[AArch64][SVE] Fix bug in lowering of fixed-length integer vector divides

[joechrisellis]

Summary

The function AArch64TargetLowering::LowerFixedLengthVectorIntDivideToSVE
previously assumed the operands were full vectors, but this is not
always true. This function would produce bogus if the division operands
are not full vectors, resulting in miscompiles when dividing 8-bit or
16-bit vectors.

The fix is to perform an extend + div + truncate for non-full vectors,
instead of the usual unpacking and unzipping logic. This is an additive
change which reduces the non-full integer vector divisions to a pattern
recognised by the existing lowering logic.

For future reference, an example of code that would miscompile before
this patch is below:

1  int8_t foo(unsigned N, int8_t *a, int8_t *b, int8_t *c) {
2      int8_t result = 0;
3      for (int i = 0; i < N; ++i) {
4          result += (a[i] / b[i]) / c[i];
5      }
6      return result;
7  }

[joechrisellis]

Fix failing llvm/test/CodeGen/AArch64/sve-fixed-length-int-rem.ll test.

[joechrisellis]

Gentle ping. 🙂

[paulwalker-arm]

llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll

0–1

Given the updates I wondered if this is still necessary?

4

OCD perhaps but can you move this and VBITS_EQ_1024 to the end of the list so the GE parts remain aligned.

74

To be precise this should be QUARTER VECTOR OR SMALLER. Same goes for the other instances that follow.

251

What is different here from the previous tests? I would have thought HALF VECTOR would carry the same meaning.

llvm/test/CodeGen/AArch64/sve-fixed-length-int-rem.ll

0–1

As above. Is this still used? I imagine most of my div.ll comments will apply to this file.

[joechrisellis]

Address review comments.

• @paulwalker-arm:
  • more accurate comments describing the fullness of vectors.
  • more consistent use of labels.
  • move EQ labels to the end of the prefix list in invocations of FileCheck.
  • removed unused VBYTES for -aarch64-sve-vector-bits-min=128.

[paulwalker-arm]

llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll

0–1

Sorry I should have been clearer. I meant to say that I didn't think any of the RUN lines require setting VBYTES because with your change the predicate patterns should be fixed based on the VBITS prefix in use.

[joechrisellis]

Address review comments:

• @paulwalker-arm:
  • remove unused VBYTES variable.

[paulwalker-arm]

The non-power-of-two RUN lines should be extended to allow maximum testing (see comments) but otherwise looks good.

llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll

18

This is missing -check-prefixes=CHECK,VBITS_EQ_256. Actually this is a universal comment, in that all the RUN lines with vector-bits-min=512-896 should include VBITS_EQ_512, vector-bits-min=1024-1920 should include VBITS_EQ_1024.

llvm/test/CodeGen/AArch64/sve-fixed-length-int-rem.ll

18

Same missing VBITS_EQ_### comment as with div tests.

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

if (EltVT == MVT::i32 || EltVT == MVT::i64)
     return LowerToPredicatedOp(Op, DAG, PredOpcode, /*OverrideNEON=*/true);
 
   // Scalable vector i8/i16 DIV is not supported. Promote it to i32.
   EVT ContainerVT = getContainerForFixedLengthVector(DAG, VT);
   EVT HalfVT = VT.getHalfNumVectorElementsVT(*DAG.getContext());
   EVT FixedWidenedVT = HalfVT.widenIntegerVectorElementType(*DAG.getContext());
   EVT ScalableWidenedVT = getContainerForFixedLengthVector(DAG, FixedWidenedVT);
 
+  // If this is not a full vector, extend, div, and truncate it.
+  EVT WidenedVT = VT.widenIntegerVectorElementType(*DAG.getContext());
+  if (DAG.getTargetLoweringInfo().isTypeLegal(WidenedVT)) {
+    unsigned ExtendOpcode = Signed ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
+    SDValue Op0 = DAG.getNode(ExtendOpcode, dl, WidenedVT, Op.getOperand(0));
+    SDValue Op1 = DAG.getNode(ExtendOpcode, dl, WidenedVT, Op.getOperand(1));
+    SDValue Div = DAG.getNode(Op.getOpcode(), dl, WidenedVT, Op0, Op1);
+    return DAG.getNode(ISD::TRUNCATE, dl, VT, Div);
+  }
+
   // Convert the operands to scalable vectors.
   SDValue Op0 = convertToScalableVector(DAG, ContainerVT, Op.getOperand(0));
   SDValue Op1 = convertToScalableVector(DAG, ContainerVT, Op.getOperand(1));
 
   // Extend the scalable operands.
   unsigned UnpkLo = Signed ? AArch64ISD::SUNPKLO : AArch64ISD::UUNPKLO;
   unsigned UnpkHi = Signed ? AArch64ISD::SUNPKHI : AArch64ISD::UUNPKHI;
   SDValue Op0Lo = DAG.getNode(UnpkLo, dl, ScalableWidenedVT, Op0);

llvm/test/CodeGen/AArch64/sve-fixed-length-int-div.ll

-; RUN: llc -aarch64-sve-vector-bits-min=128  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=16 -check-prefix=NO_SVE
-; RUN: llc -aarch64-sve-vector-bits-min=256  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=32
+; RUN: llc -aarch64-sve-vector-bits-min=128  -asm-verbose=0 < %s | FileCheck %s -check-prefix=NO_SVE
+; RUN: llc -aarch64-sve-vector-bits-min=256  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=32 -check-prefixes=CHECK,VBITS_EQ_256
 ; RUN: llc -aarch64-sve-vector-bits-min=384  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=32
-; RUN: llc -aarch64-sve-vector-bits-min=512  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=512  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512,VBITS_EQ_512
 ; RUN: llc -aarch64-sve-vector-bits-min=640  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512
 ; RUN: llc -aarch64-sve-vector-bits-min=768  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512
 ; RUN: llc -aarch64-sve-vector-bits-min=896  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512
-; RUN: llc -aarch64-sve-vector-bits-min=1024 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1024 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024,VBITS_EQ_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1152 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1280 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1408 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1536 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1664 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1792 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1920 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
-; RUN: llc -aarch64-sve-vector-bits-min=2048 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=256 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024,VBITS_GE_2048
+; RUN: llc -aarch64-sve-vector-bits-min=2048 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=256 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024,VBITS_GE_2048,VBITS_EQ_2048
 
 target triple = "aarch64-unknown-linux-gnu"
 
 ; Don't use SVE when its registers are no bigger than NEON.
 ; NO_SVE-NOT: ptrue
 
 ;
 ; SDIV
 ;
 
 ; Vector vXi8 sdiv are not legal for NEON so use SVE when available.
 define <8 x i8> @sdiv_v8i8(<8 x i8> %op1, <8 x i8> %op2) #0 {
 ; CHECK-LABEL: sdiv_v8i8:
-; CHECK: sunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,2)]]
+; CHECK: ptrue [[PG0:p[0-9]+]].s, vl8
 ; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, z1.b
 ; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, z0.b
-; CHECK-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sdivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: sdivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, z0.h
 ; CHECK-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
 ; CHECK-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: sdiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: sdiv [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1 z0.b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK:    ret
+; CHECK-NEXT: sdiv [[DIV:z[0-9]+]].s, [[PG0]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[DIV]].h, [[DIV]].h
+; CHECK-NEXT: umov [[SCALAR:w[0-9]+]], [[VEC:v[0-9]+]].h[0]
+; CHECK-NEXT: fmov s0, [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[1]
+; CHECK-NEXT: mov [[FINAL:v[0-9]+]].b[1], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[2]
+; CHECK-NEXT: mov [[FINAL]].b[2], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[3]
+; CHECK-NEXT: mov [[FINAL]].b[3], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[4]
+; CHECK-NEXT: mov [[FINAL]].b[4], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[5]
+; CHECK-NEXT: mov [[FINAL]].b[5], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[6]
+; CHECK-NEXT: mov [[FINAL]].b[6], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[7]
+; CHECK-NEXT: mov [[FINAL]].b[7], [[SCALAR]]
+; CHECK: ret
   %res = sdiv <8 x i8> %op1, %op2
   ret <8 x i8> %res
 }
 
 define <16 x i8> @sdiv_v16i8(<16 x i8> %op1, <16 x i8> %op2) #0 {
 ; CHECK-LABEL: sdiv_v16i8:
-; CHECK: sunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,4)]]
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, z1.b
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, z0.b
-; CHECK-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sdivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: sdivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: sdiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: sdiv [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1 z0.b, [[RES_LO]].b, [[RES_HI]].b
+
+; HALF VECTOR:
+; VBITS_EQ_256: ptrue [[PG:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, z1.b
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, z0.b
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
+; VBITS_EQ_256-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_256-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[RES1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: uzp1 [[RES2:z[0-9]+]].b, [[RES1]].b, [[RES1]].b
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl16
+; VBITS_GE_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, z1.b
+; VBITS_GE_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, z0.b
+; VBITS_GE_512-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
+; VBITS_GE_512-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
+; VBITS_GE_512-NEXT: sdiv [[DIV:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[RES1:z[0-9]+]].h, [[DIV]].h, [[DIV]].h
+; VBITS_GE_512-NEXT: uzp1 [[RES2:z[0-9]+]].b, [[RES1]].b, [[RES1]].b
 ; CHECK: ret
   %res = sdiv <16 x i8> %op1, %op2
   ret <16 x i8> %res
 }
 
 define void @sdiv_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
 ; CHECK-LABEL: sdiv_v32i8:
-; CHECK: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,32)]]
-; CHECK-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; CHECK-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,8)]]
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; CHECK-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; CHECK-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; CHECK-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; CHECK-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; CHECK-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: sdiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; CHECK-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; CHECK-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_256: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_EQ_256-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_256-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_256-NEXT: ptrue [[PG2:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_256-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: sdiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_256-NEXT: sdiv [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_256-NEXT: st1b { [[UZP3:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_EQ_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_512-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_512-NEXT: st1b { [[UZP2:z[0-9]+]].b }, [[PG1]], [x0]
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_1024: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_1024-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_GE_1024-NEXT: st1b { [[UZP2:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <32 x i8>, <32 x i8>* %a
   %op2 = load <32 x i8>, <32 x i8>* %b
   %res = sdiv <32 x i8> %op1, %op2
   store <32 x i8> %res, <32 x i8>* %a
   ret void
 }
 
 define void @sdiv_v64i8(<64 x i8>* %a, <64 x i8>* %b) #0 {
 ; CHECK-LABEL: sdiv_v64i8:
-; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,64)]]
-; VBITS_GE_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,16)]]
-; VBITS_GE_512-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_512-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_512-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_512-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_512-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_512-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_512-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_512-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_512-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_512-NEXT: sdiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; VBITS_GE_512-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_512-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_512-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_512-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_512-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_EQ_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_512-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: sdiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_512-NEXT: sdiv [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_512-NEXT: st1b { [[UZP3:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_EQ_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_1024-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_1024-NEXT: st1b { [[UZP2:z[0-9]+]].b }, [[PG1]], [x0]
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_2048: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_2048-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[RES1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_2048-NEXT: uzp1 [[RES2:z[0-9]+]].b, [[RES1]].b, [[RES1]].b
+; VBITS_GE_2048-NEXT: st1b { [[RES2]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <64 x i8>, <64 x i8>* %a
   %op2 = load <64 x i8>, <64 x i8>* %b
   %res = sdiv <64 x i8> %op1, %op2
   store <64 x i8> %res, <64 x i8>* %a
   ret void
 }
 
 define void @sdiv_v128i8(<128 x i8>* %a, <128 x i8>* %b) #0 {
 ; CHECK-LABEL: sdiv_v128i8:
-; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,128)]]
-; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,32)]]
-; VBITS_GE_1024-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_1024-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_1024-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_1024-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_1024-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_1024-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_1024-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_1024-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_1024-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_1024-NEXT: sdiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; VBITS_GE_1024-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_1024-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_1024-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_1024-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_1024-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].b, vl128
+; VBITS_EQ_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_1024-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: sdiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_1024-NEXT: sdiv [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_1024-NEXT: st1b { [[UZP3:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_2048: ptrue [[PG1:p[0-9]+]].b, vl128
+; VBITS_EQ_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_2048-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_2048-NEXT: st1b { [[UZP2:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <128 x i8>, <128 x i8>* %a
   %op2 = load <128 x i8>, <128 x i8>* %b
   %res = sdiv <128 x i8> %op1, %op2
   store <128 x i8> %res, <128 x i8>* %a
   ret void
 }
 
 define void @sdiv_v256i8(<256 x i8>* %a, <256 x i8>* %b) #0 {
 ; CHECK-LABEL: sdiv_v256i8:
-; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
-; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
-; VBITS_GE_2048-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_2048-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_2048-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_2048-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_2048-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_2048-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_2048-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_2048-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_2048-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_2048-NEXT: sdiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; VBITS_GE_2048-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_2048-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_2048-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_2048-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_2048-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_2048: ptrue [[PG1:p[0-9]+]].b, vl256
+; VBITS_EQ_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_2048-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: sdiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_2048-NEXT: sdiv [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_2048-NEXT: st1b { [[UZP3:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <256 x i8>, <256 x i8>* %a
   %op2 = load <256 x i8>, <256 x i8>* %b
   %res = sdiv <256 x i8> %op1, %op2
   store <256 x i8> %res, <256 x i8>* %a
   ret void
 }
 
 ; Vector vXi16 sdiv are not legal for NEON so use SVE when available.
 define <4 x i16> @sdiv_v4i16(<4 x i16> %op1, <4 x i16> %op2) #0 {
 ; CHECK-LABEL: sdiv_v4i16:
-; CHECK: sunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),2)]]
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: ret
+; CHECK: sshll v1.4s, v1.4h, #0
+; CHECK-NEXT: sshll v0.4s, v0.4h, #0
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl4
+; CHECK-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP1:z[0-9]+]].s, [[OP2:z[0-9]+]].s
+; CHECK-NEXT: mov w8, v0.s[1]
+; CHECK-NEXT: mov w9, v0.s[2]
+; CHECK-NEXT: mov w10, v0.s[3]
+; CHECK-NEXT: mov v0.h[1], w8
+; CHECK-NEXT: mov v0.h[2], w9
+; CHECK-NEXT: mov v0.h[3], w10
+; CHECK: ret
   %res = sdiv <4 x i16> %op1, %op2
   ret <4 x i16> %res
 }
 
 define <8 x i16> @sdiv_v8i16(<8 x i16> %op1, <8 x i16> %op2) #0 {
 ; CHECK-LABEL: sdiv_v8i16:
-; CHECK: sunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),4)]]
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: ret
+; CHECK: ptrue [[PG1:p[0-9]+]].s, vl8
+; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; CHECK-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; CHECK: ret
   %res = sdiv <8 x i16> %op1, %op2
   ret <8 x i16> %res
 }
 
 define void @sdiv_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
 ; CHECK-LABEL: sdiv_v16i16:
-; CHECK: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),16)]]
-; CHECK-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; CHECK-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),8)]]
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; CHECK-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; CHECK-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; CHECK-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_256: ptrue [[PG1:p[0-9]+]].h, vl16
+; VBITS_EQ_256-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_256-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_256-NEXT: ptrue [[PG2:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_256-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_256-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_512: ptrue [[PG1:p[0-9]+]].h, vl16
+; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_GE_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_512-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_512-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <16 x i16>, <16 x i16>* %a
   %op2 = load <16 x i16>, <16 x i16>* %b
   %res = sdiv <16 x i16> %op1, %op2
   store <16 x i16> %res, <16 x i16>* %a
   ret void
 }
 
 define void @sdiv_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
 ; CHECK-LABEL: sdiv_v32i16:
-; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),32)]]
-; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),16)]]
-; VBITS_GE_512-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_512-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_512-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_512-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_512-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_512-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_512-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].h, vl32
+; VBITS_EQ_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_512-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_512-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_1024: ptrue [[PG1:p[0-9]+]].h, vl32
+; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_1024-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_1024-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <32 x i16>, <32 x i16>* %a
   %op2 = load <32 x i16>, <32 x i16>* %b
   %res = sdiv <32 x i16> %op1, %op2
   store <32 x i16> %res, <32 x i16>* %a
   ret void
 }
 
 define void @sdiv_v64i16(<64 x i16>* %a, <64 x i16>* %b) #0 {
 ; CHECK-LABEL: sdiv_v64i16:
-; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),64)]]
-; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),32)]]
-; VBITS_GE_1024-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_1024-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_1024-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_1024-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_1024-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_1024-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_1024-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].h, vl64
+; VBITS_EQ_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_1024-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_1024-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_2048: ptrue [[PG1:p[0-9]+]].h, vl64
+; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_2048-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_2048-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <64 x i16>, <64 x i16>* %a
   %op2 = load <64 x i16>, <64 x i16>* %b
   %res = sdiv <64 x i16> %op1, %op2
   store <64 x i16> %res, <64 x i16>* %a
   ret void
 }
 
 define void @sdiv_v128i16(<128 x i16>* %a, <128 x i16>* %b) #0 {
 ; CHECK-LABEL: sdiv_v128i16:
-; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),128)]]
-; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),64)]]
-; VBITS_GE_2048-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_2048-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_2048-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_2048-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_2048-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_2048-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_2048-NEXT: ret
+; VBITS_EQ_2048: ptrue [[PG1:p[0-9]+]].h, vl128
+; VBITS_EQ_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_2048-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_2048-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_2048-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <128 x i16>, <128 x i16>* %a
   %op2 = load <128 x i16>, <128 x i16>* %b
   %res = sdiv <128 x i16> %op1, %op2
   store <128 x i16> %res, <128 x i16>* %a
   ret void
 }
 
 ; Vector v2i32 sdiv are not legal for NEON so use SVE when available.

 
 ;
 ; UDIV
 ;
 
 ; Vector vXi8 udiv are not legal for NEON so use SVE when available.
 define <8 x i8> @udiv_v8i8(<8 x i8> %op1, <8 x i8> %op2) #0 {
 ; CHECK-LABEL: udiv_v8i8:
-; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,2)]]
+; CHECK: ptrue [[PG0:p[0-9]+]].s, vl8
 ; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, z1.b
 ; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, z0.b
-; CHECK-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: udivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: udivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, z0.h
 ; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
 ; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: udiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: udiv [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1 z0.b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK:    ret
+; CHECK-NEXT: udiv [[DIV:z[0-9]+]].s, [[PG0]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[DIV]].h, [[DIV]].h
+; CHECK-NEXT: umov [[SCALAR:w[0-9]+]], [[VEC:v[0-9]+]].h[0]
+; CHECK-NEXT: fmov s0, [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[1]
+; CHECK-NEXT: mov [[FINAL:v[0-9]+]].b[1], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[2]
+; CHECK-NEXT: mov [[FINAL]].b[2], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[3]
+; CHECK-NEXT: mov [[FINAL]].b[3], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[4]
+; CHECK-NEXT: mov [[FINAL]].b[4], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[5]
+; CHECK-NEXT: mov [[FINAL]].b[5], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[6]
+; CHECK-NEXT: mov [[FINAL]].b[6], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[7]
+; CHECK-NEXT: mov [[FINAL]].b[7], [[SCALAR]]
+; CHECK: ret
   %res = udiv <8 x i8> %op1, %op2
   ret <8 x i8> %res
 }
 
 define <16 x i8> @udiv_v16i8(<16 x i8> %op1, <16 x i8> %op2) #0 {
 ; CHECK-LABEL: udiv_v16i8:
-; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,4)]]
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, z1.b
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, z0.b
-; CHECK-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: udivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: udivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: udiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: udiv [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1 z0.b, [[RES_LO]].b, [[RES_HI]].b
+
+; HALF VECTOR:
+; VBITS_EQ_256: ptrue [[PG:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, z1.b
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, z0.b
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
+; VBITS_EQ_256-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_256-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[RES1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: uzp1 [[RES2:z[0-9]+]].b, [[RES1]].b, [[RES1]].b
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_512: ptrue [[PG:p[0-9]+]].s, vl16
+; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, z1.b
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, z0.b
+; VBITS_GE_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, z1.h
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, z0.h
+; VBITS_GE_512-NEXT: udiv [[DIV:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[RES1:z[0-9]+]].h, [[DIV]].h, [[DIV]].h
+; VBITS_GE_512-NEXT: uzp1 [[RES2:z[0-9]+]].b, [[RES1]].b, [[RES1]].b
 ; CHECK: ret
   %res = udiv <16 x i8> %op1, %op2
   ret <16 x i8> %res
 }
 
 define void @udiv_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
 ; CHECK-LABEL: udiv_v32i8:
-; CHECK: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,32)]]
-; CHECK-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; CHECK-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,8)]]
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; CHECK-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; CHECK-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; CHECK-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; CHECK-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: udiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; CHECK-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; CHECK-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_256: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_EQ_256-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_256-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_256-NEXT: ptrue [[PG2:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_256-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: udiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_256-NEXT: udiv [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_256-NEXT: st1b { [[UZP3:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_EQ_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_512-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_512-NEXT: st1b { [[UZP2:z[0-9]+]].b }, [[PG1]], [x0]
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_1024: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_1024-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_GE_1024-NEXT: st1b { [[UZP2:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <32 x i8>, <32 x i8>* %a
   %op2 = load <32 x i8>, <32 x i8>* %b
   %res = udiv <32 x i8> %op1, %op2
   store <32 x i8> %res, <32 x i8>* %a
   ret void
 }
 
 define void @udiv_v64i8(<64 x i8>* %a, <64 x i8>* %b) #0 {
 ; CHECK-LABEL: udiv_v64i8:
-; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,64)]]
-; VBITS_GE_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,16)]]
-; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_512-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_512-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_512-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_512-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_512-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_512-NEXT: udiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; VBITS_GE_512-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_512-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_512-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_512-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_512-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_EQ_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_512-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: udiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_512-NEXT: udiv [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_512-NEXT: st1b { [[UZP3:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_EQ_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_1024-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_1024-NEXT: st1b { [[UZP2:z[0-9]+]].b }, [[PG1]], [x0]
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_2048: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_2048-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[RES1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_2048-NEXT: uzp1 [[RES2:z[0-9]+]].b, [[RES1]].b, [[RES1]].b
+; VBITS_GE_2048-NEXT: st1b { [[RES2]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <64 x i8>, <64 x i8>* %a
   %op2 = load <64 x i8>, <64 x i8>* %b
   %res = udiv <64 x i8> %op1, %op2
   store <64 x i8> %res, <64 x i8>* %a
   ret void
 }
 
 define void @udiv_v128i8(<128 x i8>* %a, <128 x i8>* %b) #0 {
 ; CHECK-LABEL: udiv_v128i8:
-; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,128)]]
-; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,32)]]
-; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_1024-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_1024-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_1024-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_1024-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_1024-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_1024-NEXT: udiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; VBITS_GE_1024-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_1024-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_1024-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_1024-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_1024-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].b, vl128
+; VBITS_EQ_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_1024-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: udiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_1024-NEXT: udiv [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_1024-NEXT: st1b { [[UZP3:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_2048: ptrue [[PG1:p[0-9]+]].b, vl128
+; VBITS_EQ_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_2048-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_2048-NEXT: st1b { [[UZP2:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <128 x i8>, <128 x i8>* %a
   %op2 = load <128 x i8>, <128 x i8>* %b
   %res = udiv <128 x i8> %op1, %op2
   store <128 x i8> %res, <128 x i8>* %a
   ret void
 }
 
 define void @udiv_v256i8(<256 x i8>* %a, <256 x i8>* %b) #0 {
 ; CHECK-LABEL: udiv_v256i8:
-; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
-; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
-; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_2048-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_2048-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_2048-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_2048-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_2048-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_2048-NEXT: udiv    [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
-; VBITS_GE_2048-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_2048-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_2048-NEXT: uzp1    [[RES:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_2048-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_2048-NEXT: ret
+; VBITS_EQ_2048: ptrue [[PG1:p[0-9]+]].b, vl256
+; VBITS_EQ_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_2048-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: udiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_2048-NEXT: udiv [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_2048-NEXT: st1b { [[UZP3:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <256 x i8>, <256 x i8>* %a
   %op2 = load <256 x i8>, <256 x i8>* %b
   %res = udiv <256 x i8> %op1, %op2
   store <256 x i8> %res, <256 x i8>* %a
   ret void
 }
 
 ; Vector vXi16 udiv are not legal for NEON so use SVE when available.
 define <4 x i16> @udiv_v4i16(<4 x i16> %op1, <4 x i16> %op2) #0 {
 ; CHECK-LABEL: udiv_v4i16:
-; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),2)]]
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: ret
+; CHECK: ushll v1.4s, v1.4h, #0
+; CHECK-NEXT: ushll v0.4s, v0.4h, #0
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl4
+; CHECK-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP1:z[0-9]+]].s, [[OP2:z[0-9]+]].s
+; CHECK-NEXT: mov w8, v0.s[1]
+; CHECK-NEXT: mov w9, v0.s[2]
+; CHECK-NEXT: mov w10, v0.s[3]
+; CHECK-NEXT: mov v0.h[1], w8
+; CHECK-NEXT: mov v0.h[2], w9
+; CHECK-NEXT: mov v0.h[3], w10
+; CHECK: ret
   %res = udiv <4 x i16> %op1, %op2
   ret <4 x i16> %res
 }
 
 define <8 x i16> @udiv_v8i16(<8 x i16> %op1, <8 x i16> %op2) #0 {
 ; CHECK-LABEL: udiv_v8i16:
-; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),4)]]
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: ret
+; CHECK: ptrue [[PG1:p[0-9]+]].s, vl8
+; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; CHECK-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; CHECK: ret
   %res = udiv <8 x i16> %op1, %op2
   ret <8 x i16> %res
 }
 
 define void @udiv_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
 ; CHECK-LABEL: udiv_v16i16:
-; CHECK: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),16)]]
-; CHECK-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; CHECK-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),8)]]
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; CHECK-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; CHECK-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; CHECK-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_256: ptrue [[PG1:p[0-9]+]].h, vl16
+; VBITS_EQ_256-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_256-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_256-NEXT: ptrue [[PG2:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_256-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_256-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_512: ptrue [[PG1:p[0-9]+]].h, vl16
+; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_512-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_512-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <16 x i16>, <16 x i16>* %a
   %op2 = load <16 x i16>, <16 x i16>* %b
   %res = udiv <16 x i16> %op1, %op2
   store <16 x i16> %res, <16 x i16>* %a
   ret void
 }
 
 define void @udiv_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
 ; CHECK-LABEL: udiv_v32i16:
-; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),32)]]
-; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),16)]]
-; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_512-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_512-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_512-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_512-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_512-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].h, vl32
+; VBITS_EQ_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_512-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_512-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_1024: ptrue [[PG1:p[0-9]+]].h, vl32
+; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_1024-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_1024-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <32 x i16>, <32 x i16>* %a
   %op2 = load <32 x i16>, <32 x i16>* %b
   %res = udiv <32 x i16> %op1, %op2
   store <32 x i16> %res, <32 x i16>* %a
   ret void
 }
 
 define void @udiv_v64i16(<64 x i16>* %a, <64 x i16>* %b) #0 {
 ; CHECK-LABEL: udiv_v64i16:
-; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),64)]]
-; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),32)]]
-; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_1024-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_1024-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_1024-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_1024-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_1024-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].h, vl64
+; VBITS_EQ_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_1024-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_1024-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_2048: ptrue [[PG1:p[0-9]+]].h, vl64
+; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_2048-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_2048-NEXT: st1h { [[UZP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <64 x i16>, <64 x i16>* %a
   %op2 = load <64 x i16>, <64 x i16>* %b
   %res = udiv <64 x i16> %op1, %op2
   store <64 x i16> %res, <64 x i16>* %a
   ret void
 }
 
 define void @udiv_v128i16(<128 x i16>* %a, <128 x i16>* %b) #0 {
 ; CHECK-LABEL: udiv_v128i16:
-; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),128)]]
-; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),64)]]
-; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_2048-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_2048-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_2048-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_2048-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_2048-NEXT: ret
+; VBITS_EQ_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),128)]]
+; VBITS_EQ_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),64)]]
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_2048-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_2048-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[RES:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; VBITS_EQ_2048-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_EQ_2048-NEXT: ret
   %op1 = load <128 x i16>, <128 x i16>* %a
   %op2 = load <128 x i16>, <128 x i16>* %b
   %res = udiv <128 x i16> %op1, %op2
   store <128 x i16> %res, <128 x i16>* %a
   ret void
 }
 
 ; Vector v2i32 udiv are not legal for NEON so use SVE when available.

llvm/test/CodeGen/AArch64/sve-fixed-length-int-rem.ll

-; RUN: llc -aarch64-sve-vector-bits-min=128  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=16 -check-prefix=NO_SVE
-; RUN: llc -aarch64-sve-vector-bits-min=256  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=32
+; RUN: llc -aarch64-sve-vector-bits-min=128  -asm-verbose=0 < %s | FileCheck %s -check-prefix=NO_SVE
+; RUN: llc -aarch64-sve-vector-bits-min=256  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=32 -check-prefixes=CHECK,VBITS_EQ_256
 ; RUN: llc -aarch64-sve-vector-bits-min=384  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=32
-; RUN: llc -aarch64-sve-vector-bits-min=512  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512
+; RUN: llc -aarch64-sve-vector-bits-min=512  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512,VBITS_EQ_512
 ; RUN: llc -aarch64-sve-vector-bits-min=640  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512
 ; RUN: llc -aarch64-sve-vector-bits-min=768  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512
 ; RUN: llc -aarch64-sve-vector-bits-min=896  -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=64 -check-prefixes=CHECK,VBITS_GE_512
-; RUN: llc -aarch64-sve-vector-bits-min=1024 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
+; RUN: llc -aarch64-sve-vector-bits-min=1024 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024,VBITS_EQ_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1152 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1280 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1408 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1536 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1664 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1792 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
 ; RUN: llc -aarch64-sve-vector-bits-min=1920 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=128 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024
-; RUN: llc -aarch64-sve-vector-bits-min=2048 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=256 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024,VBITS_GE_2048
+; RUN: llc -aarch64-sve-vector-bits-min=2048 -asm-verbose=0 < %s | FileCheck %s -D#VBYTES=256 -check-prefixes=CHECK,VBITS_GE_512,VBITS_GE_1024,VBITS_GE_2048,VBITS_EQ_2048
 
 target triple = "aarch64-unknown-linux-gnu"
 
 ; Don't use SVE when its registers are no bigger than NEON.
 ; NO_SVE-NOT: ptrue
 
 ;
 ; SREM
 ;
 
 ; Vector vXi8 sdiv are not legal for NEON so use SVE when available.
 define <8 x i8> @srem_v8i8(<8 x i8> %op1, <8 x i8> %op2) #0 {
 ; CHECK-LABEL: srem_v8i8:
-; CHECK: sunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,2)]]
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, z1.b
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, z0.b
-; CHECK-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sdivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: sdivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO:z[0-9]+]].h
-; CHECK-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO:z[0-9]+]].h
-; CHECK-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO:z[0-9]+]].h
-; CHECK-NEXT: sdiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: sdivr [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1 z2.b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK-NEXT: mls v0.8b, v2.8b, v1.8b
-; CHECK:    ret
+; CHECK: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2:z[0-9]+]].b
+; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1:z[0-9]+]].b
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl8
+; CHECK-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; CHECK-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; CHECK-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; CHECK-NEXT: umov [[SCALAR:w[0-9]+]], [[VEC:v[0-9]+]].h[0]
+; CHECK-NEXT: fmov s3, [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[1]
+; CHECK-NEXT: mov [[FINAL:v[0-9]+]].b[1], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[2]
+; CHECK-NEXT: mov [[FINAL]].b[2], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[3]
+; CHECK-NEXT: mov [[FINAL]].b[3], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[4]
+; CHECK-NEXT: mov [[FINAL]].b[4], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[5]
+; CHECK-NEXT: mov [[FINAL]].b[5], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[6]
+; CHECK-NEXT: mov [[FINAL]].b[6], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[7]
+; CHECK-NEXT: mov [[FINAL]].b[7], [[SCALAR]]
+; CHECK-NEXT: mls v0.8b, [[FINAL]].8b, v1.8b
+; CHECK: ret
   %res = srem <8 x i8> %op1, %op2
   ret <8 x i8> %res
 }
 
 define <16 x i8> @srem_v16i8(<16 x i8> %op1, <16 x i8> %op2) #0 {
 ; CHECK-LABEL: srem_v16i8:
-; CHECK: sunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,4)]]
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, z1.b
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, z0.b
-; CHECK-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sdivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: sdivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO:z[0-9]+]].h
-; CHECK-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO:z[0-9]+]].h
-; CHECK-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO:z[0-9]+]].h
-; CHECK-NEXT: sdiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: sdivr [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1 z2.b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK-NEXT: mls v0.16b, v2.16b, v1.16b
+
+; HALF VECTOR
+; VBITS_EQ_256: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: ptrue [[PG1:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_256-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_256-NEXT: mls v0.16b, v2.16b, v1.16b
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_512: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_512-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_512-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s, vl16
+; VBITS_GE_512-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_512-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_GE_512-NEXT: mls v0.16b, v2.16b, v1.16b
 ; CHECK: ret
   %res = srem <16 x i8> %op1, %op2
   ret <16 x i8> %res
 }
 
 define void @srem_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
 ; CHECK-LABEL: srem_v32i8:
-; CHECK: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,32)]]
-; CHECK-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; CHECK-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,8)]]
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; CHECK-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; CHECK-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; CHECK-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; CHECK-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; CHECK-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: sdivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; CHECK-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
-; CHECK-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
-; CHECK-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; CHECK-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_256: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_EQ_256-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_256-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_256-NEXT: ptrue [[PG2:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_256-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: sdiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_256-NEXT: sdivr [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_256-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP3]].b
+; VBITS_EQ_256-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_256-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_EQ_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_512-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_512-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_EQ_512-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_512-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_1024: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_1024-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_GE_1024-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_GE_1024-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_1024-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <32 x i8>, <32 x i8>* %a
   %op2 = load <32 x i8>, <32 x i8>* %b
   %res = srem <32 x i8> %op1, %op2
   store <32 x i8> %res, <32 x i8>* %a
   ret void
 }
 
 define void @srem_v64i8(<64 x i8>* %a, <64 x i8>* %b) #0 {
 ; CHECK-LABEL: srem_v64i8:
-; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,64)]]
-; VBITS_GE_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,16)]]
-; VBITS_GE_512-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_512-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_512-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_512-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_512-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_512-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_512-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_512-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_512-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_512-NEXT: sdivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; VBITS_GE_512-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_512-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_512-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_512-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
-; VBITS_GE_512-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
-; VBITS_GE_512-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_512-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_EQ_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_512-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: sdiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_512-NEXT: sdivr [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_512-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP3]].b
+; VBITS_EQ_512-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_512-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_EQ_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_1024-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_1024-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_EQ_1024-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_1024-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_2048: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_2048-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_GE_2048-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_GE_2048-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_2048-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <64 x i8>, <64 x i8>* %a
   %op2 = load <64 x i8>, <64 x i8>* %b
   %res = srem <64 x i8> %op1, %op2
   store <64 x i8> %res, <64 x i8>* %a
   ret void
 }
 
 define void @srem_v128i8(<128 x i8>* %a, <128 x i8>* %b) #0 {
 ; CHECK-LABEL: srem_v128i8:
-; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,128)]]
-; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,32)]]
-; VBITS_GE_1024-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_1024-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_1024-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_1024-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_1024-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_1024-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_1024-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_1024-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_1024-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_1024-NEXT: sdivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; VBITS_GE_1024-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_1024-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_1024-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_1024-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
-; VBITS_GE_1024-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
-; VBITS_GE_1024-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_1024-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].b, vl128
+; VBITS_EQ_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_1024-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: sdiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_1024-NEXT: sdivr [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_1024-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP3]].b
+; VBITS_EQ_1024-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_1024-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_2048: ptrue [[PG1:p[0-9]+]].b, vl128
+; VBITS_EQ_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_2048-NEXT: sdivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_2048-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_EQ_2048-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_2048-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <128 x i8>, <128 x i8>* %a
   %op2 = load <128 x i8>, <128 x i8>* %b
   %res = srem <128 x i8> %op1, %op2
   store <128 x i8> %res, <128 x i8>* %a
   ret void
 }
 
 define void @srem_v256i8(<256 x i8>* %a, <256 x i8>* %b) #0 {
 ; CHECK-LABEL: srem_v256i8:
-; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
-; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
-; VBITS_GE_2048-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_2048-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_2048-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_2048-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_2048-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_2048-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_2048-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_2048-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_2048-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_2048-NEXT: sdivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; VBITS_GE_2048-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_2048-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_2048-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_2048-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
-; VBITS_GE_2048-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
-; VBITS_GE_2048-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_2048-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_2048-NEXT: sdivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: sdivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: sdiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_2048-NEXT: sdivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
+; VBITS_EQ_2048-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
+; VBITS_EQ_2048-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
+; VBITS_EQ_2048-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
+; VBITS_EQ_2048-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
+; VBITS_EQ_2048-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
+; VBITS_EQ_2048-NEXT: ret
   %op1 = load <256 x i8>, <256 x i8>* %a
   %op2 = load <256 x i8>, <256 x i8>* %b
   %res = srem <256 x i8> %op1, %op2
   store <256 x i8> %res, <256 x i8>* %a
   ret void
 }
 
 ; Vector vXi16 sdiv are not legal for NEON so use SVE when available.
 define <4 x i16> @srem_v4i16(<4 x i16> %op1, <4 x i16> %op2) #0 {
 ; CHECK-LABEL: srem_v4i16:
-; CHECK: sunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),2)]]
-; CHECK-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: movprfx [[PFX:z[0-9]+]], [[OP1_LO:z[0-9]+]]
-; CHECK-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[PFX]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 z2.h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: mls v0.4h, v2.4h, v1.4h
-; CHECK-NEXT: ret
+; CHECK: sshll v2.4s, v1.4h, #0
+; CHECK-NEXT: sshll v3.4s, v0.4h, #0
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl4
+; CHECK-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, z2.s, z3.s
+; CHECK-NEXT: mov [[SCALAR1:w[0-9]+]], [[VEC:v[0-9]+]].s[1]
+; CHECK-NEXT: mov [[SCALAR2:w[0-9]+]], [[VEC]].s[2]
+; CHECK-NEXT: mov [[SCALAR3:w[0-9]+]], [[VEC]].s[3]
+; CHECK-NEXT: mov [[VEC]].h[1], [[SCALAR1]]
+; CHECK-NEXT: mov [[VEC]].h[2], [[SCALAR2]]
+; CHECK-NEXT: mov [[VEC]].h[3], [[SCALAR3]]
+; CHECK-NEXT: mls v0.4h, [[VEC]].4h, v1.4h
+; CHECK: ret
   %res = srem <4 x i16> %op1, %op2
   ret <4 x i16> %res
 }
 
 define <8 x i16> @srem_v8i16(<8 x i16> %op1, <8 x i16> %op2) #0 {
 ; CHECK-LABEL: srem_v8i16:
-; CHECK: sunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),4)]]
-; CHECK-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: movprfx [[PFX:z[0-9]+]], [[OP1_LO:z[0-9]+]]
-; CHECK-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[PFX]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 z2.h, [[RES_LO]].h, [[RES_HI]].h
+; CHECK: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl8
+; CHECK-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO]].s, [[OP1_LO]].s
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
 ; CHECK-NEXT: mls v0.8h, v2.8h, v1.8h
-; CHECK-NEXT: ret
+; CHECK: ret
   %res = srem <8 x i16> %op1, %op2
   ret <8 x i16> %res
 }
 
 define void @srem_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
 ; CHECK-LABEL: srem_v16i16:
-; CHECK: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),16)]]
-; CHECK-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; CHECK-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),8)]]
-; CHECK-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; CHECK-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; CHECK-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; CHECK-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; CHECK-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
-; CHECK-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
-; CHECK-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; CHECK-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_256: ptrue [[PG1:p[0-9]+]].h, vl16
+; VBITS_EQ_256-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_256-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_256-NEXT: ptrue [[PG2:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_256-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_256-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_256-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_256-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_256-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_EQ_256-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_EQ_256-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_512: ptrue [[PG1:p[0-9]+]].h, vl16
+; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_GE_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_512-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_512-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_GE_512-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_512-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <16 x i16>, <16 x i16>* %a
   %op2 = load <16 x i16>, <16 x i16>* %b
   %res = srem <16 x i16> %op1, %op2
   store <16 x i16> %res, <16 x i16>* %a
   ret void
 }
 
 define void @srem_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
 ; CHECK-LABEL: srem_v32i16:
-; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),32)]]
-; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),16)]]
-; VBITS_GE_512-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_512-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_512-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_512-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_512-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_512-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
-; VBITS_GE_512-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
-; VBITS_GE_512-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_512-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].h, vl32
+; VBITS_EQ_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_512-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_512-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_512-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_512-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_512-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_EQ_512-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_EQ_512-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_1024: ptrue [[PG1:p[0-9]+]].h, vl32
+; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_1024-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_1024-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_GE_1024-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_1024-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <32 x i16>, <32 x i16>* %a
   %op2 = load <32 x i16>, <32 x i16>* %b
   %res = srem <32 x i16> %op1, %op2
   store <32 x i16> %res, <32 x i16>* %a
   ret void
 }
 
 define void @srem_v64i16(<64 x i16>* %a, <64 x i16>* %b) #0 {
 ; CHECK-LABEL: srem_v64i16:
-; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),64)]]
-; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),32)]]
-; VBITS_GE_1024-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_1024-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_1024-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_1024-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_1024-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_1024-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
-; VBITS_GE_1024-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
-; VBITS_GE_1024-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_1024-NEXT: ret
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].h, vl64
+; VBITS_EQ_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_1024-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_1024-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_1024-NEXT: sdivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_1024-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_1024-NEXT: sdiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_EQ_1024-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_EQ_1024-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_2048: ptrue [[PG1:p[0-9]+]].h, vl64
+; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_2048-NEXT: sdiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_2048-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_GE_2048-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_2048-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <64 x i16>, <64 x i16>* %a
   %op2 = load <64 x i16>, <64 x i16>* %b
   %res = srem <64 x i16> %op1, %op2
   store <64 x i16> %res, <64 x i16>* %a
   ret void
 }
 
 define void @srem_v128i16(<128 x i16>* %a, <128 x i16>* %b) #0 {
 ; CHECK-LABEL: srem_v128i16:
-; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),128)]]
-; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),64)]]
-; VBITS_GE_2048-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_2048-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_2048-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_2048-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_2048-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_2048-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
-; VBITS_GE_2048-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
-; VBITS_GE_2048-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_2048-NEXT: ret
+; VBITS_EQ_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),128)]]
+; VBITS_EQ_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),64)]]
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_2048-NEXT: sunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_2048-NEXT: sunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_2048-NEXT: sdivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_2048-NEXT: sunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_2048-NEXT: sdiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; VBITS_EQ_2048-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
+; VBITS_EQ_2048-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
+; VBITS_EQ_2048-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_EQ_2048-NEXT: ret
   %op1 = load <128 x i16>, <128 x i16>* %a
   %op2 = load <128 x i16>, <128 x i16>* %b
   %res = srem <128 x i16> %op1, %op2
   store <128 x i16> %res, <128 x i16>* %a
   ret void
 }
 
 ; Vector v2i32 sdiv are not legal for NEON so use SVE when available.

 
 ;
 ; UREM
 ;
 
 ; Vector vXi8 udiv are not legal for NEON so use SVE when available.
 define <8 x i8> @urem_v8i8(<8 x i8> %op1, <8 x i8> %op2) #0 {
 ; CHECK-LABEL: urem_v8i8:
-; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,2)]]
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, z1.b
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, z0.b
-; CHECK-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: udivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: udivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO:z[0-9]+]].h
-; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO:z[0-9]+]].h
-; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO:z[0-9]+]].h
-; CHECK-NEXT: udiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: udivr [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1 z2.b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK-NEXT: mls v0.8b, v2.8b, v1.8b
-; CHECK:    ret
+; CHECK: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2:z[0-9]+]].b
+; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1:z[0-9]+]].b
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl8
+; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; CHECK-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; CHECK-NEXT: umov [[SCALAR:w[0-9]+]], [[VEC:v[0-9]+]].h[0]
+; CHECK-NEXT: fmov s3, [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[1]
+; CHECK-NEXT: mov [[FINAL:v[0-9]+]].b[1], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[2]
+; CHECK-NEXT: mov [[FINAL]].b[2], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[3]
+; CHECK-NEXT: mov [[FINAL]].b[3], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[4]
+; CHECK-NEXT: mov [[FINAL]].b[4], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[5]
+; CHECK-NEXT: mov [[FINAL]].b[5], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[6]
+; CHECK-NEXT: mov [[FINAL]].b[6], [[SCALAR]]
+; CHECK-NEXT: umov [[SCALAR]], [[VEC]].h[7]
+; CHECK-NEXT: mov [[FINAL]].b[7], [[SCALAR]]
+; CHECK-NEXT: mls v0.8b, [[FINAL]].8b, v1.8b
+; CHECK: ret
   %res = urem <8 x i8> %op1, %op2
   ret <8 x i8> %res
 }
 
 define <16 x i8> @urem_v16i8(<16 x i8> %op1, <16 x i8> %op2) #0 {
 ; CHECK-LABEL: urem_v16i8:
-; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].h, z1.b
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, z0.b
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(VBYTES,4)]]
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, z1.b
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, z0.b
-; CHECK-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: udivr [[RES_HI_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: udivr [[RES_HI_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO:z[0-9]+]].h
-; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO:z[0-9]+]].h
-; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO:z[0-9]+]].h
-; CHECK-NEXT: udiv [[RES_LO_HI:z[0-9]+]].s, [[PG]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: udivr [[RES_LO_LO:z[0-9]+]].s, [[PG]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; CHECK-NEXT: uzp1 [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1 [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1 z2.b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK-NEXT: mls v0.16b, v2.16b, v1.16b
+
+; HALF VECTOR
+; VBITS_EQ_256: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: ptrue [[PG1:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_256-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_256-NEXT: mls v0.16b, v2.16b, v1.16b
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_512: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s, vl16
+; VBITS_GE_512-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_512-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_GE_512-NEXT: mls v0.16b, v2.16b, v1.16b
 ; CHECK: ret
   %res = urem <16 x i8> %op1, %op2
   ret <16 x i8> %res
 }
 
 define void @urem_v32i8(<32 x i8>* %a, <32 x i8>* %b) #0 {
 ; CHECK-LABEL: urem_v32i8:
-; CHECK: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,32)]]
-; CHECK-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; CHECK-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,8)]]
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; CHECK-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; CHECK-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; CHECK-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; CHECK-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; CHECK-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; CHECK-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; CHECK-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; CHECK-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; CHECK-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; CHECK-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; CHECK-NEXT: udivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; CHECK-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; CHECK-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; CHECK-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; CHECK-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
-; CHECK-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
-; CHECK-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; CHECK-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_256: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_EQ_256-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_256-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_256-NEXT: ptrue [[PG2:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_256-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_256-NEXT: udiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_256-NEXT: udivr [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_256-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_256-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP3]].b
+; VBITS_EQ_256-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_256-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_EQ_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_512-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_512-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_EQ_512-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_512-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_1024: ptrue [[PG1:p[0-9]+]].b, vl32
+; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_1024-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_GE_1024-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_GE_1024-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_1024-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <32 x i8>, <32 x i8>* %a
   %op2 = load <32 x i8>, <32 x i8>* %b
   %res = urem <32 x i8> %op1, %op2
   store <32 x i8> %res, <32 x i8>* %a
   ret void
 }
 
 define void @urem_v64i8(<64 x i8>* %a, <64 x i8>* %b) #0 {
 ; CHECK-LABEL: urem_v64i8:
-; VBITS_GE_512: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,64)]]
-; VBITS_GE_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,16)]]
-; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_512-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_512-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_512-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_512-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_512-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_512-NEXT: udivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; VBITS_GE_512-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_512-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_512-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_512-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
-; VBITS_GE_512-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
-; VBITS_GE_512-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_512-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_EQ_512-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_512-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_512-NEXT: udiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_512-NEXT: udivr [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_512-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_512-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP3]].b
+; VBITS_EQ_512-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_512-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_EQ_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_1024-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_1024-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_EQ_1024-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_1024-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; QUARTER VECTOR OR SMALLER:
+; VBITS_GE_2048: ptrue [[PG1:p[0-9]+]].b, vl64
+; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_GE_2048-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_LO]].s, [[OP2_LO_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_GE_2048-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_GE_2048-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_GE_2048-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <64 x i8>, <64 x i8>* %a
   %op2 = load <64 x i8>, <64 x i8>* %b
   %res = urem <64 x i8> %op1, %op2
   store <64 x i8> %res, <64 x i8>* %a
   ret void
 }
 
 define void @urem_v128i8(<128 x i8>* %a, <128 x i8>* %b) #0 {
 ; CHECK-LABEL: urem_v128i8:
-; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,128)]]
-; VBITS_GE_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s,  vl[[#min(VBYTES,32)]]
-; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_1024-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_1024-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_1024-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_1024-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_1024-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_1024-NEXT: udivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; VBITS_GE_1024-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_1024-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_1024-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_1024-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
-; VBITS_GE_1024-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
-; VBITS_GE_1024-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_1024-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].b, vl128
+; VBITS_EQ_1024-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_1024-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_1024-NEXT: udiv [[DIV3:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_1024-NEXT: udivr [[DIV4:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP2:z[0-9]+]].h, [[DIV4]].h, [[DIV3]].h
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP3:z[0-9]+]].b, [[UZP2]].b, [[UZP1]].b
+; VBITS_EQ_1024-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP3]].b
+; VBITS_EQ_1024-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_1024-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+
+; HALF VECTOR:
+; VBITS_EQ_2048: ptrue [[PG1:p[0-9]+]].b, vl128
+; VBITS_EQ_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG1]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG1]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_HI]].s, [[OP1_LO_HI]].s
+; VBITS_EQ_2048-NEXT: udivr [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_2048-NEXT: uzp1 [[UZP2:z[0-9]+]].b, [[UZP1]].b, [[UZP1]].b
+; VBITS_EQ_2048-NEXT: mul [[OP2]].b, [[PG1]]/m, [[OP2]].b, [[UZP2]].b
+; VBITS_EQ_2048-NEXT: sub [[OP1]].b, [[PG1]]/m, [[OP1]].b, [[OP2]].b
+; VBITS_EQ_2048-NEXT: st1b { [[OP1:z[0-9]+]].b }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <128 x i8>, <128 x i8>* %a
   %op2 = load <128 x i8>, <128 x i8>* %b
   %res = urem <128 x i8> %op1, %op2
   store <128 x i8> %res, <128 x i8>* %a
   ret void
 }
 
 define void @urem_v256i8(<256 x i8>* %a, <256 x i8>* %b) #0 {
 ; CHECK-LABEL: urem_v256i8:
-; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
-; VBITS_GE_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
-; VBITS_GE_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
-; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
-; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
-; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
-; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
-; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
-; VBITS_GE_2048-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
-; VBITS_GE_2048-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_2048-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
-; VBITS_GE_2048-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
-; VBITS_GE_2048-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
-; VBITS_GE_2048-NEXT: udivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
-; VBITS_GE_2048-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
-; VBITS_GE_2048-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
-; VBITS_GE_2048-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
-; VBITS_GE_2048-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
-; VBITS_GE_2048-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
-; VBITS_GE_2048-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
-; VBITS_GE_2048-NEXT: ret
+; VBITS_EQ_2048: ptrue [[PG:p[0-9]+]].b, vl[[#min(VBYTES,256)]]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP1:z[0-9]+]].b }, [[PG]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1b { [[OP2:z[0-9]+]].b }, [[PG]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(VBYTES,64)]]
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].h, [[OP2]].b
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].h, [[OP1]].b
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_HI_HI:z[0-9]]].s, [[OP2_HI]].h
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_HI_HI:z[0-9]]].s, [[OP1_HI]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_HI_LO:z[0-9]+]].s, [[OP2_HI]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_HI_LO:z[0-9]+]].s, [[OP1_HI]].h
+; VBITS_EQ_2048-NEXT: udivr   [[RES_HI_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_HI]].s, [[OP1_HI_HI]].s
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_LO_HI:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: udivr   [[RES_HI_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI_LO]].s, [[OP1_HI_LO]].s
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_LO_HI:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO_LO:z[0-9]+]].s, [[OP2_LO]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO_LO:z[0-9]+]].s, [[OP1_LO]].h
+; VBITS_EQ_2048-NEXT: udiv    [[RES_LO_HI:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO_HI]].s, [[OP2_LO_HI]].s
+; VBITS_EQ_2048-NEXT: udivr   [[RES_LO_LO:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO_LO]].s, [[OP1_LO_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1    [[RES_HI:z[0-9]+]].h, [[RES_HI_LO]].h, [[RES_HI_HI]].h
+; VBITS_EQ_2048-NEXT: uzp1    [[RES_LO:z[0-9]+]].h, [[RES_LO_LO]].h, [[RES_LO_HI]].h
+; VBITS_EQ_2048-NEXT: uzp1    [[ZIP:z[0-9]+]].b, [[RES_LO]].b, [[RES_HI]].b
+; VBITS_EQ_2048-NEXT: mul     [[MUL:z[0-9]+]].b, [[PG]]/m, [[OP2]].b, [[ZIP]].b
+; VBITS_EQ_2048-NEXT: sub     [[RES:z[0-9]+]].b, [[PG]]/m, [[OP1]].b, [[MUL]].b
+; VBITS_EQ_2048-NEXT: st1b    { [[RES]].b }, [[PG]], [x0]
+; VBITS_EQ_2048-NEXT: ret
   %op1 = load <256 x i8>, <256 x i8>* %a
   %op2 = load <256 x i8>, <256 x i8>* %b
   %res = urem <256 x i8> %op1, %op2
   store <256 x i8> %res, <256 x i8>* %a
   ret void
 }
 
 ; Vector vXi16 udiv are not legal for NEON so use SVE when available.
 define <4 x i16> @urem_v4i16(<4 x i16> %op1, <4 x i16> %op2) #0 {
 ; CHECK-LABEL: urem_v4i16:
-; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),2)]]
-; CHECK-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: movprfx [[PFX:z[0-9]+]], [[OP1_LO:z[0-9]+]]
-; CHECK-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[PFX]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 z2.h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: mls v0.4h, v2.4h, v1.4h
-; CHECK-NEXT: ret
+; CHECK: ushll v2.4s, v1.4h, #0
+; CHECK-NEXT: ushll v3.4s, v0.4h, #0
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl4
+; CHECK-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, z2.s, z3.s
+; CHECK-NEXT: mov [[SCALAR1:w[0-9]+]], [[VEC:v[0-9]+]].s[1]
+; CHECK-NEXT: mov [[SCALAR2:w[0-9]+]], [[VEC]].s[2]
+; CHECK-NEXT: mov [[SCALAR3:w[0-9]+]], [[VEC]].s[3]
+; CHECK-NEXT: mov [[VEC]].h[1], [[SCALAR1]]
+; CHECK-NEXT: mov [[VEC]].h[2], [[SCALAR2]]
+; CHECK-NEXT: mov [[VEC]].h[3], [[SCALAR3]]
+; CHECK-NEXT: mls v0.4h, [[VEC]].4h, v1.4h
+; CHECK: ret
   %res = urem <4 x i16> %op1, %op2
   ret <4 x i16> %res
 }
 
 define <8 x i16> @urem_v8i16(<8 x i16> %op1, <8 x i16> %op2) #0 {
 ; CHECK-LABEL: urem_v8i16:
-; CHECK: uunpkhi [[OP2_HI:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, z0.h
-; CHECK-NEXT: ptrue [[PG:p[0-9]+]].s, vl[[#min(div(VBYTES,2),4)]]
-; CHECK-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, z1.h
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, z0.h
-; CHECK-NEXT: movprfx [[PFX:z[0-9]+]], [[OP1_LO:z[0-9]+]]
-; CHECK-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG]]/m, [[PFX]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 z2.h, [[RES_LO]].h, [[RES_HI]].h
+; CHECK: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl8
+; CHECK-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG1]]/m, [[OP2_LO]].s, [[OP1_LO]].s
+; CHECK-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
 ; CHECK-NEXT: mls v0.8h, v2.8h, v1.8h
-; CHECK-NEXT: ret
+; CHECK: ret
   %res = urem <8 x i16> %op1, %op2
   ret <8 x i16> %res
 }
 
 define void @urem_v16i16(<16 x i16>* %a, <16 x i16>* %b) #0 {
 ; CHECK-LABEL: urem_v16i16:
-; CHECK: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),16)]]
-; CHECK-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; CHECK-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; CHECK-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),8)]]
-; CHECK-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; CHECK-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; CHECK-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; CHECK-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; CHECK-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; CHECK-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; CHECK-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; CHECK-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
-; CHECK-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
-; CHECK-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; CHECK-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_256: ptrue [[PG1:p[0-9]+]].h, vl16
+; VBITS_EQ_256-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_256-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_256-NEXT: ptrue [[PG2:p[0-9]+]].s, vl8
+; VBITS_EQ_256-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_256-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_256-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_256-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_256-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_256-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_256-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_256-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_EQ_256-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_EQ_256-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_512: ptrue [[PG1:p[0-9]+]].h, vl16
+; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_512-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_512-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_GE_512-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_512-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <16 x i16>, <16 x i16>* %a
   %op2 = load <16 x i16>, <16 x i16>* %b
   %res = urem <16 x i16> %op1, %op2
   store <16 x i16> %res, <16 x i16>* %a
   ret void
 }
 
 define void @urem_v32i16(<32 x i16>* %a, <32 x i16>* %b) #0 {
 ; CHECK-LABEL: urem_v32i16:
-; VBITS_GE_512: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),32)]]
-; VBITS_GE_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_512-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),16)]]
-; VBITS_GE_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_512-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_512-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_512-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_512-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
-; VBITS_GE_512-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
-; VBITS_GE_512-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_512-NEXT: ret
+
+; FULL VECTOR:
+; VBITS_EQ_512: ptrue [[PG1:p[0-9]+]].h, vl32
+; VBITS_EQ_512-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_512-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_512-NEXT: ptrue [[PG2:p[0-9]+]].s, vl16
+; VBITS_EQ_512-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_512-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_512-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_512-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_512-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_512-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_512-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_512-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_EQ_512-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_EQ_512-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_1024: ptrue [[PG1:p[0-9]+]].h, vl32
+; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_1024-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_1024-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_GE_1024-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_1024-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <32 x i16>, <32 x i16>* %a
   %op2 = load <32 x i16>, <32 x i16>* %b
   %res = urem <32 x i16> %op1, %op2
   store <32 x i16> %res, <32 x i16>* %a
   ret void
 }
 
 define void @urem_v64i16(<64 x i16>* %a, <64 x i16>* %b) #0 {
 ; CHECK-LABEL: urem_v64i16:
-; VBITS_GE_1024: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),64)]]
-; VBITS_GE_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_1024-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),32)]]
-; VBITS_GE_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_1024-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_1024-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_1024-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_1024-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
-; VBITS_GE_1024-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
-; VBITS_GE_1024-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_1024-NEXT: ret
+; VBITS_EQ_1024: ptrue [[PG1:p[0-9]+]].h, vl64
+; VBITS_EQ_1024-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_EQ_1024-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_EQ_1024-NEXT: ptrue [[PG2:p[0-9]+]].s, vl32
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_1024-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_1024-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_1024-NEXT: udivr [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_1024-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_1024-NEXT: udiv [[DIV2:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_1024-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV2]].h, [[DIV1]].h
+; VBITS_EQ_1024-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_EQ_1024-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_EQ_1024-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+
+; HALF VECTOR OR SMALLER:
+; VBITS_GE_2048: ptrue [[PG1:p[0-9]+]].h, vl64
+; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG1]]/z, [x0]
+; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG1]]/z, [x1]
+; VBITS_GE_2048-NEXT: ptrue [[PG2:p[0-9]+]].s, vl64
+; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_GE_2048-NEXT: udiv [[DIV1:z[0-9]+]].s, [[PG2]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_GE_2048-NEXT: uzp1 [[UZP1:z[0-9]+]].h, [[DIV1]].h, [[DIV1]].h
+; VBITS_GE_2048-NEXT: mul [[OP2]].h, [[PG1]]/m, [[OP2]].h, [[UZP1]].h
+; VBITS_GE_2048-NEXT: sub [[OP1]].h, [[PG1]]/m, [[OP1]].h, [[OP2]].h
+; VBITS_GE_2048-NEXT: st1h { [[OP1:z[0-9]+]].h }, [[PG1]], [x0]
+; CHECK: ret
   %op1 = load <64 x i16>, <64 x i16>* %a
   %op2 = load <64 x i16>, <64 x i16>* %b
   %res = urem <64 x i16> %op1, %op2
   store <64 x i16> %res, <64 x i16>* %a
   ret void
 }
 
 define void @urem_v128i16(<128 x i16>* %a, <128 x i16>* %b) #0 {
 ; CHECK-LABEL: urem_v128i16:
-; VBITS_GE_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),128)]]
-; VBITS_GE_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
-; VBITS_GE_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
-; VBITS_GE_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),64)]]
-; VBITS_GE_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
-; VBITS_GE_2048-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
-; VBITS_GE_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
-; VBITS_GE_2048-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
-; VBITS_GE_2048-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
-; VBITS_GE_2048-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
-; VBITS_GE_2048-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
-; VBITS_GE_2048-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
-; VBITS_GE_2048-NEXT: ret
+; VBITS_EQ_2048: ptrue [[PG:p[0-9]+]].h, vl[[#min(div(VBYTES,2),128)]]
+; VBITS_EQ_2048-NEXT: ld1h { [[OP1:z[0-9]+]].h }, [[PG]]/z, [x0]
+; VBITS_EQ_2048-NEXT: ld1h { [[OP2:z[0-9]+]].h }, [[PG]]/z, [x1]
+; VBITS_EQ_2048-NEXT: ptrue [[PG1:p[0-9]+]].s, vl[[#min(div(VBYTES,2),64)]]
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP1_HI:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_2048-NEXT: uunpkhi [[OP2_HI:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_2048-NEXT: uunpklo [[OP2_LO:z[0-9]+]].s, [[OP2]].h
+; VBITS_EQ_2048-NEXT: udivr   [[RES_HI:z[0-9]+]].s, [[PG1]]/m, [[OP2_HI]].s, [[OP1_HI]].s
+; VBITS_EQ_2048-NEXT: uunpklo [[OP1_LO:z[0-9]+]].s, [[OP1]].h
+; VBITS_EQ_2048-NEXT: udiv    [[RES_LO:z[0-9]+]].s, [[PG1]]/m, [[OP1_LO]].s, [[OP2_LO]].s
+; VBITS_EQ_2048-NEXT: uzp1 [[ZIP:z[0-9]+]].h, [[RES_LO]].h, [[RES_HI]].h
+; VBITS_EQ_2048-NEXT: mul [[MUL:z[0-9]+]].h, [[PG]]/m, [[OP2]].h, [[ZIP]].h
+; VBITS_EQ_2048-NEXT: sub [[RES:z[0-9]+]].h, [[PG]]/m, [[OP1]].h, [[MUL]].h
+; VBITS_EQ_2048-NEXT: st1h { [[RES]].h }, [[PG]], [x0]
+; VBITS_EQ_2048-NEXT: ret
   %op1 = load <128 x i16>, <128 x i16>* %a
   %op2 = load <128 x i16>, <128 x i16>* %b
   %res = urem <128 x i16> %op1, %op2
   store <128 x i16> %res, <128 x i16>* %a
   ret void
 }
 
 ; Vector v2i32 udiv are not legal for NEON so use SVE when available.