13: // %else24
+; CHECK-NEXT:    add z19.d, z24.d, z3.d
+; CHECK-NEXT:    tbnz w8, #13, .LBB46_47
+; CHECK-NEXT:  .LBB46_14: // %else26
+; CHECK-NEXT:    tbnz w8, #14, .LBB46_48
+; CHECK-NEXT:  .LBB46_15: // %else28
+; CHECK-NEXT:    add z22.d, z22.d, z3.d
+; CHECK-NEXT:    tbnz w8, #15, .LBB46_49
+; CHECK-NEXT:  .LBB46_16: // %else30
+; CHECK-NEXT:    tbnz w8, #16, .LBB46_50
+; CHECK-NEXT:  .LBB46_17: // %else32
+; CHECK-NEXT:    add z16.d, z21.d, z3.d
+; CHECK-NEXT:    tbnz w8, #17, .LBB46_51
+; CHECK-NEXT:  .LBB46_18: // %else34
+; CHECK-NEXT:    tbnz w8, #18, .LBB46_52
+; CHECK-NEXT:  .LBB46_19: // %else36
+; CHECK-NEXT:    add z19.d, z20.d, z3.d
+; CHECK-NEXT:    tbnz w8, #19, .LBB46_53
+; CHECK-NEXT:  .LBB46_20: // %else38
+; CHECK-NEXT:    tbnz w8, #20, .LBB46_54
+; CHECK-NEXT:  .LBB46_21: // %else40
+; CHECK-NEXT:    add z4.d, z18.d, z3.d
+; CHECK-NEXT:    tbnz w8, #21, .LBB46_55
+; CHECK-NEXT:  .LBB46_22: // %else42
+; CHECK-NEXT:    tbnz w8, #22, .LBB46_56
+; CHECK-NEXT:  .LBB46_23: // %else44
+; CHECK-NEXT:    add z16.d, z17.d, z3.d
+; CHECK-NEXT:    tbnz w8, #23, .LBB46_57
+; CHECK-NEXT:  .LBB46_24: // %else46
+; CHECK-NEXT:    tbnz w8, #24, .LBB46_58
+; CHECK-NEXT:  .LBB46_25: // %else48
+; CHECK-NEXT:    add z2.d, z7.d, z3.d
+; CHECK-NEXT:    tbnz w8, #25, .LBB46_59
+; CHECK-NEXT:  .LBB46_26: // %else50
+; CHECK-NEXT:    tbnz w8, #26, .LBB46_60
+; CHECK-NEXT:  .LBB46_27: // %else52
+; CHECK-NEXT:    add z4.d, z6.d, z3.d
+; CHECK-NEXT:    tbnz w8, #27, .LBB46_61
+; CHECK-NEXT:  .LBB46_28: // %else54
+; CHECK-NEXT:    tbnz w8, #28, .LBB46_62
+; CHECK-NEXT:  .LBB46_29: // %else56
+; CHECK-NEXT:    add z1.d, z5.d, z3.d
+; CHECK-NEXT:    tbnz w8, #29, .LBB46_63
+; CHECK-NEXT:  .LBB46_30: // %else58
+; CHECK-NEXT:    tbnz w8, #30, .LBB46_64
+; CHECK-NEXT:  .LBB46_31: // %else60
+; CHECK-NEXT:    tbz w8, #31, .LBB46_33
+; CHECK-NEXT:  .LBB46_32: // %cond.store61
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    mov z0.s, z0.s[3]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str s0, [x8]
+; CHECK-NEXT:  .LBB46_33: // %else62
+; CHECK-NEXT:    ldp d9, d8, [sp, #16] // 16-byte Folded Reload
+; CHECK-NEXT:    add sp, sp, #32
+; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB46_34: // %cond.store
+; CHECK-NEXT:    fmov x9, d9
+; CHECK-NEXT:    str s27, [x9]
+; CHECK-NEXT:    add z31.d, z30.d, z3.d
+; CHECK-NEXT:    tbz w8, #1, .LBB46_2
+; CHECK-NEXT:  .LBB46_35: // %cond.store1
+; CHECK-NEXT:    mov z30.d, z9.d[1]
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    mov z30.s, z27.s[1]
+; CHECK-NEXT:    str s30, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB46_3
+; CHECK-NEXT:  .LBB46_36: // %cond.store3
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    mov z30.s, z27.s[2]
+; CHECK-NEXT:    str s30, [x9]
+; CHECK-NEXT:    add z30.d, z8.d, z3.d
+; CHECK-NEXT:    tbz w8, #3, .LBB46_4
+; CHECK-NEXT:  .LBB46_37: // %cond.store5
+; CHECK-NEXT:    mov z31.d, z31.d[1]
+; CHECK-NEXT:    mov z27.s, z27.s[3]
+; CHECK-NEXT:    fmov x9, d31
+; CHECK-NEXT:    str s27, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB46_5
+; CHECK-NEXT:  .LBB46_38: // %cond.store7
+; CHECK-NEXT:    fmov x9, d30
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:    add z27.d, z29.d, z3.d
+; CHECK-NEXT:    tbz w8, #5, .LBB46_6
+; CHECK-NEXT:  .LBB46_39: // %cond.store9
+; CHECK-NEXT:    mov z29.d, z30.d[1]
+; CHECK-NEXT:    fmov x9, d29
+; CHECK-NEXT:    mov z29.s, z23.s[1]
+; CHECK-NEXT:    str s29, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB46_7
+; CHECK-NEXT:  .LBB46_40: // %cond.store11
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    mov z29.s, z23.s[2]
+; CHECK-NEXT:    str s29, [x9]
+; CHECK-NEXT:    add z28.d, z28.d, z3.d
+; CHECK-NEXT:    tbz w8, #7, .LBB46_8
+; CHECK-NEXT:  .LBB46_41: // %cond.store13
+; CHECK-NEXT:    mov z27.d, z27.d[1]
+; CHECK-NEXT:    mov z23.s, z23.s[3]
+; CHECK-NEXT:    fmov x9, d27
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:    tbz w8, #8, .LBB46_9
+; CHECK-NEXT:  .LBB46_42: // %cond.store15
+; CHECK-NEXT:    fmov x9, d28
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    add z23.d, z26.d, z3.d
+; CHECK-NEXT:    tbz w8, #9, .LBB46_10
+; CHECK-NEXT:  .LBB46_43: // %cond.store17
+; CHECK-NEXT:    mov z26.d, z28.d[1]
+; CHECK-NEXT:    fmov x9, d26
+; CHECK-NEXT:    mov z26.s, z19.s[1]
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:    tbz w8, #10, .LBB46_11
+; CHECK-NEXT:  .LBB46_44: // %cond.store19
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    mov z26.s, z19.s[2]
+; CHECK-NEXT:    str s26, [x9]
+; CHECK-NEXT:    add z25.d, z25.d, z3.d
+; CHECK-NEXT:    tbz w8, #11, .LBB46_12
+; CHECK-NEXT:  .LBB46_45: // %cond.store21
+; CHECK-NEXT:    mov z23.d, z23.d[1]
+; CHECK-NEXT:    mov z19.s, z19.s[3]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    tbz w8, #12, .LBB46_13
+; CHECK-NEXT:  .LBB46_46: // %cond.store23
+; CHECK-NEXT:    fmov x9, d25
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    add z19.d, z24.d, z3.d
+; CHECK-NEXT:    tbz w8, #13, .LBB46_14
+; CHECK-NEXT:  .LBB46_47: // %cond.store25
+; CHECK-NEXT:    mov z23.d, z25.d[1]
+; CHECK-NEXT:    fmov x9, d23
+; CHECK-NEXT:    mov z23.s, z16.s[1]
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:    tbz w8, #14, .LBB46_15
+; CHECK-NEXT:  .LBB46_48: // %cond.store27
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z23.s, z16.s[2]
+; CHECK-NEXT:    str s23, [x9]
+; CHECK-NEXT:    add z22.d, z22.d, z3.d
+; CHECK-NEXT:    tbz w8, #15, .LBB46_16
+; CHECK-NEXT:  .LBB46_49: // %cond.store29
+; CHECK-NEXT:    mov z19.d, z19.d[1]
+; CHECK-NEXT:    mov z16.s, z16.s[3]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    tbz w8, #16, .LBB46_17
+; CHECK-NEXT:  .LBB46_50: // %cond.store31
+; CHECK-NEXT:    fmov x9, d22
+; CHECK-NEXT:    str s4, [x9]
+; CHECK-NEXT:    add z16.d, z21.d, z3.d
+; CHECK-NEXT:    tbz w8, #17, .LBB46_18
+; CHECK-NEXT:  .LBB46_51: // %cond.store33
+; CHECK-NEXT:    mov z19.d, z22.d[1]
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    mov z19.s, z4.s[1]
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    tbz w8, #18, .LBB46_19
+; CHECK-NEXT:  .LBB46_52: // %cond.store35
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z19.s, z4.s[2]
+; CHECK-NEXT:    str s19, [x9]
+; CHECK-NEXT:    add z19.d, z20.d, z3.d
+; CHECK-NEXT:    tbz w8, #19, .LBB46_20
+; CHECK-NEXT:  .LBB46_53: // %cond.store37
+; CHECK-NEXT:    mov z16.d, z16.d[1]
+; CHECK-NEXT:    mov z4.s, z4.s[3]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str s4, [x9]
+; CHECK-NEXT:    tbz w8, #20, .LBB46_21
+; CHECK-NEXT:  .LBB46_54: // %cond.store39
+; CHECK-NEXT:    fmov x9, d19
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    add z4.d, z18.d, z3.d
+; CHECK-NEXT:    tbz w8, #21, .LBB46_22
+; CHECK-NEXT:  .LBB46_55: // %cond.store41
+; CHECK-NEXT:    mov z16.d, z19.d[1]
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    mov z16.s, z2.s[1]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    tbz w8, #22, .LBB46_23
+; CHECK-NEXT:  .LBB46_56: // %cond.store43
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z16.s, z2.s[2]
+; CHECK-NEXT:    str s16, [x9]
+; CHECK-NEXT:    add z16.d, z17.d, z3.d
+; CHECK-NEXT:    tbz w8, #23, .LBB46_24
+; CHECK-NEXT:  .LBB46_57: // %cond.store45
+; CHECK-NEXT:    mov z4.d, z4.d[1]
+; CHECK-NEXT:    mov z2.s, z2.s[3]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #24, .LBB46_25
+; CHECK-NEXT:  .LBB46_58: // %cond.store47
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    add z2.d, z7.d, z3.d
+; CHECK-NEXT:    tbz w8, #25, .LBB46_26
+; CHECK-NEXT:  .LBB46_59: // %cond.store49
+; CHECK-NEXT:    mov z4.d, z16.d[1]
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    mov z4.s, z1.s[1]
+; CHECK-NEXT:    str s4, [x9]
+; CHECK-NEXT:    tbz w8, #26, .LBB46_27
+; CHECK-NEXT:  .LBB46_60: // %cond.store51
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z4.s, z1.s[2]
+; CHECK-NEXT:    str s4, [x9]
+; CHECK-NEXT:    add z4.d, z6.d, z3.d
+; CHECK-NEXT:    tbz w8, #27, .LBB46_28
+; CHECK-NEXT:  .LBB46_61: // %cond.store53
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    mov z1.s, z1.s[3]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str s1, [x9]
+; CHECK-NEXT:    tbz w8, #28, .LBB46_29
+; CHECK-NEXT:  .LBB46_62: // %cond.store55
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str s0, [x9]
+; CHECK-NEXT:    add z1.d, z5.d, z3.d
+; CHECK-NEXT:    tbz w8, #29, .LBB46_30
+; CHECK-NEXT:  .LBB46_63: // %cond.store57
+; CHECK-NEXT:    mov z2.d, z4.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    mov z2.s, z0.s[1]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbz w8, #30, .LBB46_31
+; CHECK-NEXT:  .LBB46_64: // %cond.store59
+; CHECK-NEXT:    fmov x9, d1
+; CHECK-NEXT:    mov z2.s, z0.s[2]
+; CHECK-NEXT:    str s2, [x9]
+; CHECK-NEXT:    tbnz w8, #31, .LBB46_32
+; CHECK-NEXT:    b .LBB46_33
   %vals = load <32 x float>, ptr %a
   %bases = load <32 x ptr>, ptr %b
   %byte_ptrs = getelementptr i8, <32 x ptr> %bases, i64 4
   %ptrs = bitcast <32 x ptr> %byte_ptrs to <32 x ptr>
   %mask = fcmp oeq <32 x float> %vals, zeroinitializer
   call void @llvm.masked.scatter.v32f32(<32 x float> %vals, <32 x ptr> %ptrs, i32 8, <32 x i1> %mask)
   ret void
 }
 
 ; extract_subvec(...(insert_subvec(a,b,c))) -> extract_subvec(bitcast(b),d) like
 ; combines can effectively unlegalise bitcast operations. This test ensures such
 ; combines do not happen after operation legalisation. When not prevented the
 ; test triggers infinite combine->legalise->combine->...
 ;
 ; NOTE: For this test to function correctly it's critical for %vals to be in a
 ; different block to the scatter store.  If not, the problematic bitcast will be
 ; removed before operation legalisation and thus not exercise the combine.
 define void @masked_scatter_bitcast_infinite_loop(ptr %a, ptr %b, i1 %cond) #0 {
 ; CHECK-LABEL: masked_scatter_bitcast_infinite_loop:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    ptrue p0.d, vl8
-; CHECK-NEXT:    ld1d { z0.d }, p0/z, [x0]
-; CHECK-NEXT:    tbz w2, #0, .LBB47_2
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    ldr q0, [x0, #48]
+; CHECK-NEXT:    ldr q1, [x0, #32]
+; CHECK-NEXT:    ldr q2, [x0, #16]
+; CHECK-NEXT:    ldr q3, [x0]
+; CHECK-NEXT:    tbz w2, #0, .LBB47_10
 ; CHECK-NEXT:  // %bb.1: // %bb.1
-; CHECK-NEXT:    ld1d { z1.d }, p0/z, [x1]
-; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, #0.0
-; CHECK-NEXT:    st1d { z0.d }, p0, [z1.d]
-; CHECK-NEXT:  .LBB47_2: // %bb.2
+; CHECK-NEXT:    adrp x8, .LCPI47_0
+; CHECK-NEXT:    ptrue p0.d, vl2
+; CHECK-NEXT:    ldr q4, [x8, :lo12:.LCPI47_0]
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z3.d, z4.d
+; CHECK-NEXT:    fcmeq p2.d, p0/z, z2.d, z4.d
+; CHECK-NEXT:    mov z5.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.d, p2/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z16.d, z5.d[1]
+; CHECK-NEXT:    fmov x8, d5
+; CHECK-NEXT:    fmov x9, d16
+; CHECK-NEXT:    fcmeq p1.d, p0/z, z1.d, z4.d
+; CHECK-NEXT:    mov z5.d, z6.d[1]
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    fcmeq p0.d, p0/z, z0.d, z4.d
+; CHECK-NEXT:    mov z4.d, p1/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    and w8, w8, #0x1
+; CHECK-NEXT:    fmov x11, d5
+; CHECK-NEXT:    bfi w8, w9, #1, #1
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    bfi w8, w10, #2, #1
+; CHECK-NEXT:    mov z7.d, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT:    mov z6.d, z4.d[1]
+; CHECK-NEXT:    bfi w8, w11, #3, #1
+; CHECK-NEXT:    fmov x10, d6
+; CHECK-NEXT:    bfi w8, w9, #4, #1
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    mov z4.d, z7.d[1]
+; CHECK-NEXT:    ldp q7, q6, [x1]
+; CHECK-NEXT:    bfi w8, w10, #5, #1
+; CHECK-NEXT:    fmov x10, d4
+; CHECK-NEXT:    and w9, w9, #0x1
+; CHECK-NEXT:    orr w8, w8, w9, lsl #6
+; CHECK-NEXT:    orr w9, w8, w10, lsl #7
+; CHECK-NEXT:    ldp q5, q4, [x1, #32]
+; CHECK-NEXT:    and w8, w9, #0xff
+; CHECK-NEXT:    tbnz w9, #0, .LBB47_11
+; CHECK-NEXT:  // %bb.2: // %else
+; CHECK-NEXT:    tbnz w8, #1, .LBB47_12
+; CHECK-NEXT:  .LBB47_3: // %else2
+; CHECK-NEXT:    tbnz w8, #2, .LBB47_13
+; CHECK-NEXT:  .LBB47_4: // %else4
+; CHECK-NEXT:    tbnz w8, #3, .LBB47_14
+; CHECK-NEXT:  .LBB47_5: // %else6
+; CHECK-NEXT:    tbnz w8, #4, .LBB47_15
+; CHECK-NEXT:  .LBB47_6: // %else8
+; CHECK-NEXT:    tbnz w8, #5, .LBB47_16
+; CHECK-NEXT:  .LBB47_7: // %else10
+; CHECK-NEXT:    tbnz w8, #6, .LBB47_17
+; CHECK-NEXT:  .LBB47_8: // %else12
+; CHECK-NEXT:    tbz w8, #7, .LBB47_10
+; CHECK-NEXT:  .LBB47_9: // %cond.store13
+; CHECK-NEXT:    mov z1.d, z4.d[1]
+; CHECK-NEXT:    mov z0.d, z0.d[1]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    str d0, [x8]
+; CHECK-NEXT:  .LBB47_10: // %bb.2
+; CHECK-NEXT:    add sp, sp, #16
 ; CHECK-NEXT:    ret
+; CHECK-NEXT:  .LBB47_11: // %cond.store
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str d3, [x9]
+; CHECK-NEXT:    tbz w8, #1, .LBB47_3
+; CHECK-NEXT:  .LBB47_12: // %cond.store1
+; CHECK-NEXT:    mov z7.d, z7.d[1]
+; CHECK-NEXT:    mov z3.d, z3.d[1]
+; CHECK-NEXT:    fmov x9, d7
+; CHECK-NEXT:    str d3, [x9]
+; CHECK-NEXT:    tbz w8, #2, .LBB47_4
+; CHECK-NEXT:  .LBB47_13: // %cond.store3
+; CHECK-NEXT:    fmov x9, d6
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:    tbz w8, #3, .LBB47_5
+; CHECK-NEXT:  .LBB47_14: // %cond.store5
+; CHECK-NEXT:    mov z3.d, z6.d[1]
+; CHECK-NEXT:    mov z2.d, z2.d[1]
+; CHECK-NEXT:    fmov x9, d3
+; CHECK-NEXT:    str d2, [x9]
+; CHECK-NEXT:    tbz w8, #4, .LBB47_6
+; CHECK-NEXT:  .LBB47_15: // %cond.store7
+; CHECK-NEXT:    fmov x9, d5
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #5, .LBB47_7
+; CHECK-NEXT:  .LBB47_16: // %cond.store9
+; CHECK-NEXT:    mov z2.d, z5.d[1]
+; CHECK-NEXT:    mov z1.d, z1.d[1]
+; CHECK-NEXT:    fmov x9, d2
+; CHECK-NEXT:    str d1, [x9]
+; CHECK-NEXT:    tbz w8, #6, .LBB47_8
+; CHECK-NEXT:  .LBB47_17: // %cond.store11
+; CHECK-NEXT:    fmov x9, d4
+; CHECK-NEXT:    str d0, [x9]
+; CHECK-NEXT:    tbnz w8, #7, .LBB47_9
+; CHECK-NEXT:    b .LBB47_10
   %vals = load volatile <8 x double>, ptr %a
   br i1 %cond, label %bb.1, label %bb.2
 
 bb.1:
   %ptrs = load <8 x ptr>, ptr %b
   %mask = fcmp oeq <8 x double> %vals, zeroinitializer
   call void @llvm.masked.scatter.v8f64(<8 x double> %vals, <8 x ptr> %ptrs, i32 8, <8 x i1> %mask)
   br label %bb.2