[AMDGPU] Propagate !noalias and !alias.scope metadata in AMDGPULowerKernelArguments for noalias arguments.

[abinavpp]

Summary

This change requires https://reviews.llvm.org/D108361.

This patch handles noalias function arguments of amdgpu_kernel during
llvm.amdgcn.kernarg.segment.ptr() emission by converting them to the equivalent
metadata representation using !noalias and !alias.scope.

[jeroen.dobbelaere]

llvm/test/CodeGen/AMDGPU/lower-kernargs.ll

1280

What is the purpose of this test ? The kernel input pointers are 'stored to undef', as such those stores see the available noalias scopes, but do not take part in it. Shouldn't the test store something to the pointers instead ?

[abinavpp]

llvm/test/CodeGen/AMDGPU/lower-kernargs.ll

1280

I'm not sure about the stores to undefs in AMDGPU/lower-kernargs.ll. This test was added in the initial commit for AMDGPULowerKernelArguments. I assume we're just checking the generated IR for the stores and not worrying about the destination. @arsenm is that right?

[arsenm]

Same purpose as D105721?

[arsenm]

llvm/test/CodeGen/AMDGPU/lower-kernargs.ll

1280

It's just a dummy use to show the handling of the value

[abinavpp]

In D108363#2960892, @arsenm wrote:

Same purpose as D105721?

Oh! I didn't notice that. @hliao , how should we move forward? Should we abandon this and https://reviews.llvm.org/D108361, or keep https://reviews.llvm.org/D108361 and use it for https://reviews.llvm.org/D105721?

[arsenm]

In D108363#2961903, @abinavpp wrote:

In D108363#2960892, @arsenm wrote:

Same purpose as D105721?

Oh! I didn't notice that. @hliao , how should we move forward? Should we abandon this and https://reviews.llvm.org/D108361, or keep https://reviews.llvm.org/D108361 and use it for https://reviews.llvm.org/D105721?

Ping. Was there any discussion of which way is better?

[abinavpp]

Rebased; Updated tests

[abinavpp]

In D108363#3442578, @arsenm wrote:

In D108363#2961903, @abinavpp wrote:

In D108363#2960892, @arsenm wrote:

Same purpose as D105721?

Oh! I didn't notice that. @hliao , how should we move forward? Should we abandon this and https://reviews.llvm.org/D108361, or keep https://reviews.llvm.org/D108361 and use it for https://reviews.llvm.org/D105721?

Ping. Was there any discussion of which way is better?

No, sorry. @hliao what do you think?

I'll check this patch with PSDB.

[abinavpp]

In D108363#3444601, @abinavpp wrote:

In D108363#3442578, @arsenm wrote:

In D108363#2961903, @abinavpp wrote:

In D108363#2960892, @arsenm wrote:

Same purpose as D105721?

Oh! I didn't notice that. @hliao , how should we move forward? Should we abandon this and https://reviews.llvm.org/D108361, or keep https://reviews.llvm.org/D108361 and use it for https://reviews.llvm.org/D105721?

Ping. Was there any discussion of which way is better?

No, sorry. @hliao what do you think?

I'll check this patch with PSDB.

PSDB's result of this patch with D108361 is a conditional go.

llvm/lib/Target/AMDGPU/AMDGPULowerKernelArguments.cpp

 //===-- AMDGPULowerKernelArguments.cpp ------------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 /// \file This pass replaces accesses to kernel arguments with loads from
 /// offsets from the kernarg base pointer.
 //
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPU.h"
 #include "GCNSubtarget.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
-#include "llvm/IR/IntrinsicsAMDGPU.h"
 #include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/IntrinsicsAMDGPU.h"
 #include "llvm/IR/MDBuilder.h"
 #include "llvm/Target/TargetMachine.h"
+#include "llvm/Transforms/Utils/Cloning.h"
 #define DEBUG_TYPE "amdgpu-lower-kernel-arguments"
 
 using namespace llvm;
 
 namespace {
 
 class AMDGPULowerKernelArguments : public FunctionPass{
 public:
bool AMDGPULowerKernelArguments::runOnFunction(Function &F) {
 
   KernArgSegment->addRetAttr(Attribute::NonNull);
   KernArgSegment->addRetAttr(
       Attribute::getWithDereferenceableBytes(Ctx, TotalKernArgSize));
 
   unsigned AS = KernArgSegment->getType()->getPointerAddressSpace();
   uint64_t ExplicitArgOffset = 0;
 
+  addAliasScopeMetadata(F);
+
   for (Argument &Arg : F.args()) {
     const bool IsByRef = Arg.hasByRefAttr();
     Type *ArgTy = IsByRef ? Arg.getParamByRefType() : Arg.getType();
     MaybeAlign ABITypeAlign = IsByRef ? Arg.getParamAlign() : None;
     if (!ABITypeAlign)
       ABITypeAlign = DL.getABITypeAlign(ArgTy);
 
     uint64_t Size = DL.getTypeSizeInBits(ArgTy);
if (PointerType *PT = dyn_cast<PointerType>(ArgTy)) {
       // FIXME: Hack. We rely on AssertZext to be able to fold DS addressing
       // modes on SI to know the high bits are 0 so pointer adds don't wrap. We
       // can't represent this with range metadata because it's only allowed for
       // integer types.
       if ((PT->getAddressSpace() == AMDGPUAS::LOCAL_ADDRESS ||
            PT->getAddressSpace() == AMDGPUAS::REGION_ADDRESS) &&
           !ST.hasUsableDSOffset())
         continue;
-
-      // FIXME: We can replace this with equivalent alias.scope/noalias
-      // metadata, but this appears to be a lot of work.
-      if (Arg.hasNoAliasAttr())
-        continue;
     }
 
     auto *VT = dyn_cast<FixedVectorType>(ArgTy);
     bool IsV3 = VT && VT->getNumElements() == 3;
     bool DoShiftOpt = Size < 32 && !ArgTy->isAggregateType();
 
     VectorType *V4Ty = nullptr;
 
if (isa<PointerType>(ArgTy)) {
         Load->setMetadata(
           LLVMContext::MD_align,
           MDNode::get(Ctx,
                       MDB.createConstant(ConstantInt::get(Builder.getInt64Ty(),
                                                           ParamAlign))));
       }
     }
 
-    // TODO: Convert noalias arg to !noalias
-
     if (DoShiftOpt) {
       Value *ExtractBits = OffsetDiff == 0 ?
         Load : Builder.CreateLShr(Load, OffsetDiff * 8);
 
       IntegerType *ArgIntTy = Builder.getIntNTy(Size);
       Value *Trunc = Builder.CreateTrunc(ExtractBits, ArgIntTy);
       Value *NewVal = Builder.CreateBitCast(Trunc, ArgTy,
                                             Arg.getName() + ".load");

llvm/test/CodeGen/AMDGPU/GlobalISel/cvt_f32_ubyte.ll

 ; GCN-NEXT:    s_setpc_b64 s[30:31]
   %cvt = uitofp i8 %arg0 to double
   ret double %cvt
 }
 
 define amdgpu_kernel void @load_i8_to_f32(float addrspace(1)* noalias %out, i8 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_i8_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_i8_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_ashrrev_i32_e32 v3, 31, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s2
 ; VI-NEXT:    v_mov_b32_e32 v2, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, v2, v3, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v0

 
 ; This should not be adding instructions to shift into the correct
 ; position in the word for the component.
 
 ; FIXME: Packing bytes
 define amdgpu_kernel void @load_v4i8_to_v4f32_unaligned(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_v4i8_to_v4f32_unaligned:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[4:7], 0 addr64 offset:1
 ; SI-NEXT:    buffer_load_ubyte v3, v[0:1], s[4:7], 0 addr64 offset:3
 ; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[4:7], 0 addr64 offset:2
 ; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v2
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_lshlrev_b32_e32 v2, 24, v3
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_lshlrev_b32_e32 v3, 16, v4
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_or_b32_e32 v1, v2, v3
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_and_b32_e32 v1, 0xff, v0
 ; SI-NEXT:    v_bfe_u32 v2, v0, 8, 8
 ; SI-NEXT:    v_bfe_u32 v4, v0, 16, 8
 ; SI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v1
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v1, v2
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v4
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_v4i8_to_v4f32_unaligned:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 1, v0
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 2, v0

 ;   %cvt = uitofp <8 x i8> %load to <8 x float>
 ;   store <8 x float> %cvt, <8 x float> addrspace(1)* %out, align 16
 ;   ret void
 ; }
 
 define amdgpu_kernel void @i8_zext_inreg_i32_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: i8_zext_inreg_i32_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
 ; SI-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: i8_zext_inreg_i32_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
; VI-NEXT:    s_endpgm
   %cvt = uitofp i32 %inreg to float
   store float %cvt, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @i8_zext_inreg_hi1_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: i8_zext_inreg_hi1_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bfe_u32 v0, v0, 8, 8
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: i8_zext_inreg_hi1_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1
; VI-NEXT:    s_endpgm
   ret void
 }
 
 ; We don't get these ones because of the zext, but instcombine removes
 ; them so it shouldn't really matter.
 define amdgpu_kernel void @i8_zext_i32_to_f32(float addrspace(1)* noalias %out, i8 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: i8_zext_i32_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: i8_zext_i32_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_ashrrev_i32_e32 v3, 31, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s2
 ; VI-NEXT:    v_mov_b32_e32 v2, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, v2, v3, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v0
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i8, i8 addrspace(1)* %in, i32 %tid
   %load = load i8, i8 addrspace(1)* %gep, align 1
   %ext = zext i8 %load to i32
   %cvt = uitofp i32 %ext to float
   store float %cvt, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v4i8_zext_v4i32_to_v4f32(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v4i8_zext_v4i32_to_v4f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[4:7], 0 addr64 offset:1
 ; SI-NEXT:    buffer_load_ubyte v3, v[0:1], s[4:7], 0 addr64 offset:3
 ; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[4:7], 0 addr64 offset:2
 ; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v2
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_lshlrev_b32_e32 v2, 24, v3
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_lshlrev_b32_e32 v3, 16, v4
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_or_b32_e32 v1, v2, v3
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_and_b32_e32 v1, 0xff, v0
 ; SI-NEXT:    v_bfe_u32 v2, v0, 8, 8
 ; SI-NEXT:    v_bfe_u32 v4, v0, 16, 8
 ; SI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v1
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v1, v2
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v4
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v4i8_zext_v4i32_to_v4f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 1, v0
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 2, v0
; VI-NEXT:    s_endpgm
   %cvt = uitofp <4 x i32> %ext to <4 x float>
   store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @extract_byte0_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: extract_byte0_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: extract_byte0_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v0
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
   %val = load i32, i32 addrspace(1)* %gep
   %and = and i32 %val, 255
   %cvt = uitofp i32 %and to float
   store float %cvt, float addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @extract_byte1_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: extract_byte1_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bfe_u32 v0, v0, 8, 8
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: extract_byte1_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1
; VI-NEXT:    s_endpgm
   %cvt = uitofp i32 %and to float
   store float %cvt, float addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @extract_byte2_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: extract_byte2_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bfe_u32 v0, v0, 16, 8
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: extract_byte2_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_sdwa v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_2
; VI-NEXT:    s_endpgm
   %cvt = uitofp i32 %and to float
   store float %cvt, float addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @extract_byte3_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: extract_byte3_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: extract_byte3_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte3_e32 v2, v0

llvm/test/CodeGen/AMDGPU/always-uniform.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple amdgcn-amdhsa -mcpu=fiji -amdgpu-scalarize-global-loads -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
 
 declare i32 @llvm.amdgcn.workitem.id.x()
 declare i32 @llvm.amdgcn.readfirstlane(i32)
 
-; GCN-LABEL: readfirstlane_uniform
-; GCN:  s_load_dwordx2 s[[[IN_ADDR:[0-9]+]]:1], s[4:5], 0x0
-; GCN:  v_readfirstlane_b32 s[[SCALAR:[0-9]+]], v0
-; GCN:  s_add_u32 s[[LOAD_ADDR:[0-9]+]], s[[IN_ADDR]], s[[SCALAR]]
-; GCN:  s_load_dword s{{[0-9]+}}, s[[[LOAD_ADDR]]
-
 define amdgpu_kernel void @readfirstlane_uniform(float addrspace(1)* noalias nocapture readonly, float addrspace(1)* noalias nocapture readonly) {
+; GCN-LABEL: readfirstlane_uniform:
+; GCN:       ; %bb.0:
+; GCN-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
+; GCN-NEXT:    v_readfirstlane_b32 s4, v0
+; GCN-NEXT:    s_mov_b32 s5, 0
+; GCN-NEXT:    s_lshl_b64 s[4:5], s[4:5], 2
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    s_add_u32 s0, s0, s4
+; GCN-NEXT:    s_addc_u32 s1, s1, s5
+; GCN-NEXT:    s_load_dword s4, s[0:1], 0x0
+; GCN-NEXT:    s_add_u32 s0, s2, 40
+; GCN-NEXT:    s_addc_u32 s1, s3, 0
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_mov_b32_e32 v1, s1
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_mov_b32_e32 v2, s4
+; GCN-NEXT:    flat_store_dword v[0:1], v2
+; GCN-NEXT:    s_endpgm
   %tid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %scalar = tail call i32 @llvm.amdgcn.readfirstlane(i32 %tid)
   %idx = zext i32 %scalar to i64
   %gep0 = getelementptr inbounds float, float addrspace(1)* %0, i64 %idx
   %val = load float, float addrspace(1)* %gep0, align 4
   %gep1 = getelementptr inbounds float, float addrspace(1)* %1, i64 10
   store float %val, float addrspace(1)* %gep1, align 4
   ret void
 }

llvm/test/CodeGen/AMDGPU/bitreverse.ll

; GISEL-NEXT:    s_endpgm
   %brev = call i16 @llvm.bitreverse.i16(i16 %val) #1
   store i16 %brev, i16 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_brev_i16(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %valptr) #0 {
 ; SI-LABEL: v_brev_i16:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ushort v0, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bfrev_b32_e32 v0, v0
 ; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; FLAT-LABEL: v_brev_i16:
 ; FLAT:       ; %bb.0:
-; FLAT-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; FLAT-NEXT:    s_mov_b32 s3, 0xf000
-; FLAT-NEXT:    s_mov_b32 s2, -1
-; FLAT-NEXT:    s_mov_b32 s6, s2
-; FLAT-NEXT:    s_mov_b32 s7, s3
-; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
-; FLAT-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
-; FLAT-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; FLAT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; FLAT-NEXT:    s_mov_b32 s7, 0xf000
+; FLAT-NEXT:    s_mov_b32 s6, -1
+; FLAT-NEXT:    s_mov_b32 s10, s6
+; FLAT-NEXT:    s_mov_b32 s11, s7
+; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
+; FLAT-NEXT:    s_mov_b32 s8, s2
+; FLAT-NEXT:    s_mov_b32 s9, s3
+; FLAT-NEXT:    buffer_load_ushort v0, off, s[8:11], 0
+; FLAT-NEXT:    s_mov_b32 s4, s0
+; FLAT-NEXT:    s_mov_b32 s5, s1
 ; FLAT-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT-NEXT:    v_bfrev_b32_e32 v0, v0
 ; FLAT-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
-; FLAT-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; FLAT-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; FLAT-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: v_brev_i16:
 ; GISEL:       ; %bb.0:
-; GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    v_mov_b32_e32 v0, s2
 ; GISEL-NEXT:    v_mov_b32_e32 v1, s3
 ; GISEL-NEXT:    flat_load_ushort v0, v[0:1]
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    v_bfrev_b32_e32 v0, v0
 ; GISEL-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
 ; GISEL-NEXT:    v_mov_b32_e32 v0, s0
; GISEL-NEXT:    s_endpgm
   %brev = call i32 @llvm.bitreverse.i32(i32 %val) #1
   store i32 %brev, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_brev_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) #0 {
 ; SI-LABEL: v_brev_i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bfrev_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; FLAT-LABEL: v_brev_i32:
 ; FLAT:       ; %bb.0:
-; FLAT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; FLAT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; FLAT-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; FLAT-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
 ; FLAT-NEXT:    v_mov_b32_e32 v1, s3
 ; FLAT-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; FLAT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; FLAT-NEXT:    flat_load_dword v0, v[0:1]
 ; FLAT-NEXT:    s_mov_b32 s3, 0xf000
 ; FLAT-NEXT:    s_mov_b32 s2, -1
 ; FLAT-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT-NEXT:    v_bfrev_b32_e32 v0, v0
 ; FLAT-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; FLAT-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: v_brev_i32:
 ; GISEL:       ; %bb.0:
-; GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 2, v0
-; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    v_mov_b32_e32 v0, s2
 ; GISEL-NEXT:    v_mov_b32_e32 v1, s3
 ; GISEL-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GISEL-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GISEL-NEXT:    flat_load_dword v0, v[0:1]
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    v_bfrev_b32_e32 v2, v0
; GISEL-NEXT:    s_endpgm
   %brev = call <2 x i32> @llvm.bitreverse.v2i32(<2 x i32> %val) #1
   store <2 x i32> %brev, <2 x i32> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_brev_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrspace(1)* noalias %valptr) #0 {
 ; SI-LABEL: v_brev_v2i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bfrev_b32_e32 v1, v1
 ; SI-NEXT:    v_bfrev_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; FLAT-LABEL: v_brev_v2i32:
 ; FLAT:       ; %bb.0:
-; FLAT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; FLAT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; FLAT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; FLAT-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
 ; FLAT-NEXT:    v_mov_b32_e32 v1, s3
 ; FLAT-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; FLAT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; FLAT-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; FLAT-NEXT:    s_mov_b32 s3, 0xf000
 ; FLAT-NEXT:    s_mov_b32 s2, -1
 ; FLAT-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT-NEXT:    v_bfrev_b32_e32 v1, v1
 ; FLAT-NEXT:    v_bfrev_b32_e32 v0, v0
 ; FLAT-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; FLAT-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: v_brev_v2i32:
 ; GISEL:       ; %bb.0:
-; GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    v_mov_b32_e32 v0, s2
 ; GISEL-NEXT:    v_mov_b32_e32 v1, s3
 ; GISEL-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GISEL-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GISEL-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; GISEL-NEXT:    v_mov_b32_e32 v3, s1
 ; GISEL-NEXT:    v_mov_b32_e32 v2, s0
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    v_bfrev_b32_e32 v0, v0
 ; GISEL-NEXT:    v_bfrev_b32_e32 v1, v1
 ; GISEL-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %valptr, i32 %tid
   %val = load <2 x i32>, <2 x i32> addrspace(1)* %gep
   %brev = call <2 x i32> @llvm.bitreverse.v2i32(<2 x i32> %val) #1
   store <2 x i32> %brev, <2 x i32> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_brev_i64(i64 addrspace(1)* noalias %out, i64 %val) #0 {
 ; SI-LABEL: s_brev_i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_brev_b64 s[4:5], s[4:5]
-; SI-NEXT:    v_mov_b32_e32 v0, s4
-; SI-NEXT:    v_mov_b32_e32 v1, s5
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
+; SI-NEXT:    s_brev_b64 s[0:1], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v0, s0
+; SI-NEXT:    v_mov_b32_e32 v1, s1
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; FLAT-LABEL: s_brev_i64:
 ; FLAT:       ; %bb.0:
-; FLAT-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; FLAT-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; FLAT-NEXT:    s_mov_b32 s3, 0xf000
-; FLAT-NEXT:    s_mov_b32 s2, -1
-; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
-; FLAT-NEXT:    s_brev_b64 s[4:5], s[4:5]
-; FLAT-NEXT:    v_mov_b32_e32 v0, s4
-; FLAT-NEXT:    v_mov_b32_e32 v1, s5
-; FLAT-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; FLAT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; FLAT-NEXT:    s_mov_b32 s7, 0xf000
+; FLAT-NEXT:    s_mov_b32 s6, -1
+; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
+; FLAT-NEXT:    s_mov_b32 s4, s0
+; FLAT-NEXT:    s_mov_b32 s5, s1
+; FLAT-NEXT:    s_brev_b64 s[0:1], s[2:3]
+; FLAT-NEXT:    v_mov_b32_e32 v0, s0
+; FLAT-NEXT:    v_mov_b32_e32 v1, s1
+; FLAT-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; FLAT-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: s_brev_i64:
 ; GISEL:       ; %bb.0:
-; GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    s_brev_b64 s[2:3], s[2:3]
 ; GISEL-NEXT:    v_mov_b32_e32 v0, s2
 ; GISEL-NEXT:    v_mov_b32_e32 v3, s1
 ; GISEL-NEXT:    v_mov_b32_e32 v1, s3
 ; GISEL-NEXT:    v_mov_b32_e32 v2, s0
 ; GISEL-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GISEL-NEXT:    s_endpgm
   %brev = call i64 @llvm.bitreverse.i64(i64 %val) #1
   store i64 %brev, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_brev_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %valptr) #0 {
 ; SI-LABEL: v_brev_i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bfrev_b32_e32 v2, v0
 ; SI-NEXT:    v_bfrev_b32_e32 v1, v1
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[1:2], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx2 v[1:2], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; FLAT-LABEL: v_brev_i64:
 ; FLAT:       ; %bb.0:
-; FLAT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; FLAT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; FLAT-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; FLAT-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
 ; FLAT-NEXT:    v_mov_b32_e32 v1, s3
 ; FLAT-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; FLAT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; FLAT-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; FLAT-NEXT:    s_mov_b32 s3, 0xf000
 ; FLAT-NEXT:    s_mov_b32 s2, -1
 ; FLAT-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT-NEXT:    v_bfrev_b32_e32 v2, v0
 ; FLAT-NEXT:    v_bfrev_b32_e32 v1, v1
 ; FLAT-NEXT:    buffer_store_dwordx2 v[1:2], off, s[0:3], 0
 ; FLAT-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: v_brev_i64:
 ; GISEL:       ; %bb.0:
-; GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    v_mov_b32_e32 v0, s2
 ; GISEL-NEXT:    v_mov_b32_e32 v1, s3
 ; GISEL-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GISEL-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GISEL-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; GISEL-NEXT:    v_mov_b32_e32 v4, s1
 ; GISEL-NEXT:    v_mov_b32_e32 v3, s0
; GISEL-NEXT:    s_endpgm
   %brev = call <2 x i64> @llvm.bitreverse.v2i64(<2 x i64> %val) #1
   store <2 x i64> %brev, <2 x i64> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_brev_v2i64(<2 x i64> addrspace(1)* noalias %out, <2 x i64> addrspace(1)* noalias %valptr) #0 {
 ; SI-LABEL: v_brev_v2i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bfrev_b32_e32 v4, v2
 ; SI-NEXT:    v_bfrev_b32_e32 v3, v3
 ; SI-NEXT:    v_bfrev_b32_e32 v2, v0
 ; SI-NEXT:    v_bfrev_b32_e32 v1, v1
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[1:4], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[1:4], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; FLAT-LABEL: v_brev_v2i64:
 ; FLAT:       ; %bb.0:
-; FLAT-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; FLAT-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; FLAT-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; FLAT-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; FLAT-NEXT:    s_waitcnt lgkmcnt(0)
 ; FLAT-NEXT:    v_mov_b32_e32 v1, s3
 ; FLAT-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; FLAT-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; FLAT-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
 ; FLAT-NEXT:    s_mov_b32 s3, 0xf000
 ; FLAT-NEXT:    s_mov_b32 s2, -1
 ; FLAT-NEXT:    s_waitcnt vmcnt(0)
 ; FLAT-NEXT:    v_bfrev_b32_e32 v4, v2
 ; FLAT-NEXT:    v_bfrev_b32_e32 v3, v3
 ; FLAT-NEXT:    v_bfrev_b32_e32 v2, v0
 ; FLAT-NEXT:    v_bfrev_b32_e32 v1, v1
 ; FLAT-NEXT:    buffer_store_dwordx4 v[1:4], off, s[0:3], 0
 ; FLAT-NEXT:    s_endpgm
 ;
 ; GISEL-LABEL: v_brev_v2i64:
 ; GISEL:       ; %bb.0:
-; GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GISEL-NEXT:    v_lshlrev_b32_e32 v2, 4, v0
-; GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GISEL-NEXT:    v_mov_b32_e32 v0, s2
 ; GISEL-NEXT:    v_mov_b32_e32 v1, s3
 ; GISEL-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; GISEL-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GISEL-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
 ; GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GISEL-NEXT:    v_bfrev_b32_e32 v4, v1

llvm/test/CodeGen/AMDGPU/cluster_stores.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -mcpu=gfx900 -mattr=-xnack -verify-machineinstrs -debug-only=machine-scheduler < %s 2> %t | FileCheck --enable-var-scope --check-prefix=GFX9 %s
 ; RUN: FileCheck --enable-var-scope --check-prefix=DBG %s < %t
 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -debug-only=machine-scheduler < %s 2> %t | FileCheck --enable-var-scope --check-prefix=GFX10 %s
 ; RUN: FileCheck --enable-var-scope --check-prefix=DBG %s < %t
 ; REQUIRES: asserts
 
 ; FIXME: Verifier error with xnack enabled.
 
 ; DBG-LABEL: cluster_load_cluster_store:
 
-; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 8
-; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 8
+; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
 ; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
 ; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
 ; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
 ; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
 
-; DBG: Cluster ld/st SU(1) - SU(2)
+; DBG: Cluster ld/st SU({{[0-9]+}}) - SU({{[0-9]+}})
 
 ; DBG: Cluster ld/st SU([[L1:[0-9]+]]) - SU([[L2:[0-9]+]])
 ; DBG: Cluster ld/st SU([[L2]]) - SU([[L3:[0-9]+]])
-; DBG: Cluster ld/st SU([[L3]]) - SU([[L4:[0-9]+]])
 
 ; DBG-NOT: Cluster ld/st
 
 define amdgpu_kernel void @cluster_load_cluster_store(i32* noalias %lb, i32* noalias %sb) {
 ; GFX9-LABEL: cluster_load_cluster_store:
 ; GFX9:       ; %bb.0: ; %bb
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v0, s2
-; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX9-NEXT:    flat_load_dword v3, v[0:1] offset:8
 ; GFX9-NEXT:    flat_load_dword v4, v[0:1] offset:16
 ; GFX9-NEXT:    flat_load_dword v5, v[0:1] offset:24
-; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    flat_store_dword v[0:1], v2
 ; GFX9-NEXT:    flat_store_dword v[0:1], v3 offset:8
 ; GFX9-NEXT:    flat_store_dword v[0:1], v4 offset:16
 ; GFX9-NEXT:    flat_store_dword v[0:1], v5 offset:24
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: cluster_load_cluster_store:
 ; GFX10:       ; %bb.0: ; %bb
-; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s2, 8
-; GFX10-NEXT:    s_addc_u32 s1, s3, 0
-; GFX10-NEXT:    s_add_u32 s6, s2, 16
-; GFX10-NEXT:    v_mov_b32_e32 v3, s1
-; GFX10-NEXT:    s_addc_u32 s7, s3, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s2
-; GFX10-NEXT:    v_mov_b32_e32 v2, s0
-; GFX10-NEXT:    s_add_u32 s0, s2, 24
-; GFX10-NEXT:    v_mov_b32_e32 v1, s3
-; GFX10-NEXT:    s_addc_u32 s1, s3, 0
+; GFX10-NEXT:    s_add_u32 s4, s0, 8
+; GFX10-NEXT:    s_addc_u32 s5, s1, 0
+; GFX10-NEXT:    s_add_u32 s6, s0, 16
+; GFX10-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-NEXT:    s_addc_u32 s7, s1, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    s_add_u32 s0, s0, 24
+; GFX10-NEXT:    v_mov_b32_e32 v2, s4
+; GFX10-NEXT:    s_addc_u32 s1, s1, 0
+; GFX10-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v5, s7
 ; GFX10-NEXT:    v_mov_b32_e32 v7, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v6, s0
 ; GFX10-NEXT:    s_clause 0x3
 ; GFX10-NEXT:    flat_load_dword v8, v[0:1]
 ; GFX10-NEXT:    flat_load_dword v9, v[2:3]
 ; GFX10-NEXT:    flat_load_dword v10, v[4:5]
 ; GFX10-NEXT:    flat_load_dword v11, v[6:7]
-; GFX10-NEXT:    s_add_u32 s0, s4, 8
-; GFX10-NEXT:    s_addc_u32 s1, s5, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s4
+; GFX10-NEXT:    s_add_u32 s0, s2, 8
+; GFX10-NEXT:    s_addc_u32 s1, s3, 0
+; GFX10-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s0
-; GFX10-NEXT:    s_add_u32 s0, s4, 16
-; GFX10-NEXT:    s_addc_u32 s1, s5, 0
-; GFX10-NEXT:    s_add_u32 s2, s4, 24
-; GFX10-NEXT:    v_mov_b32_e32 v1, s5
-; GFX10-NEXT:    s_addc_u32 s3, s5, 0
+; GFX10-NEXT:    s_add_u32 s0, s2, 16
+; GFX10-NEXT:    s_addc_u32 s1, s3, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-NEXT:    s_add_u32 s2, s2, 24
+; GFX10-NEXT:    s_addc_u32 s3, s3, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s0
 ; GFX10-NEXT:    v_mov_b32_e32 v7, s3
 ; GFX10-NEXT:    v_mov_b32_e32 v6, s2
 ; GFX10-NEXT:    s_waitcnt vmcnt(3) lgkmcnt(3)
 ; GFX10-NEXT:    flat_store_dword v[0:1], v8
 ; GFX10-NEXT:    s_waitcnt vmcnt(2) lgkmcnt(3)
 ; GFX10-NEXT:    flat_store_dword v[2:3], v9
bb:
   %sa3 = getelementptr inbounds i32, i32* %sb, i32 6
   store i32 %ld3, i32* %sa3
 
   ret void
 }
 
 ; DBG-LABEL: cluster_load_valu_cluster_store:
 
-; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 8
-; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 8
+; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 16
 ; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
 ; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
 ; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
 ; DBG: Num BaseOps: {{[1-9]+}}, Offset: {{[0-9]+}}, OffsetIsScalable: {{[01]}}, Width: 4
 
-; DBG: Cluster ld/st SU(1) - SU(2)
+; DBG: Cluster ld/st SU({{[0-9]+}}) - SU({{[0-9]+}})
 
 ; DBG: Cluster ld/st SU([[L1:[0-9]+]]) - SU([[L2:[0-9]+]])
 ; DBG: Cluster ld/st SU([[L2]]) - SU([[L3:[0-9]+]])
-; DBG: Cluster ld/st SU([[L3]]) - SU([[L4:[0-9]+]])
 
 ; DBG-NOT: Cluster ld/st
 
 define amdgpu_kernel void @cluster_load_valu_cluster_store(i32* noalias %lb, i32* noalias %sb) {
 ; GFX9-LABEL: cluster_load_valu_cluster_store:
 ; GFX9:       ; %bb.0: ; %bb
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v0, s2
-; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    flat_load_dword v2, v[0:1]
 ; GFX9-NEXT:    flat_load_dword v3, v[0:1] offset:8
 ; GFX9-NEXT:    flat_load_dword v4, v[0:1] offset:16
 ; GFX9-NEXT:    flat_load_dword v5, v[0:1] offset:24
-; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    flat_store_dword v[0:1], v2
 ; GFX9-NEXT:    v_add_u32_e32 v2, 1, v3
 ; GFX9-NEXT:    flat_store_dword v[0:1], v4 offset:16
 ; GFX9-NEXT:    flat_store_dword v[0:1], v2 offset:8
 ; GFX9-NEXT:    flat_store_dword v[0:1], v5 offset:24
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: cluster_load_valu_cluster_store:
 ; GFX10:       ; %bb.0: ; %bb
-; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_add_u32 s0, s2, 8
-; GFX10-NEXT:    s_addc_u32 s1, s3, 0
-; GFX10-NEXT:    s_add_u32 s6, s2, 16
-; GFX10-NEXT:    v_mov_b32_e32 v3, s1
-; GFX10-NEXT:    v_mov_b32_e32 v2, s0
-; GFX10-NEXT:    s_addc_u32 s7, s3, 0
-; GFX10-NEXT:    s_add_u32 s0, s2, 24
-; GFX10-NEXT:    v_mov_b32_e32 v0, s2
-; GFX10-NEXT:    s_addc_u32 s1, s3, 0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-NEXT:    s_add_u32 s4, s0, 8
+; GFX10-NEXT:    s_addc_u32 s5, s1, 0
+; GFX10-NEXT:    v_mov_b32_e32 v2, s4
+; GFX10-NEXT:    s_add_u32 s6, s0, 16
+; GFX10-NEXT:    v_mov_b32_e32 v3, s5
+; GFX10-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-NEXT:    s_addc_u32 s7, s1, 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s1
+; GFX10-NEXT:    s_add_u32 s0, s0, 24
+; GFX10-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-NEXT:    v_mov_b32_e32 v4, s6
 ; GFX10-NEXT:    v_mov_b32_e32 v5, s7
 ; GFX10-NEXT:    flat_load_dword v6, v[2:3]
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX10-NEXT:    s_clause 0x2
 ; GFX10-NEXT:    flat_load_dword v8, v[0:1]
 ; GFX10-NEXT:    flat_load_dword v9, v[4:5]
 ; GFX10-NEXT:    flat_load_dword v10, v[2:3]
-; GFX10-NEXT:    s_add_u32 s0, s4, 8
-; GFX10-NEXT:    s_addc_u32 s1, s5, 0
-; GFX10-NEXT:    s_add_u32 s2, s4, 16
+; GFX10-NEXT:    s_add_u32 s0, s2, 8
+; GFX10-NEXT:    s_addc_u32 s1, s3, 0
+; GFX10-NEXT:    s_add_u32 s4, s2, 16
 ; GFX10-NEXT:    v_mov_b32_e32 v3, s1
-; GFX10-NEXT:    s_addc_u32 s3, s5, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s4
+; GFX10-NEXT:    s_addc_u32 s5, s3, 0
+; GFX10-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX10-NEXT:    v_mov_b32_e32 v2, s0
-; GFX10-NEXT:    s_add_u32 s0, s4, 24
-; GFX10-NEXT:    v_mov_b32_e32 v1, s5
-; GFX10-NEXT:    v_mov_b32_e32 v5, s3
-; GFX10-NEXT:    s_addc_u32 s1, s5, 0
-; GFX10-NEXT:    v_mov_b32_e32 v4, s2
+; GFX10-NEXT:    s_add_u32 s0, s2, 24
+; GFX10-NEXT:    v_mov_b32_e32 v1, s3
+; GFX10-NEXT:    v_mov_b32_e32 v4, s4
+; GFX10-NEXT:    s_addc_u32 s1, s3, 0
+; GFX10-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX10-NEXT:    s_waitcnt vmcnt(3) lgkmcnt(3)
 ; GFX10-NEXT:    v_add_nc_u32_e32 v11, 1, v6
 ; GFX10-NEXT:    v_mov_b32_e32 v7, s1
 ; GFX10-NEXT:    v_mov_b32_e32 v6, s0
 ; GFX10-NEXT:    s_waitcnt vmcnt(2) lgkmcnt(2)
 ; GFX10-NEXT:    flat_store_dword v[0:1], v8
 ; GFX10-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(2)
 ; GFX10-NEXT:    flat_store_dword v[4:5], v9

llvm/test/CodeGen/AMDGPU/ctlz.ll

; GFX10-GISEL-NEXT:    s_endpgm
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
   store i32 %ctlz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBH_UINT * T0.W, T0.X,
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    32(4.484155e-44), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_ctlz_i32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i32:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep, align 4
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
   store i32 %ctlz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_v2i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v1
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v1, 32, v1
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_v2i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     FFBH_UINT * T0.W, T0.X,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    32(4.484155e-44), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_ctlz_v2i32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_v2i32:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %valptr, i32 %tid
   %val = load <2 x i32>, <2 x i32> addrspace(1)* %in.gep, align 8
   %ctlz = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %val, i1 false) nounwind readnone
   store <2 x i32> %ctlz, <2 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_v4i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v3, v3
 ; SI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v1
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v3, 32, v3
 ; SI-NEXT:    v_min_u32_e32 v2, 32, v2
 ; SI-NEXT:    v_min_u32_e32 v1, 32, v1
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_v4i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     FFBH_UINT * T1.W, T0.X,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    32(4.484155e-44), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_ctlz_v4i32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v3, v3
 ; GFX10-NEXT:    v_ffbh_u32_e32 v2, v2
 ; GFX10-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v3, 32, v3
 ; GFX10-NEXT:    v_min_u32_e32 v2, 32, v2
 ; GFX10-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_v4i32:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v2, v2
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v3, v3
; GFX10-GISEL-NEXT:    s_endpgm
   %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %val, i1 false) nounwind readnone
   store <4 x i32> %ctlz, <4 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i8:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
 ; SI-NEXT:    v_subrev_i32_e32 v0, vcc, 24, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i8:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
-; VI-NEXT:    s_mov_b32 s6, s2
-; VI-NEXT:    s_mov_b32 s7, s3
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
+; VI-NEXT:    s_mov_b32 s10, s6
+; VI-NEXT:    s_mov_b32 s11, s7
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s8, s2
+; VI-NEXT:    s_mov_b32 s9, s3
+; VI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
+; VI-NEXT:    s_mov_b32 s4, s0
+; VI-NEXT:    s_mov_b32 s5, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; VI-NEXT:    v_min_u32_e32 v0, 32, v0
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, -16, v0
 ; VI-NEXT:    v_add_u16_e32 v0, -8, v0
-; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; VI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_i8:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 15, @9, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT MSKOR T0.XW, T1.X

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_ctlz_i8:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-NEXT:    v_add_nc_u32_e32 v1, -16, v1
 ; GFX10-NEXT:    v_add_nc_u16 v1, v1, -8
 ; GFX10-NEXT:    global_store_byte v0, v1, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i8:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-GISEL-NEXT:    v_subrev_nc_u32_e32 v1, 24, v1
 ; GFX10-GISEL-NEXT:    global_store_byte v0, v1, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
; GFX10-GISEL-NEXT:    s_endpgm
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
   store i64 %ctlz, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_ctlz_i64_trunc(i32 addrspace(1)* noalias %out, i64 %val) nounwind {
 ; SI-LABEL: s_ctlz_i64_trunc:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_flbit_i32_b32 s4, s4
-; SI-NEXT:    s_min_u32 s4, s4, 0xffffffdf
-; SI-NEXT:    s_flbit_i32_b32 s5, s5
-; SI-NEXT:    s_add_i32 s4, s4, 32
-; SI-NEXT:    v_mov_b32_e32 v0, s5
-; SI-NEXT:    v_min3_u32 v0, s4, v0, 64
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
+; SI-NEXT:    s_flbit_i32_b32 s0, s2
+; SI-NEXT:    s_min_u32 s0, s0, 0xffffffdf
+; SI-NEXT:    s_flbit_i32_b32 s1, s3
+; SI-NEXT:    s_add_i32 s0, s0, 32
+; SI-NEXT:    v_mov_b32_e32 v0, s1
+; SI-NEXT:    v_min3_u32 v0, s0, v0, 64
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: s_ctlz_i64_trunc:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_flbit_i32_b32 s4, s4
-; VI-NEXT:    v_add_u32_e64 v0, s[6:7], s4, 32 clamp
-; VI-NEXT:    s_flbit_i32_b32 s4, s5
-; VI-NEXT:    v_min3_u32 v0, v0, s4, 64
-; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; VI-NEXT:    s_mov_b32 s4, s0
+; VI-NEXT:    s_flbit_i32_b32 s0, s2
+; VI-NEXT:    s_mov_b32 s5, s1
+; VI-NEXT:    v_add_u32_e64 v0, s[0:1], s0, 32 clamp
+; VI-NEXT:    s_flbit_i32_b32 s0, s3
+; VI-NEXT:    v_min3_u32 v0, v0, s0, 64
+; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: s_ctlz_i64_trunc:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 8, @4, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
 ; EG-NEXT:    CF_END
 ; EG-NEXT:    PAD
 ; EG-NEXT:    ALU clause starting at 4:
 ; EG-NEXT:     FFBH_UINT * T0.W, KC0[2].W,
 ; EG-NEXT:     CNDE_INT * T0.W, KC0[2].W, literal.x, PV.W,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     FFBH_UINT T1.W, KC0[3].X,
 ; EG-NEXT:     ADD_INT * T0.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, KC0[3].X, PS, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: s_ctlz_i64_trunc:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_flbit_i32_b32 s0, s2
-; GFX10-NEXT:    v_add_nc_u32_e64 v0, s0, 32 clamp
-; GFX10-NEXT:    s_flbit_i32_b32 s0, s3
-; GFX10-NEXT:    v_min3_u32 v0, v0, s0, 64
-; GFX10-NEXT:    global_store_dword v1, v0, s[4:5]
+; GFX10-NEXT:    s_flbit_i32_b32 s2, s2
+; GFX10-NEXT:    v_add_nc_u32_e64 v0, s2, 32 clamp
+; GFX10-NEXT:    s_flbit_i32_b32 s2, s3
+; GFX10-NEXT:    v_min3_u32 v0, v0, s2, 64
+; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: s_ctlz_i64_trunc:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_clause 0x1
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-GISEL-NEXT:    s_flbit_i32_b64 s0, s[2:3]
-; GFX10-GISEL-NEXT:    s_min_u32 s0, s0, 64
-; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[4:5]
+; GFX10-GISEL-NEXT:    s_flbit_i32_b64 s2, s[2:3]
+; GFX10-GISEL-NEXT:    s_min_u32 s2, s2, 64
+; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
   %trunc = trunc i64 %ctlz to i32
   store i32 %trunc, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctlz_i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; SI-NEXT:    v_min_u32_e32 v2, 0xffffffdf, v2
 ; SI-NEXT:    v_add_i32_e32 v2, vcc, 32, v2
 ; SI-NEXT:    v_ffbh_u32_e32 v3, v3
 ; SI-NEXT:    v_min3_u32 v2, v2, v3, 64
 ; SI-NEXT:    v_mov_b32_e32 v3, v1
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
+; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i64:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v3, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    v_mov_b32_e32 v2, 0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v3
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v4, s1
 ; VI-NEXT:    v_add_u32_e32 v3, vcc, s0, v3

 ; EG-NEXT:     CNDE_INT T0.X, T0.Y, PS, PV.W,
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     ADD_INT * T0.W, KC0[2].Y, T0.W,
 ; EG-NEXT:     LSHR * T1.X, PV.W, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_ctlz_i64:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-NEXT:    v_add_nc_u32_e64 v0, v0, 32 clamp
 ; GFX10-NEXT:    v_min3_u32 v0, v0, v1, 64
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i64:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_add_nc_u32_e64 v0, v0, 32 clamp
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, v1, v0
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 64, v0
 ; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
   %out.gep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
   %val = load i64, i64 addrspace(1)* %in.gep
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
   store i64 %ctlz, i64 addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_i64_trunc(i32 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctlz_i64_trunc:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v2, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dwordx2 v[3:4], v[1:2], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v3
 ; SI-NEXT:    v_min_u32_e32 v0, 0xffffffdf, v0
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, 32, v0
 ; SI-NEXT:    v_ffbh_u32_e32 v3, v4
 ; SI-NEXT:    v_min3_u32 v0, v0, v3, 64
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, v[1:2], s[4:7], 0 addr64
+; SI-NEXT:    buffer_store_dword v0, v[1:2], s[0:3], 0 addr64
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i64_trunc:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v2, s3
 ; VI-NEXT:    v_add_u32_e32 v1, vcc, s2, v1
 ; VI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[1:2], v[1:2]
 ; VI-NEXT:    v_add_u32_e32 v3, vcc, s0, v0
 ; VI-NEXT:    v_mov_b32_e32 v4, s1

 ; EG-NEXT:    2(2.802597e-45), 32(4.484155e-44)
 ; EG-NEXT:     CNDE_INT T0.X, T1.Y, PS, PV.W,
 ; EG-NEXT:     ADD_INT * T0.W, KC0[2].Y, PV.Z,
 ; EG-NEXT:     LSHR * T1.X, PV.W, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_ctlz_i64_trunc:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx2 v[1:2], v1, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-NEXT:    v_ffbh_u32_e32 v2, v2
 ; GFX10-NEXT:    v_add_nc_u32_e64 v1, v1, 32 clamp
 ; GFX10-NEXT:    v_min3_u32 v1, v1, v2, 64
 ; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i64_trunc:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dwordx2 v[1:2], v1, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v2, v2
 ; GFX10-GISEL-NEXT:    v_add_nc_u32_e64 v1, v1, 32 clamp
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, v2, v1
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 64, v1
 ; GFX10-GISEL-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
   %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
   %val = load i64, i64 addrspace(1)* %in.gep
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
   %trunc = trunc i64 %ctlz to i32
   store i32 %trunc, i32 addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_i32_sel_eq_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i32_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i32_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     CNDE_INT * T0.W, T0.X, literal.x, PV.W,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_ctlz_i32_sel_eq_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i32_sel_eq_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
 ; GFX10-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e64 v0, v1, -1, vcc_lo
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp eq i32 %val, 0
   %sel = select i1 %cmp, i32 -1, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_i32_sel_ne_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i32_sel_ne_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i32_sel_ne_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     CNDE_INT * T0.W, T0.X, literal.x, PV.W,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_ctlz_i32_sel_ne_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i32_sel_ne_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
 ; GFX10-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e32 v0, -1, v1, vcc_lo
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp ne i32 %val, 0
   %sel = select i1 %cmp, i32 %ctlz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; TODO: Should be able to eliminate select here as well.
 define amdgpu_kernel void @v_ctlz_i32_sel_eq_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i32_sel_eq_bitwidth:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 32, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i32_sel_eq_bitwidth:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     SETE_INT * T1.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, PV.W, T0.W, literal.x,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_ctlz_i32_sel_eq_bitwidth:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 32, v0
 ; GFX10-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc_lo
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i32_sel_eq_bitwidth:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 32, v0
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e64 v0, v0, -1, vcc_lo
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp eq i32 %ctlz, 32
   %sel = select i1 %cmp, i32 -1, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_i32_sel_ne_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i32_sel_ne_bitwidth:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 32, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i32_sel_ne_bitwidth:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     SETNE_INT * T1.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, PV.W, literal.x, T0.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_ctlz_i32_sel_ne_bitwidth:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 32, v0
 ; GFX10-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc_lo
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i32_sel_ne_bitwidth:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 32, v0
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc_lo
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp ne i32 %ctlz, 32
   %sel = select i1 %cmp, i32 %ctlz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
  define amdgpu_kernel void @v_ctlz_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i8_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i8_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_ctlz_i8_sel_eq_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    global_store_byte v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i8_sel_eq_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_ashrrev_i32_e32 v3, 31, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX10-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, v1, v0
 ; GFX10-GISEL-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v2, v3, vcc_lo
 ; GFX10-GISEL-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
; GFX10-GISEL-NEXT:    s_endpgm
   %sel = select i1 %cmp, i8 -1, i8 %ctlz
   store i8 %sel, i8 addrspace(1)* %out
   ret void
 }
 
  define amdgpu_kernel void @v_ctlz_i16_sel_eq_neg1(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i16_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ushort v0, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i16_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
-; VI-NEXT:    s_mov_b32 s6, s2
-; VI-NEXT:    s_mov_b32 s7, s3
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
+; VI-NEXT:    s_mov_b32 s10, s6
+; VI-NEXT:    s_mov_b32 s11, s7
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s8, s2
+; VI-NEXT:    s_mov_b32 s9, s3
+; VI-NEXT:    buffer_load_ushort v0, off, s[8:11], 0
 ; VI-NEXT:    v_mov_b32_e32 v1, 0xffff
+; VI-NEXT:    s_mov_b32 s4, s0
+; VI-NEXT:    s_mov_b32 s5, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_ffbh_u32_e32 v2, v0
 ; VI-NEXT:    v_min_u32_e32 v2, 32, v2
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, -16, v2
 ; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; VI-NEXT:    v_cndmask_b32_e32 v0, v1, v2, vcc
-; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; VI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_i16_sel_eq_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 12, @9, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT MSKOR T0.XW, T1.X

 ; EG-NEXT:    65535(9.183409e-41), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_ctlz_i16_sel_eq_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ushort v1, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v2, v1
 ; GFX10-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 0, v1
 ; GFX10-NEXT:    v_min_u32_e32 v2, 32, v2
 ; GFX10-NEXT:    v_add_nc_u32_e32 v2, -16, v2
 ; GFX10-NEXT:    v_cndmask_b32_e32 v1, 0xffff, v2, vcc_lo
 ; GFX10-NEXT:    global_store_short v0, v1, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i16_sel_eq_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_ushort v1, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-GISEL-NEXT:    s_mov_b32 s2, 0xffff
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbh_u32_e32 v2, v1
 ; GFX10-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v2, 32, v2
; GFX10-GISEL-NEXT:    s_endpgm
   store i16 %sel, i16 addrspace(1)* %out
   ret void
 }
 
 ; FIXME: Need to handle non-uniform case for function below (load without gep).
 define amdgpu_kernel void @v_ctlz_i7_sel_eq_neg1(i7 addrspace(1)* noalias %out, i7 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_i7_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; SI-NEXT:    v_and_b32_e32 v0, 0x7f, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_i7_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_ctlz_i7_sel_eq_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX10-NEXT:    v_and_b32_e32 v0, 0x7f, v0
 ; GFX10-NEXT:    global_store_byte v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_ctlz_i7_sel_eq_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_ashrrev_i32_e32 v3, 31, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX10-GISEL-NEXT:    s_movk_i32 s2, 0x7f
 ; GFX10-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, v1, v0
 ; GFX10-GISEL-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v2, v3, vcc_lo
 ; GFX10-GISEL-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/ctlz_zero_undef.ll

; GFX9-GISEL-NEXT:    s_endpgm
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   store i32 %ctlz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v2, v0
+; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 2, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 2, @11, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
 ; EG-NEXT:    CF_END
 ; EG-NEXT:    PAD
 ; EG-NEXT:    Fetch clause starting at 6:
 ; EG-NEXT:     VTX_READ_32 T0.X, T0.X, 0, #1
 ; EG-NEXT:    ALU clause starting at 8:
 ; EG-NEXT:     LSHL * T0.W, T0.X, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ; EG-NEXT:     ADD_INT * T0.X, KC0[2].Z, PV.W,
 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBH_UINT T0.X, T0.X,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i32:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep, align 4
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   store i32 %ctlz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_v2i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v1
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_v2i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v3, s1
 ; VI-NEXT:    v_mov_b32_e32 v2, s0
+; VI-NEXT:    v_mov_b32_e32 v3, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_ffbh_u32_e32 v1, v1
 ; VI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_v2i32:
 ; EG:       ; %bb.0:

 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBH_UINT * T0.Y, T0.Y,
 ; EG-NEXT:     FFBH_UINT T0.X, T0.X,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_v2i32:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX9-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %valptr, i32 %tid
   %val = load <2 x i32>, <2 x i32> addrspace(1)* %in.gep, align 8
   %ctlz = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %val, i1 true) nounwind readnone
   store <2 x i32> %ctlz, <2 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_v4i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v3, v3
 ; SI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v1
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_v4i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v5, s1
 ; VI-NEXT:    v_mov_b32_e32 v4, s0
+; VI-NEXT:    v_mov_b32_e32 v5, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_ffbh_u32_e32 v3, v3
 ; VI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; VI-NEXT:    v_ffbh_u32_e32 v1, v1
 ; VI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; VI-NEXT:    s_endpgm
 ;

 ; EG-NEXT:     FFBH_UINT * T0.Z, T0.Z,
 ; EG-NEXT:     FFBH_UINT * T0.Y, T0.Y,
 ; EG-NEXT:     FFBH_UINT T0.X, T0.X,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_v4i32:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v2, v2
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v3, v3
 ; GFX9-GISEL-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %valptr, i32 %tid
   %val = load <4 x i32>, <4 x i32> addrspace(1)* %in.gep, align 16
   %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %val, i1 true) nounwind readnone
   store <4 x i32> %ctlz, <4 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i8:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
 ; SI-NEXT:    v_subrev_i32_e32 v0, vcc, 24, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i8:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v0, v0
-; VI-NEXT:    v_add_u32_e32 v0, vcc, -16, v0
-; VI-NEXT:    v_add_u16_e32 v2, -8, v0
+; VI-NEXT:    flat_load_ubyte v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v2, v2
+; VI-NEXT:    v_add_u32_e32 v2, vcc, -16, v2
+; VI-NEXT:    v_add_u16_e32 v2, -8, v2
 ; VI-NEXT:    flat_store_byte v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i8:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 14, @9, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i8:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_ashrrev_i32_e32 v3, 31, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX9-GISEL-NEXT:    v_add_co_u32_e32 v0, vcc, v1, v0
 ; GFX9-GISEL-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v3, vcc
 ; GFX9-GISEL-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
; GFX9-GISEL-NEXT:    s_endpgm
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 true)
   store i64 %ctlz, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_ctlz_zero_undef_i64_trunc(i32 addrspace(1)* noalias %out, i64 %val) nounwind {
 ; SI-LABEL: s_ctlz_zero_undef_i64_trunc:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_flbit_i32_b32 s2, s4
-; SI-NEXT:    s_flbit_i32_b32 s4, s5
+; SI-NEXT:    s_flbit_i32_b32 s2, s2
+; SI-NEXT:    s_flbit_i32_b32 s3, s3
 ; SI-NEXT:    s_add_i32 s2, s2, 32
-; SI-NEXT:    s_min_u32 s4, s2, s4
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    v_mov_b32_e32 v0, s4
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    s_min_u32 s2, s2, s3
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
+; SI-NEXT:    v_mov_b32_e32 v0, s2
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: s_ctlz_zero_undef_i64_trunc:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_flbit_i32_b32 s2, s2
-; VI-NEXT:    s_flbit_i32_b32 s3, s3
-; VI-NEXT:    s_add_i32 s2, s2, 32
-; VI-NEXT:    s_min_u32 s2, s2, s3
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    s_flbit_i32_b32 s0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
-; VI-NEXT:    v_mov_b32_e32 v2, s2
+; VI-NEXT:    s_add_i32 s0, s0, 32
+; VI-NEXT:    s_flbit_i32_b32 s1, s3
+; VI-NEXT:    s_min_u32 s0, s0, s1
+; VI-NEXT:    v_mov_b32_e32 v2, s0
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: s_ctlz_zero_undef_i64_trunc:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 6, @4, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
 ; EG-NEXT:    CF_END
 ; EG-NEXT:    PAD
 ; EG-NEXT:    ALU clause starting at 4:
 ; EG-NEXT:     FFBH_UINT * T0.W, KC0[2].W,
 ; EG-NEXT:     FFBH_UINT T1.W, KC0[3].X,
 ; EG-NEXT:     ADD_INT * T0.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, KC0[3].X, PS, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: s_ctlz_zero_undef_i64_trunc:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-GISEL-NEXT:    s_flbit_i32_b64 s0, s[2:3]
-; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[4:5]
+; GFX9-GISEL-NEXT:    s_flbit_i32_b64 s2, s[2:3]
+; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 true)
   %trunc = trunc i64 %ctlz to i32
   store i32 %trunc, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; SI-NEXT:    v_add_i32_e32 v2, vcc, 32, v2
 ; SI-NEXT:    v_ffbh_u32_e32 v3, v3
 ; SI-NEXT:    v_min_u32_e32 v2, v2, v3
 ; SI-NEXT:    v_mov_b32_e32 v3, v1
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
+; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i64:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v3, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    v_mov_b32_e32 v2, 0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v3
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v4, s1
 ; VI-NEXT:    v_add_u32_e32 v3, vcc, s0, v3

 ; EG-NEXT:     CNDE_INT T0.X, T0.Y, PS, PV.W,
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     ADD_INT * T0.W, KC0[2].Y, T0.W,
 ; EG-NEXT:     LSHR * T1.X, PV.W, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i64:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v0, v0
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX9-GISEL-NEXT:    v_add_u32_e32 v0, 32, v0
 ; GFX9-GISEL-NEXT:    v_min_u32_e32 v0, v1, v0
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
   %out.gep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
   %val = load i64, i64 addrspace(1)* %in.gep
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 true)
   store i64 %ctlz, i64 addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_i64_trunc(i32 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i64_trunc:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v2, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dwordx2 v[3:4], v[1:2], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v3
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, 32, v0
 ; SI-NEXT:    v_ffbh_u32_e32 v3, v4
 ; SI-NEXT:    v_min_u32_e32 v0, v0, v3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, v[1:2], s[4:7], 0 addr64
+; SI-NEXT:    buffer_store_dword v0, v[1:2], s[0:3], 0 addr64
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i64_trunc:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v2, s3
 ; VI-NEXT:    v_add_u32_e32 v1, vcc, s2, v1
 ; VI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[1:2], v[1:2]
 ; VI-NEXT:    v_mov_b32_e32 v4, s1
 ; VI-NEXT:    v_add_u32_e32 v3, vcc, s0, v0

 ; EG-NEXT:    2(2.802597e-45), 32(4.484155e-44)
 ; EG-NEXT:     CNDE_INT T0.X, T1.Y, PS, PV.W,
 ; EG-NEXT:     ADD_INT * T0.W, KC0[2].Y, PV.Z,
 ; EG-NEXT:     LSHR * T1.X, PV.W, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i64_trunc:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dwordx2 v[1:2], v1, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v1
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v2, v2
 ; GFX9-GISEL-NEXT:    v_add_u32_e32 v1, 32, v1
 ; GFX9-GISEL-NEXT:    v_min_u32_e32 v1, v2, v1
 ; GFX9-GISEL-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
   %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
   %val = load i64, i64 addrspace(1)* %in.gep
   %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 true)
   %trunc = trunc i64 %ctlz to i32
   store i32 %trunc, i32 addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_i32_sel_eq_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i32_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i32_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v2, v0
+; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i32_sel_eq_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 2, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 3, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBH_UINT * T0.W, T0.X,
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i32_sel_eq_neg1:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
 ; GFX9-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e64 v0, v1, -1, vcc
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp eq i32 %val, 0
   %sel = select i1 %cmp, i32 -1, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_i32_sel_ne_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i32_sel_ne_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i32_sel_ne_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v2, v0
+; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i32_sel_ne_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 2, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 3, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBH_UINT * T0.W, T0.X,
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i32_sel_ne_neg1:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
 ; GFX9-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e32 v0, -1, v1, vcc
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp ne i32 %val, 0
   %sel = select i1 %cmp, i32 %ctlz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i8_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i8_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v2, v0
+; VI-NEXT:    flat_load_ubyte v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v2, v2
 ; VI-NEXT:    flat_store_byte v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i8_sel_eq_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 12, @9, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i8_sel_eq_neg1:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_ashrrev_i32_e32 v3, 31, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX9-GISEL-NEXT:    v_add_co_u32_e32 v0, vcc, v1, v0
 ; GFX9-GISEL-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v3, vcc
 ; GFX9-GISEL-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
; GFX9-GISEL-NEXT:    s_endpgm
   %sel = select i1 %cmp, i8 -1, i8 %ctlz
   store i8 %sel, i8 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_ctlz_zero_undef_i32_sel_eq_neg1_two_use(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i32_sel_eq_neg1_two_use:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v0
 ; SI-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; SI-NEXT:    v_cndmask_b32_e64 v0, 0, 1, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v1, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v1, off, s[4:7], 0
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i32_sel_eq_neg1_two_use:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     FFBH_UINT * T0.W, T0.X,
 ; EG-NEXT:    0(0.000000e+00), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T3.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i32_sel_eq_neg1_two_use:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v2, v0
 ; GFX9-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e64 v0, v2, -1, vcc
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc
; GFX9-GISEL-NEXT:    s_endpgm
   store volatile i1 %cmp, i1 addrspace(1)* undef
   ret void
 }
 
 ; Selected on wrong constant
 define amdgpu_kernel void @v_ctlz_zero_undef_i32_sel_eq_0(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i32_sel_eq_0:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i32_sel_eq_0:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v1, v0
-; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, 0, v1, vcc
+; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v3, v2
+; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, 0, v3, vcc
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i32_sel_eq_0:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 2, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 3, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBH_UINT * T0.W, T0.X,
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, 0.0, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i32_sel_eq_0:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
 ; GFX9-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v0
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e64 v0, v1, 0, vcc
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp eq i32 %val, 0
   %sel = select i1 %cmp, i32 0, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; Selected on wrong constant
 define amdgpu_kernel void @v_ctlz_zero_undef_i32_sel_ne_0(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i32_sel_ne_0:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i32_sel_ne_0:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v1, v0
-; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, 0, v1, vcc
+; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v3, v2
+; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, 0, v3, vcc
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i32_sel_ne_0:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 2, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 3, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBH_UINT * T0.W, T0.X,
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, 0.0, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i32_sel_ne_0:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
 ; GFX9-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp ne i32 %val, 0
   %sel = select i1 %cmp, i32 %ctlz, i32 0
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; Compare on wrong constant
 define amdgpu_kernel void @v_ctlz_zero_undef_i32_sel_eq_cmp_non0(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i32_sel_eq_cmp_non0:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 1, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i32_sel_eq_cmp_non0:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v1, v0
-; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 1, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, 0, v1, vcc
+; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v3, v2
+; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 1, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, 0, v3, vcc
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i32_sel_eq_cmp_non0:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 2, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 4, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:     FFBH_UINT T0.W, T0.X,
 ; EG-NEXT:     SETE_INT * T1.W, T0.X, 1,
 ; EG-NEXT:     CNDE_INT T0.X, PS, PV.W, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i32_sel_eq_cmp_non0:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
 ; GFX9-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v0
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e64 v0, v1, 0, vcc
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone
   %cmp = icmp eq i32 %val, 1
   %sel = select i1 %cmp, i32 0, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; Selected on wrong constant
 define amdgpu_kernel void @v_ctlz_zero_undef_i32_sel_ne_cmp_non0(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_ctlz_zero_undef_i32_sel_ne_cmp_non0:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbh_u32_e32 v1, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 1, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctlz_zero_undef_i32_sel_ne_cmp_non0:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbh_u32_e32 v1, v0
-; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 1, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, 0, v1, vcc
+; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbh_u32_e32 v3, v2
+; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 1, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, 0, v3, vcc
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctlz_zero_undef_i32_sel_ne_cmp_non0:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 2, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 4, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:     FFBH_UINT T0.W, T0.X,
 ; EG-NEXT:     SETNE_INT * T1.W, T0.X, 1,
 ; EG-NEXT:     CNDE_INT T0.X, PS, 0.0, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_ctlz_zero_undef_i32_sel_ne_cmp_non0:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbh_u32_e32 v1, v0
 ; GFX9-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc, 1, v0
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]

llvm/test/CodeGen/AMDGPU/ctpop16.ll

; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   store i16 %ctpop, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 ; XXX - Why 0 in register?
 define amdgpu_kernel void @v_ctpop_i16(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctpop_i16:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, 0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_i16:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %ctpop = call i16 @llvm.ctpop.i16(i16 %val) nounwind readnone
   store i16 %ctpop, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_add_chain_i16(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in0, i16 addrspace(1)* noalias %in1) nounwind {
 ; SI-LABEL: v_ctpop_add_chain_i16:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; SI-NEXT:    s_mov_b32 s11, 0xf000
+; SI-NEXT:    s_mov_b32 s14, 0
+; SI-NEXT:    s_mov_b32 s15, s11
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b64 s[10:11], s[6:7]
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v2, v[0:1], s[4:7], 0 addr64 glc
+; SI-NEXT:    s_mov_b64 s[2:3], s[14:15]
+; SI-NEXT:    buffer_load_ushort v2, v[0:1], s[12:15], 0 addr64 glc
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64 glc
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[0:3], 0 addr64 glc
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s10, -1
+; SI-NEXT:    s_mov_b32 s8, s4
+; SI-NEXT:    s_mov_b32 s9, s5
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e32 v0, v2, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[8:11], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_add_chain_i16:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x34
+; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; VI-NEXT:    v_mov_b32_e32 v1, s7
+; VI-NEXT:    v_add_u32_e32 v0, vcc, s6, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    v_mov_b32_e32 v3, s5
-; VI-NEXT:    v_add_u32_e32 v2, vcc, s4, v2
+; VI-NEXT:    v_mov_b32_e32 v3, s1
+; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v2
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1] glc
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    flat_load_ushort v1, v[2:3] glc
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    v_bcnt_u32_b32 v1, v1, 0
 ; VI-NEXT:    v_bcnt_u32_b32 v0, v0, v1
-; VI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; VI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctpop_add_chain_i16:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 1, @12, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @8
 ; EG-NEXT:    ALU 0, @14, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @10
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %add = add i16 %ctpop0, %ctpop1
   store i16 %add, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_add_sgpr_i16(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in, i16 %sval) nounwind {
 ; SI-LABEL: v_ctpop_add_sgpr_i16:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dword s8, s[0:1], 0xd
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT:    s_load_dword s12, s[0:1], 0xd
 ; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s3
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[8:9], s[6:7]
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
 ; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, s8
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, s12
 ; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_add_sgpr_i16:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dword s4, s[0:1], 0x34
+; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; VI-NEXT:    s_load_dword s0, s[0:1], 0x34
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
+; VI-NEXT:    v_mov_b32_e32 v1, s7
+; VI-NEXT:    v_add_u32_e32 v0, vcc, s6, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_bcnt_u32_b32 v0, v0, s4
-; VI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; VI-NEXT:    v_bcnt_u32_b32 v0, v0, s0
+; VI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctpop_add_sgpr_i16:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 1, @12, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @8
 ; EG-NEXT:    ALU 0, @14, KC0[], KC1[]
 ; EG-NEXT:    TEX 0 @10
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %add = add i16 %ctpop, %sval
   store i16 %add, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_v2i16(<2 x i16> addrspace(1)* noalias %out, <2 x i16> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctpop_v2i16:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v0
 ; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v1, v1, 0
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_v2i16:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %ctpop = call <2 x i16> @llvm.ctpop.v2i16(<2 x i16> %val) nounwind readnone
   store <2 x i16> %ctpop, <2 x i16> addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_v4i16(<4 x i16> addrspace(1)* noalias %out, <4 x i16> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctpop_v4i16:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s4, 0xffff
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s2, 0xffff
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_and_b32_e32 v2, s4, v0
+; SI-NEXT:    v_and_b32_e32 v2, s2, v0
 ; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v3, s4, v1
+; SI-NEXT:    v_and_b32_e32 v3, s2, v1
 ; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v1, v1, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v3, v3, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v2, v2, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; SI-NEXT:    v_or_b32_e32 v1, v3, v1
 ; SI-NEXT:    v_or_b32_e32 v0, v2, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_v4i16:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_mov_b32 s4, 0xffff
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
; EG-NEXT:     MOV * T8.X, T4.X,
   %ctpop = call <4 x i16> @llvm.ctpop.v4i16(<4 x i16> %val) nounwind readnone
   store <4 x i16> %ctpop, <4 x i16> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_v8i16(<8 x i16> addrspace(1)* noalias %out, <8 x i16> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctpop_v8i16:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s3
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s4, 0xffff
+; SI-NEXT:    s_mov_b64 s[8:9], s[6:7]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, 0xffff
 ; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_and_b32_e32 v4, s4, v0
+; SI-NEXT:    v_and_b32_e32 v4, s6, v0
 ; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v5, s4, v1
+; SI-NEXT:    v_and_b32_e32 v5, s6, v1
 ; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; SI-NEXT:    v_and_b32_e32 v6, s4, v2
+; SI-NEXT:    v_and_b32_e32 v6, s6, v2
 ; SI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
-; SI-NEXT:    v_and_b32_e32 v7, s4, v3
+; SI-NEXT:    v_and_b32_e32 v7, s6, v3
 ; SI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v3, v3, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v2, v2, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v1, v1, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v7, v7, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v6, v6, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v5, v5, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v4, v4, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
 ; SI-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; SI-NEXT:    v_or_b32_e32 v3, v7, v3
 ; SI-NEXT:    v_or_b32_e32 v2, v6, v2
 ; SI-NEXT:    v_or_b32_e32 v1, v5, v1
 ; SI-NEXT:    v_or_b32_e32 v0, v4, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_v8i16:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_mov_b32 s4, 0xffff
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
; EG-NEXT:     MOV * T0.Z, T8.X,
   %ctpop = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %val) nounwind readnone
   store <8 x i16> %ctpop, <8 x i16> addrspace(1)* %out, align 32
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_v16i16(<16 x i16> addrspace(1)* noalias %out, <16 x i16> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctpop_v16i16:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s3
 ; SI-NEXT:    v_lshlrev_b32_e32 v4, 5, v0
-; SI-NEXT:    v_mov_b32_e32 v5, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[4:5], s[4:7], 0 addr64 offset:16
-; SI-NEXT:    buffer_load_dwordx4 v[4:7], v[4:5], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s4, 0xffff
+; SI-NEXT:    s_mov_b64 s[8:9], s[6:7]
+; SI-NEXT:    v_mov_b32_e32 v5, 0
+; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[4:5], s[8:11], 0 addr64 offset:16
+; SI-NEXT:    buffer_load_dwordx4 v[4:7], v[4:5], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, 0xffff
 ; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
 ; SI-NEXT:    s_waitcnt vmcnt(1)
-; SI-NEXT:    v_and_b32_e32 v8, s4, v0
+; SI-NEXT:    v_and_b32_e32 v8, s6, v0
 ; SI-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v9, s4, v1
+; SI-NEXT:    v_and_b32_e32 v9, s6, v1
 ; SI-NEXT:    v_lshrrev_b32_e32 v1, 16, v1
-; SI-NEXT:    v_and_b32_e32 v10, s4, v2
+; SI-NEXT:    v_and_b32_e32 v10, s6, v2
 ; SI-NEXT:    v_lshrrev_b32_e32 v2, 16, v2
-; SI-NEXT:    v_and_b32_e32 v11, s4, v3
+; SI-NEXT:    v_and_b32_e32 v11, s6, v3
 ; SI-NEXT:    v_lshrrev_b32_e32 v3, 16, v3
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_and_b32_e32 v12, s4, v4
+; SI-NEXT:    v_and_b32_e32 v12, s6, v4
 ; SI-NEXT:    v_lshrrev_b32_e32 v4, 16, v4
-; SI-NEXT:    v_and_b32_e32 v13, s4, v5
+; SI-NEXT:    v_and_b32_e32 v13, s6, v5
 ; SI-NEXT:    v_lshrrev_b32_e32 v5, 16, v5
-; SI-NEXT:    v_and_b32_e32 v14, s4, v6
+; SI-NEXT:    v_and_b32_e32 v14, s6, v6
 ; SI-NEXT:    v_lshrrev_b32_e32 v6, 16, v6
-; SI-NEXT:    v_and_b32_e32 v15, s4, v7
+; SI-NEXT:    v_and_b32_e32 v15, s6, v7
 ; SI-NEXT:    v_lshrrev_b32_e32 v7, 16, v7
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v7, v7, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v6, v6, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v5, v5, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v4, v4, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v3, v3, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v2, v2, 0
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v1, v1, 0

 ; SI-NEXT:    v_or_b32_e32 v3, v15, v7
 ; SI-NEXT:    v_or_b32_e32 v2, v14, v6
 ; SI-NEXT:    v_or_b32_e32 v1, v13, v5
 ; SI-NEXT:    v_or_b32_e32 v0, v12, v4
 ; SI-NEXT:    v_or_b32_e32 v7, v11, v16
 ; SI-NEXT:    v_or_b32_e32 v6, v10, v17
 ; SI-NEXT:    v_or_b32_e32 v5, v9, v18
 ; SI-NEXT:    v_or_b32_e32 v4, v8, v19
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_v16i16:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 5, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_mov_b32 s4, 0xffff
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx4 v[0:3], v[4:5]
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 16, v4
 ; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v5, vcc
; EG-NEXT:     MOV * T20.Z, T16.X, BS:VEC_120/SCL_212
   %ctpop = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %val) nounwind readnone
   store <16 x i16> %ctpop, <16 x i16> addrspace(1)* %out, align 32
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_i16_add_inline_constant(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctpop_i16_add_inline_constant:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, 4
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_i16_add_inline_constant:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %add = add i16 %ctpop, 4
   store i16 %add, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_i16_add_inline_constant_inv(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctpop_i16_add_inline_constant_inv:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, 4
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_i16_add_inline_constant_inv:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %add = add i16 4, %ctpop
   store i16 %add, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_i16_add_literal(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_ctpop_i16_add_literal:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_movk_i32 s4, 0x3e7
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_movk_i32 s0, 0x3e7
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, s4
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, s0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_i16_add_literal:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_movk_i32 s4, 0x3e7
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %add = add i16 %ctpop, 999
   store i16 %add, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_i16_add_var(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in, i16 %const) nounwind {
 ; SI-LABEL: v_ctpop_i16_add_var:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dword s8, s[0:1], 0xd
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT:    s_load_dword s12, s[0:1], 0xd
 ; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s3
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[8:9], s[6:7]
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
 ; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, s8
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, s12
 ; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_i16_add_var:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dword s4, s[0:1], 0x34
+; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; VI-NEXT:    s_load_dword s0, s[0:1], 0x34
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
+; VI-NEXT:    v_mov_b32_e32 v1, s7
+; VI-NEXT:    v_add_u32_e32 v0, vcc, s6, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_bcnt_u32_b32 v0, v0, s4
-; VI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; VI-NEXT:    v_bcnt_u32_b32 v0, v0, s0
+; VI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctpop_i16_add_var:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 1, @12, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @8
 ; EG-NEXT:    ALU 0, @14, KC0[], KC1[]
 ; EG-NEXT:    TEX 0 @10
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %add = add i16 %ctpop, %const
   store i16 %add, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_i16_add_var_inv(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in, i16 %const) nounwind {
 ; SI-LABEL: v_ctpop_i16_add_var_inv:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dword s8, s[0:1], 0xd
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT:    s_load_dword s12, s[0:1], 0xd
 ; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s3
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[8:9], s[6:7]
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
 ; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, s8
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    v_bcnt_u32_b32_e64 v0, v0, s12
 ; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_i16_add_var_inv:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dword s4, s[0:1], 0x34
+; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; VI-NEXT:    s_load_dword s0, s[0:1], 0x34
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
+; VI-NEXT:    v_mov_b32_e32 v1, s7
+; VI-NEXT:    v_add_u32_e32 v0, vcc, s6, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_bcnt_u32_b32 v0, v0, s4
-; VI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; VI-NEXT:    v_bcnt_u32_b32 v0, v0, s0
+; VI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctpop_i16_add_var_inv:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 1, @12, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @8
 ; EG-NEXT:    ALU 0, @14, KC0[], KC1[]
 ; EG-NEXT:    TEX 0 @10
; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
   %add = add i16 %const, %ctpop
   store i16 %add, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_ctpop_i16_add_vvar_inv(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %in, i16 addrspace(1)* noalias %constptr) nounwind {
 ; SI-LABEL: v_ctpop_i16_add_vvar_inv:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; SI-NEXT:    s_mov_b32 s11, 0xf000
+; SI-NEXT:    s_mov_b32 s14, 0
+; SI-NEXT:    s_mov_b32 s15, s11
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[12:13], s[6:7]
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b64 s[10:11], s[6:7]
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v2, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[2:3], s[14:15]
+; SI-NEXT:    buffer_load_ushort v2, v[0:1], s[12:15], 0 addr64
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[0:3], 0 addr64
+; SI-NEXT:    s_mov_b32 s10, -1
+; SI-NEXT:    s_mov_b32 s8, s4
+; SI-NEXT:    s_mov_b32 s9, s5
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_bcnt_u32_b32_e32 v0, v2, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[8:11], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_ctpop_i16_add_vvar_inv:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x34
+; VI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
 ; VI-NEXT:    v_lshlrev_b32_e32 v2, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v2
+; VI-NEXT:    v_mov_b32_e32 v1, s7
+; VI-NEXT:    v_add_u32_e32 v0, vcc, s6, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v3, v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v1, s5
-; VI-NEXT:    v_add_u32_e32 v0, vcc, s4, v2
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_bcnt_u32_b32 v0, v3, v0
-; VI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; VI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_ctpop_i16_add_vvar_inv:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 1, @12, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @8
 ; EG-NEXT:    ALU 0, @14, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @10

llvm/test/CodeGen/AMDGPU/cttz.ll

; GFX10-GISEL-NEXT:    s_endpgm
   %cttz = call i32 @llvm.cttz.i32(i32 %val, i1 false) nounwind readnone
   store i32 %cttz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBL_INT * T0.W, T0.X,
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    32(4.484155e-44), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_cttz_i32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i32:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep, align 4
   %cttz = call i32 @llvm.cttz.i32(i32 %val, i1 false) nounwind readnone
   store i32 %cttz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_v2i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v1, v1
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v1, 32, v1
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_v2i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     FFBL_INT * T0.W, T0.X,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    32(4.484155e-44), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_cttz_v2i32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_v2i32:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %valptr, i32 %tid
   %val = load <2 x i32>, <2 x i32> addrspace(1)* %in.gep, align 8
   %cttz = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %val, i1 false) nounwind readnone
   store <2 x i32> %cttz, <2 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_v4i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v3, v3
 ; SI-NEXT:    v_ffbl_b32_e32 v2, v2
 ; SI-NEXT:    v_ffbl_b32_e32 v1, v1
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v3, 32, v3
 ; SI-NEXT:    v_min_u32_e32 v2, 32, v2
 ; SI-NEXT:    v_min_u32_e32 v1, 32, v1
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_v4i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     FFBL_INT * T1.W, T0.X,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    32(4.484155e-44), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_cttz_v4i32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v3, v3
 ; GFX10-NEXT:    v_ffbl_b32_e32 v2, v2
 ; GFX10-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v3, 32, v3
 ; GFX10-NEXT:    v_min_u32_e32 v2, 32, v2
 ; GFX10-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_v4i32:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v2, v2
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v3, v3
; GFX10-GISEL-NEXT:    s_endpgm
   %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %val, i1 false) nounwind readnone
   store <4 x i32> %cttz, <4 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i8:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v0, 0x100, v0
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i8:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
-; VI-NEXT:    s_mov_b32 s6, s2
-; VI-NEXT:    s_mov_b32 s7, s3
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
+; VI-NEXT:    s_mov_b32 s10, s6
+; VI-NEXT:    s_mov_b32 s11, s7
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s8, s2
+; VI-NEXT:    s_mov_b32 s9, s3
+; VI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
+; VI-NEXT:    s_mov_b32 s4, s0
+; VI-NEXT:    s_mov_b32 s5, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_or_b32_e32 v0, 0x100, v0
 ; VI-NEXT:    v_ffbl_b32_e32 v0, v0
-; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; VI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_i8:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 14, @9, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT MSKOR T0.XW, T1.X

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_cttz_i8:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_or_b32_e32 v1, 0x100, v1
 ; GFX10-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-NEXT:    global_store_byte v0, v1, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i8:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_or_b32_e32 v1, 0x100, v1
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    global_store_byte v0, v1, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %val = load i8, i8 addrspace(1)* %valptr
; GFX10-GISEL-NEXT:    s_endpgm
   %cttz = call i64 @llvm.cttz.i64(i64 %val, i1 false)
   store i64 %cttz, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_cttz_i64_trunc(i32 addrspace(1)* noalias %out, i64 %val) nounwind {
 ; SI-LABEL: s_cttz_i64_trunc:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_ff1_i32_b32 s5, s5
-; SI-NEXT:    s_min_u32 s5, s5, 0xffffffdf
-; SI-NEXT:    s_add_i32 s5, s5, 32
-; SI-NEXT:    s_ff1_i32_b32 s4, s4
-; SI-NEXT:    v_mov_b32_e32 v0, s5
-; SI-NEXT:    v_min3_u32 v0, s4, v0, 64
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
+; SI-NEXT:    s_ff1_i32_b32 s0, s3
+; SI-NEXT:    s_min_u32 s0, s0, 0xffffffdf
+; SI-NEXT:    s_add_i32 s0, s0, 32
+; SI-NEXT:    s_ff1_i32_b32 s1, s2
+; SI-NEXT:    v_mov_b32_e32 v0, s0
+; SI-NEXT:    v_min3_u32 v0, s1, v0, 64
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: s_cttz_i64_trunc:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_ff1_i32_b32 s5, s5
-; VI-NEXT:    v_add_u32_e64 v0, s[6:7], s5, 32 clamp
-; VI-NEXT:    s_ff1_i32_b32 s4, s4
-; VI-NEXT:    v_min3_u32 v0, s4, v0, 64
-; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; VI-NEXT:    s_mov_b32 s4, s0
+; VI-NEXT:    s_ff1_i32_b32 s0, s3
+; VI-NEXT:    s_mov_b32 s5, s1
+; VI-NEXT:    v_add_u32_e64 v0, s[0:1], s0, 32 clamp
+; VI-NEXT:    s_ff1_i32_b32 s0, s2
+; VI-NEXT:    v_min3_u32 v0, s0, v0, 64
+; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: s_cttz_i64_trunc:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 8, @4, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
 ; EG-NEXT:    CF_END
 ; EG-NEXT:    PAD
 ; EG-NEXT:    ALU clause starting at 4:
 ; EG-NEXT:     FFBL_INT * T0.W, KC0[3].X,
 ; EG-NEXT:     CNDE_INT * T0.W, KC0[3].X, literal.x, PV.W,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     FFBL_INT T1.W, KC0[2].W,
 ; EG-NEXT:     ADD_INT * T0.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, KC0[2].W, PS, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: s_cttz_i64_trunc:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_clause 0x1
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_ff1_i32_b32 s0, s3
-; GFX10-NEXT:    v_add_nc_u32_e64 v0, s0, 32 clamp
-; GFX10-NEXT:    s_ff1_i32_b32 s0, s2
-; GFX10-NEXT:    v_min3_u32 v0, s0, v0, 64
-; GFX10-NEXT:    global_store_dword v1, v0, s[4:5]
+; GFX10-NEXT:    s_ff1_i32_b32 s3, s3
+; GFX10-NEXT:    s_ff1_i32_b32 s2, s2
+; GFX10-NEXT:    v_add_nc_u32_e64 v0, s3, 32 clamp
+; GFX10-NEXT:    v_min3_u32 v0, s2, v0, 64
+; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: s_cttz_i64_trunc:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_clause 0x1
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-GISEL-NEXT:    s_ff1_i32_b64 s0, s[2:3]
-; GFX10-GISEL-NEXT:    s_min_u32 s0, s0, 64
-; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s0
-; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[4:5]
+; GFX10-GISEL-NEXT:    s_ff1_i32_b64 s2, s[2:3]
+; GFX10-GISEL-NEXT:    s_min_u32 s2, s2, 64
+; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, s2
+; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %cttz = call i64 @llvm.cttz.i64(i64 %val, i1 false)
   %trunc = trunc i64 %cttz to i32
   store i32 %trunc, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_cttz_i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v3, v3
 ; SI-NEXT:    v_min_u32_e32 v3, 0xffffffdf, v3
 ; SI-NEXT:    v_add_i32_e32 v3, vcc, 32, v3
 ; SI-NEXT:    v_ffbl_b32_e32 v2, v2
 ; SI-NEXT:    v_min3_u32 v2, v2, v3, 64
 ; SI-NEXT:    v_mov_b32_e32 v3, v1
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[4:7], 0 addr64
+; SI-NEXT:    buffer_store_dwordx2 v[2:3], v[0:1], s[0:3], 0 addr64
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i64:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v3, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    v_mov_b32_e32 v2, 0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v3
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v4, s1
 ; VI-NEXT:    v_add_u32_e32 v3, vcc, s0, v3

 ; EG-NEXT:     CNDE_INT T0.X, T0.X, PS, PV.W,
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     ADD_INT * T0.W, KC0[2].Y, T0.W,
 ; EG-NEXT:     LSHR * T1.X, PV.W, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_cttz_i64:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    v_add_nc_u32_e64 v1, v1, 32 clamp
 ; GFX10-NEXT:    v_min3_u32 v0, v0, v1, 64
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i64:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v2, 3, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_add_nc_u32_e64 v1, v1, 32 clamp
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, v0, v1
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 64, v0
 ; GFX10-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
   %out.gep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
   %val = load i64, i64 addrspace(1)* %in.gep
   %cttz = call i64 @llvm.cttz.i64(i64 %val, i1 false)
   store i64 %cttz, i64 addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i64_trunc(i32 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v_cttz_i64_trunc:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v2, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b64 s[4:5], s[2:3]
 ; SI-NEXT:    buffer_load_dwordx2 v[3:4], v[1:2], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 2, v0
+; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v4
 ; SI-NEXT:    v_min_u32_e32 v0, 0xffffffdf, v0
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, 32, v0
 ; SI-NEXT:    v_ffbl_b32_e32 v3, v3
 ; SI-NEXT:    v_min3_u32 v0, v3, v0, 64
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, v[1:2], s[4:7], 0 addr64
+; SI-NEXT:    buffer_store_dword v0, v[1:2], s[0:3], 0 addr64
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i64_trunc:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v2, s3
 ; VI-NEXT:    v_add_u32_e32 v1, vcc, s2, v1
 ; VI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[1:2], v[1:2]
 ; VI-NEXT:    v_add_u32_e32 v3, vcc, s0, v0
 ; VI-NEXT:    v_mov_b32_e32 v4, s1

 ; EG-NEXT:    2(2.802597e-45), 32(4.484155e-44)
 ; EG-NEXT:     CNDE_INT T0.X, T1.X, PS, PV.W,
 ; EG-NEXT:     ADD_INT * T0.W, KC0[2].Y, PV.Z,
 ; EG-NEXT:     LSHR * T1.X, PV.W, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_cttz_i64_trunc:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx2 v[1:2], v1, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v2, v2
 ; GFX10-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-NEXT:    v_add_nc_u32_e64 v2, v2, 32 clamp
 ; GFX10-NEXT:    v_min3_u32 v1, v1, v2, 64
 ; GFX10-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i64_trunc:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v1, 3, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dwordx2 v[1:2], v1, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v2, v2
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX10-GISEL-NEXT:    v_add_nc_u32_e64 v2, v2, 32 clamp
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, v1, v2
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 64, v1
 ; GFX10-GISEL-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
   %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
   %val = load i64, i64 addrspace(1)* %in.gep
   %cttz = call i64 @llvm.cttz.i64(i64 %val, i1 false)
   %trunc = trunc i64 %cttz to i32
   store i32 %trunc, i32 addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i32_sel_eq_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i32_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i32_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     CNDE_INT * T0.W, T0.X, literal.x, PV.W,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_cttz_i32_sel_eq_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i32_sel_eq_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v1, v0
 ; GFX10-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e64 v0, v1, -1, vcc_lo
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %cttz = call i32 @llvm.cttz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp eq i32 %val, 0
   %sel = select i1 %cmp, i32 -1, i32 %cttz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i32_sel_ne_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i32_sel_ne_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i32_sel_ne_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     CNDE_INT * T0.W, T0.X, literal.x, PV.W,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.X, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_cttz_i32_sel_ne_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i32_sel_ne_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v1, v0
 ; GFX10-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v1, 32, v1
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e32 v0, -1, v1, vcc_lo
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %cttz = call i32 @llvm.cttz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp ne i32 %val, 0
   %sel = select i1 %cmp, i32 %cttz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; TODO: Should be able to eliminate select here as well.
 define amdgpu_kernel void @v_cttz_i32_sel_eq_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i32_sel_eq_bitwidth:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 32, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i32_sel_eq_bitwidth:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     SETE_INT * T1.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, PV.W, T0.W, literal.x,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_cttz_i32_sel_eq_bitwidth:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 32, v0
 ; GFX10-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc_lo
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i32_sel_eq_bitwidth:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 32, v0
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e64 v0, v0, -1, vcc_lo
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %cttz = call i32 @llvm.cttz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp eq i32 %cttz, 32
   %sel = select i1 %cmp, i32 -1, i32 %cttz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i32_sel_ne_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i32_sel_ne_bitwidth:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 32, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i32_sel_ne_bitwidth:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:     SETNE_INT * T1.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, PV.W, literal.x, T0.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX10-LABEL: v_cttz_i32_sel_ne_bitwidth:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 32, v0
 ; GFX10-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc_lo
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i32_sel_ne_bitwidth:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-GISEL-NEXT:    v_min_u32_e32 v0, 32, v0
 ; GFX10-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc_lo, 32, v0
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc_lo
 ; GFX10-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep
   %cttz = call i32 @llvm.cttz.i32(i32 %val, i1 false) nounwind readnone
   %cmp = icmp ne i32 %cttz, 32
   %sel = select i1 %cmp, i32 %cttz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
  define amdgpu_kernel void @v_cttz_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i8_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i8_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_cttz_i8_sel_eq_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    global_store_byte v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i8_sel_eq_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_ashrrev_i32_e32 v3, 31, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX10-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, v1, v0
 ; GFX10-GISEL-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v2, v3, vcc_lo
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10-GISEL-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
; GFX10-GISEL-NEXT:    s_endpgm
   %sel = select i1 %cmp, i8 -1, i8 %cttz
   store i8 %sel, i8 addrspace(1)* %out
   ret void
 }
 
  define amdgpu_kernel void @v_cttz_i16_sel_eq_neg1(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i16_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ushort v0, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i16_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
-; VI-NEXT:    s_mov_b32 s6, s2
-; VI-NEXT:    s_mov_b32 s7, s3
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
+; VI-NEXT:    s_mov_b32 s10, s6
+; VI-NEXT:    s_mov_b32 s11, s7
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_load_ushort v0, off, s[4:7], 0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_mov_b32 s8, s2
+; VI-NEXT:    s_mov_b32 s9, s3
+; VI-NEXT:    buffer_load_ushort v0, off, s[8:11], 0
 ; VI-NEXT:    v_mov_b32_e32 v1, 0xffff
+; VI-NEXT:    s_mov_b32 s4, s0
+; VI-NEXT:    s_mov_b32 s5, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_or_b32_e32 v2, 0x10000, v0
 ; VI-NEXT:    v_ffbl_b32_e32 v2, v2
 ; VI-NEXT:    v_min_u32_e32 v2, 32, v2
 ; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v0
 ; VI-NEXT:    v_cndmask_b32_e32 v0, v1, v2, vcc
-; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; VI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_i16_sel_eq_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 12, @9, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT MSKOR T0.XW, T1.X

 ; EG-NEXT:    65535(9.183409e-41), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_cttz_i16_sel_eq_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ushort v1, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_or_b32_e32 v2, 0x10000, v1
 ; GFX10-NEXT:    v_cmp_ne_u16_e32 vcc_lo, 0, v1
 ; GFX10-NEXT:    v_ffbl_b32_e32 v2, v2
 ; GFX10-NEXT:    v_min_u32_e32 v2, 32, v2
 ; GFX10-NEXT:    v_cndmask_b32_e32 v1, 0xffff, v2, vcc_lo
 ; GFX10-NEXT:    global_store_short v0, v1, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i16_sel_eq_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    global_load_ushort v1, v0, s[2:3]
 ; GFX10-GISEL-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-GISEL-NEXT:    s_mov_b32 s2, 0xffff
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-GISEL-NEXT:    v_or_b32_e32 v2, 0x10000, v1
 ; GFX10-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc_lo, 0, v1
 ; GFX10-GISEL-NEXT:    v_ffbl_b32_e32 v2, v2
 ; GFX10-GISEL-NEXT:    v_and_b32_e32 v2, s2, v2
 ; GFX10-GISEL-NEXT:    v_cndmask_b32_e64 v1, v2, s2, vcc_lo
 ; GFX10-GISEL-NEXT:    global_store_short v0, v1, s[0:1]
 ; GFX10-GISEL-NEXT:    s_endpgm
   %val = load i16, i16 addrspace(1)* %valptr
   %cttz = call i16 @llvm.cttz.i16(i16 %val, i1 false) nounwind readnone
   %cmp = icmp eq i16 %val, 0
   %sel = select i1 %cmp, i16 -1, i16 %cttz
   store i16 %sel, i16 addrspace(1)* %out
   ret void
 }
 
 ; FIXME: Need to handle non-uniform case for function below (load without gep).
 define amdgpu_kernel void @v_cttz_i7_sel_eq_neg1(i7 addrspace(1)* noalias %out, i7 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_i7_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; SI-NEXT:    v_and_b32_e32 v0, 0x7f, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i7_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX10-LABEL: v_cttz_i7_sel_eq_neg1:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX10-NEXT:    v_and_b32_e32 v0, 0x7f, v0
 ; GFX10-NEXT:    global_store_byte v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX10-GISEL-LABEL: v_cttz_i7_sel_eq_neg1:
 ; GFX10-GISEL:       ; %bb.0:
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    v_ashrrev_i32_e32 v3, 31, v0
-; GFX10-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX10-GISEL-NEXT:    v_mov_b32_e32 v2, s3
 ; GFX10-GISEL-NEXT:    s_movk_i32 s2, 0x7f
 ; GFX10-GISEL-NEXT:    v_add_co_u32 v0, vcc_lo, v1, v0
 ; GFX10-GISEL-NEXT:    v_add_co_ci_u32_e32 v1, vcc_lo, v2, v3, vcc_lo
 ; GFX10-GISEL-NEXT:    global_load_ubyte v0, v[0:1], off
 ; GFX10-GISEL-NEXT:    s_waitcnt vmcnt(0)

llvm/test/CodeGen/AMDGPU/cttz_zero_undef.ll

; GFX9-GISEL-NEXT:    s_endpgm
   %cttz = call i32 @llvm.cttz.i32(i32 %val, i1 true) nounwind readnone
   store i32 %cttz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_zero_undef_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_zero_undef_i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_zero_undef_i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_dword v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbl_b32_e32 v2, v0
+; VI-NEXT:    flat_load_dword v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbl_b32_e32 v2, v2
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_zero_undef_i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 2, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 2, @11, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
 ; EG-NEXT:    CF_END
 ; EG-NEXT:    PAD
 ; EG-NEXT:    Fetch clause starting at 6:
 ; EG-NEXT:     VTX_READ_32 T0.X, T0.X, 0, #1
 ; EG-NEXT:    ALU clause starting at 8:
 ; EG-NEXT:     LSHL * T0.W, T0.X, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ; EG-NEXT:     ADD_INT * T0.X, KC0[2].Z, PV.W,
 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBL_INT T0.X, T0.X,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_zero_undef_i32:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX9-GISEL-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr i32, i32 addrspace(1)* %valptr, i32 %tid
   %val = load i32, i32 addrspace(1)* %in.gep, align 4
   %cttz = call i32 @llvm.cttz.i32(i32 %val, i1 true) nounwind readnone
   store i32 %cttz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_zero_undef_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_zero_undef_v2i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v1, v1
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_zero_undef_v2i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[0:1], v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v3, s1
 ; VI-NEXT:    v_mov_b32_e32 v2, s0
+; VI-NEXT:    v_mov_b32_e32 v3, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_ffbl_b32_e32 v1, v1
 ; VI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_zero_undef_v2i32:
 ; EG:       ; %bb.0:

 ; EG-NEXT:    ALU clause starting at 11:
 ; EG-NEXT:     FFBL_INT * T0.Y, T0.Y,
 ; EG-NEXT:     FFBL_INT T0.X, T0.X,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_zero_undef_v2i32:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dwordx2 v[0:1], v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX9-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %in.gep = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %valptr, i32 %tid
   %val = load <2 x i32>, <2 x i32> addrspace(1)* %in.gep, align 8
   %cttz = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %val, i1 true) nounwind readnone
   store <2 x i32> %cttz, <2 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_zero_undef_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %valptr) nounwind {
 ; SI-LABEL: v_cttz_zero_undef_v4i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_dwordx4 v[0:3], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v3, v3
 ; SI-NEXT:    v_ffbl_b32_e32 v2, v2
 ; SI-NEXT:    v_ffbl_b32_e32 v1, v1
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_zero_undef_v4i32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx4 v[0:3], v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v5, s1
 ; VI-NEXT:    v_mov_b32_e32 v4, s0
+; VI-NEXT:    v_mov_b32_e32 v5, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_ffbl_b32_e32 v3, v3
 ; VI-NEXT:    v_ffbl_b32_e32 v2, v2
 ; VI-NEXT:    v_ffbl_b32_e32 v1, v1
 ; VI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; VI-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; VI-NEXT:    s_endpgm
 ;

 ; EG-NEXT:     FFBL_INT * T0.Z, T0.Z,
 ; EG-NEXT:     FFBL_INT * T0.Y, T0.Y,
 ; EG-NEXT:     FFBL_INT T0.X, T0.X,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_zero_undef_v4i32:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_lshlrev_b32_e32 v0, 4, v0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_dwordx4 v[0:3], v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v0, v0
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v1, v1
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v2, v2
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v3, v3
; GFX9-GISEL-NEXT:    s_endpgm
   %ret = select i1 %cttz_ret, i32 %cttz, i32 32
   store i32 %cttz, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @s_cttz_zero_undef_i64_with_select(i64 addrspace(1)* noalias %out, i64 %val) nounwind {
 ; SI-LABEL: s_cttz_zero_undef_i64_with_select:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_ff1_i32_b32 s5, s5
-; SI-NEXT:    s_ff1_i32_b32 s4, s4
-; SI-NEXT:    s_add_i32 s5, s5, 32
-; SI-NEXT:    s_min_u32 s4, s4, s5
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_ff1_i32_b32 s3, s3
+; SI-NEXT:    s_ff1_i32_b32 s2, s2
+; SI-NEXT:    s_add_i32 s3, s3, 32
+; SI-NEXT:    s_min_u32 s2, s2, s3
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s4
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
+; SI-NEXT:    v_mov_b32_e32 v0, s2
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: s_cttz_zero_undef_i64_with_select:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; VI-NEXT:    v_mov_b32_e32 v1, 0
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    v_mov_b32_e32 v3, 0
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_ff1_i32_b32 s3, s3
-; VI-NEXT:    s_ff1_i32_b32 s2, s2
-; VI-NEXT:    s_add_i32 s3, s3, 32
-; VI-NEXT:    s_min_u32 s2, s2, s3
-; VI-NEXT:    v_mov_b32_e32 v3, s1
-; VI-NEXT:    v_mov_b32_e32 v0, s2
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    s_ff1_i32_b32 s0, s3
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_ff1_i32_b32 s1, s2
+; VI-NEXT:    s_add_i32 s0, s0, 32
+; VI-NEXT:    s_min_u32 s0, s1, s0
 ; VI-NEXT:    v_mov_b32_e32 v2, s0
-; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
+; VI-NEXT:    flat_store_dwordx2 v[0:1], v[2:3]
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: s_cttz_zero_undef_i64_with_select:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 7, @4, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
 ; EG-NEXT:    CF_END
 ; EG-NEXT:    PAD
 ; EG-NEXT:    ALU clause starting at 4:
 ; EG-NEXT:     FFBL_INT * T0.W, KC0[3].X,
 ; EG-NEXT:     FFBL_INT T1.W, KC0[2].W,
 ; EG-NEXT:     ADD_INT * T0.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, KC0[2].W, PS, PV.W,
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: s_cttz_zero_undef_i64_with_select:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-GISEL-NEXT:    s_ff1_i32_b64 s0, s[2:3]
-; GFX9-GISEL-NEXT:    s_bfe_u64 s[0:1], s[0:1], 0x200000
-; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, s1
-; GFX9-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[4:5]
+; GFX9-GISEL-NEXT:    s_ff1_i32_b64 s2, s[2:3]
+; GFX9-GISEL-NEXT:    s_bfe_u64 s[2:3], s[2:3], 0x200000
+; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-GISEL-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %cttz = tail call i64 @llvm.cttz.i64(i64 %val, i1 true) nounwind readnone
   %cttz_ret = icmp ne i64 %val, 0
   %ret = select i1 %cttz_ret, i64 %cttz, i64 32
   store i64 %cttz, i64 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_zero_undef_i8_with_select(i8 addrspace(1)* noalias %out, i8 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_zero_undef_i8_with_select:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v1, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, 32, v1, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_zero_undef_i8_with_select:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ffbl_b32_e32 v1, v0
-; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, 32, v1, vcc
+; VI-NEXT:    flat_load_ubyte v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_ffbl_b32_e32 v3, v2
+; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, 32, v3, vcc
 ; VI-NEXT:    flat_store_byte v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_zero_undef_i8_with_select:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 14, @9, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_zero_undef_i8_with_select:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v2, v1
 ; GFX9-GISEL-NEXT:    v_and_b32_e32 v2, 0xff, v2
 ; GFX9-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v1
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e32 v1, 32, v2, vcc
 ; GFX9-GISEL-NEXT:    global_store_byte v0, v1, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %val = load i8, i8 addrspace(1)* %arrayidx, align 1
   %cttz = tail call i8 @llvm.cttz.i8(i8 %val, i1 true) nounwind readnone
   %cttz_ret = icmp ne i8 %val, 0
   %ret = select i1 %cttz_ret, i8 %cttz, i8 32
   store i8 %ret, i8 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_zero_undef_i16_with_select(i16 addrspace(1)* noalias %out, i16 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_zero_undef_i16_with_select:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0 offset:1
-; SI-NEXT:    buffer_load_ubyte v1, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0 offset:1
+; SI-NEXT:    buffer_load_ubyte v1, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v1
 ; SI-NEXT:    v_ffbl_b32_e32 v1, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, 32, v1, vcc
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_zero_undef_i16_with_select:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    s_add_u32 s4, s2, 1
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v2, s4
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v3, s5
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    flat_load_ubyte v2, v[2:3]
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(1)
-; VI-NEXT:    v_lshlrev_b16_e32 v1, 8, v2
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_or_b32_e32 v0, v1, v0
-; VI-NEXT:    v_ffbl_b32_e32 v1, v0
-; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, 32, v1, vcc
+; VI-NEXT:    flat_load_ubyte v3, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(1)
+; VI-NEXT:    v_lshlrev_b16_e32 v2, 8, v2
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_or_b32_e32 v2, v2, v3
+; VI-NEXT:    v_ffbl_b32_e32 v3, v2
+; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, 32, v3, vcc
 ; VI-NEXT:    flat_store_short v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_zero_undef_i16_with_select:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 14, @9, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    65535(9.183409e-41), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_zero_undef_i16_with_select:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:1
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_lshl_or_b32 v1, v2, 8, v1
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v2, v1
 ; GFX9-GISEL-NEXT:    v_and_b32_e32 v2, 0xffff, v2
 ; GFX9-GISEL-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v1
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e32 v1, 32, v2, vcc
 ; GFX9-GISEL-NEXT:    global_store_short v0, v1, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %val = load i16, i16 addrspace(1)* %arrayidx, align 1
   %cttz = tail call i16 @llvm.cttz.i16(i16 %val, i1 true) nounwind readnone
   %cttz_ret = icmp ne i16 %val, 0
   %ret = select i1 %cttz_ret, i16 %cttz, i16 32
   store i16 %ret, i16 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_zero_undef_i32_with_select(i32 addrspace(1)* noalias %out, i32 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_zero_undef_i32_with_select:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0 offset:1
-; SI-NEXT:    buffer_load_ubyte v1, off, s[4:7], 0 offset:3
-; SI-NEXT:    buffer_load_ubyte v2, off, s[4:7], 0
-; SI-NEXT:    buffer_load_ubyte v3, off, s[4:7], 0 offset:2
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0 offset:1
+; SI-NEXT:    buffer_load_ubyte v1, off, s[8:11], 0 offset:3
+; SI-NEXT:    buffer_load_ubyte v2, off, s[8:11], 0
+; SI-NEXT:    buffer_load_ubyte v3, off, s[8:11], 0 offset:2
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v2
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v1, v1, v3
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_zero_undef_i32_with_select:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_add_u32 s4, s2, 3
+; VI-NEXT:    s_add_u32 s4, s2, 1
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v2, s4
 ; VI-NEXT:    v_mov_b32_e32 v3, s5
-; VI-NEXT:    s_add_u32 s4, s2, 2
+; VI-NEXT:    s_add_u32 s4, s2, 3
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    s_add_u32 s2, s2, 1
-; VI-NEXT:    s_addc_u32 s3, s3, 0
+; VI-NEXT:    s_add_u32 s2, s2, 2
 ; VI-NEXT:    v_mov_b32_e32 v4, s4
-; VI-NEXT:    v_mov_b32_e32 v7, s3
+; VI-NEXT:    s_addc_u32 s3, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v5, s5
+; VI-NEXT:    v_mov_b32_e32 v7, s3
 ; VI-NEXT:    v_mov_b32_e32 v6, s2
 ; VI-NEXT:    flat_load_ubyte v2, v[2:3]
-; VI-NEXT:    flat_load_ubyte v3, v[4:5]
-; VI-NEXT:    flat_load_ubyte v4, v[6:7]
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
+; VI-NEXT:    flat_load_ubyte v3, v[0:1]
+; VI-NEXT:    flat_load_ubyte v4, v[4:5]
+; VI-NEXT:    flat_load_ubyte v5, v[6:7]
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    s_waitcnt vmcnt(3)
-; VI-NEXT:    v_lshlrev_b32_e32 v1, 8, v2
+; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
 ; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v2, v2, v3
 ; VI-NEXT:    s_waitcnt vmcnt(1)
-; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v4
+; VI-NEXT:    v_lshlrev_b32_e32 v3, 8, v4
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_or_b32_e32 v0, v2, v0
-; VI-NEXT:    v_or_b32_e32 v0, v1, v0
-; VI-NEXT:    v_ffbl_b32_e32 v0, v0
-; VI-NEXT:    v_min_u32_e32 v2, 32, v0
-; VI-NEXT:    v_mov_b32_e32 v0, s0
-; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v2, v3, v2
+; VI-NEXT:    v_ffbl_b32_e32 v2, v2
+; VI-NEXT:    v_min_u32_e32 v2, 32, v2
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_zero_undef_i32_with_select:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @10, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 1 @6
 ; EG-NEXT:    ALU 6, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:     OR_INT * T0.W, PV.W, T0.X,
 ; EG-NEXT:     FFBL_INT * T1.W, PV.W,
 ; EG-NEXT:     CNDE_INT T0.X, T0.W, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    32(4.484155e-44), 2(2.802597e-45)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_zero_undef_i32_with_select:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:1
 ; GFX9-GISEL-NEXT:    global_load_ubyte v3, v0, s[2:3] offset:3
 ; GFX9-GISEL-NEXT:    global_load_ubyte v4, v0, s[2:3] offset:2
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(2)
 ; GFX9-GISEL-NEXT:    v_lshl_or_b32 v1, v2, 8, v1
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(1)
; GFX9-GISEL-NEXT:    s_endpgm
   %ret = select i1 %cttz_ret, i32 %cttz, i32 32
   store i32 %ret, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_zero_undef_i64_with_select(i64 addrspace(1)* noalias %out, i64 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_zero_undef_i64_with_select:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s3, 0xf000
 ; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, s2
+; SI-NEXT:    s_mov_b32 s11, s3
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
-; SI-NEXT:    buffer_load_ubyte v2, off, s[4:7], 0 offset:1
-; SI-NEXT:    buffer_load_ubyte v3, off, s[4:7], 0 offset:2
-; SI-NEXT:    buffer_load_ubyte v4, off, s[4:7], 0 offset:3
-; SI-NEXT:    buffer_load_ubyte v5, off, s[4:7], 0 offset:4
-; SI-NEXT:    buffer_load_ubyte v6, off, s[4:7], 0 offset:5
-; SI-NEXT:    buffer_load_ubyte v7, off, s[4:7], 0 offset:7
-; SI-NEXT:    buffer_load_ubyte v8, off, s[4:7], 0 offset:6
+; SI-NEXT:    s_mov_b32 s8, s6
+; SI-NEXT:    s_mov_b32 s9, s7
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
+; SI-NEXT:    buffer_load_ubyte v2, off, s[8:11], 0 offset:1
+; SI-NEXT:    buffer_load_ubyte v3, off, s[8:11], 0 offset:2
+; SI-NEXT:    buffer_load_ubyte v4, off, s[8:11], 0 offset:3
+; SI-NEXT:    buffer_load_ubyte v5, off, s[8:11], 0 offset:4
+; SI-NEXT:    buffer_load_ubyte v6, off, s[8:11], 0 offset:5
+; SI-NEXT:    buffer_load_ubyte v7, off, s[8:11], 0 offset:7
+; SI-NEXT:    buffer_load_ubyte v8, off, s[8:11], 0 offset:6
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_lshlrev_b32_e32 v6, 8, v6
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_lshlrev_b32_e32 v7, 8, v7
 ; SI-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
 ; SI-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
 ; SI-NEXT:    v_or_b32_e32 v5, v6, v5
 ; SI-NEXT:    s_waitcnt vmcnt(0)

 ; SI-NEXT:    v_min_u32_e32 v0, v4, v0
 ; SI-NEXT:    v_cmp_ne_u64_e32 vcc, 0, v[2:3]
 ; SI-NEXT:    v_cndmask_b32_e32 v0, 32, v0, vcc
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_zero_undef_i64_with_select:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    s_add_u32 s4, s2, 5
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v0, s4
 ; VI-NEXT:    v_mov_b32_e32 v1, s5
 ; VI-NEXT:    s_add_u32 s4, s2, 4
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v2, s4
 ; VI-NEXT:    v_mov_b32_e32 v3, s5
 ; VI-NEXT:    s_add_u32 s4, s2, 7
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v4, s4
 ; VI-NEXT:    v_mov_b32_e32 v5, s5
 ; VI-NEXT:    s_add_u32 s4, s2, 6
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v7, s5
 ; VI-NEXT:    v_mov_b32_e32 v6, s4
-; VI-NEXT:    s_add_u32 s4, s2, 3
+; VI-NEXT:    s_add_u32 s4, s2, 1
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v9, s5
 ; VI-NEXT:    v_mov_b32_e32 v8, s4
-; VI-NEXT:    s_add_u32 s4, s2, 2
+; VI-NEXT:    s_add_u32 s4, s2, 3
+; VI-NEXT:    v_mov_b32_e32 v11, s3
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
-; VI-NEXT:    v_mov_b32_e32 v11, s5
-; VI-NEXT:    v_mov_b32_e32 v10, s4
+; VI-NEXT:    v_mov_b32_e32 v10, s2
 ; VI-NEXT:    flat_load_ubyte v12, v[0:1]
 ; VI-NEXT:    flat_load_ubyte v13, v[2:3]
 ; VI-NEXT:    flat_load_ubyte v4, v[4:5]
 ; VI-NEXT:    flat_load_ubyte v5, v[6:7]
-; VI-NEXT:    s_add_u32 s4, s2, 1
-; VI-NEXT:    flat_load_ubyte v6, v[8:9]
-; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v0, s4
-; VI-NEXT:    v_mov_b32_e32 v2, s2
+; VI-NEXT:    s_add_u32 s2, s2, 2
 ; VI-NEXT:    v_mov_b32_e32 v1, s5
-; VI-NEXT:    v_mov_b32_e32 v3, s3
+; VI-NEXT:    flat_load_ubyte v6, v[8:9]
 ; VI-NEXT:    flat_load_ubyte v7, v[10:11]
+; VI-NEXT:    s_addc_u32 s3, s3, 0
+; VI-NEXT:    v_mov_b32_e32 v2, s2
+; VI-NEXT:    v_mov_b32_e32 v3, s3
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    flat_load_ubyte v2, v[2:3]
+; VI-NEXT:    flat_load_ubyte v8, v[2:3]
 ; VI-NEXT:    v_mov_b32_e32 v1, 0
+; VI-NEXT:    v_mov_b32_e32 v2, s0
+; VI-NEXT:    v_mov_b32_e32 v3, s1
 ; VI-NEXT:    s_waitcnt vmcnt(7)
-; VI-NEXT:    v_lshlrev_b32_e32 v3, 8, v12
+; VI-NEXT:    v_lshlrev_b32_e32 v9, 8, v12
 ; VI-NEXT:    s_waitcnt vmcnt(6)
-; VI-NEXT:    v_or_b32_e32 v3, v3, v13
+; VI-NEXT:    v_or_b32_e32 v9, v9, v13
 ; VI-NEXT:    s_waitcnt vmcnt(5)
 ; VI-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
 ; VI-NEXT:    s_waitcnt vmcnt(4)
 ; VI-NEXT:    v_or_b32_sdwa v4, v4, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; VI-NEXT:    v_or_b32_e32 v3, v4, v3
+; VI-NEXT:    v_or_b32_e32 v5, v4, v9
 ; VI-NEXT:    s_waitcnt vmcnt(3)
 ; VI-NEXT:    v_lshlrev_b32_e32 v4, 8, v6
-; VI-NEXT:    v_ffbl_b32_e32 v5, v3
-; VI-NEXT:    v_add_u32_e32 v5, vcc, 32, v5
+; VI-NEXT:    v_ffbl_b32_e32 v6, v5
 ; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    v_or_b32_sdwa v4, v4, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v4, v4, v7
+; VI-NEXT:    v_add_u32_e32 v6, vcc, 32, v6
 ; VI-NEXT:    s_waitcnt vmcnt(1)
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_or_b32_e32 v0, v0, v2
-; VI-NEXT:    v_or_b32_e32 v2, v4, v0
-; VI-NEXT:    v_ffbl_b32_e32 v0, v2
-; VI-NEXT:    v_cmp_ne_u64_e32 vcc, 0, v[2:3]
-; VI-NEXT:    v_min_u32_e32 v0, v0, v5
-; VI-NEXT:    v_mov_b32_e32 v3, s1
+; VI-NEXT:    v_or_b32_sdwa v0, v0, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v4, v0, v4
+; VI-NEXT:    v_ffbl_b32_e32 v0, v4
+; VI-NEXT:    v_cmp_ne_u64_e32 vcc, 0, v[4:5]
+; VI-NEXT:    v_min_u32_e32 v0, v0, v6
 ; VI-NEXT:    v_cndmask_b32_e32 v0, 32, v0, vcc
-; VI-NEXT:    v_mov_b32_e32 v2, s0
 ; VI-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_zero_undef_i64_with_select:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @14, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 3 @6
 ; EG-NEXT:    ALU 16, @15, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:     OR_INT * T1.W, T0.Y, PV.Z,
 ; EG-NEXT:     CNDE_INT T0.X, PS, literal.x, PV.W,
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    32(4.484155e-44), 2(2.802597e-45)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_zero_undef_i64_with_select:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v0, v1, s[2:3]
 ; GFX9-GISEL-NEXT:    global_load_ubyte v2, v1, s[2:3] offset:1
 ; GFX9-GISEL-NEXT:    global_load_ubyte v3, v1, s[2:3] offset:2
 ; GFX9-GISEL-NEXT:    global_load_ubyte v4, v1, s[2:3] offset:3
 ; GFX9-GISEL-NEXT:    global_load_ubyte v5, v1, s[2:3] offset:4
 ; GFX9-GISEL-NEXT:    global_load_ubyte v6, v1, s[2:3] offset:5
 ; GFX9-GISEL-NEXT:    global_load_ubyte v7, v1, s[2:3] offset:7
; GFX9-GISEL-NEXT:    s_endpgm
   %ret = select i1 %cttz_ret, i64 %cttz, i64 32
   store i64 %ret, i64 addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i32_sel_eq_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_i32_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0 offset:1
-; SI-NEXT:    buffer_load_ubyte v1, off, s[4:7], 0 offset:3
-; SI-NEXT:    buffer_load_ubyte v2, off, s[4:7], 0
-; SI-NEXT:    buffer_load_ubyte v3, off, s[4:7], 0 offset:2
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0 offset:1
+; SI-NEXT:    buffer_load_ubyte v1, off, s[8:11], 0 offset:3
+; SI-NEXT:    buffer_load_ubyte v2, off, s[8:11], 0
+; SI-NEXT:    buffer_load_ubyte v3, off, s[8:11], 0 offset:2
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v2
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v1, v1, v3
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i32_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_add_u32 s4, s2, 3
+; VI-NEXT:    s_add_u32 s4, s2, 1
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v2, s4
 ; VI-NEXT:    v_mov_b32_e32 v3, s5
-; VI-NEXT:    s_add_u32 s4, s2, 2
+; VI-NEXT:    s_add_u32 s4, s2, 3
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    s_add_u32 s2, s2, 1
-; VI-NEXT:    s_addc_u32 s3, s3, 0
+; VI-NEXT:    s_add_u32 s2, s2, 2
 ; VI-NEXT:    v_mov_b32_e32 v4, s4
-; VI-NEXT:    v_mov_b32_e32 v7, s3
+; VI-NEXT:    s_addc_u32 s3, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v5, s5
+; VI-NEXT:    v_mov_b32_e32 v7, s3
 ; VI-NEXT:    v_mov_b32_e32 v6, s2
 ; VI-NEXT:    flat_load_ubyte v2, v[2:3]
-; VI-NEXT:    flat_load_ubyte v3, v[4:5]
-; VI-NEXT:    flat_load_ubyte v4, v[6:7]
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
+; VI-NEXT:    flat_load_ubyte v3, v[0:1]
+; VI-NEXT:    flat_load_ubyte v4, v[4:5]
+; VI-NEXT:    flat_load_ubyte v5, v[6:7]
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    s_waitcnt vmcnt(3)
-; VI-NEXT:    v_lshlrev_b32_e32 v1, 8, v2
+; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
 ; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v2, v2, v3
 ; VI-NEXT:    s_waitcnt vmcnt(1)
-; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v4
+; VI-NEXT:    v_lshlrev_b32_e32 v3, 8, v4
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_or_b32_e32 v0, v2, v0
-; VI-NEXT:    v_or_b32_e32 v0, v1, v0
-; VI-NEXT:    v_ffbl_b32_e32 v2, v0
-; VI-NEXT:    v_mov_b32_e32 v0, s0
-; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v2, v3, v2
+; VI-NEXT:    v_ffbl_b32_e32 v2, v2
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_i32_sel_eq_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @10, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 1 @6
 ; EG-NEXT:    ALU 8, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:     CNDE_INT * T1.W, T0.W, literal.x, PV.W,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.W, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_i32_sel_eq_neg1:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:1
 ; GFX9-GISEL-NEXT:    global_load_ubyte v3, v0, s[2:3] offset:3
 ; GFX9-GISEL-NEXT:    global_load_ubyte v4, v0, s[2:3] offset:2
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(2)
 ; GFX9-GISEL-NEXT:    v_lshl_or_b32 v1, v2, 8, v1
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(1)
; GFX9-GISEL-NEXT:    s_endpgm
   %sel = select i1 %cmp, i32 -1, i32 %ctlz
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i32_sel_ne_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_i32_sel_ne_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0 offset:1
-; SI-NEXT:    buffer_load_ubyte v1, off, s[4:7], 0 offset:3
-; SI-NEXT:    buffer_load_ubyte v2, off, s[4:7], 0
-; SI-NEXT:    buffer_load_ubyte v3, off, s[4:7], 0 offset:2
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0 offset:1
+; SI-NEXT:    buffer_load_ubyte v1, off, s[8:11], 0 offset:3
+; SI-NEXT:    buffer_load_ubyte v2, off, s[8:11], 0
+; SI-NEXT:    buffer_load_ubyte v3, off, s[8:11], 0 offset:2
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v2
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v1, v1, v3
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i32_sel_ne_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_add_u32 s4, s2, 3
+; VI-NEXT:    s_add_u32 s4, s2, 1
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v2, s4
 ; VI-NEXT:    v_mov_b32_e32 v3, s5
-; VI-NEXT:    s_add_u32 s4, s2, 2
+; VI-NEXT:    s_add_u32 s4, s2, 3
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    s_add_u32 s2, s2, 1
-; VI-NEXT:    s_addc_u32 s3, s3, 0
+; VI-NEXT:    s_add_u32 s2, s2, 2
 ; VI-NEXT:    v_mov_b32_e32 v4, s4
-; VI-NEXT:    v_mov_b32_e32 v7, s3
+; VI-NEXT:    s_addc_u32 s3, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v5, s5
+; VI-NEXT:    v_mov_b32_e32 v7, s3
 ; VI-NEXT:    v_mov_b32_e32 v6, s2
 ; VI-NEXT:    flat_load_ubyte v2, v[2:3]
-; VI-NEXT:    flat_load_ubyte v3, v[4:5]
-; VI-NEXT:    flat_load_ubyte v4, v[6:7]
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
+; VI-NEXT:    flat_load_ubyte v3, v[0:1]
+; VI-NEXT:    flat_load_ubyte v4, v[4:5]
+; VI-NEXT:    flat_load_ubyte v5, v[6:7]
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    s_waitcnt vmcnt(3)
-; VI-NEXT:    v_lshlrev_b32_e32 v1, 8, v2
+; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
 ; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v2, v2, v3
 ; VI-NEXT:    s_waitcnt vmcnt(1)
-; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v4
+; VI-NEXT:    v_lshlrev_b32_e32 v3, 8, v4
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_or_b32_e32 v0, v2, v0
-; VI-NEXT:    v_or_b32_e32 v0, v1, v0
-; VI-NEXT:    v_ffbl_b32_e32 v2, v0
-; VI-NEXT:    v_mov_b32_e32 v0, s0
-; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v2, v3, v2
+; VI-NEXT:    v_ffbl_b32_e32 v2, v2
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_i32_sel_ne_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @10, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 1 @6
 ; EG-NEXT:    ALU 8, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:     CNDE_INT * T1.W, T0.W, literal.x, PV.W,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, T0.W, literal.x, PV.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_i32_sel_ne_neg1:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:1
 ; GFX9-GISEL-NEXT:    global_load_ubyte v3, v0, s[2:3] offset:3
 ; GFX9-GISEL-NEXT:    global_load_ubyte v4, v0, s[2:3] offset:2
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(2)
 ; GFX9-GISEL-NEXT:    v_lshl_or_b32 v1, v2, 8, v1
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(1)
; GFX9-GISEL-NEXT:    s_endpgm
   %sel = select i1 %cmp, i32 %ctlz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @v_cttz_i32_sel_ne_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_i32_sel_ne_bitwidth:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0 offset:1
-; SI-NEXT:    buffer_load_ubyte v1, off, s[4:7], 0 offset:3
-; SI-NEXT:    buffer_load_ubyte v2, off, s[4:7], 0
-; SI-NEXT:    buffer_load_ubyte v3, off, s[4:7], 0 offset:2
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0 offset:1
+; SI-NEXT:    buffer_load_ubyte v1, off, s[8:11], 0 offset:3
+; SI-NEXT:    buffer_load_ubyte v2, off, s[8:11], 0
+; SI-NEXT:    buffer_load_ubyte v3, off, s[8:11], 0 offset:2
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v2
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v1, v1, v3
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
 ; SI-NEXT:    v_min_u32_e32 v0, 32, v0
 ; SI-NEXT:    v_cmp_ne_u32_e32 vcc, 32, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i32_sel_ne_bitwidth:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_add_u32 s4, s2, 3
+; VI-NEXT:    s_add_u32 s4, s2, 1
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v2, s4
 ; VI-NEXT:    v_mov_b32_e32 v3, s5
-; VI-NEXT:    s_add_u32 s4, s2, 2
+; VI-NEXT:    s_add_u32 s4, s2, 3
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    s_add_u32 s2, s2, 1
-; VI-NEXT:    s_addc_u32 s3, s3, 0
+; VI-NEXT:    s_add_u32 s2, s2, 2
 ; VI-NEXT:    v_mov_b32_e32 v4, s4
-; VI-NEXT:    v_mov_b32_e32 v7, s3
+; VI-NEXT:    s_addc_u32 s3, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v5, s5
+; VI-NEXT:    v_mov_b32_e32 v7, s3
 ; VI-NEXT:    v_mov_b32_e32 v6, s2
 ; VI-NEXT:    flat_load_ubyte v2, v[2:3]
-; VI-NEXT:    flat_load_ubyte v3, v[4:5]
-; VI-NEXT:    flat_load_ubyte v4, v[6:7]
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
+; VI-NEXT:    flat_load_ubyte v3, v[0:1]
+; VI-NEXT:    flat_load_ubyte v4, v[4:5]
+; VI-NEXT:    flat_load_ubyte v5, v[6:7]
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    s_waitcnt vmcnt(3)
-; VI-NEXT:    v_lshlrev_b32_e32 v1, 8, v2
+; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
 ; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v2, v2, v3
 ; VI-NEXT:    s_waitcnt vmcnt(1)
-; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v4
+; VI-NEXT:    v_lshlrev_b32_e32 v3, 8, v4
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_or_b32_e32 v0, v2, v0
-; VI-NEXT:    v_or_b32_e32 v0, v1, v0
-; VI-NEXT:    v_ffbl_b32_e32 v0, v0
-; VI-NEXT:    v_min_u32_e32 v0, 32, v0
-; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 32, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, -1, v0, vcc
-; VI-NEXT:    v_mov_b32_e32 v0, s0
-; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_or_b32_sdwa v3, v3, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_or_b32_e32 v2, v3, v2
+; VI-NEXT:    v_ffbl_b32_e32 v2, v2
+; VI-NEXT:    v_min_u32_e32 v2, 32, v2
+; VI-NEXT:    v_cmp_ne_u32_e32 vcc, 32, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, -1, v2, vcc
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_i32_sel_ne_bitwidth:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @10, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 1 @6
 ; EG-NEXT:    ALU 10, @11, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:     SETNE_INT * T1.W, PV.W, literal.x,
 ; EG-NEXT:    32(4.484155e-44), 0(0.000000e+00)
 ; EG-NEXT:     CNDE_INT T0.X, PV.W, literal.x, T0.W,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.y,
 ; EG-NEXT:    -1(nan), 2(2.802597e-45)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_i32_sel_ne_bitwidth:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:1
 ; GFX9-GISEL-NEXT:    global_load_ubyte v3, v0, s[2:3] offset:3
 ; GFX9-GISEL-NEXT:    global_load_ubyte v4, v0, s[2:3] offset:2
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(2)
 ; GFX9-GISEL-NEXT:    v_lshl_or_b32 v1, v2, 8, v1
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(1)
; GFX9-GISEL-NEXT:    s_endpgm
   %sel = select i1 %cmp, i32 %ctlz, i32 -1
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
  define amdgpu_kernel void @v_cttz_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_i8_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_byte v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i8_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    v_mov_b32_e32 v3, 0xff
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v1, 0xff
-; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_or_b32_e32 v2, 0x100, v0
-; VI-NEXT:    v_ffbl_b32_e32 v2, v2
-; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, v1, v2, vcc
+; VI-NEXT:    flat_load_ubyte v2, v[0:1]
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    s_waitcnt vmcnt(0)
+; VI-NEXT:    v_or_b32_e32 v4, 0x100, v2
+; VI-NEXT:    v_ffbl_b32_e32 v4, v4
+; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, v3, v4, vcc
 ; VI-NEXT:    flat_store_byte v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_i8_sel_eq_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 12, @9, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    255(3.573311e-43), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_i8_sel_eq_neg1:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v2, 0xff
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_or_b32_e32 v3, 0x100, v1
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v3, v3
 ; GFX9-GISEL-NEXT:    v_and_b32_e32 v3, 0xff, v3
 ; GFX9-GISEL-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
 ; GFX9-GISEL-NEXT:    v_cndmask_b32_e32 v1, v3, v2, vcc
 ; GFX9-GISEL-NEXT:    global_store_byte v0, v1, s[0:1]
 ; GFX9-GISEL-NEXT:    s_endpgm
   %val = load i8, i8 addrspace(1)* %arrayidx, align 1
   %ctlz = call i8 @llvm.cttz.i8(i8 %val, i1 false) nounwind readnone
   %cmp = icmp eq i8 %val, 0
   %sel = select i1 %cmp, i8 -1, i8 %ctlz
   store i8 %sel, i8 addrspace(1)* %out
   ret void
 }
 
  define amdgpu_kernel void @v_cttz_i16_sel_eq_neg1(i16 addrspace(1)* noalias %out, i16 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; SI-LABEL: v_cttz_i16_sel_eq_neg1:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_mov_b32 s6, s2
-; SI-NEXT:    s_mov_b32 s7, s3
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[4:7], 0 offset:1
-; SI-NEXT:    buffer_load_ubyte v1, off, s[4:7], 0
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s10, s6
+; SI-NEXT:    s_mov_b32 s11, s7
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s8, s2
+; SI-NEXT:    s_mov_b32 s9, s3
+; SI-NEXT:    buffer_load_ubyte v0, off, s[8:11], 0 offset:1
+; SI-NEXT:    buffer_load_ubyte v1, off, s[8:11], 0
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_or_b32_e32 v0, v0, v1
 ; SI-NEXT:    v_ffbl_b32_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_short v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_short v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_cttz_i16_sel_eq_neg1:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    v_mov_b32_e32 v4, 0xffff
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    s_add_u32 s4, s2, 1
 ; VI-NEXT:    s_addc_u32 s5, s3, 0
 ; VI-NEXT:    v_mov_b32_e32 v2, s4
 ; VI-NEXT:    v_mov_b32_e32 v0, s2
 ; VI-NEXT:    v_mov_b32_e32 v3, s5
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    flat_load_ubyte v2, v[2:3]
-; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    v_mov_b32_e32 v1, 0xffff
+; VI-NEXT:    flat_load_ubyte v3, v[0:1]
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    s_waitcnt vmcnt(1)
 ; VI-NEXT:    v_lshlrev_b16_e32 v2, 8, v2
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_or_b32_e32 v0, v2, v0
-; VI-NEXT:    v_or_b32_e32 v2, 0x10000, v0
-; VI-NEXT:    v_ffbl_b32_e32 v2, v2
-; VI-NEXT:    v_min_u32_e32 v2, 32, v2
-; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v0
-; VI-NEXT:    v_cndmask_b32_e32 v2, v1, v2, vcc
-; VI-NEXT:    v_mov_b32_e32 v0, s0
-; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_or_b32_e32 v2, v2, v3
+; VI-NEXT:    v_or_b32_e32 v3, 0x10000, v2
+; VI-NEXT:    v_ffbl_b32_e32 v3, v3
+; VI-NEXT:    v_min_u32_e32 v3, 32, v3
+; VI-NEXT:    v_cmp_ne_u16_e32 vcc, 0, v2
+; VI-NEXT:    v_cndmask_b32_e32 v2, v4, v3, vcc
 ; VI-NEXT:    flat_store_short v[0:1], v2
 ; VI-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_cttz_i16_sel_eq_neg1:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 12, @9, KC0[CB0:0-32], KC1[]

 ; EG-NEXT:    65535(9.183409e-41), 0(0.000000e+00)
 ; EG-NEXT:     MOV T0.Y, 0.0,
 ; EG-NEXT:     MOV * T0.Z, 0.0,
 ; EG-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; EG-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX9-GISEL-LABEL: v_cttz_i16_sel_eq_neg1:
 ; GFX9-GISEL:       ; %bb.0:
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-GISEL-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-GISEL-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-GISEL-NEXT:    v_mov_b32_e32 v3, 0xffff
 ; GFX9-GISEL-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-GISEL-NEXT:    global_load_ubyte v1, v0, s[2:3]
 ; GFX9-GISEL-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:1
 ; GFX9-GISEL-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-GISEL-NEXT:    v_lshl_or_b32 v1, v2, 8, v1
 ; GFX9-GISEL-NEXT:    v_or_b32_e32 v2, 0x10000, v1
 ; GFX9-GISEL-NEXT:    v_ffbl_b32_e32 v2, v2

llvm/test/CodeGen/AMDGPU/cvt_f32_ubyte.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-- -mcpu=tahiti -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,SI
 ; RUN: llc -mtriple=amdgcn-- -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GCN,VI
 ; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GFX10
-; RUN: llc -mtriple=amdgcn-- -mcpu=gfx908 -start-before=amdgpu-isel -verify-machineinstrs < %s | FileCheck %s -check-prefixes=GFX9
 
 declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
 declare i32 @llvm.amdgcn.workitem.id.y() nounwind readnone
 
 define float @v_uitofp_i32_to_f32_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_i32_to_f32_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_i32_to_f32_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_i32_to_f32_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %masked = and i32 %arg0, 255
   %cvt = uitofp i32 %masked to float
   ret float %cvt
 }
 
 define float @v_sitofp_i32_to_f32_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_sitofp_i32_to_f32_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_sitofp_i32_to_f32_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_sitofp_i32_to_f32_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %masked = and i32 %arg0, 255
   %cvt = sitofp i32 %masked to float
   ret float %cvt
 }
 
 define float @v_uitofp_to_f32_lshr7_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_to_f32_lshr7_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_lshrrev_b32_e32 v0, 7, v0
 ; GCN-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_to_f32_lshr7_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v0, 7, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f32_lshr7_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 7, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.7 = lshr i32 %arg0, 7
   %masked = and i32 %lshr.7, 255
   %cvt = uitofp i32 %masked to float
   ret float %cvt
 }
 
 define float @v_uitofp_to_f32_lshr8_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_to_f32_lshr8_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_to_f32_lshr8_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f32_lshr8_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.8 = lshr i32 %arg0, 8
   %masked = and i32 %lshr.8, 255
   %cvt = uitofp i32 %masked to float
   ret float %cvt
 }
 
 define float @v_uitofp_to_f32_multi_use_lshr8_mask255(i32 %arg0) nounwind {
 ; SI-LABEL: v_uitofp_to_f32_multi_use_lshr8_mask255:

 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v[0:1], v1, off
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f32_multi_use_lshr8_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v1, 8, v0
-; GFX9-NEXT:    global_store_dword v[0:1], v1, off
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.8 = lshr i32 %arg0, 8
   store i32 %lshr.8, i32 addrspace(1)* undef
   %masked = and i32 %lshr.8, 255
   %cvt = uitofp i32 %masked to float
   ret float %cvt
 }
 
 define float @v_uitofp_to_f32_lshr16_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_to_f32_lshr16_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_to_f32_lshr16_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f32_lshr16_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.16 = lshr i32 %arg0, 16
   %masked = and i32 %lshr.16, 255
   %cvt = uitofp i32 %masked to float
   ret float %cvt
 }
 
 define float @v_uitofp_to_f32_lshr24_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_to_f32_lshr24_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_to_f32_lshr24_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f32_lshr24_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.16 = lshr i32 %arg0, 24
   %masked = and i32 %lshr.16, 255
   %cvt = uitofp i32 %masked to float
   ret float %cvt
 }
 
 define float @v_uitofp_i8_to_f32(i8 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_i8_to_f32:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_i8_to_f32:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_i8_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %cvt = uitofp i8 %arg0 to float
   ret float %cvt
 }
 
 define <2 x float> @v_uitofp_v2i8_to_v2f32(i16 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_v2i8_to_v2f32:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_ubyte0_e32 v2, v0
 ; GCN-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GCN-NEXT:    v_mov_b32_e32 v0, v2
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_v2i8_to_v2f32:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v2, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v2
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_v2i8_to_v2f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v2, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
-; GFX9-NEXT:    v_mov_b32_e32 v0, v2
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val = bitcast i16 %arg0 to <2 x i8>
   %cvt = uitofp <2 x i8> %val to <2 x float>
   ret <2 x float> %cvt
 }
 
 define <3 x float> @v_uitofp_v3i8_to_v3f32(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_v3i8_to_v3f32:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_ubyte0_e32 v3, v0
 ; GCN-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GCN-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; GCN-NEXT:    v_mov_b32_e32 v0, v3
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_v3i8_to_v3f32:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v3, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v3
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_v3i8_to_v3f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v3, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
-; GFX9-NEXT:    v_mov_b32_e32 v0, v3
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %trunc = trunc i32 %arg0 to i24
   %val = bitcast i24 %trunc to <3 x i8>
   %cvt = uitofp <3 x i8> %val to <3 x float>
   ret <3 x float> %cvt
 }
 
 define <4 x float> @v_uitofp_v4i8_to_v4f32(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_v4i8_to_v4f32:

 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v4, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_v4i8_to_v4f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v4, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
-; GFX9-NEXT:    v_mov_b32_e32 v0, v4
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %val = bitcast i32 %arg0 to <4 x i8>
   %cvt = uitofp <4 x i8> %val to <4 x float>
   ret <4 x float> %cvt
 }
 
 define <4 x float> @v_uitofp_unpack_i32_to_v4f32(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_unpack_i32_to_v4f32:
 ; GCN:       ; %bb.0:

 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v4, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_unpack_i32_to_v4f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v4, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
-; GFX9-NEXT:    v_mov_b32_e32 v0, v4
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %mask.arg0 = and i32 %arg0, 255
   %cvt0 = uitofp i32 %mask.arg0 to float
 
   %lshr.8 = lshr i32 %arg0, 8
   %mask.lshr.8 = and i32 %lshr.8, 255
   %cvt1 = uitofp i32 %mask.lshr.8 to float
 
   %lshr.16 = lshr i32 %arg0, 16

 ;
 ; GFX10-LABEL: v_uitofp_i32_to_f16_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_i32_to_f16_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %masked = and i32 %arg0, 255
   %cvt = uitofp i32 %masked to half
   ret half %cvt
 }
 
 define half @v_sitofp_i32_to_f16_mask255(i32 %arg0) nounwind {
 ; SI-LABEL: v_sitofp_i32_to_f16_mask255:
 ; SI:       ; %bb.0:

 ;
 ; GFX10-LABEL: v_sitofp_i32_to_f16_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_sitofp_i32_to_f16_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %masked = and i32 %arg0, 255
   %cvt = sitofp i32 %masked to half
   ret half %cvt
 }
 
 define half @v_uitofp_to_f16_lshr8_mask255(i32 %arg0) nounwind {
 ; SI-LABEL: v_uitofp_to_f16_lshr8_mask255:
 ; SI:       ; %bb.0:

 ;
 ; GFX10-LABEL: v_uitofp_to_f16_lshr8_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
 ; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f16_lshr8_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
-; GFX9-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.8 = lshr i32 %arg0, 8
   %masked = and i32 %lshr.8, 255
   %cvt = uitofp i32 %masked to half
   ret half %cvt
 }
 
 define half @v_uitofp_to_f16_lshr16_mask255(i32 %arg0) nounwind {
 ; SI-LABEL: v_uitofp_to_f16_lshr16_mask255:

 ;
 ; GFX10-LABEL: v_uitofp_to_f16_lshr16_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
 ; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f16_lshr16_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
-; GFX9-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.16 = lshr i32 %arg0, 16
   %masked = and i32 %lshr.16, 255
   %cvt = uitofp i32 %masked to half
   ret half %cvt
 }
 
 define half @v_uitofp_to_f16_lshr24_mask255(i32 %arg0) nounwind {
 ; SI-LABEL: v_uitofp_to_f16_lshr24_mask255:

 ;
 ; GFX10-LABEL: v_uitofp_to_f16_lshr24_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
 ; GFX10-NEXT:    v_cvt_f16_f32_e32 v0, v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f16_lshr24_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
-; GFX9-NEXT:    v_cvt_f16_f32_e32 v0, v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.16 = lshr i32 %arg0, 24
   %masked = and i32 %lshr.16, 255
   %cvt = uitofp i32 %masked to half
   ret half %cvt
 }
 
 define half @v_uitofp_i8_to_f16(i8 %arg0) nounwind {
 ; SI-LABEL: v_uitofp_i8_to_f16:

 ; VI-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_i8_to_f16:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_cvt_f16_u16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_i8_to_f16:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_cvt_f16_u16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %cvt = uitofp i8 %arg0 to half
   ret half %cvt
 }
 
 define double @v_uitofp_i32_to_f64_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_i32_to_f64_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GCN-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_i32_to_f64_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_and_b32_e32 v0, 0xff, v0
 ; GFX10-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_i32_to_f64_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xff, v0
-; GFX9-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %masked = and i32 %arg0, 255
   %cvt = uitofp i32 %masked to double
   ret double %cvt
 }
 
 define double @v_uitofp_to_f64_lshr8_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_to_f64_lshr8_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_bfe_u32 v0, v0, 8, 8
 ; GCN-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_to_f64_lshr8_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_bfe_u32 v0, v0, 8, 8
 ; GFX10-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f64_lshr8_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_bfe_u32 v0, v0, 8, 8
-; GFX9-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.8 = lshr i32 %arg0, 8
   %masked = and i32 %lshr.8, 255
   %cvt = uitofp i32 %masked to double
   ret double %cvt
 }
 
 define double @v_uitofp_to_f64_lshr16_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_to_f64_lshr16_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_bfe_u32 v0, v0, 16, 8
 ; GCN-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_to_f64_lshr16_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_bfe_u32 v0, v0, 16, 8
 ; GFX10-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f64_lshr16_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_bfe_u32 v0, v0, 16, 8
-; GFX9-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.16 = lshr i32 %arg0, 16
   %masked = and i32 %lshr.16, 255
   %cvt = uitofp i32 %masked to double
   ret double %cvt
 }
 
 define double @v_uitofp_to_f64_lshr24_mask255(i32 %arg0) nounwind {
 ; GCN-LABEL: v_uitofp_to_f64_lshr24_mask255:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GCN-NEXT:    v_lshrrev_b32_e32 v0, 24, v0
 ; GCN-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GCN-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_uitofp_to_f64_lshr24_mask255:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v0, 24, v0
 ; GFX10-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_to_f64_lshr24_mask255:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 24, v0
-; GFX9-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %lshr.16 = lshr i32 %arg0, 24
   %masked = and i32 %lshr.16, 255
   %cvt = uitofp i32 %masked to double
   ret double %cvt
 }
 
 define double @v_uitofp_i8_to_f64(i8 %arg0) nounwind {
 ; SI-LABEL: v_uitofp_i8_to_f64:

 ; GFX10-LABEL: v_uitofp_i8_to_f64:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0xffff
 ; GFX10-NEXT:    v_and_b32_sdwa v0, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
 ; GFX10-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
-;
-; GFX9-LABEL: v_uitofp_i8_to_f64:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s4, 0xffff
-; GFX9-NEXT:    v_and_b32_sdwa v0, s4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-NEXT:    v_cvt_f64_u32_e32 v[0:1], v0
-; GFX9-NEXT:    s_setpc_b64 s[30:31]
   %cvt = uitofp i8 %arg0 to double
   ret double %cvt
 }
 
 define amdgpu_kernel void @load_i8_to_f32(float addrspace(1)* noalias %out, i8 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_i8_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_i8_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: load_i8_to_f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: load_i8_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i8, i8 addrspace(1)* %in, i32 %tid
   %load = load i8, i8 addrspace(1)* %gep, align 1
   %cvt = uitofp i8 %load to float
   store float %cvt, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @load_v2i8_to_v2f32(<2 x float> addrspace(1)* noalias %out, <2 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_v2i8_to_v2f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_ushort v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_v2i8_to_v2f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: load_v2i8_to_v2f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 1, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ushort v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: load_v2i8_to_v2f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 1, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_ushort v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <2 x i8>, <2 x i8> addrspace(1)* %in, i32 %tid
   %load = load <2 x i8>, <2 x i8> addrspace(1)* %gep, align 2
   %cvt = uitofp <2 x i8> %load to <2 x float>
   store <2 x float> %cvt, <2 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @load_v3i8_to_v3f32(<3 x float> addrspace(1)* noalias %out, <3 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_v3i8_to_v3f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v2, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v2, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v2
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v2
 ; SI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v2
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v2, off, s[0:3], 0 offset:8
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v2, off, s[4:7], 0 offset:8
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_v3i8_to_v3f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; VI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    buffer_store_dwordx3 v[0:2], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: load_v3i8_to_v3f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    global_store_dwordx3 v3, v[0:2], s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: load_v3i8_to_v3f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx3 v3, v[0:2], s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <3 x i8>, <3 x i8> addrspace(1)* %in, i32 %tid
   %load = load <3 x i8>, <3 x i8> addrspace(1)* %gep, align 4
   %cvt = uitofp <3 x i8> %load to <3 x float>
   store <3 x float> %cvt, <3 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @load_v4i8_to_v4f32(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_v4i8_to_v4f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
 ; SI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_v4i8_to_v4f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
 ; VI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; VI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: load_v4i8_to_v4f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: load_v4i8_to_v4f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid
   %load = load <4 x i8>, <4 x i8> addrspace(1)* %gep, align 4
   %cvt = uitofp <4 x i8> %load to <4 x float>
   store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 ; This should not be adding instructions to shift into the correct
 ; position in the word for the component.
 
 ; FIXME: Packing bytes
 define amdgpu_kernel void @load_v4i8_to_v4f32_unaligned(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_v4i8_to_v4f32_unaligned:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[4:7], 0 addr64 offset:3
-; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[4:7], 0 addr64 offset:2
-; SI-NEXT:    buffer_load_ubyte v5, v[0:1], s[4:7], 0 addr64 offset:1
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[8:11], 0 addr64 offset:3
+; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[8:11], 0 addr64 offset:2
+; SI-NEXT:    buffer_load_ubyte v5, v[0:1], s[8:11], 0 addr64 offset:1
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v3, v2
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v4
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v1, v5
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_v4i8_to_v4f32_unaligned:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 2, v0
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 3, v0
 ; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc

 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v4
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v1, v1
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: load_v4i8_to_v4f32_unaligned:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v6, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_clause 0x3
 ; GFX10-NEXT:    global_load_ubyte v1, v0, s[2:3] offset:3
 ; GFX10-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:2
 ; GFX10-NEXT:    global_load_ubyte v4, v0, s[2:3] offset:1
 ; GFX10-NEXT:    global_load_ubyte v5, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(3)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v3, v1
 ; GFX10-NEXT:    s_waitcnt vmcnt(2)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v2, v2
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v1, v4
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v5
 ; GFX10-NEXT:    global_store_dwordx4 v6, v[0:3], s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: load_v4i8_to_v4f32_unaligned:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v7, 0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v2, v[0:1], off offset:3
-; GFX9-NEXT:    global_load_ubyte v4, v[0:1], off offset:2
-; GFX9-NEXT:    global_load_ubyte v5, v[0:1], off offset:1
-; GFX9-NEXT:    global_load_ubyte v6, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    s_waitcnt vmcnt(3)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v3, v2
-; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v2, v4
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v1, v5
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v6
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx4 v7, v[0:3], s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid
   %load = load <4 x i8>, <4 x i8> addrspace(1)* %gep, align 1
   %cvt = uitofp <4 x i8> %load to <4 x float>
   store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 ; FIXME: Need to handle non-uniform case for function below (load without gep).
 ; Instructions still emitted to repack bytes for add use.
 define amdgpu_kernel void @load_v4i8_to_v4f32_2_uses(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %out2, <4 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_v4i8_to_v4f32_2_uses:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
 ; SI-NEXT:    s_mov_b32 s3, 0xf000
 ; SI-NEXT:    s_mov_b32 s6, 0
 ; SI-NEXT:    s_mov_b32 s7, s3
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    buffer_load_dword v4, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xb
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_movk_i32 s8, 0xff
-; SI-NEXT:    s_mov_b32 s6, s2
+; SI-NEXT:    s_movk_i32 s12, 0xff
+; SI-NEXT:    s_mov_b32 s10, s2
+; SI-NEXT:    s_mov_b32 s11, s3
+; SI-NEXT:    s_waitcnt lgkmcnt(0)
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
+; SI-NEXT:    s_mov_b32 s8, s6
+; SI-NEXT:    s_mov_b32 s9, s7
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_lshrrev_b32_e32 v5, 16, v4
 ; SI-NEXT:    v_lshrrev_b32_e32 v6, 24, v4
 ; SI-NEXT:    v_and_b32_e32 v7, 0xff00, v4
 ; SI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v4
 ; SI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v4
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v4
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v4
 ; SI-NEXT:    v_add_i32_e32 v4, vcc, 9, v4
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_waitcnt expcnt(0)
-; SI-NEXT:    v_and_b32_e32 v0, s8, v4
+; SI-NEXT:    v_and_b32_e32 v0, s12, v4
 ; SI-NEXT:    v_add_i32_e32 v2, vcc, 9, v5
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v6
 ; SI-NEXT:    v_or_b32_e32 v0, v7, v0
-; SI-NEXT:    v_and_b32_e32 v2, s8, v2
+; SI-NEXT:    v_and_b32_e32 v2, s12, v2
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, 0x900, v0
 ; SI-NEXT:    v_or_b32_e32 v1, v1, v2
 ; SI-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; SI-NEXT:    v_or_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, 0x9000000, v0
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_v4i8_to_v4f32_2_uses:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
+; VI-NEXT:    s_mov_b32 s7, 0xf000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    v_mov_b32_e32 v5, 9
-; VI-NEXT:    s_movk_i32 s8, 0x900
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v4, v[0:1]
-; VI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
-; VI-NEXT:    s_mov_b32 s6, s2
-; VI-NEXT:    s_mov_b32 s7, s3
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; VI-NEXT:    s_movk_i32 s12, 0x900
+; VI-NEXT:    s_mov_b32 s10, s6
+; VI-NEXT:    s_mov_b32 s11, s7
+; VI-NEXT:    s_waitcnt lgkmcnt(0)
+; VI-NEXT:    s_mov_b32 s4, s0
+; VI-NEXT:    s_mov_b32 s5, s1
+; VI-NEXT:    s_mov_b32 s8, s2
+; VI-NEXT:    s_mov_b32 s9, s3
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_lshrrev_b32_e32 v6, 24, v4
 ; VI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v4
 ; VI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v4
 ; VI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v4
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v4
 ; VI-NEXT:    v_and_b32_e32 v7, 0xffffff00, v4
 ; VI-NEXT:    v_add_u16_e32 v8, 9, v4
 ; VI-NEXT:    v_add_u16_sdwa v4, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; VI-NEXT:    s_nop 0
 ; VI-NEXT:    v_lshlrev_b16_e32 v1, 8, v6
 ; VI-NEXT:    v_or_b32_sdwa v0, v7, v8 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
 ; VI-NEXT:    v_or_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; VI-NEXT:    v_mov_b32_e32 v2, s8
-; VI-NEXT:    v_add_u16_e32 v0, s8, v0
+; VI-NEXT:    v_mov_b32_e32 v2, s12
+; VI-NEXT:    v_add_u16_e32 v0, s12, v0
 ; VI-NEXT:    v_add_u16_sdwa v1, v1, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; VI-NEXT:    v_or_b32_e32 v0, v0, v1
-; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; VI-NEXT:    buffer_store_dword v0, off, s[8:11], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: load_v4i8_to_v4f32_2_uses:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x34
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 24
+; GFX10-NEXT:    s_movk_i32 s4, 0x900
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
-; GFX10-NEXT:    s_clause 0x1
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX10-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
-; GFX10-NEXT:    s_movk_i32 s0, 0x900
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_lshrrev_b32_e32 v2, 16, v0
 ; GFX10-NEXT:    v_lshrrev_b32_sdwa v1, v1, v0 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; GFX10-NEXT:    v_and_b32_e32 v3, 0xffffff00, v0
 ; GFX10-NEXT:    v_add_nc_u16 v4, v0, 9
 ; GFX10-NEXT:    v_add_nc_u16 v2, v2, 9
 ; GFX10-NEXT:    v_or_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
 ; GFX10-NEXT:    v_or_b32_sdwa v2, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
 ; GFX10-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v3, v0
-; GFX10-NEXT:    v_add_nc_u16 v1, v1, s0
-; GFX10-NEXT:    v_add_nc_u16 v5, v2, s0
+; GFX10-NEXT:    v_add_nc_u16 v1, v1, s4
+; GFX10-NEXT:    v_add_nc_u16 v5, v2, s4
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v2, v0
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v6, 16, v1
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    v_or_b32_sdwa v5, v5, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    global_store_dwordx4 v4, v[0:3], s[2:3]
-; GFX10-NEXT:    global_store_dword v4, v5, s[4:5]
+; GFX10-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
+; GFX10-NEXT:    global_store_dword v4, v5, s[2:3]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: load_v4i8_to_v4f32_2_uses:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x34
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-NEXT:    v_mov_b32_e32 v6, 9
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v4, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2c
-; GFX9-NEXT:    s_movk_i32 s4, 0x900
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_lshrrev_b32_e32 v7, 24, v4
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v3, v4
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v2, v4
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v4
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v4
-; GFX9-NEXT:    v_and_b32_e32 v8, 0xffffff00, v4
-; GFX9-NEXT:    v_add_u16_e32 v9, 9, v4
-; GFX9-NEXT:    v_add_u16_sdwa v4, v4, v6 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx4 v5, v[0:3], s[0:1]
-; GFX9-NEXT:    s_nop 0
-; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v7
-; GFX9-NEXT:    v_or_b32_sdwa v0, v8, v9 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-NEXT:    v_or_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
-; GFX9-NEXT:    v_add_u16_e32 v0, s4, v0
-; GFX9-NEXT:    v_add_u16_sdwa v1, v1, s4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GFX9-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX9-NEXT:    global_store_dword v5, v0, s[2:3]
-; GFX9-NEXT:    s_endpgm
   %tid.x = call i32 @llvm.amdgcn.workitem.id.x()
   %in.ptr = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid.x
   %load = load <4 x i8>, <4 x i8> addrspace(1)* %in.ptr, align 4
   %cvt = uitofp <4 x i8> %load to <4 x float>
   store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
   %add = add <4 x i8> %load, <i8 9, i8 9, i8 9, i8 9> ; Second use of %load
   store <4 x i8> %add, <4 x i8> addrspace(1)* %out2, align 4
   ret void
 }
 
 ; Make sure this doesn't crash.
 define amdgpu_kernel void @load_v7i8_to_v7f32(<7 x float> addrspace(1)* noalias %out, <7 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_v7i8_to_v7f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[4:7], 0 addr64 offset:3
-; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[4:7], 0 addr64 offset:2
-; SI-NEXT:    buffer_load_ubyte v5, v[0:1], s[4:7], 0 addr64 offset:1
-; SI-NEXT:    buffer_load_ubyte v6, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    buffer_load_ubyte v7, v[0:1], s[4:7], 0 addr64 offset:5
-; SI-NEXT:    buffer_load_ubyte v8, v[0:1], s[4:7], 0 addr64 offset:4
-; SI-NEXT:    buffer_load_ubyte v9, v[0:1], s[4:7], 0 addr64 offset:6
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[8:11], 0 addr64 offset:3
+; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[8:11], 0 addr64 offset:2
+; SI-NEXT:    buffer_load_ubyte v5, v[0:1], s[8:11], 0 addr64 offset:1
+; SI-NEXT:    buffer_load_ubyte v6, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    buffer_load_ubyte v7, v[0:1], s[8:11], 0 addr64 offset:5
+; SI-NEXT:    buffer_load_ubyte v8, v[0:1], s[8:11], 0 addr64 offset:4
+; SI-NEXT:    buffer_load_ubyte v9, v[0:1], s[8:11], 0 addr64 offset:6
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(6)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v3, v2
 ; SI-NEXT:    s_waitcnt vmcnt(5)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v4
 ; SI-NEXT:    s_waitcnt vmcnt(4)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v1, v5
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v6
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v5, v7
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v4, v8
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v6, v9
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v6, off, s[0:3], 0 offset:24
-; SI-NEXT:    buffer_store_dwordx2 v[4:5], off, s[0:3], 0 offset:16
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v6, off, s[4:7], 0 offset:24
+; SI-NEXT:    buffer_store_dwordx2 v[4:5], off, s[4:7], 0 offset:16
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_v7i8_to_v7f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 5, v0
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ubyte v10, v[2:3]
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 6, v0

 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v4, v9
 ; VI-NEXT:    buffer_store_dwordx3 v[4:6], off, s[0:3], 0 offset:16
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: load_v7i8_to_v7f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
 ; GFX10-NEXT:    v_mov_b32_e32 v4, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v7, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_clause 0x5
 ; GFX10-NEXT:    global_load_ubyte v5, v0, s[2:3] offset:6
 ; GFX10-NEXT:    global_load_ubyte v1, v0, s[2:3] offset:3
 ; GFX10-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:2
 ; GFX10-NEXT:    global_load_ubyte v6, v0, s[2:3] offset:1
 ; GFX10-NEXT:    global_load_short_d16 v4, v0, s[2:3] offset:4
 ; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(4)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v3, v1
 ; GFX10-NEXT:    s_waitcnt vmcnt(3)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v2, v2
 ; GFX10-NEXT:    s_waitcnt vmcnt(2)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v1, v6
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v6, v5
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v5, v4
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v4, v4
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    global_store_dwordx3 v7, v[4:6], s[0:1] offset:16
 ; GFX10-NEXT:    global_store_dwordx4 v7, v[0:3], s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: load_v7i8_to_v7f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 3, v[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_ushort v2, v[0:1], off offset:4
-; GFX9-NEXT:    global_load_ubyte v3, v[0:1], off offset:6
-; GFX9-NEXT:    global_load_ubyte v4, v[0:1], off offset:3
-; GFX9-NEXT:    global_load_ubyte v5, v[0:1], off offset:2
-; GFX9-NEXT:    global_load_ubyte v7, v[0:1], off offset:1
-; GFX9-NEXT:    global_load_ubyte v8, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    s_waitcnt vmcnt(5)
-; GFX9-NEXT:    v_and_b32_e32 v10, 0xffff, v2
-; GFX9-NEXT:    s_waitcnt vmcnt(4)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v6, v3
-; GFX9-NEXT:    s_waitcnt vmcnt(3)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v3, v4
-; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v2, v5
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v1, v7
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v8
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v5, v10
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v4, v10
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx4 v9, v[0:3], s[0:1]
-; GFX9-NEXT:    global_store_dwordx3 v9, v[4:6], s[0:1] offset:16
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <7 x i8>, <7 x i8> addrspace(1)* %in, i32 %tid
   %load = load <7 x i8>, <7 x i8> addrspace(1)* %gep, align 1
   %cvt = uitofp <7 x i8> %load to <7 x float>
   store <7 x float> %cvt, <7 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @load_v8i8_to_v8f32(<8 x float> addrspace(1)* noalias %out, <8 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: load_v8i8_to_v8f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dwordx2 v[7:8], v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dwordx2 v[7:8], v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v7
 ; SI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v7
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v7
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v7
 ; SI-NEXT:    v_cvt_f32_ubyte3_e32 v7, v8
 ; SI-NEXT:    v_cvt_f32_ubyte2_e32 v6, v8
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v5, v8
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v4, v8
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[4:7], 0 offset:16
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: load_v8i8_to_v8f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dwordx2 v[7:8], v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v7
 ; VI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v7
 ; VI-NEXT:    v_cvt_f32_ubyte1_e32 v1, v7
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v7
 ; VI-NEXT:    v_cvt_f32_ubyte3_e32 v7, v8
 ; VI-NEXT:    v_cvt_f32_ubyte2_e32 v6, v8
 ; VI-NEXT:    v_cvt_f32_ubyte1_e32 v5, v8
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v4, v8
 ; VI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: load_v8i8_to_v8f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 3, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v10, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dwordx2 v[8:9], v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v7, v9
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v6, v9
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v5, v9
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v4, v9
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v3, v8
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v2, v8
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v1, v8
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v8
 ; GFX10-NEXT:    global_store_dwordx4 v10, v[4:7], s[0:1] offset:16
 ; GFX10-NEXT:    global_store_dwordx4 v10, v[0:3], s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: load_v8i8_to_v8f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 3, v[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v9, 0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dwordx2 v[7:8], v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v3, v7
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v2, v7
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v1, v7
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v7
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v7, v8
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v6, v8
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v5, v8
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v4, v8
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx4 v9, v[4:7], s[0:1] offset:16
-; GFX9-NEXT:    global_store_dwordx4 v9, v[0:3], s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <8 x i8>, <8 x i8> addrspace(1)* %in, i32 %tid
   %load = load <8 x i8>, <8 x i8> addrspace(1)* %gep, align 8
   %cvt = uitofp <8 x i8> %load to <8 x float>
   store <8 x float> %cvt, <8 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @i8_zext_inreg_i32_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: i8_zext_inreg_i32_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, 2, v0
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: i8_zext_inreg_i32_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, 2, v0
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: i8_zext_inreg_i32_to_f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_add_nc_u32_e32 v0, 2, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: i8_zext_inreg_i32_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_add_u32_e32 v0, 2, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
   %load = load i32, i32 addrspace(1)* %gep, align 4
   %add = add i32 %load, 2
   %inreg = and i32 %add, 255
   %cvt = uitofp i32 %inreg to float
   store float %cvt, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @i8_zext_inreg_hi1_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: i8_zext_inreg_hi1_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: i8_zext_inreg_hi1_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: i8_zext_inreg_hi1_to_f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: i8_zext_inreg_hi1_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
   %load = load i32, i32 addrspace(1)* %gep, align 4
   %inreg = and i32 %load, 65280
   %shr = lshr i32 %inreg, 8
   %cvt = uitofp i32 %shr to float
   store float %cvt, float addrspace(1)* %out, align 4
   ret void
 }
 
 ; We don't get these ones because of the zext, but instcombine removes
 ; them so it shouldn't really matter.
 define amdgpu_kernel void @i8_zext_i32_to_f32(float addrspace(1)* noalias %out, i8 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: i8_zext_i32_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
 ; SI-NEXT:    v_mov_b32_e32 v1, 0
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: i8_zext_i32_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: i8_zext_i32_to_f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_ubyte v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: i8_zext_i32_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i8, i8 addrspace(1)* %in, i32 %tid
   %load = load i8, i8 addrspace(1)* %gep, align 1
   %ext = zext i8 %load to i32
   %cvt = uitofp i32 %ext to float
   store float %cvt, float addrspace(1)* %out, align 4
   ret void
 }
 
 define amdgpu_kernel void @v4i8_zext_v4i32_to_v4f32(<4 x float> addrspace(1)* noalias %out, <4 x i8> addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: v4i8_zext_v4i32_to_v4f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[4:7], 0 addr64 offset:3
-; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[4:7], 0 addr64 offset:2
-; SI-NEXT:    buffer_load_ubyte v5, v[0:1], s[4:7], 0 addr64 offset:1
-; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_ubyte v2, v[0:1], s[8:11], 0 addr64 offset:3
+; SI-NEXT:    buffer_load_ubyte v4, v[0:1], s[8:11], 0 addr64 offset:2
+; SI-NEXT:    buffer_load_ubyte v5, v[0:1], s[8:11], 0 addr64 offset:1
+; SI-NEXT:    buffer_load_ubyte v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(3)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v3, v2
 ; SI-NEXT:    s_waitcnt vmcnt(2)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v4
 ; SI-NEXT:    s_waitcnt vmcnt(1)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v1, v5
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v4i8_zext_v4i32_to_v4f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 2, v0
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 3, v0
 ; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc

 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v4
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v1, v1
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: v4i8_zext_v4i32_to_v4f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v6, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_clause 0x3
 ; GFX10-NEXT:    global_load_ubyte v1, v0, s[2:3] offset:3
 ; GFX10-NEXT:    global_load_ubyte v2, v0, s[2:3] offset:2
 ; GFX10-NEXT:    global_load_ubyte v4, v0, s[2:3] offset:1
 ; GFX10-NEXT:    global_load_ubyte v5, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(3)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v3, v1
 ; GFX10-NEXT:    s_waitcnt vmcnt(2)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v2, v2
 ; GFX10-NEXT:    s_waitcnt vmcnt(1)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v1, v4
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v5
 ; GFX10-NEXT:    global_store_dwordx4 v6, v[0:3], s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: v4i8_zext_v4i32_to_v4f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v7, 0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_ubyte v2, v[0:1], off offset:3
-; GFX9-NEXT:    global_load_ubyte v4, v[0:1], off offset:2
-; GFX9-NEXT:    global_load_ubyte v5, v[0:1], off offset:1
-; GFX9-NEXT:    global_load_ubyte v6, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    s_waitcnt vmcnt(3)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v3, v2
-; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v2, v4
-; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v1, v5
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v6
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx4 v7, v[0:3], s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid
   %load = load <4 x i8>, <4 x i8> addrspace(1)* %gep, align 1
   %ext = zext <4 x i8> %load to <4 x i32>
   %cvt = uitofp <4 x i32> %ext to <4 x float>
   store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @extract_byte0_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: extract_byte0_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: extract_byte0_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: extract_byte0_to_f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: extract_byte0_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
   %val = load i32, i32 addrspace(1)* %gep
   %and = and i32 %val, 255
   %cvt = uitofp i32 %and to float
   store float %cvt, float addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @extract_byte1_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: extract_byte1_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: extract_byte1_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: extract_byte1_to_f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: extract_byte1_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte1_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
   %val = load i32, i32 addrspace(1)* %gep
   %srl = lshr i32 %val, 8
   %and = and i32 %srl, 255
   %cvt = uitofp i32 %and to float
   store float %cvt, float addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @extract_byte2_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: extract_byte2_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: extract_byte2_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: extract_byte2_to_f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: extract_byte2_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte2_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
   %val = load i32, i32 addrspace(1)* %gep
   %srl = lshr i32 %val, 16
   %and = and i32 %srl, 255
   %cvt = uitofp i32 %and to float
   store float %cvt, float addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @extract_byte3_to_f32(float addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in) nounwind {
 ; SI-LABEL: extract_byte3_to_f32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xb
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s6, 0
-; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s10, 0
+; SI-NEXT:    s_mov_b32 s11, s7
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; SI-NEXT:    v_mov_b32_e32 v1, 0
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_dword v0, v[0:1], s[4:7], 0 addr64
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b64 s[8:9], s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, 0
+; SI-NEXT:    buffer_load_dword v0, v[0:1], s[8:11], 0 addr64
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: extract_byte3_to_f32:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s3
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s2, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    flat_load_dword v0, v[0:1]
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: extract_byte3_to_f32:
 ; GFX10:       ; %bb.0:
-; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    v_lshlrev_b32_e32 v0, 2, v0
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[2:3]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
 ; GFX10-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: extract_byte3_to_f32:
-; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_cvt_f32_ubyte3_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
-; GFX9-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr i32, i32 addrspace(1)* %in, i32 %tid
   %val = load i32, i32 addrspace(1)* %gep
   %srl = lshr i32 %val, 24
   %and = and i32 %srl, 255
   %cvt = uitofp i32 %and to float
   store float %cvt, float addrspace(1)* %out
   ret void

 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_load_dword v0, v0, s[0:1]
 ; GFX10-NEXT:    s_waitcnt vmcnt(0)
 ; GFX10-NEXT:    v_or_b32_e32 v0, 0x80000001, v0
 ; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v1, v0
 ; GFX10-NEXT:    v_add_f32_e32 v0, v0, v1
 ; GFX10-NEXT:    global_store_dword v2, v0, s[2:3]
 ; GFX10-NEXT:    s_endpgm
-;
-; GFX9-LABEL: cvt_ubyte0_or_multiuse:
-; GFX9:       ; %bb.0: ; %bb
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x24
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x2c
-; GFX9-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[0:1], 2, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    global_load_dword v0, v[0:1], off
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_or_b32_e32 v0, 0x80000001, v0
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v2, v0
-; GFX9-NEXT:    v_add_f32_e32 v0, v0, v2
-; GFX9-NEXT:    global_store_dword v1, v0, s[2:3]
-; GFX9-NEXT:    s_endpgm
 bb:
   %lid = tail call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 %lid
   %load = load i32, i32 addrspace(1)* %gep
   %or = or i32 %load, -2147483647
   %and = and i32 %or, 255
   %uitofp = uitofp i32 %and to float
   %cast = bitcast i32 %or to float
   %add = fadd float %cast, %uitofp
   store float %add, float addrspace(1)* %out
   ret void
 }
-
-%Vec = type { [4 x i8] }
-
-define amdgpu_kernel void @cvt_f32_ubyte0_vector() local_unnamed_addr {
-; SI-LABEL: cvt_f32_ubyte0_vector:
-; SI:       ; %bb.0: ; %entry
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
-; SI-NEXT:    s_mov_b32 s3, 0xf000
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:3
-; SI-NEXT:    buffer_load_ubyte v1, off, s[0:3], 0 offset:2
-; SI-NEXT:    buffer_load_ubyte v2, off, s[0:3], 0 offset:1
-; SI-NEXT:    buffer_load_ubyte v3, off, s[0:3], 0
-; SI-NEXT:    s_load_dword s0, s[0:1], 0x0
-; SI-NEXT:    s_waitcnt vmcnt(3)
-; SI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    v_fma_f32 v0, s0, v0, 0.5
-; SI-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; SI-NEXT:    s_waitcnt vmcnt(2)
-; SI-NEXT:    buffer_store_byte v1, off, s[0:3], 0
-; SI-NEXT:    s_waitcnt vmcnt(2)
-; SI-NEXT:    buffer_store_byte v2, off, s[0:3], 0
-; SI-NEXT:    s_waitcnt vmcnt(2)
-; SI-NEXT:    buffer_store_byte v3, off, s[0:3], 0
-; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
-; SI-NEXT:  .LBB40_1: ; %for.body.i
-; SI-NEXT:    ; =>This Inner Loop Header: Depth=1
-; SI-NEXT:    s_branch .LBB40_1
-;
-; VI-LABEL: cvt_f32_ubyte0_vector:
-; VI:       ; %bb.0: ; %entry
-; VI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
-; VI-NEXT:    s_mov_b32 s3, 0xf000
-; VI-NEXT:    s_mov_b32 s2, -1
-; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0 offset:3
-; VI-NEXT:    buffer_load_ubyte v1, off, s[0:3], 0 offset:2
-; VI-NEXT:    buffer_load_ubyte v2, off, s[0:3], 0 offset:1
-; VI-NEXT:    buffer_load_ubyte v3, off, s[0:3], 0
-; VI-NEXT:    s_load_dword s0, s[0:1], 0x0
-; VI-NEXT:    s_waitcnt vmcnt(3)
-; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    v_mul_f32_e32 v0, s0, v0
-; VI-NEXT:    v_add_f32_e32 v0, 0.5, v0
-; VI-NEXT:    v_cvt_i32_f32_e32 v0, v0
-; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    buffer_store_byte v1, off, s[0:3], 0
-; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    buffer_store_byte v2, off, s[0:3], 0
-; VI-NEXT:    s_waitcnt vmcnt(2)
-; VI-NEXT:    buffer_store_byte v3, off, s[0:3], 0
-; VI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
-; VI-NEXT:  .LBB40_1: ; %for.body.i
-; VI-NEXT:    ; =>This Inner Loop Header: Depth=1
-; VI-NEXT:    s_branch .LBB40_1
-;
-; GFX10-LABEL: cvt_f32_ubyte0_vector:
-; GFX10:       ; %bb.0: ; %entry
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
-; GFX10-NEXT:    v_mov_b32_e32 v0, 0
-; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_clause 0x3
-; GFX10-NEXT:    global_load_ubyte v1, v0, s[0:1] offset:3
-; GFX10-NEXT:    global_load_ubyte v2, v0, s[0:1] offset:2
-; GFX10-NEXT:    global_load_ubyte v3, v0, s[0:1] offset:1
-; GFX10-NEXT:    global_load_ubyte v4, v0, s[0:1]
-; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
-; GFX10-NEXT:    s_load_dword s0, s[0:1], 0x0
-; GFX10-NEXT:    s_waitcnt vmcnt(3)
-; GFX10-NEXT:    v_cvt_f32_ubyte0_e32 v0, v1
-; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    v_fma_f32 v0, s0, v0, 0.5
-; GFX10-NEXT:    v_cvt_i32_f32_e32 v0, v0
-; GFX10-NEXT:    s_waitcnt vmcnt(2)
-; GFX10-NEXT:    global_store_byte v[0:1], v2, off
-; GFX10-NEXT:    s_waitcnt vmcnt(1)
-; GFX10-NEXT:    global_store_byte v[0:1], v3, off
-; GFX10-NEXT:    s_waitcnt vmcnt(0)
-; GFX10-NEXT:    global_store_byte v[0:1], v4, off
-; GFX10-NEXT:    global_store_byte v[0:1], v0, off
-; GFX10-NEXT:  .LBB40_1: ; %for.body.i
-; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    s_branch .LBB40_1
-;
-; GFX9-LABEL: cvt_f32_ubyte0_vector:
-; GFX9:       ; %bb.0: ; %entry
-; GFX9-NEXT:  .LBB40_1: ; %for.body.i
-; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT:    global_load_dwordx2 v[0:1], v[0:1], off
-; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    global_load_ubyte v2, v[0:1], off offset:3
-; GFX9-NEXT:    global_load_dword v3, v[0:1], off
-; GFX9-NEXT:    global_load_ubyte v4, v[0:1], off offset:2
-; GFX9-NEXT:    global_load_ubyte v5, v[0:1], off offset:1
-; GFX9-NEXT:    global_load_ubyte v6, v[0:1], off
-; GFX9-NEXT:    s_waitcnt vmcnt(4)
-; GFX9-NEXT:    v_cvt_f32_ubyte0_e32 v0, v2
-; GFX9-NEXT:    s_waitcnt vmcnt(3)
-; GFX9-NEXT:    v_fma_f32 v0, v3, v0, 0.5
-; GFX9-NEXT:    v_cvt_i32_f32_e32 v0, v0
-; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    global_store_byte v[0:1], v4, off
-; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    global_store_byte v[0:1], v5, off
-; GFX9-NEXT:    s_waitcnt vmcnt(2)
-; GFX9-NEXT:    global_store_byte v[0:1], v6, off
-; GFX9-NEXT:    global_store_byte v[0:1], v0, off
-; GFX9-NEXT:    s_branch .LBB40_1
-entry:
-  br label %for.body.i
-
-for.body.i:                                       ; preds = %for.body.i, %entry
-  %retval.sroa.0.0.copyload = load %Vec*, %Vec* addrspace(1)* undef, align 8
-  %add.ptr = getelementptr inbounds %Vec, %Vec* %retval.sroa.0.0.copyload, i64 undef
-  %retval.sroa.0.0..sroa_cast = bitcast %Vec* %add.ptr to i32*
-  %retval.sroa.0.0..sroa_cast_adr = addrspacecast i32* %retval.sroa.0.0..sroa_cast to i32 addrspace(1)*
-  %retval.sroa.0.0.copyload.i = load i32, i32 addrspace(1)* %retval.sroa.0.0..sroa_cast_adr, align 1
-  %p1.sroa.6.0.extract.shift = lshr i32 %retval.sroa.0.0.copyload.i, 24
-  %p1.sroa.6.0.extract.trunc = trunc i32 %p1.sroa.6.0.extract.shift to i8
-  %conv12 = uitofp i8 %p1.sroa.6.0.extract.trunc to float
-  %0 = load float, float addrspace(1)* undef, align 8
-  %mul = fmul contract float %0, %conv12
-  %add = fadd contract float %mul, 5.000000e-01
-  %conv13 = fptoui float %add to i8
-  %retval.sroa.4.0.insert.ext = zext i8 %conv13 to i32
-  %retval.sroa.4.0.insert.shift = shl nuw i32 %retval.sroa.4.0.insert.ext, 24
-  %retval.sroa.3.0.insert.ext = and i32 %retval.sroa.0.0.copyload.i, 16711680
-  %retval.sroa.3.0.insert.insert = or i32 %retval.sroa.4.0.insert.shift, %retval.sroa.3.0.insert.ext
-  %retval.sroa.2.0.insert.ext = and i32 %retval.sroa.0.0.copyload.i, 65280
-  %retval.sroa.2.0.insert.insert = or i32 %retval.sroa.3.0.insert.insert, %retval.sroa.2.0.insert.ext
-  %retval.sroa.0.0.insert.ext = and i32 %retval.sroa.0.0.copyload.i, 255
-  %retval.sroa.0.0.insert.insert = or i32 %retval.sroa.2.0.insert.insert, %retval.sroa.0.0.insert.ext
-  store i32 %retval.sroa.0.0.insert.insert, i32 addrspace(1)* undef, align 1
-  br label %for.body.i
-}

llvm/test/CodeGen/AMDGPU/disable_form_clauses.ll

 ; RUN: llc -march=amdgcn -mcpu=gfx902 -verify-machineinstrs -stop-after=si-form-memory-clauses < %s | FileCheck -check-prefix=GCN %s
 
 ; GCN-LABEL: {{^}}name:{{[  ]*}}vector_clause
-; GCN: LOAD_DWORDX2
-; GCN-NEXT: LOAD_DWORDX2
-; GCN-NEXT: KILL
+; GCN: LOAD_DWORDX4
 define amdgpu_kernel void @vector_clause(<4 x i32> addrspace(1)* noalias nocapture readonly %arg, <4 x i32> addrspace(1)* noalias nocapture %arg1) {
 bb:
   %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
   %tmp2 = zext i32 %tmp to i64
   %tmp3 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 %tmp2
   %tmp4 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp3, align 16
   %tmp5 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 %tmp2
   %tmp6 = add nuw nsw i64 %tmp2, 1
bb:
   store <4 x i32> %tmp12, <4 x i32> addrspace(1)* %tmp13, align 16
   store <4 x i32> %tmp16, <4 x i32> addrspace(1)* %tmp17, align 16
   ret void
 }
 
 declare i32 @llvm.amdgcn.workitem.id.x()
 
 attributes #0 = { "amdgpu-max-memory-clause"="1" }
-

llvm/test/CodeGen/AMDGPU/llvm.memcpy.ll

 
 ; FUNC-LABEL: {{^}}test_small_memcpy_i64_lds_to_lds_align4:
 ; SI: ds_read2_b32
 ; SI: ds_read2_b32
 ; SI: ds_read2_b32
 ; SI: ds_read2_b32
 
 ; SI: ds_write2_b32
-; SI: ds_write2_b32
-; SI: ds_write2_b32
-; SI: ds_write2_b32
 
 ; SI: s_endpgm
 define amdgpu_kernel void @test_small_memcpy_i64_lds_to_lds_align4(i64 addrspace(3)* noalias %out, i64 addrspace(3)* noalias %in) nounwind {
   %bcin = bitcast i64 addrspace(3)* %in to i8 addrspace(3)*
   %bcout = bitcast i64 addrspace(3)* %out to i8 addrspace(3)*
   call void @llvm.memcpy.p3i8.p3i8.i32(i8 addrspace(3)* align 4 %bcout, i8 addrspace(3)* align 4 %bcin, i32 32, i1 false) nounwind
   ret void
 }
 
 ; FIXME: Use 64-bit ops
 ; FUNC-LABEL: {{^}}test_small_memcpy_i64_lds_to_lds_align8:
 
 ; SI: ds_read2_b64
 ; SI: ds_read2_b64
 
 ; SI: ds_write2_b64
-; SI: ds_write2_b64
 
 ; SI-DAG: s_endpgm
 define amdgpu_kernel void @test_small_memcpy_i64_lds_to_lds_align8(i64 addrspace(3)* noalias %out, i64 addrspace(3)* noalias %in) nounwind {
   %bcin = bitcast i64 addrspace(3)* %in to i8 addrspace(3)*
   %bcout = bitcast i64 addrspace(3)* %out to i8 addrspace(3)*
   call void @llvm.memcpy.p3i8.p3i8.i32(i8 addrspace(3)* align 8 %bcout, i8 addrspace(3)* align 8 %bcin, i32 32, i1 false) nounwind
   ret void
 }

llvm/test/CodeGen/AMDGPU/lower-kernargs.ll

 ;
   store volatile i8 addrspace(1)* %ptr, i8 addrspace(1)* addrspace(1)* undef
   ret void
 }
 
 define amdgpu_kernel void @kern_noalias_global_ptr(i8 addrspace(1)* noalias %ptr) #0 {
 ; HSA-LABEL: @kern_noalias_global_ptr(
 ; HSA-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(64) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
-; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR:%.*]], i8 addrspace(1)* addrspace(1)* undef, align 8
+; HSA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_NOALIAS_GLOBAL_PTR_KERNARG_SEGMENT]], i64 0
+; HSA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
+; HSA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
+; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef, align 8, !noalias !4
 ; HSA-NEXT:    ret void
 ;
 ; MESA-LABEL: @kern_noalias_global_ptr(
 ; MESA-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(64) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
-; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR:%.*]], i8 addrspace(1)* addrspace(1)* undef, align 8
+; MESA-NEXT:    [[PTR_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_NOALIAS_GLOBAL_PTR_KERNARG_SEGMENT]], i64 36
+; MESA-NEXT:    [[PTR_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
+; MESA-NEXT:    [[PTR_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
+; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR_LOAD]], i8 addrspace(1)* addrspace(1)* undef, align 8, !noalias !4
 ; MESA-NEXT:    ret void
 ;
   store volatile i8 addrspace(1)* %ptr, i8 addrspace(1)* addrspace(1)* undef
   ret void
 }
 
 define amdgpu_kernel void @kern_noalias_global_ptr_x2(i8 addrspace(1)* noalias %ptr0, i8 addrspace(1)* noalias %ptr1) #0 {
 ; HSA-LABEL: @kern_noalias_global_ptr_x2(
 ; HSA-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(72) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
-; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR0:%.*]], i8 addrspace(1)* addrspace(1)* undef, align 8
-; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR1:%.*]], i8 addrspace(1)* addrspace(1)* undef, align 8
+; HSA-NEXT:    [[PTR0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT]], i64 0
+; HSA-NEXT:    [[PTR0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR0_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
+; HSA-NEXT:    [[PTR0_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR0_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
+; HSA-NEXT:    [[PTR1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT]], i64 8
+; HSA-NEXT:    [[PTR1_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR1_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
+; HSA-NEXT:    [[PTR1_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR1_KERNARG_OFFSET_CAST]], align 8, !invariant.load !0
+; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR0_LOAD]], i8 addrspace(1)* addrspace(1)* undef, align 8, !noalias !7
+; HSA-NEXT:    store volatile i8 addrspace(1)* [[PTR1_LOAD]], i8 addrspace(1)* addrspace(1)* undef, align 8, !noalias !7
 ; HSA-NEXT:    ret void
 ;
 ; MESA-LABEL: @kern_noalias_global_ptr_x2(
 ; MESA-NEXT:    [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(72) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
-; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR0:%.*]], i8 addrspace(1)* addrspace(1)* undef, align 8
-; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR1:%.*]], i8 addrspace(1)* addrspace(1)* undef, align 8
+; MESA-NEXT:    [[PTR0_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT]], i64 36
+; MESA-NEXT:    [[PTR0_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR0_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
+; MESA-NEXT:    [[PTR0_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR0_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
+; MESA-NEXT:    [[PTR1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[KERN_NOALIAS_GLOBAL_PTR_X2_KERNARG_SEGMENT]], i64 44
+; MESA-NEXT:    [[PTR1_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[PTR1_KERNARG_OFFSET]] to i8 addrspace(1)* addrspace(4)*
+; MESA-NEXT:    [[PTR1_LOAD:%.*]] = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(4)* [[PTR1_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
+; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR0_LOAD]], i8 addrspace(1)* addrspace(1)* undef, align 8, !noalias !7
+; MESA-NEXT:    store volatile i8 addrspace(1)* [[PTR1_LOAD]], i8 addrspace(1)* addrspace(1)* undef, align 8, !noalias !7
 ; MESA-NEXT:    ret void
 ;
   store volatile i8 addrspace(1)* %ptr0, i8 addrspace(1)* addrspace(1)* undef
   store volatile i8 addrspace(1)* %ptr1, i8 addrspace(1)* addrspace(1)* undef
   ret void
 }
 
 define amdgpu_kernel void @struct_i8_i8_arg({i8, i8} %in) #0 {
 ; HSA-LABEL: @struct_i8_i8_arg(
 ; HSA-NEXT:  entry:
 ; HSA-NEXT:    [[STRUCT_I8_I8_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(64) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()

 ; MESA-NEXT:    store i32 [[IN]], i32 addrspace(1)* [[OUT_LOAD]], align 4
 ; MESA-NEXT:    ret void
 ;
   %in = load i32, i32 addrspace(1)* %in.byref
   store i32 %in, i32 addrspace(1)* %out, align 4
   ret void
 }
 
+define amdgpu_kernel void @noalias_byref_global_i32_arg(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* noalias byref(i32) %in.byref) {
+; HSA-LABEL: @noalias_byref_global_i32_arg(
+; HSA-NEXT:    [[NOALIAS_BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(72) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
+; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[NOALIAS_BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT]], i64 0
+; HSA-NEXT:    [[OUT_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[OUT_KERNARG_OFFSET]] to i32 addrspace(1)* addrspace(4)*
+; HSA-NEXT:    [[OUT_LOAD:%.*]] = load i32 addrspace(1)*, i32 addrspace(1)* addrspace(4)* [[OUT_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
+; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[NOALIAS_BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT]], i64 8
+; HSA-NEXT:    [[TMP1:%.*]] = addrspacecast i8 addrspace(4)* [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to i32 addrspace(1)*
+; HSA-NEXT:    [[IN:%.*]] = load i32, i32 addrspace(1)* [[TMP1]], align 4, !alias.scope !11
+; HSA-NEXT:    store i32 [[IN]], i32 addrspace(1)* [[OUT_LOAD]], align 4, !noalias !11
+; HSA-NEXT:    ret void
+;
+; MESA-LABEL: @noalias_byref_global_i32_arg(
+; MESA-NEXT:    [[NOALIAS_BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(68) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
+; MESA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[NOALIAS_BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT]], i64 36
+; MESA-NEXT:    [[OUT_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[OUT_KERNARG_OFFSET]] to i32 addrspace(1)* addrspace(4)*
+; MESA-NEXT:    [[OUT_LOAD:%.*]] = load i32 addrspace(1)*, i32 addrspace(1)* addrspace(4)* [[OUT_KERNARG_OFFSET_CAST]], align 4, !invariant.load !0
+; MESA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[NOALIAS_BYREF_GLOBAL_I32_ARG_KERNARG_SEGMENT]], i64 44
+; MESA-NEXT:    [[TMP1:%.*]] = addrspacecast i8 addrspace(4)* [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to i32 addrspace(1)*
+; MESA-NEXT:    [[IN:%.*]] = load i32, i32 addrspace(1)* [[TMP1]], align 4, !alias.scope !11
+; MESA-NEXT:    store i32 [[IN]], i32 addrspace(1)* [[OUT_LOAD]], align 4, !noalias !11
+; MESA-NEXT:    ret void
+;
+  %in = load i32, i32 addrspace(1)* %in.byref
+  store i32 %in, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
 define amdgpu_kernel void @byref_flat_i32_arg(i32 addrspace(1)* nocapture %out, i32* byref(i32) %in.byref) {
 ; HSA-LABEL: @byref_flat_i32_arg(
 ; HSA-NEXT:    [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(72) i8 addrspace(4)* @llvm.amdgcn.kernarg.segment.ptr()
 ; HSA-NEXT:    [[OUT_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT]], i64 0
 ; HSA-NEXT:    [[OUT_KERNARG_OFFSET_CAST:%.*]] = bitcast i8 addrspace(4)* [[OUT_KERNARG_OFFSET]] to i32 addrspace(1)* addrspace(4)*
 ; HSA-NEXT:    [[OUT_LOAD:%.*]] = load i32 addrspace(1)*, i32 addrspace(1)* addrspace(4)* [[OUT_KERNARG_OFFSET_CAST]], align 16, !invariant.load !0
 ; HSA-NEXT:    [[IN_BYREF_BYVAL_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, i8 addrspace(4)* [[BYREF_FLAT_I32_ARG_KERNARG_SEGMENT]], i64 8
 ; HSA-NEXT:    [[TMP1:%.*]] = addrspacecast i8 addrspace(4)* [[IN_BYREF_BYVAL_KERNARG_OFFSET]] to i32*

 attributes #0 = { nounwind "target-cpu"="kaveri" }
 attributes #1 = { nounwind "target-cpu"="kaveri" "amdgpu-implicitarg-num-bytes"="40" }
 attributes #2 = { nounwind "target-cpu"="tahiti" }
 
 ; GCN: 0 = !{}
 ; GCN: !1 = !{i64 42}
 ; GCN: !2 = !{i64 128}
 ; GCN: !3 = !{i64 1024}
+; GCN: !4 = !{!5}
+; GCN: !5 = distinct !{!5, !6, !"kern_noalias_global_ptr: %ptr"}
+; GCN: !6 = distinct !{!6, !"kern_noalias_global_ptr"}
+; GCN: !7 = !{!8, !10}
+; GCN: !8 = distinct !{!8, !9, !"kern_noalias_global_ptr_x2: %ptr0"}
+; GCN: !9 = distinct !{!9, !"kern_noalias_global_ptr_x2"}
+; GCN: !10 = distinct !{!10, !9, !"kern_noalias_global_ptr_x2: %ptr1"}
+; GCN: !11 = !{!12}
+; GCN: !12 = distinct !{!12, !13, !"noalias_byref_global_i32_arg: %in.byref"}
+; GCN: !13 = distinct !{!13, !"noalias_byref_global_i32_arg"}

llvm/test/CodeGen/AMDGPU/memory_clause.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -mcpu=gfx902 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
 ; RUN: llc -march=amdgcn -mcpu=gfx1030 -mattr=+enable-flat-scratch -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-SCRATCH %s
 
 define amdgpu_kernel void @vector_clause(<4 x i32> addrspace(1)* noalias nocapture readonly %arg, <4 x i32> addrspace(1)* noalias nocapture %arg1) {
 ; GCN-LABEL: vector_clause:
 ; GCN:       ; %bb.0: ; %bb
-; GCN-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GCN-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x2c
+; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GCN-NEXT:    v_lshlrev_b32_e32 v16, 4, v0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    global_load_dwordx4 v[0:3], v16, s[2:3]
-; GCN-NEXT:    global_load_dwordx4 v[4:7], v16, s[2:3] offset:16
-; GCN-NEXT:    global_load_dwordx4 v[8:11], v16, s[2:3] offset:32
-; GCN-NEXT:    global_load_dwordx4 v[12:15], v16, s[2:3] offset:48
+; GCN-NEXT:    global_load_dwordx4 v[0:3], v16, s[0:1]
+; GCN-NEXT:    global_load_dwordx4 v[4:7], v16, s[0:1] offset:16
+; GCN-NEXT:    global_load_dwordx4 v[8:11], v16, s[0:1] offset:32
+; GCN-NEXT:    global_load_dwordx4 v[12:15], v16, s[0:1] offset:48
 ; GCN-NEXT:    s_waitcnt vmcnt(3)
-; GCN-NEXT:    global_store_dwordx4 v16, v[0:3], s[4:5]
+; GCN-NEXT:    global_store_dwordx4 v16, v[0:3], s[2:3]
 ; GCN-NEXT:    s_waitcnt vmcnt(3)
-; GCN-NEXT:    global_store_dwordx4 v16, v[4:7], s[4:5] offset:16
+; GCN-NEXT:    global_store_dwordx4 v16, v[4:7], s[2:3] offset:16
 ; GCN-NEXT:    s_waitcnt vmcnt(3)
-; GCN-NEXT:    global_store_dwordx4 v16, v[8:11], s[4:5] offset:32
+; GCN-NEXT:    global_store_dwordx4 v16, v[8:11], s[2:3] offset:32
 ; GCN-NEXT:    s_waitcnt vmcnt(3)
-; GCN-NEXT:    global_store_dwordx4 v16, v[12:15], s[4:5] offset:48
+; GCN-NEXT:    global_store_dwordx4 v16, v[12:15], s[2:3] offset:48
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-SCRATCH-LABEL: vector_clause:
 ; GCN-SCRATCH:       ; %bb.0: ; %bb
-; GCN-SCRATCH-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GCN-SCRATCH-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GCN-SCRATCH-NEXT:    v_lshlrev_b32_e32 v16, 4, v0
-; GCN-SCRATCH-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x2c
 ; GCN-SCRATCH-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-SCRATCH-NEXT:    s_clause 0x3
-; GCN-SCRATCH-NEXT:    global_load_dwordx4 v[0:3], v16, s[2:3]
-; GCN-SCRATCH-NEXT:    global_load_dwordx4 v[4:7], v16, s[2:3] offset:16
-; GCN-SCRATCH-NEXT:    global_load_dwordx4 v[8:11], v16, s[2:3] offset:32
-; GCN-SCRATCH-NEXT:    global_load_dwordx4 v[12:15], v16, s[2:3] offset:48
+; GCN-SCRATCH-NEXT:    global_load_dwordx4 v[0:3], v16, s[0:1]
+; GCN-SCRATCH-NEXT:    global_load_dwordx4 v[4:7], v16, s[0:1] offset:16
+; GCN-SCRATCH-NEXT:    global_load_dwordx4 v[8:11], v16, s[0:1] offset:32
+; GCN-SCRATCH-NEXT:    global_load_dwordx4 v[12:15], v16, s[0:1] offset:48
 ; GCN-SCRATCH-NEXT:    s_waitcnt vmcnt(3)
-; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[0:3], s[0:1]
+; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[0:3], s[2:3]
 ; GCN-SCRATCH-NEXT:    s_waitcnt vmcnt(2)
-; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[4:7], s[0:1] offset:16
+; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[4:7], s[2:3] offset:16
 ; GCN-SCRATCH-NEXT:    s_waitcnt vmcnt(1)
-; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[8:11], s[0:1] offset:32
+; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[8:11], s[2:3] offset:32
 ; GCN-SCRATCH-NEXT:    s_waitcnt vmcnt(0)
-; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[12:15], s[0:1] offset:48
+; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[12:15], s[2:3] offset:48
 ; GCN-SCRATCH-NEXT:    s_endpgm
 bb:
   %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
   %tmp2 = zext i32 %tmp to i64
   %tmp3 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 %tmp2
   %tmp4 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp3, align 16
   %tmp5 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 %tmp2
   %tmp6 = add nuw nsw i64 %tmp2, 1
bb:
   store <4 x i32> %tmp12, <4 x i32> addrspace(1)* %tmp13, align 16
   store <4 x i32> %tmp16, <4 x i32> addrspace(1)* %tmp17, align 16
   ret void
 }
 
 define amdgpu_kernel void @scalar_clause(<4 x i32> addrspace(1)* noalias nocapture readonly %arg, <4 x i32> addrspace(1)* noalias nocapture %arg1) {
 ; GCN-LABEL: scalar_clause:
 ; GCN:       ; %bb.0: ; %bb
-; GCN-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GCN-NEXT:    s_load_dwordx2 s[16:17], s[0:1], 0x2c
+; GCN-NEXT:    s_load_dwordx4 s[16:19], s[0:1], 0x24
 ; GCN-NEXT:    v_mov_b32_e32 v16, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_load_dwordx16 s[0:15], s[2:3], 0x0
+; GCN-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    v_mov_b32_e32 v1, s1
 ; GCN-NEXT:    v_mov_b32_e32 v2, s2
 ; GCN-NEXT:    v_mov_b32_e32 v3, s3
 ; GCN-NEXT:    v_mov_b32_e32 v4, s4
 ; GCN-NEXT:    v_mov_b32_e32 v5, s5
 ; GCN-NEXT:    v_mov_b32_e32 v6, s6
 ; GCN-NEXT:    v_mov_b32_e32 v7, s7
 ; GCN-NEXT:    v_mov_b32_e32 v8, s8
 ; GCN-NEXT:    v_mov_b32_e32 v9, s9
 ; GCN-NEXT:    v_mov_b32_e32 v10, s10
 ; GCN-NEXT:    v_mov_b32_e32 v11, s11
 ; GCN-NEXT:    v_mov_b32_e32 v12, s12
 ; GCN-NEXT:    v_mov_b32_e32 v13, s13
 ; GCN-NEXT:    v_mov_b32_e32 v14, s14
 ; GCN-NEXT:    v_mov_b32_e32 v15, s15
-; GCN-NEXT:    global_store_dwordx4 v16, v[0:3], s[16:17]
-; GCN-NEXT:    global_store_dwordx4 v16, v[4:7], s[16:17] offset:16
-; GCN-NEXT:    global_store_dwordx4 v16, v[8:11], s[16:17] offset:32
-; GCN-NEXT:    global_store_dwordx4 v16, v[12:15], s[16:17] offset:48
+; GCN-NEXT:    global_store_dwordx4 v16, v[0:3], s[18:19]
+; GCN-NEXT:    global_store_dwordx4 v16, v[4:7], s[18:19] offset:16
+; GCN-NEXT:    global_store_dwordx4 v16, v[8:11], s[18:19] offset:32
+; GCN-NEXT:    global_store_dwordx4 v16, v[12:15], s[18:19] offset:48
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-SCRATCH-LABEL: scalar_clause:
 ; GCN-SCRATCH:       ; %bb.0: ; %bb
-; GCN-SCRATCH-NEXT:    s_clause 0x1
-; GCN-SCRATCH-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GCN-SCRATCH-NEXT:    s_load_dwordx2 s[16:17], s[0:1], 0x2c
+; GCN-SCRATCH-NEXT:    s_load_dwordx4 s[16:19], s[0:1], 0x24
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v16, 0
 ; GCN-SCRATCH-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-SCRATCH-NEXT:    s_load_dwordx16 s[0:15], s[2:3], 0x0
+; GCN-SCRATCH-NEXT:    s_load_dwordx16 s[0:15], s[16:17], 0x0
 ; GCN-SCRATCH-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v1, s1
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v2, s2
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v3, s3
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v4, s4
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v5, s5
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v6, s6
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v7, s7
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v8, s8
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v9, s9
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v10, s10
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v11, s11
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v12, s12
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v13, s13
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v14, s14
 ; GCN-SCRATCH-NEXT:    v_mov_b32_e32 v15, s15
-; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[0:3], s[16:17]
-; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[4:7], s[16:17] offset:16
-; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[8:11], s[16:17] offset:32
-; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[12:15], s[16:17] offset:48
+; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[0:3], s[18:19]
+; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[4:7], s[18:19] offset:16
+; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[8:11], s[18:19] offset:32
+; GCN-SCRATCH-NEXT:    global_store_dwordx4 v16, v[12:15], s[18:19] offset:48
 ; GCN-SCRATCH-NEXT:    s_endpgm
 bb:
   %tmp = load <4 x i32>, <4 x i32> addrspace(1)* %arg, align 16
   %tmp2 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 1
   %tmp3 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp2, align 16
   %tmp4 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg1, i64 1
   %tmp5 = getelementptr inbounds <4 x i32>, <4 x i32> addrspace(1)* %arg, i64 2
   %tmp6 = load <4 x i32>, <4 x i32> addrspace(1)* %tmp5, align 16

llvm/test/CodeGen/AMDGPU/select-constant-cttz.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
 
 declare i32 @llvm.cttz.i32(i32, i1) nounwind readnone
 declare i32 @llvm.amdgcn.sffbh.i32(i32) nounwind readnone speculatable
 define amdgpu_kernel void @select_constant_cttz(i32 addrspace(1)* noalias %out, i32 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; GCN-LABEL: select_constant_cttz:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xb
+; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_load_dword s2, s[2:3], 0x0
-; GCN-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; GCN-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-NEXT:    s_mov_b32 s6, -1
+; GCN-NEXT:    s_mov_b32 s4, s0
+; GCN-NEXT:    s_mov_b32 s5, s1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_lshr_b32 s0, 1, s2
 ; GCN-NEXT:    s_ff1_i32_b32 s0, s0
 ; GCN-NEXT:    s_cmp_lg_u32 s2, 0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    s_cselect_b64 s[2:3], -1, 0
 ; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, -1, s[2:3]
 ; GCN-NEXT:    v_ffbh_i32_e32 v1, v0