[llvm] 6454391 - AMDGPU/GlobalISel: Widen s1 SGPR constants during regbankselect
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 10 11:45:31 PST 2023
Author: Matt Arsenault
Date: 2023-01-10T14:45:23-05:00
New Revision: 6454391b3111993f277fb2570d28f59699b7dae2
URL: https://github.com/llvm/llvm-project/commit/6454391b3111993f277fb2570d28f59699b7dae2
DIFF: https://github.com/llvm/llvm-project/commit/6454391b3111993f277fb2570d28f59699b7dae2.diff
LOG: AMDGPU/GlobalISel: Widen s1 SGPR constants during regbankselect
To unambiguously interpret these as 32-bit SGPRs, we need to widen
these to s32. This was selecting to a copy from a 64-bit SGPR to a
32-bit SGPR for wave64.
Added:
llvm/test/CodeGen/AMDGPU/GlobalISel/br-constant-invalid-sgpr-copy.ll
Modified:
llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
llvm/test/CodeGen/AMDGPU/GlobalISel/bool-legalization.ll
llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-constant.mir
llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.set.inactive.ll
llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll
llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.kill.mir
llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.wqm.demote.mir
llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-xor.mir
llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index 3b322d87ab51..bffbb3d78464 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -2117,6 +2117,35 @@ void AMDGPURegisterBankInfo::applyMappingImpl(
unsigned Opc = MI.getOpcode();
MachineRegisterInfo &MRI = OpdMapper.getMRI();
switch (Opc) {
+ case AMDGPU::G_CONSTANT: {
+ Register DstReg = MI.getOperand(0).getReg();
+ LLT DstTy = MRI.getType(DstReg);
+ if (DstTy != LLT::scalar(1))
+ break;
+
+ const RegisterBank *DstBank =
+ OpdMapper.getInstrMapping().getOperandMapping(0).BreakDown[0].RegBank;
+ if (DstBank == &AMDGPU::VCCRegBank)
+ break;
+ SmallVector<Register, 1> DefRegs(OpdMapper.getVRegs(0));
+ if (DefRegs.empty())
+ DefRegs.push_back(DstReg);
+
+ MachineIRBuilder B(MI);
+ B.setInsertPt(*MI.getParent(), ++MI.getIterator());
+
+ Register NewDstReg = MRI.createGenericVirtualRegister(LLT::scalar(32));
+ LLVMContext &Ctx = B.getMF().getFunction().getContext();
+
+ MI.getOperand(0).setReg(NewDstReg);
+ uint64_t ConstVal = MI.getOperand(1).getCImm()->getZExtValue();
+ MI.getOperand(1).setCImm(
+ ConstantInt::get(IntegerType::getInt32Ty(Ctx), ConstVal));
+
+ MRI.setRegBank(NewDstReg, *DstBank);
+ B.buildTrunc(DefRegs[0], NewDstReg);
+ return;
+ }
case AMDGPU::G_PHI: {
Register DstReg = MI.getOperand(0).getReg();
LLT DstTy = MRI.getType(DstReg);
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/bool-legalization.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/bool-legalization.ll
index 7b3bc41a8259..aeb6d7d9dc71 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/bool-legalization.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/bool-legalization.ll
@@ -68,7 +68,7 @@ define amdgpu_kernel void @sgpr_trunc_brcond(i32 %cond) {
; WAVE64: ; %bb.0: ; %entry
; WAVE64-NEXT: s_load_dword s0, s[0:1], 0x24
; WAVE64-NEXT: s_waitcnt lgkmcnt(0)
-; WAVE64-NEXT: s_xor_b32 s0, s0, -1
+; WAVE64-NEXT: s_xor_b32 s0, s0, 1
; WAVE64-NEXT: s_and_b32 s0, s0, 1
; WAVE64-NEXT: s_cmp_lg_u32 s0, 0
; WAVE64-NEXT: s_cbranch_scc1 .LBB3_2
@@ -85,7 +85,7 @@ define amdgpu_kernel void @sgpr_trunc_brcond(i32 %cond) {
; WAVE32: ; %bb.0: ; %entry
; WAVE32-NEXT: s_load_dword s0, s[0:1], 0x24
; WAVE32-NEXT: s_waitcnt lgkmcnt(0)
-; WAVE32-NEXT: s_xor_b32 s0, s0, -1
+; WAVE32-NEXT: s_xor_b32 s0, s0, 1
; WAVE32-NEXT: s_and_b32 s0, s0, 1
; WAVE32-NEXT: s_cmp_lg_u32 s0, 0
; WAVE32-NEXT: s_cbranch_scc1 .LBB3_2
@@ -116,7 +116,7 @@ define amdgpu_kernel void @brcond_sgpr_trunc_and(i32 %cond0, i32 %cond1) {
; WAVE64-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; WAVE64-NEXT: s_waitcnt lgkmcnt(0)
; WAVE64-NEXT: s_and_b32 s0, s0, s1
-; WAVE64-NEXT: s_xor_b32 s0, s0, -1
+; WAVE64-NEXT: s_xor_b32 s0, s0, 1
; WAVE64-NEXT: s_and_b32 s0, s0, 1
; WAVE64-NEXT: s_cmp_lg_u32 s0, 0
; WAVE64-NEXT: s_cbranch_scc1 .LBB4_2
@@ -134,7 +134,7 @@ define amdgpu_kernel void @brcond_sgpr_trunc_and(i32 %cond0, i32 %cond1) {
; WAVE32-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; WAVE32-NEXT: s_waitcnt lgkmcnt(0)
; WAVE32-NEXT: s_and_b32 s0, s0, s1
-; WAVE32-NEXT: s_xor_b32 s0, s0, -1
+; WAVE32-NEXT: s_xor_b32 s0, s0, 1
; WAVE32-NEXT: s_and_b32 s0, s0, 1
; WAVE32-NEXT: s_cmp_lg_u32 s0, 0
; WAVE32-NEXT: s_cbranch_scc1 .LBB4_2
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/br-constant-invalid-sgpr-copy.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/br-constant-invalid-sgpr-copy.ll
new file mode 100644
index 000000000000..351b0234b58d
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/br-constant-invalid-sgpr-copy.ll
@@ -0,0 +1,142 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -global-isel -O0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=WAVE64 %s
+; RUN: llc -global-isel -O0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1031 < %s | FileCheck -check-prefix=WAVE32 %s
+
+; This was mishandling the constant true and false values used as a
+; scalar branch condition.
+
+define void @br_false() {
+; WAVE64-LABEL: br_false:
+; WAVE64: ; %bb.0: ; %.exit
+; WAVE64-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; WAVE64-NEXT: .LBB0_1: ; %bb0
+; WAVE64-NEXT: ; =>This Inner Loop Header: Depth=1
+; WAVE64-NEXT: s_mov_b32 s4, 1
+; WAVE64-NEXT: s_cmp_lg_u32 s4, 0
+; WAVE64-NEXT: s_cbranch_scc1 .LBB0_1
+; WAVE64-NEXT: ; %bb.2: ; %.exit5
+; WAVE64-NEXT: s_setpc_b64 s[30:31]
+;
+; WAVE32-LABEL: br_false:
+; WAVE32: ; %bb.0: ; %.exit
+; WAVE32-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; WAVE32-NEXT: s_waitcnt_vscnt null, 0x0
+; WAVE32-NEXT: .LBB0_1: ; %bb0
+; WAVE32-NEXT: ; =>This Inner Loop Header: Depth=1
+; WAVE32-NEXT: s_mov_b32 s4, 1
+; WAVE32-NEXT: s_cmp_lg_u32 s4, 0
+; WAVE32-NEXT: s_cbranch_scc1 .LBB0_1
+; WAVE32-NEXT: ; %bb.2: ; %.exit5
+; WAVE32-NEXT: s_setpc_b64 s[30:31]
+.exit:
+ br label %bb0
+
+bb0:
+ br i1 false, label %.exit5, label %bb0
+
+.exit5:
+ ret void
+}
+
+define void @br_true() {
+; WAVE64-LABEL: br_true:
+; WAVE64: ; %bb.0: ; %.exit
+; WAVE64-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; WAVE64-NEXT: .LBB1_1: ; %bb0
+; WAVE64-NEXT: ; =>This Inner Loop Header: Depth=1
+; WAVE64-NEXT: s_mov_b32 s4, 0
+; WAVE64-NEXT: s_cmp_lg_u32 s4, 0
+; WAVE64-NEXT: s_cbranch_scc1 .LBB1_1
+; WAVE64-NEXT: ; %bb.2: ; %.exit5
+; WAVE64-NEXT: s_setpc_b64 s[30:31]
+;
+; WAVE32-LABEL: br_true:
+; WAVE32: ; %bb.0: ; %.exit
+; WAVE32-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; WAVE32-NEXT: s_waitcnt_vscnt null, 0x0
+; WAVE32-NEXT: .LBB1_1: ; %bb0
+; WAVE32-NEXT: ; =>This Inner Loop Header: Depth=1
+; WAVE32-NEXT: s_mov_b32 s4, 0
+; WAVE32-NEXT: s_cmp_lg_u32 s4, 0
+; WAVE32-NEXT: s_cbranch_scc1 .LBB1_1
+; WAVE32-NEXT: ; %bb.2: ; %.exit5
+; WAVE32-NEXT: s_setpc_b64 s[30:31]
+.exit:
+ br label %bb0
+
+bb0:
+ br i1 true, label %.exit5, label %bb0
+
+.exit5:
+ ret void
+}
+
+define void @br_undef() {
+; WAVE64-LABEL: br_undef:
+; WAVE64: ; %bb.0: ; %.exit
+; WAVE64-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; WAVE64-NEXT: .LBB2_1: ; %bb0
+; WAVE64-NEXT: ; =>This Inner Loop Header: Depth=1
+; WAVE64-NEXT: ; implicit-def: $sgpr4
+; WAVE64-NEXT: s_and_b32 s4, s4, 1
+; WAVE64-NEXT: s_cmp_lg_u32 s4, 0
+; WAVE64-NEXT: s_cbranch_scc1 .LBB2_1
+; WAVE64-NEXT: ; %bb.2: ; %.exit5
+; WAVE64-NEXT: s_setpc_b64 s[30:31]
+;
+; WAVE32-LABEL: br_undef:
+; WAVE32: ; %bb.0: ; %.exit
+; WAVE32-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; WAVE32-NEXT: s_waitcnt_vscnt null, 0x0
+; WAVE32-NEXT: .LBB2_1: ; %bb0
+; WAVE32-NEXT: ; =>This Inner Loop Header: Depth=1
+; WAVE32-NEXT: ; implicit-def: $sgpr4
+; WAVE32-NEXT: s_and_b32 s4, s4, 1
+; WAVE32-NEXT: s_cmp_lg_u32 s4, 0
+; WAVE32-NEXT: s_cbranch_scc1 .LBB2_1
+; WAVE32-NEXT: ; %bb.2: ; %.exit5
+; WAVE32-NEXT: s_setpc_b64 s[30:31]
+.exit:
+ br label %bb0
+
+bb0:
+ br i1 undef, label %.exit5, label %bb0
+
+.exit5:
+ ret void
+}
+
+define void @br_poison() {
+; WAVE64-LABEL: br_poison:
+; WAVE64: ; %bb.0: ; %.exit
+; WAVE64-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; WAVE64-NEXT: .LBB3_1: ; %bb0
+; WAVE64-NEXT: ; =>This Inner Loop Header: Depth=1
+; WAVE64-NEXT: ; implicit-def: $sgpr4
+; WAVE64-NEXT: s_and_b32 s4, s4, 1
+; WAVE64-NEXT: s_cmp_lg_u32 s4, 0
+; WAVE64-NEXT: s_cbranch_scc1 .LBB3_1
+; WAVE64-NEXT: ; %bb.2: ; %.exit5
+; WAVE64-NEXT: s_setpc_b64 s[30:31]
+;
+; WAVE32-LABEL: br_poison:
+; WAVE32: ; %bb.0: ; %.exit
+; WAVE32-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; WAVE32-NEXT: s_waitcnt_vscnt null, 0x0
+; WAVE32-NEXT: .LBB3_1: ; %bb0
+; WAVE32-NEXT: ; =>This Inner Loop Header: Depth=1
+; WAVE32-NEXT: ; implicit-def: $sgpr4
+; WAVE32-NEXT: s_and_b32 s4, s4, 1
+; WAVE32-NEXT: s_cmp_lg_u32 s4, 0
+; WAVE32-NEXT: s_cbranch_scc1 .LBB3_1
+; WAVE32-NEXT: ; %bb.2: ; %.exit5
+; WAVE32-NEXT: s_setpc_b64 s[30:31]
+.exit:
+ br label %bb0
+
+bb0:
+ br i1 poison, label %.exit5, label %bb0
+
+.exit5:
+ ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-constant.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-constant.mir
index ebd6405778af..e968ac207311 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-constant.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-constant.mir
@@ -584,3 +584,53 @@ body: |
%7:vgpr(p999) = G_CONSTANT i64 18446744004990098135
S_ENDPGM 0, implicit %0 , implicit %1 , implicit %2, implicit %3, implicit %4, implicit %5, implicit %6, implicit %7
...
+
+---
+name: zext_sgpr_s1_to_sgpr_s32
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+ ; WAVE64-LABEL: name: zext_sgpr_s1_to_sgpr_s32
+ ; WAVE64: bb.0:
+ ; WAVE64-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000)
+ ; WAVE64-NEXT: {{ $}}
+ ; WAVE64-NEXT: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 -1
+ ; WAVE64-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY [[S_MOV_B64_]]
+ ; WAVE64-NEXT: [[S_AND_B32_:%[0-9]+]]:sreg_32 = S_AND_B32 [[COPY]], 1, implicit-def $scc
+ ; WAVE64-NEXT: $scc = COPY [[S_AND_B32_]]
+ ; WAVE64-NEXT: S_CBRANCH_SCC1 %bb.1, implicit $scc
+ ; WAVE64-NEXT: S_BRANCH %bb.2
+ ; WAVE64-NEXT: {{ $}}
+ ; WAVE64-NEXT: bb.1:
+ ; WAVE64-NEXT: successors: %bb.2(0x80000000)
+ ; WAVE64-NEXT: {{ $}}
+ ; WAVE64-NEXT: {{ $}}
+ ; WAVE64-NEXT: bb.2:
+ ; WAVE32-LABEL: name: zext_sgpr_s1_to_sgpr_s32
+ ; WAVE32: bb.0:
+ ; WAVE32-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000)
+ ; WAVE32-NEXT: {{ $}}
+ ; WAVE32-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
+ ; WAVE32-NEXT: [[S_AND_B32_:%[0-9]+]]:sreg_32 = S_AND_B32 [[S_MOV_B32_]], 1, implicit-def $scc
+ ; WAVE32-NEXT: $scc = COPY [[S_AND_B32_]]
+ ; WAVE32-NEXT: S_CBRANCH_SCC1 %bb.1, implicit $scc
+ ; WAVE32-NEXT: S_BRANCH %bb.2
+ ; WAVE32-NEXT: {{ $}}
+ ; WAVE32-NEXT: bb.1:
+ ; WAVE32-NEXT: successors: %bb.2(0x80000000)
+ ; WAVE32-NEXT: {{ $}}
+ ; WAVE32-NEXT: {{ $}}
+ ; WAVE32-NEXT: bb.2:
+ bb.0:
+ %0:sgpr(s1) = G_CONSTANT i1 true
+ %1:sgpr(s32) = G_ZEXT %0
+ G_BRCOND %1, %bb.1
+ G_BR %bb.2
+
+ bb.1:
+
+ bb.2:
+
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.set.inactive.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.set.inactive.ll
index 95ede542ddd7..8506efc1d978 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.set.inactive.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.set.inactive.ll
@@ -48,6 +48,7 @@ define amdgpu_kernel void @set_inactive_scc(ptr addrspace(1) %out, i32 %in, <4 x
; GCN-NEXT: s_buffer_load_dword s2, s[4:7], 0x0
; GCN-NEXT: s_load_dword s3, s[0:1], 0x2c
; GCN-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GCN-NEXT: s_mov_b32 s4, 1
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_cmp_lg_u32 s2, 56
; GCN-NEXT: s_cselect_b32 s2, 1, 0
@@ -63,19 +64,16 @@ define amdgpu_kernel void @set_inactive_scc(ptr addrspace(1) %out, i32 %in, <4 x
; GCN-NEXT: s_mov_b32 s3, 0xf000
; GCN-NEXT: s_mov_b32 s4, 0
; GCN-NEXT: buffer_store_dword v1, off, s[0:3], 0
-; GCN-NEXT: s_branch .LBB2_3
-; GCN-NEXT: .LBB2_2:
-; GCN-NEXT: s_mov_b32 s4, -1
-; GCN-NEXT: .LBB2_3: ; %Flow
-; GCN-NEXT: s_xor_b32 s2, s4, -1
+; GCN-NEXT: .LBB2_2: ; %Flow
+; GCN-NEXT: s_xor_b32 s2, s4, 1
; GCN-NEXT: s_and_b32 s2, s2, 1
; GCN-NEXT: s_cmp_lg_u32 s2, 0
-; GCN-NEXT: s_cbranch_scc1 .LBB2_5
-; GCN-NEXT: ; %bb.4: ; %.zero
+; GCN-NEXT: s_cbranch_scc1 .LBB2_4
+; GCN-NEXT: ; %bb.3: ; %.zero
; GCN-NEXT: s_mov_b32 s2, -1
; GCN-NEXT: s_mov_b32 s3, 0xf000
; GCN-NEXT: buffer_store_dword v0, off, s[0:3], 0
-; GCN-NEXT: .LBB2_5: ; %.exit
+; GCN-NEXT: .LBB2_4: ; %.exit
; GCN-NEXT: s_endpgm
%val = call i32 @llvm.amdgcn.s.buffer.load.i32(<4 x i32> %desc, i32 0, i32 0)
%cmp = icmp eq i32 %val, 56
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll
index c7e5931c110a..2214d6909528 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/localizer.ll
@@ -8,9 +8,9 @@ define amdgpu_kernel void @localize_constants(i1 %cond) {
; GFX9-LABEL: localize_constants:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_load_dword s1, s[4:5], 0x0
-; GFX9-NEXT: s_mov_b32 s0, -1
+; GFX9-NEXT: s_mov_b32 s0, 1
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-NEXT: s_xor_b32 s1, s1, -1
+; GFX9-NEXT: s_xor_b32 s1, s1, 1
; GFX9-NEXT: s_and_b32 s1, s1, 1
; GFX9-NEXT: s_cmp_lg_u32 s1, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB0_2
@@ -35,7 +35,7 @@ define amdgpu_kernel void @localize_constants(i1 %cond) {
; GFX9-NEXT: global_store_dword v[0:1], v0, off
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: .LBB0_2: ; %Flow
-; GFX9-NEXT: s_xor_b32 s0, s0, -1
+; GFX9-NEXT: s_xor_b32 s0, s0, 1
; GFX9-NEXT: s_and_b32 s0, s0, 1
; GFX9-NEXT: s_cmp_lg_u32 s0, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB0_4
@@ -96,9 +96,9 @@ define amdgpu_kernel void @localize_globals(i1 %cond) {
; GFX9-LABEL: localize_globals:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_load_dword s1, s[4:5], 0x0
-; GFX9-NEXT: s_mov_b32 s0, -1
+; GFX9-NEXT: s_mov_b32 s0, 1
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
-; GFX9-NEXT: s_xor_b32 s1, s1, -1
+; GFX9-NEXT: s_xor_b32 s1, s1, 1
; GFX9-NEXT: s_and_b32 s1, s1, 1
; GFX9-NEXT: s_cmp_lg_u32 s1, 0
; GFX9-NEXT: s_cbranch_scc0 .LBB1_2
@@ -120,7 +120,7 @@ define amdgpu_kernel void @localize_globals(i1 %cond) {
; GFX9-NEXT: global_store_dword v0, v1, s[4:5]
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: .LBB1_2: ; %Flow
-; GFX9-NEXT: s_xor_b32 s0, s0, -1
+; GFX9-NEXT: s_xor_b32 s0, s0, 1
; GFX9-NEXT: s_and_b32 s0, s0, 1
; GFX9-NEXT: s_cmp_lg_u32 s0, 0
; GFX9-NEXT: s_cbranch_scc1 .LBB1_4
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.kill.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.kill.mir
index d419a21d0d96..8f2cabcce8a5 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.kill.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.kill.mir
@@ -51,8 +51,9 @@ legalized: true
body: |
bb.0:
; CHECK-LABEL: name: kill_constant_true
- ; CHECK: [[C:%[0-9]+]]:sgpr(s1) = G_CONSTANT i1 true
- ; CHECK-NEXT: [[COPY:%[0-9]+]]:vcc(s1) = COPY [[C]](s1)
+ ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[C]](s32)
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vcc(s1) = COPY [[TRUNC]](s1)
; CHECK-NEXT: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.kill), [[COPY]](s1)
%0:_(s1) = G_CONSTANT i1 true
G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.kill), %0
@@ -65,8 +66,9 @@ legalized: true
body: |
bb.0:
; CHECK-LABEL: name: kill_constant_false
- ; CHECK: [[C:%[0-9]+]]:sgpr(s1) = G_CONSTANT i1 false
- ; CHECK-NEXT: [[COPY:%[0-9]+]]:vcc(s1) = COPY [[C]](s1)
+ ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[C]](s32)
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vcc(s1) = COPY [[TRUNC]](s1)
; CHECK-NEXT: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.kill), [[COPY]](s1)
%0:_(s1) = G_CONSTANT i1 false
G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.kill), %0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.wqm.demote.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.wqm.demote.mir
index babec48ab6ce..20803f55b943 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.wqm.demote.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-amdgcn.wqm.demote.mir
@@ -51,8 +51,9 @@ legalized: true
body: |
bb.0:
; CHECK-LABEL: name: wqm_demote_constant_true
- ; CHECK: [[C:%[0-9]+]]:sgpr(s1) = G_CONSTANT i1 true
- ; CHECK-NEXT: [[COPY:%[0-9]+]]:vcc(s1) = COPY [[C]](s1)
+ ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[C]](s32)
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vcc(s1) = COPY [[TRUNC]](s1)
; CHECK-NEXT: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.wqm.demote), [[COPY]](s1)
%0:_(s1) = G_CONSTANT i1 true
G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.wqm.demote), %0
@@ -65,8 +66,9 @@ legalized: true
body: |
bb.0:
; CHECK-LABEL: name: wqm_demote_constant_false
- ; CHECK: [[C:%[0-9]+]]:sgpr(s1) = G_CONSTANT i1 false
- ; CHECK-NEXT: [[COPY:%[0-9]+]]:vcc(s1) = COPY [[C]](s1)
+ ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[C]](s32)
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vcc(s1) = COPY [[TRUNC]](s1)
; CHECK-NEXT: G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.wqm.demote), [[COPY]](s1)
%0:_(s1) = G_CONSTANT i1 false
G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.amdgcn.wqm.demote), %0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-xor.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-xor.mir
index 930a1d4e1153..4b6bd57c3a34 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-xor.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-xor.mir
@@ -833,8 +833,9 @@ body: |
; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
; CHECK-NEXT: [[ICMP:%[0-9]+]]:vcc(s1) = G_ICMP intpred(ne), [[COPY]](s32), [[COPY1]]
- ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s1) = G_CONSTANT i1 true
- ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vcc(s1) = COPY [[C1]](s1)
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:sgpr(s1) = G_TRUNC [[C1]](s32)
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:vcc(s1) = COPY [[TRUNC]](s1)
; CHECK-NEXT: [[XOR:%[0-9]+]]:vcc(s1) = G_XOR [[ICMP]], [[COPY2]]
; CHECK-NEXT: S_NOP 0, implicit [[XOR]](s1)
%0:_(s32) = COPY $vgpr0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
index 64ab731adb1f..2f422cd9510d 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
@@ -192,11 +192,12 @@ declare i32 @llvm.amdgcn.readfirstlane(i32)
define amdgpu_ps i64 @s_sdiv_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-LABEL: s_sdiv_i64:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_or_b64 s[6:7], s[2:3], s[4:5]
-; CHECK-NEXT: s_mov_b32 s0, 0
-; CHECK-NEXT: s_mov_b32 s1, -1
-; CHECK-NEXT: s_and_b64 s[6:7], s[6:7], s[0:1]
-; CHECK-NEXT: v_cmp_ne_u64_e64 vcc, s[6:7], 0
+; CHECK-NEXT: s_mov_b32 s6, 0
+; CHECK-NEXT: s_or_b64 s[0:1], s[2:3], s[4:5]
+; CHECK-NEXT: s_mov_b32 s7, -1
+; CHECK-NEXT: s_and_b64 s[0:1], s[0:1], s[6:7]
+; CHECK-NEXT: v_cmp_ne_u64_e64 vcc, s[0:1], 0
+; CHECK-NEXT: s_mov_b32 s0, 1
; CHECK-NEXT: s_cbranch_vccz .LBB1_2
; CHECK-NEXT: ; %bb.1:
; CHECK-NEXT: s_ashr_i32 s6, s3, 31
@@ -326,12 +327,12 @@ define amdgpu_ps i64 @s_sdiv_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-NEXT: s_xor_b64 s[0:1], s[6:7], s[8:9]
; CHECK-NEXT: v_xor_b32_e32 v0, s0, v0
; CHECK-NEXT: v_subrev_i32_e32 v0, vcc, s0, v0
-; CHECK-NEXT: s_mov_b32 s1, 0
+; CHECK-NEXT: s_mov_b32 s0, 0
; CHECK-NEXT: s_branch .LBB1_3
; CHECK-NEXT: .LBB1_2:
; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1
; CHECK-NEXT: .LBB1_3: ; %Flow
-; CHECK-NEXT: s_xor_b32 s0, s1, -1
+; CHECK-NEXT: s_xor_b32 s0, s0, 1
; CHECK-NEXT: s_and_b32 s0, s0, 1
; CHECK-NEXT: s_cmp_lg_u32 s0, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB1_5
@@ -1091,7 +1092,7 @@ define i64 @v_sdiv_i64_pow2k_denom(i64 %num) {
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v4, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], 0, v5, v[1:2]
; CHECK-NEXT: v_mov_b32_e32 v6, 0x1000
-; CHECK-NEXT: s_bfe_i32 s6, -1, 0x10000
+; CHECK-NEXT: s_bfe_i32 s6, 1, 0x10000
; CHECK-NEXT: v_subb_u32_e64 v2, s[4:5], v9, v1, vcc
; CHECK-NEXT: v_sub_i32_e64 v1, s[4:5], v9, v1
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
@@ -1104,7 +1105,7 @@ define i64 @v_sdiv_i64_pow2k_denom(i64 %num) {
; CHECK-NEXT: v_cndmask_b32_e64 v2, v8, v4, s[4:5]
; CHECK-NEXT: v_add_i32_e32 v4, vcc, 1, v5
; CHECK-NEXT: v_addc_u32_e32 v8, vcc, 0, v3, vcc
-; CHECK-NEXT: s_bfe_i32 s4, -1, 0x10000
+; CHECK-NEXT: s_bfe_i32 s4, 1, 0x10000
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v0, v6
; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
; CHECK-NEXT: v_mov_b32_e32 v6, s4
@@ -1404,7 +1405,7 @@ define <2 x i64> @v_sdiv_v2i64_pow2k_denom(<2 x i64> %num) {
; CGP-NEXT: s_movk_i32 s7, 0x1000
; CGP-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; CGP-NEXT: v_rcp_iflag_f32_e32 v4, v4
-; CGP-NEXT: s_bfe_i32 s8, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s8, 1, 0x10000
; CGP-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; CGP-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; CGP-NEXT: v_trunc_f32_e32 v6, v5
@@ -1517,7 +1518,7 @@ define <2 x i64> @v_sdiv_v2i64_pow2k_denom(<2 x i64> %num) {
; CGP-NEXT: v_trunc_f32_e32 v6, v6
; CGP-NEXT: v_mac_f32_e32 v1, 0xcf800000, v6
; CGP-NEXT: v_cvt_u32_f32_e32 v13, v1
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s4, 1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v5
; CGP-NEXT: v_mov_b32_e32 v15, s4
; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v13, 0
@@ -1622,7 +1623,7 @@ define <2 x i64> @v_sdiv_v2i64_pow2k_denom(<2 x i64> %num) {
; CGP-NEXT: v_subb_u32_e64 v4, s[4:5], v12, v3, vcc
; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v12, v3
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; CGP-NEXT: s_bfe_i32 s6, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s6, 1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
@@ -1632,7 +1633,7 @@ define <2 x i64> @v_sdiv_v2i64_pow2k_denom(<2 x i64> %num) {
; CGP-NEXT: v_cndmask_b32_e64 v4, v7, v6, s[4:5]
; CGP-NEXT: v_add_i32_e32 v6, vcc, 1, v11
; CGP-NEXT: v_addc_u32_e32 v7, vcc, 0, v13, vcc
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s4, 1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; CGP-NEXT: v_mov_b32_e32 v5, s4
@@ -1755,7 +1756,7 @@ define i64 @v_sdiv_i64_oddk_denom(i64 %num) {
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v4, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], 0, v5, v[1:2]
; CHECK-NEXT: v_mov_b32_e32 v6, 0x12d8fb
-; CHECK-NEXT: s_bfe_i32 s6, -1, 0x10000
+; CHECK-NEXT: s_bfe_i32 s6, 1, 0x10000
; CHECK-NEXT: v_subb_u32_e64 v2, s[4:5], v9, v1, vcc
; CHECK-NEXT: v_sub_i32_e64 v1, s[4:5], v9, v1
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
@@ -1768,7 +1769,7 @@ define i64 @v_sdiv_i64_oddk_denom(i64 %num) {
; CHECK-NEXT: v_cndmask_b32_e64 v2, v8, v4, s[4:5]
; CHECK-NEXT: v_add_i32_e32 v4, vcc, 1, v5
; CHECK-NEXT: v_addc_u32_e32 v8, vcc, 0, v3, vcc
-; CHECK-NEXT: s_bfe_i32 s4, -1, 0x10000
+; CHECK-NEXT: s_bfe_i32 s4, 1, 0x10000
; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v0, v6
; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
; CHECK-NEXT: v_mov_b32_e32 v6, s4
@@ -2068,7 +2069,7 @@ define <2 x i64> @v_sdiv_v2i64_oddk_denom(<2 x i64> %num) {
; CGP-NEXT: s_mov_b32 s7, 0x12d8fb
; CGP-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; CGP-NEXT: v_rcp_iflag_f32_e32 v4, v4
-; CGP-NEXT: s_bfe_i32 s8, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s8, 1, 0x10000
; CGP-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; CGP-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; CGP-NEXT: v_trunc_f32_e32 v6, v5
@@ -2181,7 +2182,7 @@ define <2 x i64> @v_sdiv_v2i64_oddk_denom(<2 x i64> %num) {
; CGP-NEXT: v_trunc_f32_e32 v6, v6
; CGP-NEXT: v_mac_f32_e32 v1, 0xcf800000, v6
; CGP-NEXT: v_cvt_u32_f32_e32 v13, v1
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s4, 1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v5
; CGP-NEXT: v_mov_b32_e32 v15, s4
; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v13, 0
@@ -2286,7 +2287,7 @@ define <2 x i64> @v_sdiv_v2i64_oddk_denom(<2 x i64> %num) {
; CGP-NEXT: v_subb_u32_e64 v4, s[4:5], v12, v3, vcc
; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v12, v3
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; CGP-NEXT: s_bfe_i32 s6, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s6, 1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
@@ -2296,7 +2297,7 @@ define <2 x i64> @v_sdiv_v2i64_oddk_denom(<2 x i64> %num) {
; CGP-NEXT: v_cndmask_b32_e64 v4, v7, v6, s[4:5]
; CGP-NEXT: v_add_i32_e32 v6, vcc, 1, v11
; CGP-NEXT: v_addc_u32_e32 v7, vcc, 0, v13, vcc
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s4, 1, 0x10000
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v2, v5
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
; CGP-NEXT: v_mov_b32_e32 v5, s4
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
index 91e435aad7ae..a40fe242e758 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll
@@ -188,11 +188,12 @@ declare i32 @llvm.amdgcn.readfirstlane(i32)
define amdgpu_ps i64 @s_srem_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-LABEL: s_srem_i64:
; CHECK: ; %bb.0:
-; CHECK-NEXT: s_or_b64 s[6:7], s[2:3], s[4:5]
-; CHECK-NEXT: s_mov_b32 s0, 0
-; CHECK-NEXT: s_mov_b32 s1, -1
-; CHECK-NEXT: s_and_b64 s[6:7], s[6:7], s[0:1]
-; CHECK-NEXT: v_cmp_ne_u64_e64 vcc, s[6:7], 0
+; CHECK-NEXT: s_mov_b32 s6, 0
+; CHECK-NEXT: s_or_b64 s[0:1], s[2:3], s[4:5]
+; CHECK-NEXT: s_mov_b32 s7, -1
+; CHECK-NEXT: s_and_b64 s[0:1], s[0:1], s[6:7]
+; CHECK-NEXT: v_cmp_ne_u64_e64 vcc, s[0:1], 0
+; CHECK-NEXT: s_mov_b32 s0, 1
; CHECK-NEXT: s_cbranch_vccz .LBB1_2
; CHECK-NEXT: ; %bb.1:
; CHECK-NEXT: s_ashr_i32 s6, s3, 31
@@ -320,12 +321,12 @@ define amdgpu_ps i64 @s_srem_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
; CHECK-NEXT: v_xor_b32_e32 v0, s6, v0
; CHECK-NEXT: v_subrev_i32_e32 v0, vcc, s6, v0
-; CHECK-NEXT: s_mov_b32 s1, 0
+; CHECK-NEXT: s_mov_b32 s0, 0
; CHECK-NEXT: s_branch .LBB1_3
; CHECK-NEXT: .LBB1_2:
; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1
; CHECK-NEXT: .LBB1_3: ; %Flow
-; CHECK-NEXT: s_xor_b32 s0, s1, -1
+; CHECK-NEXT: s_xor_b32 s0, s0, 1
; CHECK-NEXT: s_and_b32 s0, s0, 1
; CHECK-NEXT: s_cmp_lg_u32 s0, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB1_5
@@ -1073,27 +1074,25 @@ define i64 @v_srem_i64_pow2k_denom(i64 %num) {
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v4, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], 0, v5, v[1:2]
; CHECK-NEXT: v_mov_b32_e32 v3, 0x1000
-; CHECK-NEXT: s_bfe_i32 s6, -1, 0x10000
+; CHECK-NEXT: s_bfe_i32 s6, 1, 0x10000
; CHECK-NEXT: v_subb_u32_e64 v2, s[4:5], v9, v1, vcc
; CHECK-NEXT: v_sub_i32_e64 v1, s[4:5], v9, v1
+; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
+; CHECK-NEXT: v_sub_i32_e32 v6, vcc, v0, v3
+; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v3
+; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v6, v3
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, -1, s[4:5]
; CHECK-NEXT: v_mov_b32_e32 v5, s6
; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v2
-; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; CHECK-NEXT: v_cndmask_b32_e64 v4, v5, v4, s[4:5]
-; CHECK-NEXT: v_sub_i32_e32 v5, vcc, v0, v3
-; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; CHECK-NEXT: s_bfe_i32 s4, -1, 0x10000
-; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v5, v3
-; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, -1, vcc
-; CHECK-NEXT: v_mov_b32_e32 v8, s4
+; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1
-; CHECK-NEXT: v_cndmask_b32_e32 v6, v8, v6, vcc
-; CHECK-NEXT: v_sub_i32_e32 v3, vcc, v5, v3
+; CHECK-NEXT: v_cndmask_b32_e64 v4, v5, v4, s[4:5]
+; CHECK-NEXT: v_cndmask_b32_e32 v5, v5, v8, vcc
+; CHECK-NEXT: v_sub_i32_e32 v3, vcc, v6, v3
; CHECK-NEXT: v_subbrev_u32_e32 v8, vcc, 0, v1, vcc
-; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
-; CHECK-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc
+; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
+; CHECK-NEXT: v_cndmask_b32_e32 v3, v6, v3, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v8, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
@@ -1382,7 +1381,7 @@ define <2 x i64> @v_srem_v2i64_pow2k_denom(<2 x i64> %num) {
; CGP-NEXT: s_movk_i32 s7, 0x1000
; CGP-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; CGP-NEXT: v_rcp_iflag_f32_e32 v4, v4
-; CGP-NEXT: s_bfe_i32 s8, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s8, 1, 0x10000
; CGP-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; CGP-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; CGP-NEXT: v_trunc_f32_e32 v6, v5
@@ -1471,159 +1470,153 @@ define <2 x i64> @v_srem_v2i64_pow2k_denom(<2 x i64> %num) {
; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; CGP-NEXT: v_add_i32_e32 v5, vcc, v7, v5
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v5, v[1:2]
-; CGP-NEXT: v_mov_b32_e32 v5, 0x1000
+; CGP-NEXT: v_sub_i32_e32 v9, vcc, v10, v0
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v8, v[6:7]
-; CGP-NEXT: v_sub_i32_e32 v8, vcc, v10, v0
-; CGP-NEXT: v_subb_u32_e64 v9, s[4:5], v11, v6, vcc
+; CGP-NEXT: v_mov_b32_e32 v5, 0x1000
+; CGP-NEXT: v_cvt_f32_ubyte0_e32 v7, 0
+; CGP-NEXT: v_subb_u32_e64 v10, s[4:5], v11, v6, vcc
; CGP-NEXT: v_sub_i32_e64 v0, s[4:5], v11, v6
-; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v8, v5
+; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v9, v5
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, -1, s[4:5]
; CGP-NEXT: v_mov_b32_e32 v6, s8
-; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v9
-; CGP-NEXT: v_cndmask_b32_e64 v10, v6, v1, s[4:5]
+; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v10
+; CGP-NEXT: v_cndmask_b32_e64 v11, v6, v1, s[4:5]
; CGP-NEXT: v_cvt_f32_u32_e32 v1, 0x1000
-; CGP-NEXT: v_cvt_f32_ubyte0_e32 v6, 0
; CGP-NEXT: v_subbrev_u32_e32 v0, vcc, 0, v0, vcc
-; CGP-NEXT: v_mac_f32_e32 v1, 0x4f800000, v6
+; CGP-NEXT: v_sub_i32_e32 v12, vcc, v9, v5
+; CGP-NEXT: v_mac_f32_e32 v1, 0x4f800000, v7
; CGP-NEXT: v_rcp_iflag_f32_e32 v1, v1
-; CGP-NEXT: v_sub_i32_e32 v11, vcc, v8, v5
-; CGP-NEXT: v_subbrev_u32_e32 v12, vcc, 0, v0, vcc
+; CGP-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v0, vcc
+; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v12, v5
; CGP-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v1
; CGP-NEXT: v_mul_f32_e32 v1, 0x2f800000, v0
-; CGP-NEXT: v_trunc_f32_e32 v6, v1
-; CGP-NEXT: v_mac_f32_e32 v0, 0xcf800000, v6
-; CGP-NEXT: v_cvt_u32_f32_e32 v13, v0
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
-; CGP-NEXT: v_mov_b32_e32 v14, s4
-; CGP-NEXT: v_cvt_u32_f32_e32 v15, v6
-; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v13, 0
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v11, v5
-; CGP-NEXT: v_cndmask_b32_e64 v7, 0, -1, vcc
-; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v12
-; CGP-NEXT: v_cndmask_b32_e32 v14, v14, v7, vcc
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v15, v[1:2]
-; CGP-NEXT: v_sub_i32_e32 v1, vcc, v11, v5
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v13, v[6:7]
-; CGP-NEXT: v_subbrev_u32_e32 v16, vcc, 0, v12, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14
-; CGP-NEXT: v_cndmask_b32_e32 v7, v11, v1, vcc
+; CGP-NEXT: v_trunc_f32_e32 v7, v1
+; CGP-NEXT: v_mac_f32_e32 v0, 0xcf800000, v7
+; CGP-NEXT: v_cvt_u32_f32_e32 v14, v0
+; CGP-NEXT: v_cvt_u32_f32_e32 v15, v7
+; CGP-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
+; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v13
+; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v14, 0
+; CGP-NEXT: v_cndmask_b32_e32 v16, v6, v8, vcc
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], s6, v15, v[1:2]
+; CGP-NEXT: v_sub_i32_e32 v1, vcc, v12, v5
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], -1, v14, v[7:8]
+; CGP-NEXT: v_subbrev_u32_e32 v17, vcc, 0, v13, vcc
+; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v16
+; CGP-NEXT: v_cndmask_b32_e32 v8, v12, v1, vcc
; CGP-NEXT: v_mul_lo_u32 v1, v15, v0
-; CGP-NEXT: v_mul_lo_u32 v11, v13, v6
-; CGP-NEXT: v_mul_hi_u32 v14, v13, v0
-; CGP-NEXT: v_cndmask_b32_e32 v12, v12, v16, vcc
+; CGP-NEXT: v_mul_lo_u32 v12, v14, v7
+; CGP-NEXT: v_mul_hi_u32 v16, v14, v0
+; CGP-NEXT: v_cndmask_b32_e32 v13, v13, v17, vcc
; CGP-NEXT: v_mul_hi_u32 v0, v15, v0
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v11
-; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v14
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v12
+; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v16
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v14, v15, v6
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
-; CGP-NEXT: v_mul_hi_u32 v11, v13, v6
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v14, v0
-; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v11
-; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v14, v11
-; CGP-NEXT: v_mul_hi_u32 v6, v15, v6
+; CGP-NEXT: v_mul_lo_u32 v16, v15, v7
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v12, v1
+; CGP-NEXT: v_mul_hi_u32 v12, v14, v7
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v16, v0
+; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v12
+; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v16, v12
+; CGP-NEXT: v_mul_hi_u32 v7, v15, v7
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v6, v1
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v13, v0
-; CGP-NEXT: v_addc_u32_e32 v13, vcc, v15, v1, vcc
-; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v11, 0
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
-; CGP-NEXT: v_cndmask_b32_e32 v8, v8, v7, vcc
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v13, v[1:2]
-; CGP-NEXT: v_xor_b32_e32 v1, v8, v4
-; CGP-NEXT: v_ashrrev_i32_e32 v8, 31, v3
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v11, v[6:7]
-; CGP-NEXT: v_cndmask_b32_e32 v9, v9, v12, vcc
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v12, v1
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v7, v1
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v14, v0
+; CGP-NEXT: v_addc_u32_e32 v14, vcc, v15, v1, vcc
+; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v12, 0
+; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v11
+; CGP-NEXT: v_cndmask_b32_e32 v9, v9, v8, vcc
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], s6, v14, v[1:2]
+; CGP-NEXT: v_xor_b32_e32 v1, v9, v4
+; CGP-NEXT: v_ashrrev_i32_e32 v9, 31, v3
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], -1, v12, v[7:8]
+; CGP-NEXT: v_cndmask_b32_e32 v10, v10, v13, vcc
+; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v9
+; CGP-NEXT: v_addc_u32_e32 v3, vcc, v3, v9, vcc
+; CGP-NEXT: v_xor_b32_e32 v11, v2, v9
+; CGP-NEXT: v_mul_lo_u32 v2, v14, v0
+; CGP-NEXT: v_mul_lo_u32 v8, v12, v7
+; CGP-NEXT: v_xor_b32_e32 v13, v3, v9
+; CGP-NEXT: v_mul_hi_u32 v3, v12, v0
+; CGP-NEXT: v_mul_hi_u32 v0, v14, v0
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v8
-; CGP-NEXT: v_addc_u32_e32 v3, vcc, v3, v8, vcc
-; CGP-NEXT: v_xor_b32_e32 v10, v2, v8
-; CGP-NEXT: v_mul_lo_u32 v2, v13, v0
-; CGP-NEXT: v_mul_lo_u32 v7, v11, v6
-; CGP-NEXT: v_xor_b32_e32 v12, v3, v8
-; CGP-NEXT: v_mul_hi_u32 v3, v11, v0
-; CGP-NEXT: v_mul_hi_u32 v0, v13, v0
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v7
-; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; CGP-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v3, v13, v6
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v7, v2
-; CGP-NEXT: v_mul_hi_u32 v7, v11, v6
+; CGP-NEXT: v_mul_lo_u32 v3, v14, v7
+; CGP-NEXT: v_add_i32_e32 v2, vcc, v8, v2
+; CGP-NEXT: v_mul_hi_u32 v8, v12, v7
; CGP-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
-; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v7
-; CGP-NEXT: v_mul_hi_u32 v6, v13, v6
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v8
+; CGP-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v8
+; CGP-NEXT: v_mul_hi_u32 v7, v14, v7
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v3, v2
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v6, v2
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v11, v0
-; CGP-NEXT: v_addc_u32_e32 v2, vcc, v13, v2, vcc
-; CGP-NEXT: v_mul_lo_u32 v3, v12, v0
-; CGP-NEXT: v_mul_lo_u32 v6, v10, v2
-; CGP-NEXT: v_mul_hi_u32 v7, v10, v0
-; CGP-NEXT: v_mul_hi_u32 v0, v12, v0
-; CGP-NEXT: v_xor_b32_e32 v9, v9, v4
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v6
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v2, vcc, v7, v2
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v12, v0
+; CGP-NEXT: v_addc_u32_e32 v2, vcc, v14, v2, vcc
+; CGP-NEXT: v_mul_lo_u32 v3, v13, v0
+; CGP-NEXT: v_mul_lo_u32 v7, v11, v2
+; CGP-NEXT: v_mul_hi_u32 v8, v11, v0
+; CGP-NEXT: v_mul_hi_u32 v0, v13, v0
+; CGP-NEXT: v_xor_b32_e32 v10, v10, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v7
+; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v8
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v7, v12, v2
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v6, v3
-; CGP-NEXT: v_mul_hi_u32 v6, v10, v2
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
+; CGP-NEXT: v_mul_lo_u32 v8, v13, v2
+; CGP-NEXT: v_add_i32_e32 v3, vcc, v7, v3
+; CGP-NEXT: v_mul_hi_u32 v7, v11, v2
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v8, v0
+; CGP-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v6
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v6, vcc, v7, v6
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v0, v3
-; CGP-NEXT: v_mul_hi_u32 v7, v12, v2
-; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s7, v11, 0
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v8, v7
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v0, v3
+; CGP-NEXT: v_mul_hi_u32 v8, v13, v2
+; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s7, v12, 0
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v6, v0
-; CGP-NEXT: v_add_i32_e32 v6, vcc, v7, v0
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v8, v0
; CGP-NEXT: v_mov_b32_e32 v0, v3
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v6, v[0:1]
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], s7, v7, v[0:1]
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v1, v4
-; CGP-NEXT: v_subb_u32_e32 v1, vcc, v9, v4, vcc
-; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], 0, v11, v[6:7]
-; CGP-NEXT: v_sub_i32_e32 v2, vcc, v10, v2
-; CGP-NEXT: v_subb_u32_e64 v4, s[4:5], v12, v3, vcc
-; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v12, v3
-; CGP-NEXT: s_bfe_i32 s6, -1, 0x10000
-; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
-; CGP-NEXT: v_mov_b32_e32 v7, s6
-; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
+; CGP-NEXT: v_subb_u32_e32 v1, vcc, v10, v4, vcc
+; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], 0, v12, v[7:8]
+; CGP-NEXT: v_sub_i32_e32 v2, vcc, v11, v2
+; CGP-NEXT: v_subb_u32_e64 v4, s[4:5], v13, v3, vcc
+; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v13, v3
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; CGP-NEXT: v_cndmask_b32_e64 v6, v7, v6, s[4:5]
-; CGP-NEXT: v_sub_i32_e32 v7, vcc, v2, v5
+; CGP-NEXT: v_sub_i32_e32 v8, vcc, v2, v5
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v7, v5
-; CGP-NEXT: v_cndmask_b32_e64 v9, 0, -1, vcc
-; CGP-NEXT: v_mov_b32_e32 v10, s4
+; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
+; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v8, v5
+; CGP-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
+; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
+; CGP-NEXT: v_cndmask_b32_e64 v10, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3
-; CGP-NEXT: v_cndmask_b32_e32 v9, v10, v9, vcc
-; CGP-NEXT: v_sub_i32_e32 v5, vcc, v7, v5
+; CGP-NEXT: v_cndmask_b32_e64 v7, v6, v7, s[4:5]
+; CGP-NEXT: v_cndmask_b32_e32 v6, v6, v10, vcc
+; CGP-NEXT: v_sub_i32_e32 v5, vcc, v8, v5
; CGP-NEXT: v_subbrev_u32_e32 v10, vcc, 0, v3, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9
-; CGP-NEXT: v_cndmask_b32_e32 v5, v7, v5, vcc
-; CGP-NEXT: v_cndmask_b32_e32 v3, v3, v10, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
+; CGP-NEXT: v_cndmask_b32_e32 v5, v8, v5, vcc
+; CGP-NEXT: v_cndmask_b32_e32 v3, v3, v10, vcc
+; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v7
; CGP-NEXT: v_cndmask_b32_e32 v2, v2, v5, vcc
; CGP-NEXT: v_cndmask_b32_e32 v3, v4, v3, vcc
-; CGP-NEXT: v_xor_b32_e32 v2, v2, v8
-; CGP-NEXT: v_xor_b32_e32 v3, v3, v8
-; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v8
-; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v8, vcc
+; CGP-NEXT: v_xor_b32_e32 v2, v2, v9
+; CGP-NEXT: v_xor_b32_e32 v3, v3, v9
+; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v9
+; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v9, vcc
; CGP-NEXT: s_setpc_b64 s[30:31]
%result = srem <2 x i64> %num, <i64 4096, i64 4096>
ret <2 x i64> %result
@@ -1729,27 +1722,25 @@ define i64 @v_srem_i64_oddk_denom(i64 %num) {
; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v4, v0
; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], 0, v5, v[1:2]
; CHECK-NEXT: v_mov_b32_e32 v3, 0x12d8fb
-; CHECK-NEXT: s_bfe_i32 s6, -1, 0x10000
+; CHECK-NEXT: s_bfe_i32 s6, 1, 0x10000
; CHECK-NEXT: v_subb_u32_e64 v2, s[4:5], v9, v1, vcc
; CHECK-NEXT: v_sub_i32_e64 v1, s[4:5], v9, v1
+; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
+; CHECK-NEXT: v_sub_i32_e32 v6, vcc, v0, v3
+; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v3
+; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v6, v3
; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, -1, s[4:5]
; CHECK-NEXT: v_mov_b32_e32 v5, s6
; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v2
-; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; CHECK-NEXT: v_cndmask_b32_e64 v4, v5, v4, s[4:5]
-; CHECK-NEXT: v_sub_i32_e32 v5, vcc, v0, v3
-; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; CHECK-NEXT: s_bfe_i32 s4, -1, 0x10000
-; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v5, v3
-; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, -1, vcc
-; CHECK-NEXT: v_mov_b32_e32 v8, s4
+; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1
-; CHECK-NEXT: v_cndmask_b32_e32 v6, v8, v6, vcc
-; CHECK-NEXT: v_sub_i32_e32 v3, vcc, v5, v3
+; CHECK-NEXT: v_cndmask_b32_e64 v4, v5, v4, s[4:5]
+; CHECK-NEXT: v_cndmask_b32_e32 v5, v5, v8, vcc
+; CHECK-NEXT: v_sub_i32_e32 v3, vcc, v6, v3
; CHECK-NEXT: v_subbrev_u32_e32 v8, vcc, 0, v1, vcc
-; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
-; CHECK-NEXT: v_cndmask_b32_e32 v3, v5, v3, vcc
+; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
+; CHECK-NEXT: v_cndmask_b32_e32 v3, v6, v3, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v8, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
@@ -2038,7 +2029,7 @@ define <2 x i64> @v_srem_v2i64_oddk_denom(<2 x i64> %num) {
; CGP-NEXT: s_mov_b32 s7, 0x12d8fb
; CGP-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
; CGP-NEXT: v_rcp_iflag_f32_e32 v4, v4
-; CGP-NEXT: s_bfe_i32 s8, -1, 0x10000
+; CGP-NEXT: s_bfe_i32 s8, 1, 0x10000
; CGP-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
; CGP-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
; CGP-NEXT: v_trunc_f32_e32 v6, v5
@@ -2127,159 +2118,153 @@ define <2 x i64> @v_srem_v2i64_oddk_denom(<2 x i64> %num) {
; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v5
; CGP-NEXT: v_add_i32_e32 v5, vcc, v7, v5
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v5, v[1:2]
-; CGP-NEXT: v_mov_b32_e32 v5, 0x12d8fb
+; CGP-NEXT: v_sub_i32_e32 v9, vcc, v10, v0
; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v8, v[6:7]
-; CGP-NEXT: v_sub_i32_e32 v8, vcc, v10, v0
-; CGP-NEXT: v_subb_u32_e64 v9, s[4:5], v11, v6, vcc
+; CGP-NEXT: v_mov_b32_e32 v5, 0x12d8fb
+; CGP-NEXT: v_cvt_f32_ubyte0_e32 v7, 0
+; CGP-NEXT: v_subb_u32_e64 v10, s[4:5], v11, v6, vcc
; CGP-NEXT: v_sub_i32_e64 v0, s[4:5], v11, v6
-; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v8, v5
+; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v9, v5
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, -1, s[4:5]
; CGP-NEXT: v_mov_b32_e32 v6, s8
-; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v9
-; CGP-NEXT: v_cndmask_b32_e64 v10, v6, v1, s[4:5]
+; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v10
+; CGP-NEXT: v_cndmask_b32_e64 v11, v6, v1, s[4:5]
; CGP-NEXT: v_cvt_f32_u32_e32 v1, 0x12d8fb
-; CGP-NEXT: v_cvt_f32_ubyte0_e32 v6, 0
; CGP-NEXT: v_subbrev_u32_e32 v0, vcc, 0, v0, vcc
-; CGP-NEXT: v_mac_f32_e32 v1, 0x4f800000, v6
+; CGP-NEXT: v_sub_i32_e32 v12, vcc, v9, v5
+; CGP-NEXT: v_mac_f32_e32 v1, 0x4f800000, v7
; CGP-NEXT: v_rcp_iflag_f32_e32 v1, v1
-; CGP-NEXT: v_sub_i32_e32 v11, vcc, v8, v5
-; CGP-NEXT: v_subbrev_u32_e32 v12, vcc, 0, v0, vcc
+; CGP-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v0, vcc
+; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v12, v5
; CGP-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v1
; CGP-NEXT: v_mul_f32_e32 v1, 0x2f800000, v0
-; CGP-NEXT: v_trunc_f32_e32 v6, v1
-; CGP-NEXT: v_mac_f32_e32 v0, 0xcf800000, v6
-; CGP-NEXT: v_cvt_u32_f32_e32 v13, v0
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
-; CGP-NEXT: v_mov_b32_e32 v14, s4
-; CGP-NEXT: v_cvt_u32_f32_e32 v15, v6
-; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v13, 0
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v11, v5
-; CGP-NEXT: v_cndmask_b32_e64 v7, 0, -1, vcc
-; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v12
-; CGP-NEXT: v_cndmask_b32_e32 v14, v14, v7, vcc
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v15, v[1:2]
-; CGP-NEXT: v_sub_i32_e32 v1, vcc, v11, v5
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v13, v[6:7]
-; CGP-NEXT: v_subbrev_u32_e32 v16, vcc, 0, v12, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14
-; CGP-NEXT: v_cndmask_b32_e32 v7, v11, v1, vcc
+; CGP-NEXT: v_trunc_f32_e32 v7, v1
+; CGP-NEXT: v_mac_f32_e32 v0, 0xcf800000, v7
+; CGP-NEXT: v_cvt_u32_f32_e32 v14, v0
+; CGP-NEXT: v_cvt_u32_f32_e32 v15, v7
+; CGP-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
+; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v13
+; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v14, 0
+; CGP-NEXT: v_cndmask_b32_e32 v16, v6, v8, vcc
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], s6, v15, v[1:2]
+; CGP-NEXT: v_sub_i32_e32 v1, vcc, v12, v5
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], -1, v14, v[7:8]
+; CGP-NEXT: v_subbrev_u32_e32 v17, vcc, 0, v13, vcc
+; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v16
+; CGP-NEXT: v_cndmask_b32_e32 v8, v12, v1, vcc
; CGP-NEXT: v_mul_lo_u32 v1, v15, v0
-; CGP-NEXT: v_mul_lo_u32 v11, v13, v6
-; CGP-NEXT: v_mul_hi_u32 v14, v13, v0
-; CGP-NEXT: v_cndmask_b32_e32 v12, v12, v16, vcc
+; CGP-NEXT: v_mul_lo_u32 v12, v14, v7
+; CGP-NEXT: v_mul_hi_u32 v16, v14, v0
+; CGP-NEXT: v_cndmask_b32_e32 v13, v13, v17, vcc
; CGP-NEXT: v_mul_hi_u32 v0, v15, v0
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v11
-; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v14
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v12
+; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v16
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v14, v15, v6
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
-; CGP-NEXT: v_mul_hi_u32 v11, v13, v6
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v14, v0
-; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v11
-; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v14, v11
-; CGP-NEXT: v_mul_hi_u32 v6, v15, v6
+; CGP-NEXT: v_mul_lo_u32 v16, v15, v7
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v12, v1
+; CGP-NEXT: v_mul_hi_u32 v12, v14, v7
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v16, v0
+; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v12
+; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v16, v12
+; CGP-NEXT: v_mul_hi_u32 v7, v15, v7
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v11, v1
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v6, v1
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v13, v0
-; CGP-NEXT: v_addc_u32_e32 v13, vcc, v15, v1, vcc
-; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v11, 0
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
-; CGP-NEXT: v_cndmask_b32_e32 v8, v8, v7, vcc
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v13, v[1:2]
-; CGP-NEXT: v_xor_b32_e32 v1, v8, v4
-; CGP-NEXT: v_ashrrev_i32_e32 v8, 31, v3
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], -1, v11, v[6:7]
-; CGP-NEXT: v_cndmask_b32_e32 v9, v9, v12, vcc
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v12, v1
+; CGP-NEXT: v_add_i32_e32 v1, vcc, v7, v1
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v14, v0
+; CGP-NEXT: v_addc_u32_e32 v14, vcc, v15, v1, vcc
+; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v12, 0
+; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v11
+; CGP-NEXT: v_cndmask_b32_e32 v9, v9, v8, vcc
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], s6, v14, v[1:2]
+; CGP-NEXT: v_xor_b32_e32 v1, v9, v4
+; CGP-NEXT: v_ashrrev_i32_e32 v9, 31, v3
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], -1, v12, v[7:8]
+; CGP-NEXT: v_cndmask_b32_e32 v10, v10, v13, vcc
+; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v9
+; CGP-NEXT: v_addc_u32_e32 v3, vcc, v3, v9, vcc
+; CGP-NEXT: v_xor_b32_e32 v11, v2, v9
+; CGP-NEXT: v_mul_lo_u32 v2, v14, v0
+; CGP-NEXT: v_mul_lo_u32 v8, v12, v7
+; CGP-NEXT: v_xor_b32_e32 v13, v3, v9
+; CGP-NEXT: v_mul_hi_u32 v3, v12, v0
+; CGP-NEXT: v_mul_hi_u32 v0, v14, v0
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v8
-; CGP-NEXT: v_addc_u32_e32 v3, vcc, v3, v8, vcc
-; CGP-NEXT: v_xor_b32_e32 v10, v2, v8
-; CGP-NEXT: v_mul_lo_u32 v2, v13, v0
-; CGP-NEXT: v_mul_lo_u32 v7, v11, v6
-; CGP-NEXT: v_xor_b32_e32 v12, v3, v8
-; CGP-NEXT: v_mul_hi_u32 v3, v11, v0
-; CGP-NEXT: v_mul_hi_u32 v0, v13, v0
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v7
-; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; CGP-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v3
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v3, v13, v6
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v7, v2
-; CGP-NEXT: v_mul_hi_u32 v7, v11, v6
+; CGP-NEXT: v_mul_lo_u32 v3, v14, v7
+; CGP-NEXT: v_add_i32_e32 v2, vcc, v8, v2
+; CGP-NEXT: v_mul_hi_u32 v8, v12, v7
; CGP-NEXT: v_add_i32_e32 v0, vcc, v3, v0
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
-; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v7
-; CGP-NEXT: v_mul_hi_u32 v6, v13, v6
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v8
+; CGP-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v8
+; CGP-NEXT: v_mul_hi_u32 v7, v14, v7
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v2
; CGP-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v2, vcc, v3, v2
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v6, v2
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v11, v0
-; CGP-NEXT: v_addc_u32_e32 v2, vcc, v13, v2, vcc
-; CGP-NEXT: v_mul_lo_u32 v3, v12, v0
-; CGP-NEXT: v_mul_lo_u32 v6, v10, v2
-; CGP-NEXT: v_mul_hi_u32 v7, v10, v0
-; CGP-NEXT: v_mul_hi_u32 v0, v12, v0
-; CGP-NEXT: v_xor_b32_e32 v9, v9, v4
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v6
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v2, vcc, v7, v2
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v12, v0
+; CGP-NEXT: v_addc_u32_e32 v2, vcc, v14, v2, vcc
+; CGP-NEXT: v_mul_lo_u32 v3, v13, v0
+; CGP-NEXT: v_mul_lo_u32 v7, v11, v2
+; CGP-NEXT: v_mul_hi_u32 v8, v11, v0
+; CGP-NEXT: v_mul_hi_u32 v0, v13, v0
+; CGP-NEXT: v_xor_b32_e32 v10, v10, v4
; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v7
+; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v8
; CGP-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v7, v12, v2
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v6, v3
-; CGP-NEXT: v_mul_hi_u32 v6, v10, v2
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
+; CGP-NEXT: v_mul_lo_u32 v8, v13, v2
+; CGP-NEXT: v_add_i32_e32 v3, vcc, v7, v3
+; CGP-NEXT: v_mul_hi_u32 v7, v11, v2
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v8, v0
+; CGP-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v6
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v6, vcc, v7, v6
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v0, v3
-; CGP-NEXT: v_mul_hi_u32 v7, v12, v2
-; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s7, v11, 0
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v8, v7
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v0, v3
+; CGP-NEXT: v_mul_hi_u32 v8, v13, v2
+; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], s7, v12, 0
; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v6, v0
-; CGP-NEXT: v_add_i32_e32 v6, vcc, v7, v0
+; CGP-NEXT: v_add_i32_e32 v0, vcc, v7, v0
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v8, v0
; CGP-NEXT: v_mov_b32_e32 v0, v3
-; CGP-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v6, v[0:1]
+; CGP-NEXT: v_mad_u64_u32 v[7:8], s[4:5], s7, v7, v[0:1]
; CGP-NEXT: v_sub_i32_e32 v0, vcc, v1, v4
-; CGP-NEXT: v_subb_u32_e32 v1, vcc, v9, v4, vcc
-; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], 0, v11, v[6:7]
-; CGP-NEXT: v_sub_i32_e32 v2, vcc, v10, v2
-; CGP-NEXT: v_subb_u32_e64 v4, s[4:5], v12, v3, vcc
-; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v12, v3
-; CGP-NEXT: s_bfe_i32 s6, -1, 0x10000
-; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
-; CGP-NEXT: v_mov_b32_e32 v7, s6
-; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
+; CGP-NEXT: v_subb_u32_e32 v1, vcc, v10, v4, vcc
+; CGP-NEXT: v_mad_u64_u32 v[3:4], s[4:5], 0, v12, v[7:8]
+; CGP-NEXT: v_sub_i32_e32 v2, vcc, v11, v2
+; CGP-NEXT: v_subb_u32_e64 v4, s[4:5], v13, v3, vcc
+; CGP-NEXT: v_sub_i32_e64 v3, s[4:5], v13, v3
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; CGP-NEXT: v_cndmask_b32_e64 v6, v7, v6, s[4:5]
-; CGP-NEXT: v_sub_i32_e32 v7, vcc, v2, v5
+; CGP-NEXT: v_sub_i32_e32 v8, vcc, v2, v5
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v7, v5
-; CGP-NEXT: v_cndmask_b32_e64 v9, 0, -1, vcc
-; CGP-NEXT: v_mov_b32_e32 v10, s4
+; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
+; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v8, v5
+; CGP-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
+; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
+; CGP-NEXT: v_cndmask_b32_e64 v10, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3
-; CGP-NEXT: v_cndmask_b32_e32 v9, v10, v9, vcc
-; CGP-NEXT: v_sub_i32_e32 v5, vcc, v7, v5
+; CGP-NEXT: v_cndmask_b32_e64 v7, v6, v7, s[4:5]
+; CGP-NEXT: v_cndmask_b32_e32 v6, v6, v10, vcc
+; CGP-NEXT: v_sub_i32_e32 v5, vcc, v8, v5
; CGP-NEXT: v_subbrev_u32_e32 v10, vcc, 0, v3, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9
-; CGP-NEXT: v_cndmask_b32_e32 v5, v7, v5, vcc
-; CGP-NEXT: v_cndmask_b32_e32 v3, v3, v10, vcc
; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
+; CGP-NEXT: v_cndmask_b32_e32 v5, v8, v5, vcc
+; CGP-NEXT: v_cndmask_b32_e32 v3, v3, v10, vcc
+; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v7
; CGP-NEXT: v_cndmask_b32_e32 v2, v2, v5, vcc
; CGP-NEXT: v_cndmask_b32_e32 v3, v4, v3, vcc
-; CGP-NEXT: v_xor_b32_e32 v2, v2, v8
-; CGP-NEXT: v_xor_b32_e32 v3, v3, v8
-; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v8
-; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v8, vcc
+; CGP-NEXT: v_xor_b32_e32 v2, v2, v9
+; CGP-NEXT: v_xor_b32_e32 v3, v3, v9
+; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v9
+; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v9, vcc
; CGP-NEXT: s_setpc_b64 s[30:31]
%result = srem <2 x i64> %num, <i64 1235195, i64 1235195>
ret <2 x i64> %result
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
index 0f26b6478791..93e8a43ba773 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll
@@ -186,10 +186,11 @@ declare i32 @llvm.amdgcn.readfirstlane(i32)
define amdgpu_ps i64 @s_udiv_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-LABEL: s_udiv_i64:
; CHECK: ; %bb.0:
+; CHECK-NEXT: s_mov_b32 s4, 1
; CHECK-NEXT: s_or_b64 s[6:7], s[0:1], s[2:3]
-; CHECK-NEXT: s_mov_b32 s4, 0
-; CHECK-NEXT: s_mov_b32 s5, -1
-; CHECK-NEXT: s_and_b64 s[6:7], s[6:7], s[4:5]
+; CHECK-NEXT: s_mov_b32 s8, 0
+; CHECK-NEXT: s_mov_b32 s9, -1
+; CHECK-NEXT: s_and_b64 s[6:7], s[6:7], s[8:9]
; CHECK-NEXT: v_cmp_ne_u64_e64 vcc, s[6:7], 0
; CHECK-NEXT: v_cvt_f32_u32_e32 v2, s2
; CHECK-NEXT: s_cbranch_vccz .LBB1_2
@@ -316,12 +317,12 @@ define amdgpu_ps i64 @s_udiv_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-NEXT: v_cndmask_b32_e32 v0, v9, v5, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; CHECK-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc
-; CHECK-NEXT: s_mov_b32 s5, 0
+; CHECK-NEXT: s_mov_b32 s4, 0
; CHECK-NEXT: s_branch .LBB1_3
; CHECK-NEXT: .LBB1_2:
; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1
; CHECK-NEXT: .LBB1_3: ; %Flow
-; CHECK-NEXT: s_xor_b32 s1, s5, -1
+; CHECK-NEXT: s_xor_b32 s1, s4, 1
; CHECK-NEXT: s_and_b32 s1, s1, 1
; CHECK-NEXT: s_cmp_lg_u32 s1, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB1_5
@@ -1971,10 +1972,10 @@ define <2 x i64> @v_udiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_add_i32_e32 v20, vcc, v20, v6
; GISEL-NEXT: v_and_b32_e32 v6, 0xffffff, v0
; GISEL-NEXT: v_and_b32_e32 v0, 0xffffff, v2
-; GISEL-NEXT: s_bfe_i32 s4, -1, 0x10000
-; GISEL-NEXT: s_bfe_i32 s5, -1, 0x10000
-; GISEL-NEXT: s_bfe_i32 s6, -1, 0x10000
-; GISEL-NEXT: s_bfe_i32 s7, -1, 0x10000
+; GISEL-NEXT: s_bfe_i32 s4, 1, 0x10000
+; GISEL-NEXT: s_bfe_i32 s5, 1, 0x10000
+; GISEL-NEXT: s_bfe_i32 s6, 1, 0x10000
+; GISEL-NEXT: s_bfe_i32 s7, 1, 0x10000
; GISEL-NEXT: v_add_i32_e32 v2, vcc, v18, v15
; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v14, vcc, v14, v15
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
index 261482c97e3e..61a9f261c8eb 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll
@@ -183,10 +183,11 @@ declare i32 @llvm.amdgcn.readfirstlane(i32)
define amdgpu_ps i64 @s_urem_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-LABEL: s_urem_i64:
; CHECK: ; %bb.0:
+; CHECK-NEXT: s_mov_b32 s4, 1
; CHECK-NEXT: s_or_b64 s[6:7], s[0:1], s[2:3]
-; CHECK-NEXT: s_mov_b32 s4, 0
-; CHECK-NEXT: s_mov_b32 s5, -1
-; CHECK-NEXT: s_and_b64 s[6:7], s[6:7], s[4:5]
+; CHECK-NEXT: s_mov_b32 s8, 0
+; CHECK-NEXT: s_mov_b32 s9, -1
+; CHECK-NEXT: s_and_b64 s[6:7], s[6:7], s[8:9]
; CHECK-NEXT: v_cmp_ne_u64_e64 vcc, s[6:7], 0
; CHECK-NEXT: v_cvt_f32_u32_e32 v2, s2
; CHECK-NEXT: s_cbranch_vccz .LBB1_2
@@ -312,12 +313,12 @@ define amdgpu_ps i64 @s_urem_i64(i64 inreg %num, i64 inreg %den) {
; CHECK-NEXT: v_cndmask_b32_e32 v0, v3, v6, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1
; CHECK-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc
-; CHECK-NEXT: s_mov_b32 s5, 0
+; CHECK-NEXT: s_mov_b32 s4, 0
; CHECK-NEXT: s_branch .LBB1_3
; CHECK-NEXT: .LBB1_2:
; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1
; CHECK-NEXT: .LBB1_3: ; %Flow
-; CHECK-NEXT: s_xor_b32 s1, s5, -1
+; CHECK-NEXT: s_xor_b32 s1, s4, 1
; CHECK-NEXT: s_and_b32 s1, s1, 1
; CHECK-NEXT: s_cmp_lg_u32 s1, 0
; CHECK-NEXT: s_cbranch_scc1 .LBB1_5
@@ -973,121 +974,119 @@ define i64 @v_urem_i64_oddk_denom(i64 %num) {
; CHECK-NEXT: v_cvt_f32_u32_e32 v3, 0x12d8fb
; CHECK-NEXT: v_cvt_f32_ubyte0_e32 v4, 0
; CHECK-NEXT: s_mov_b32 s5, 0xffed2705
-; CHECK-NEXT: s_bfe_i32 s6, -1, 0x10000
-; CHECK-NEXT: s_bfe_i32 s7, -1, 0x10000
+; CHECK-NEXT: s_bfe_i32 s6, 1, 0x10000
; CHECK-NEXT: v_mac_f32_e32 v3, 0x4f800000, v4
; CHECK-NEXT: v_mov_b32_e32 v4, s6
-; CHECK-NEXT: v_mov_b32_e32 v5, s7
; CHECK-NEXT: v_rcp_iflag_f32_e32 v3, v3
; CHECK-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3
-; CHECK-NEXT: v_mul_f32_e32 v6, 0x2f800000, v3
-; CHECK-NEXT: v_trunc_f32_e32 v6, v6
-; CHECK-NEXT: v_mac_f32_e32 v3, 0xcf800000, v6
-; CHECK-NEXT: v_cvt_u32_f32_e32 v6, v6
+; CHECK-NEXT: v_mul_f32_e32 v5, 0x2f800000, v3
+; CHECK-NEXT: v_trunc_f32_e32 v5, v5
+; CHECK-NEXT: v_mac_f32_e32 v3, 0xcf800000, v5
+; CHECK-NEXT: v_cvt_u32_f32_e32 v5, v5
; CHECK-NEXT: v_cvt_u32_f32_e32 v3, v3
-; CHECK-NEXT: v_mul_lo_u32 v7, v6, s5
-; CHECK-NEXT: v_mul_lo_u32 v8, v3, s5
-; CHECK-NEXT: v_mul_hi_u32 v9, s5, v3
-; CHECK-NEXT: v_sub_i32_e32 v7, vcc, v7, v3
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v9
-; CHECK-NEXT: v_mul_lo_u32 v9, v6, v8
-; CHECK-NEXT: v_mul_hi_u32 v10, v3, v8
-; CHECK-NEXT: v_mul_hi_u32 v8, v6, v8
-; CHECK-NEXT: v_mul_lo_u32 v11, v3, v7
-; CHECK-NEXT: v_mul_lo_u32 v12, v6, v7
-; CHECK-NEXT: v_mul_hi_u32 v13, v3, v7
-; CHECK-NEXT: v_mul_hi_u32 v7, v6, v7
-; CHECK-NEXT: v_add_i32_e32 v9, vcc, v9, v11
-; CHECK-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v8, vcc, v12, v8
-; CHECK-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v9, vcc, v9, v10
-; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v8, vcc, v8, v13
-; CHECK-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v9, vcc, v11, v9
-; CHECK-NEXT: v_add_i32_e32 v10, vcc, v12, v10
-; CHECK-NEXT: v_add_i32_e32 v8, vcc, v8, v9
-; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v9, vcc, v10, v9
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v9
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v8
-; CHECK-NEXT: v_addc_u32_e32 v6, vcc, v6, v7, vcc
+; CHECK-NEXT: v_mul_lo_u32 v6, v5, s5
; CHECK-NEXT: v_mul_lo_u32 v7, v3, s5
; CHECK-NEXT: v_mul_hi_u32 v8, s5, v3
-; CHECK-NEXT: v_mul_lo_u32 v9, v6, s5
-; CHECK-NEXT: v_mul_lo_u32 v10, v6, v7
-; CHECK-NEXT: v_mul_hi_u32 v11, v3, v7
-; CHECK-NEXT: v_mul_hi_u32 v7, v6, v7
-; CHECK-NEXT: v_sub_i32_e32 v9, vcc, v9, v3
-; CHECK-NEXT: v_add_i32_e32 v8, vcc, v9, v8
-; CHECK-NEXT: v_mul_lo_u32 v9, v3, v8
-; CHECK-NEXT: v_mul_lo_u32 v12, v6, v8
-; CHECK-NEXT: v_mul_hi_u32 v13, v3, v8
-; CHECK-NEXT: v_mul_hi_u32 v8, v6, v8
-; CHECK-NEXT: v_add_i32_e32 v9, vcc, v10, v9
+; CHECK-NEXT: v_sub_i32_e32 v6, vcc, v6, v3
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v6, v8
+; CHECK-NEXT: v_mul_lo_u32 v8, v5, v7
+; CHECK-NEXT: v_mul_hi_u32 v9, v3, v7
+; CHECK-NEXT: v_mul_hi_u32 v7, v5, v7
+; CHECK-NEXT: v_mul_lo_u32 v10, v3, v6
+; CHECK-NEXT: v_mul_lo_u32 v11, v5, v6
+; CHECK-NEXT: v_mul_hi_u32 v12, v3, v6
+; CHECK-NEXT: v_mul_hi_u32 v6, v5, v6
+; CHECK-NEXT: v_add_i32_e32 v8, vcc, v8, v10
; CHECK-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v12, v7
-; CHECK-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v9, vcc, v9, v11
-; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v13
+; CHECK-NEXT: v_add_i32_e32 v7, vcc, v11, v7
; CHECK-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v9, vcc, v10, v9
-; CHECK-NEXT: v_add_i32_e32 v10, vcc, v12, v11
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v9
-; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v9, vcc, v10, v9
; CHECK-NEXT: v_add_i32_e32 v8, vcc, v8, v9
+; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v12
+; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v8, vcc, v10, v8
+; CHECK-NEXT: v_add_i32_e32 v9, vcc, v11, v9
+; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v8
+; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v8, vcc, v9, v8
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v6, v8
; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v7
-; CHECK-NEXT: v_addc_u32_e32 v6, vcc, v6, v8, vcc
-; CHECK-NEXT: v_mul_lo_u32 v7, v1, v3
-; CHECK-NEXT: v_mul_hi_u32 v8, v0, v3
-; CHECK-NEXT: v_mul_hi_u32 v3, v1, v3
-; CHECK-NEXT: v_mul_lo_u32 v9, v0, v6
-; CHECK-NEXT: v_mul_lo_u32 v10, v1, v6
-; CHECK-NEXT: v_mul_hi_u32 v11, v0, v6
-; CHECK-NEXT: v_mul_hi_u32 v6, v1, v6
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v9
+; CHECK-NEXT: v_addc_u32_e32 v5, vcc, v5, v6, vcc
+; CHECK-NEXT: v_mul_lo_u32 v6, v3, s5
+; CHECK-NEXT: v_mul_hi_u32 v7, s5, v3
+; CHECK-NEXT: v_mul_lo_u32 v8, v5, s5
+; CHECK-NEXT: v_mul_lo_u32 v9, v5, v6
+; CHECK-NEXT: v_mul_hi_u32 v10, v3, v6
+; CHECK-NEXT: v_mul_hi_u32 v6, v5, v6
+; CHECK-NEXT: v_sub_i32_e32 v8, vcc, v8, v3
+; CHECK-NEXT: v_add_i32_e32 v7, vcc, v8, v7
+; CHECK-NEXT: v_mul_lo_u32 v8, v3, v7
+; CHECK-NEXT: v_mul_lo_u32 v11, v5, v7
+; CHECK-NEXT: v_mul_hi_u32 v12, v3, v7
+; CHECK-NEXT: v_mul_hi_u32 v7, v5, v7
+; CHECK-NEXT: v_add_i32_e32 v8, vcc, v9, v8
; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, v10, v3
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v11, v6
+; CHECK-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v8, vcc, v8, v10
+; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v6, v12
; CHECK-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v8, vcc, v9, v8
+; CHECK-NEXT: v_add_i32_e32 v9, vcc, v11, v10
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v6, v8
+; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v8, vcc, v9, v8
; CHECK-NEXT: v_add_i32_e32 v7, vcc, v7, v8
-; CHECK-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v11
+; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v6
+; CHECK-NEXT: v_addc_u32_e32 v5, vcc, v5, v7, vcc
+; CHECK-NEXT: v_mul_lo_u32 v6, v1, v3
+; CHECK-NEXT: v_mul_hi_u32 v7, v0, v3
+; CHECK-NEXT: v_mul_hi_u32 v3, v1, v3
+; CHECK-NEXT: v_mul_lo_u32 v8, v0, v5
+; CHECK-NEXT: v_mul_lo_u32 v9, v1, v5
+; CHECK-NEXT: v_mul_hi_u32 v10, v0, v5
+; CHECK-NEXT: v_mul_hi_u32 v5, v1, v5
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v6, v8
; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v9, v7
-; CHECK-NEXT: v_add_i32_e32 v8, vcc, v10, v8
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v7
+; CHECK-NEXT: v_add_i32_e32 v3, vcc, v9, v3
+; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v6, v7
+; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v10
; CHECK-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v8, v7
-; CHECK-NEXT: v_mul_lo_u32 v8, v3, s4
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v8, v6
+; CHECK-NEXT: v_add_i32_e32 v7, vcc, v9, v7
+; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v6
+; CHECK-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
+; CHECK-NEXT: v_add_i32_e32 v6, vcc, v7, v6
+; CHECK-NEXT: v_mul_lo_u32 v7, v3, s4
; CHECK-NEXT: v_mul_hi_u32 v3, s4, v3
-; CHECK-NEXT: v_add_i32_e32 v6, vcc, v6, v7
-; CHECK-NEXT: v_mul_lo_u32 v6, v6, s4
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, v6, v3
-; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v8
-; CHECK-NEXT: v_subb_u32_e64 v6, s[4:5], v1, v3, vcc
+; CHECK-NEXT: v_add_i32_e32 v5, vcc, v5, v6
+; CHECK-NEXT: v_mul_lo_u32 v5, v5, s4
+; CHECK-NEXT: v_add_i32_e32 v3, vcc, v5, v3
+; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v7
+; CHECK-NEXT: v_subb_u32_e64 v5, s[4:5], v1, v3, vcc
; CHECK-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v3
; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v2
; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, -1, s[4:5]
-; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v6
+; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v5
; CHECK-NEXT: v_cndmask_b32_e64 v3, v4, v3, s[4:5]
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; CHECK-NEXT: v_sub_i32_e32 v4, vcc, v0, v2
+; CHECK-NEXT: v_sub_i32_e32 v6, vcc, v0, v2
; CHECK-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v4, v2
+; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v6, v2
; CHECK-NEXT: v_cndmask_b32_e64 v7, 0, -1, vcc
-; CHECK-NEXT: v_sub_i32_e32 v2, vcc, v4, v2
+; CHECK-NEXT: v_sub_i32_e32 v2, vcc, v6, v2
; CHECK-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1
-; CHECK-NEXT: v_cndmask_b32_e64 v5, v5, v7, s[4:5]
+; CHECK-NEXT: v_cndmask_b32_e64 v4, v4, v7, s[4:5]
; CHECK-NEXT: v_subbrev_u32_e32 v7, vcc, 0, v1, vcc
-; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
-; CHECK-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc
+; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
+; CHECK-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc
; CHECK-NEXT: v_cndmask_b32_e32 v1, v1, v7, vcc
; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
; CHECK-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; CHECK-NEXT: v_cndmask_b32_e32 v1, v6, v1, vcc
+; CHECK-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc
; CHECK-NEXT: s_setpc_b64 s[30:31]
%result = urem i64 %num, 1235195
ret i64 %result
@@ -1097,68 +1096,66 @@ define <2 x i64> @v_urem_v2i64_oddk_denom(<2 x i64> %num) {
; GISEL-LABEL: v_urem_v2i64_oddk_denom:
; GISEL: ; %bb.0:
; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: s_mov_b32 s8, 0x12d8fb
+; GISEL-NEXT: s_mov_b32 s6, 0x12d8fb
; GISEL-NEXT: v_mov_b32_e32 v4, 0x12d8fb
-; GISEL-NEXT: v_cvt_f32_u32_e32 v7, 0x12d8fb
+; GISEL-NEXT: v_cvt_f32_u32_e32 v6, 0x12d8fb
; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v5, 0
-; GISEL-NEXT: s_sub_u32 s6, 0, 0x12d8fb
-; GISEL-NEXT: v_madmk_f32 v6, v5, 0x4f800000, v7
-; GISEL-NEXT: s_subb_u32 s7, 0, 0
-; GISEL-NEXT: s_bfe_i32 s4, -1, 0x10000
-; GISEL-NEXT: s_bfe_i32 s5, -1, 0x10000
-; GISEL-NEXT: v_mac_f32_e32 v7, 0x4f800000, v5
-; GISEL-NEXT: v_rcp_iflag_f32_e32 v8, v6
-; GISEL-NEXT: v_mov_b32_e32 v6, s4
-; GISEL-NEXT: v_mov_b32_e32 v5, s5
+; GISEL-NEXT: s_sub_u32 s7, 0, 0x12d8fb
+; GISEL-NEXT: v_madmk_f32 v7, v5, 0x4f800000, v6
+; GISEL-NEXT: s_subb_u32 s8, 0, 0
+; GISEL-NEXT: s_bfe_i32 s4, 1, 0x10000
+; GISEL-NEXT: v_mac_f32_e32 v6, 0x4f800000, v5
; GISEL-NEXT: v_rcp_iflag_f32_e32 v7, v7
+; GISEL-NEXT: v_mov_b32_e32 v5, s4
+; GISEL-NEXT: v_rcp_iflag_f32_e32 v6, v6
; GISEL-NEXT: s_sub_u32 s9, 0, 0x12d8fb
-; GISEL-NEXT: v_mul_f32_e32 v8, 0x5f7ffffc, v8
; GISEL-NEXT: v_mul_f32_e32 v7, 0x5f7ffffc, v7
+; GISEL-NEXT: v_mul_f32_e32 v6, 0x5f7ffffc, v6
; GISEL-NEXT: s_subb_u32 s10, 0, 0
-; GISEL-NEXT: s_bfe_i32 s11, -1, 0x10000
-; GISEL-NEXT: s_bfe_i32 s12, -1, 0x10000
-; GISEL-NEXT: v_mul_f32_e32 v9, 0x2f800000, v8
-; GISEL-NEXT: v_mul_f32_e32 v10, 0x2f800000, v7
+; GISEL-NEXT: s_bfe_i32 s4, 1, 0x10000
+; GISEL-NEXT: v_mul_f32_e32 v8, 0x2f800000, v7
+; GISEL-NEXT: v_mul_f32_e32 v9, 0x2f800000, v6
+; GISEL-NEXT: v_mov_b32_e32 v10, s4
+; GISEL-NEXT: v_trunc_f32_e32 v8, v8
; GISEL-NEXT: v_trunc_f32_e32 v9, v9
-; GISEL-NEXT: v_trunc_f32_e32 v10, v10
-; GISEL-NEXT: v_mac_f32_e32 v8, 0xcf800000, v9
-; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v9
-; GISEL-NEXT: v_mac_f32_e32 v7, 0xcf800000, v10
-; GISEL-NEXT: v_cvt_u32_f32_e32 v10, v10
+; GISEL-NEXT: v_mac_f32_e32 v7, 0xcf800000, v8
; GISEL-NEXT: v_cvt_u32_f32_e32 v8, v8
-; GISEL-NEXT: v_mul_lo_u32 v11, s6, v9
+; GISEL-NEXT: v_mac_f32_e32 v6, 0xcf800000, v9
+; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v9
; GISEL-NEXT: v_cvt_u32_f32_e32 v7, v7
-; GISEL-NEXT: v_mul_lo_u32 v12, s9, v10
-; GISEL-NEXT: v_mul_lo_u32 v13, s6, v8
-; GISEL-NEXT: v_mul_lo_u32 v14, s7, v8
-; GISEL-NEXT: v_mul_hi_u32 v15, s6, v8
-; GISEL-NEXT: v_mul_lo_u32 v16, s9, v7
-; GISEL-NEXT: v_mul_lo_u32 v17, s10, v7
-; GISEL-NEXT: v_mul_hi_u32 v18, s9, v7
+; GISEL-NEXT: v_mul_lo_u32 v11, s7, v8
+; GISEL-NEXT: v_cvt_u32_f32_e32 v6, v6
+; GISEL-NEXT: v_mul_lo_u32 v12, s9, v9
+; GISEL-NEXT: v_mul_lo_u32 v13, s7, v7
+; GISEL-NEXT: v_mul_lo_u32 v14, s8, v7
+; GISEL-NEXT: v_mul_hi_u32 v15, s7, v7
+; GISEL-NEXT: v_mul_lo_u32 v16, s9, v6
+; GISEL-NEXT: v_mul_lo_u32 v17, s10, v6
+; GISEL-NEXT: v_mul_hi_u32 v18, s9, v6
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v14, v11
-; GISEL-NEXT: v_mul_lo_u32 v14, v9, v13
-; GISEL-NEXT: v_mul_hi_u32 v19, v8, v13
-; GISEL-NEXT: v_mul_hi_u32 v13, v9, v13
+; GISEL-NEXT: v_mul_lo_u32 v14, v8, v13
+; GISEL-NEXT: v_mul_hi_u32 v19, v7, v13
+; GISEL-NEXT: v_mul_hi_u32 v13, v8, v13
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v17, v12
-; GISEL-NEXT: v_mul_lo_u32 v17, v10, v16
+; GISEL-NEXT: v_mul_lo_u32 v17, v9, v16
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v11, v15
-; GISEL-NEXT: v_mul_hi_u32 v15, v7, v16
-; GISEL-NEXT: v_mul_hi_u32 v16, v10, v16
+; GISEL-NEXT: v_mul_hi_u32 v15, v6, v16
+; GISEL-NEXT: v_mul_hi_u32 v16, v9, v16
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v18
-; GISEL-NEXT: v_mul_lo_u32 v18, v7, v12
+; GISEL-NEXT: v_mul_lo_u32 v18, v6, v12
; GISEL-NEXT: v_add_i32_e32 v17, vcc, v17, v18
; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v17, v15
-; GISEL-NEXT: v_mul_lo_u32 v15, v8, v11
-; GISEL-NEXT: v_mul_lo_u32 v17, v9, v11
+; GISEL-NEXT: v_mul_lo_u32 v15, v7, v11
+; GISEL-NEXT: v_mul_lo_u32 v17, v8, v11
; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v14, v15
; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v14, v19
-; GISEL-NEXT: v_mul_hi_u32 v14, v8, v11
-; GISEL-NEXT: v_mul_hi_u32 v11, v9, v11
+; GISEL-NEXT: v_mul_hi_u32 v14, v7, v11
+; GISEL-NEXT: v_mul_hi_u32 v11, v8, v11
; GISEL-NEXT: v_cndmask_b32_e64 v19, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v15, s[4:5], v15, v19
-; GISEL-NEXT: v_mul_lo_u32 v19, v10, v12
+; GISEL-NEXT: v_mul_lo_u32 v19, v9, v12
; GISEL-NEXT: v_add_i32_e64 v13, s[4:5], v17, v13
; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v13, s[4:5], v13, v14
@@ -1166,8 +1163,8 @@ define <2 x i64> @v_urem_v2i64_oddk_denom(<2 x i64> %num) {
; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v17, v14
; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v17, vcc, v18, v17
-; GISEL-NEXT: v_mul_hi_u32 v18, v7, v12
-; GISEL-NEXT: v_mul_hi_u32 v12, v10, v12
+; GISEL-NEXT: v_mul_hi_u32 v18, v6, v12
+; GISEL-NEXT: v_mul_hi_u32 v12, v9, v12
; GISEL-NEXT: v_add_i32_e32 v16, vcc, v19, v16
; GISEL-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v16, vcc, v16, v18
@@ -1181,41 +1178,42 @@ define <2 x i64> @v_urem_v2i64_oddk_denom(<2 x i64> %num) {
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v18, v17
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v11, v14
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v15
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v13
-; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v9, v11, vcc
-; GISEL-NEXT: v_mul_lo_u32 v11, s6, v8
-; GISEL-NEXT: v_mul_lo_u32 v13, s7, v8
-; GISEL-NEXT: v_mul_hi_u32 v14, s6, v8
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v16
-; GISEL-NEXT: v_addc_u32_e32 v10, vcc, v10, v12, vcc
-; GISEL-NEXT: v_mul_lo_u32 v12, s9, v7
-; GISEL-NEXT: v_mul_lo_u32 v15, s10, v7
-; GISEL-NEXT: v_mul_hi_u32 v16, s9, v7
-; GISEL-NEXT: v_mul_lo_u32 v17, s6, v9
-; GISEL-NEXT: v_mul_lo_u32 v18, v9, v11
-; GISEL-NEXT: v_mul_hi_u32 v19, v8, v11
-; GISEL-NEXT: v_mul_hi_u32 v11, v9, v11
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v13
+; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v8, v11, vcc
+; GISEL-NEXT: v_mul_lo_u32 v11, s7, v7
+; GISEL-NEXT: v_mul_lo_u32 v13, s8, v7
+; GISEL-NEXT: v_mul_hi_u32 v14, s7, v7
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v16
+; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v9, v12, vcc
+; GISEL-NEXT: v_mul_lo_u32 v12, s9, v6
+; GISEL-NEXT: v_mul_lo_u32 v15, s10, v6
+; GISEL-NEXT: v_mul_hi_u32 v16, s9, v6
+; GISEL-NEXT: v_mul_lo_u32 v17, s7, v8
+; GISEL-NEXT: v_mul_lo_u32 v18, v8, v11
+; GISEL-NEXT: v_mul_hi_u32 v19, v7, v11
+; GISEL-NEXT: v_mul_hi_u32 v11, v8, v11
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v17
-; GISEL-NEXT: v_mul_lo_u32 v17, s9, v10
+; GISEL-NEXT: v_mul_lo_u32 v17, s9, v9
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v15, v17
-; GISEL-NEXT: v_mul_lo_u32 v17, v10, v12
+; GISEL-NEXT: v_mul_lo_u32 v17, v9, v12
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v14
-; GISEL-NEXT: v_mul_hi_u32 v14, v7, v12
-; GISEL-NEXT: v_mul_hi_u32 v12, v10, v12
+; GISEL-NEXT: v_mul_hi_u32 v14, v6, v12
+; GISEL-NEXT: v_mul_hi_u32 v12, v9, v12
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v15, v16
-; GISEL-NEXT: v_mul_lo_u32 v16, v7, v15
+; GISEL-NEXT: v_mul_lo_u32 v16, v6, v15
; GISEL-NEXT: v_add_i32_e32 v16, vcc, v17, v16
; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v14, vcc, v16, v14
-; GISEL-NEXT: v_mul_lo_u32 v14, v8, v13
-; GISEL-NEXT: v_mul_lo_u32 v16, v9, v13
+; GISEL-NEXT: v_mul_lo_u32 v14, v7, v13
+; GISEL-NEXT: v_mul_lo_u32 v16, v8, v13
; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v18, v14
; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v14, v19
-; GISEL-NEXT: v_mul_hi_u32 v14, v8, v13
+; GISEL-NEXT: v_mul_hi_u32 v14, v7, v13
+; GISEL-NEXT: v_mul_hi_u32 v13, v8, v13
; GISEL-NEXT: v_cndmask_b32_e64 v19, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v18, s[4:5], v18, v19
-; GISEL-NEXT: v_mul_lo_u32 v19, v10, v15
+; GISEL-NEXT: v_mul_lo_u32 v19, v9, v15
; GISEL-NEXT: v_add_i32_e64 v11, s[4:5], v16, v11
; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v11, s[4:5], v11, v14
@@ -1223,364 +1221,355 @@ define <2 x i64> @v_urem_v2i64_oddk_denom(<2 x i64> %num) {
; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v16, v14
; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v16, vcc, v17, v16
-; GISEL-NEXT: v_mul_hi_u32 v17, v7, v15
+; GISEL-NEXT: v_mul_hi_u32 v17, v6, v15
+; GISEL-NEXT: v_mul_hi_u32 v15, v9, v15
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v19, v12
; GISEL-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v17
; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v17, vcc, v19, v17
-; GISEL-NEXT: v_mov_b32_e32 v19, s11
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v11, v18
; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v14, vcc, v14, v18
-; GISEL-NEXT: v_mov_b32_e32 v18, s12
-; GISEL-NEXT: v_mul_hi_u32 v13, v9, v13
-; GISEL-NEXT: v_mul_hi_u32 v15, v10, v15
; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v16
; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v14, vcc, v14, v18
; GISEL-NEXT: v_add_i32_e32 v16, vcc, v17, v16
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v14
; GISEL-NEXT: v_add_i32_e32 v14, vcc, v15, v16
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v11
-; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v9, v13, vcc
-; GISEL-NEXT: v_mul_lo_u32 v11, v1, v8
-; GISEL-NEXT: v_mul_hi_u32 v13, v0, v8
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v11
+; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v8, v13, vcc
+; GISEL-NEXT: v_mul_lo_u32 v11, v1, v7
+; GISEL-NEXT: v_mul_hi_u32 v13, v0, v7
+; GISEL-NEXT: v_mul_hi_u32 v7, v1, v7
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v12
+; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v9, v14, vcc
+; GISEL-NEXT: v_mul_lo_u32 v12, v3, v6
+; GISEL-NEXT: v_mul_hi_u32 v14, v2, v6
+; GISEL-NEXT: v_mul_hi_u32 v6, v3, v6
+; GISEL-NEXT: v_mul_lo_u32 v15, v0, v8
+; GISEL-NEXT: v_mul_lo_u32 v16, v1, v8
+; GISEL-NEXT: v_mul_hi_u32 v17, v0, v8
; GISEL-NEXT: v_mul_hi_u32 v8, v1, v8
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v12
-; GISEL-NEXT: v_addc_u32_e32 v10, vcc, v10, v14, vcc
-; GISEL-NEXT: v_mul_lo_u32 v12, v3, v7
-; GISEL-NEXT: v_mul_hi_u32 v14, v2, v7
-; GISEL-NEXT: v_mul_hi_u32 v7, v3, v7
-; GISEL-NEXT: v_mul_lo_u32 v15, v0, v9
-; GISEL-NEXT: v_mul_lo_u32 v16, v1, v9
-; GISEL-NEXT: v_mul_hi_u32 v17, v0, v9
-; GISEL-NEXT: v_mul_hi_u32 v9, v1, v9
+; GISEL-NEXT: v_mul_lo_u32 v18, v2, v9
+; GISEL-NEXT: v_mul_lo_u32 v19, v3, v9
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v11, v15
; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v11, v13
-; GISEL-NEXT: v_mul_lo_u32 v11, v2, v10
-; GISEL-NEXT: v_mul_lo_u32 v13, v3, v10
-; GISEL-NEXT: v_add_i32_e64 v11, s[4:5], v12, v11
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v11, s[4:5], v11, v14
-; GISEL-NEXT: v_mul_hi_u32 v11, v2, v10
-; GISEL-NEXT: v_mul_hi_u32 v10, v3, v10
-; GISEL-NEXT: v_add_i32_e64 v8, s[6:7], v16, v8
-; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, s[6:7]
-; GISEL-NEXT: v_add_i32_e64 v7, s[6:7], v13, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[6:7]
-; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v17
-; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v15, vcc, v15, v16
+; GISEL-NEXT: v_mul_hi_u32 v11, v2, v9
+; GISEL-NEXT: v_mul_hi_u32 v9, v3, v9
+; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v16, v7
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v12, v18
; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v11
+; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v19, v6
+; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v17
+; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v14
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v11
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v14, vcc, v14, v17
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v16
-; GISEL-NEXT: v_add_i32_e32 v11, vcc, v13, v11
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v15
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v12
+; GISEL-NEXT: v_add_i32_e32 v14, vcc, v15, v19
+; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v17
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v16, v12
+; GISEL-NEXT: v_add_i32_e32 v11, vcc, v18, v11
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v14
+; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v12
; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v13, vcc, v14, v13
-; GISEL-NEXT: v_mul_lo_u32 v14, v8, s8
-; GISEL-NEXT: v_mul_hi_u32 v8, s8, v8
+; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v14
+; GISEL-NEXT: v_mul_lo_u32 v14, v7, s6
+; GISEL-NEXT: v_mul_hi_u32 v7, s6, v7
; GISEL-NEXT: v_add_i32_e32 v11, vcc, v11, v12
-; GISEL-NEXT: v_mul_lo_u32 v12, v7, s8
-; GISEL-NEXT: v_mul_hi_u32 v7, s8, v7
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v13
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
-; GISEL-NEXT: v_mul_lo_u32 v9, v9, s8
-; GISEL-NEXT: v_mul_lo_u32 v10, v10, s8
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v10, v7
+; GISEL-NEXT: v_mul_lo_u32 v12, v6, s6
+; GISEL-NEXT: v_mul_hi_u32 v6, s6, v6
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v13
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v11
+; GISEL-NEXT: v_mul_lo_u32 v8, v8, s6
+; GISEL-NEXT: v_mul_lo_u32 v9, v9, s6
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v8, v7
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v9, v6
; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v14
-; GISEL-NEXT: v_subb_u32_e64 v9, s[4:5], v1, v8, vcc
-; GISEL-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v8
+; GISEL-NEXT: v_subb_u32_e64 v8, s[4:5], v1, v7, vcc
+; GISEL-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v7
; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
; GISEL-NEXT: v_sub_i32_e64 v2, s[4:5], v2, v12
-; GISEL-NEXT: v_subb_u32_e64 v10, s[6:7], v3, v7, s[4:5]
-; GISEL-NEXT: v_sub_i32_e64 v3, s[6:7], v3, v7
+; GISEL-NEXT: v_subb_u32_e64 v9, s[6:7], v3, v6, s[4:5]
+; GISEL-NEXT: v_sub_i32_e64 v3, s[6:7], v3, v6
; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v2, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[6:7]
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v6, v6, v8, s[6:7]
+; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[6:7]
+; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v8
+; GISEL-NEXT: v_cndmask_b32_e64 v7, v5, v7, s[6:7]
; GISEL-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v10
-; GISEL-NEXT: v_cndmask_b32_e32 v7, v19, v7, vcc
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v9
+; GISEL-NEXT: v_cndmask_b32_e32 v6, v5, v6, vcc
; GISEL-NEXT: v_subbrev_u32_e64 v3, vcc, 0, v3, s[4:5]
-; GISEL-NEXT: v_sub_i32_e32 v8, vcc, v0, v4
+; GISEL-NEXT: v_sub_i32_e32 v11, vcc, v0, v4
; GISEL-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v8, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, -1, vcc
-; GISEL-NEXT: v_sub_i32_e32 v12, vcc, v2, v4
+; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v11, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, -1, vcc
+; GISEL-NEXT: v_sub_i32_e32 v13, vcc, v2, v4
; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v12, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, -1, vcc
+; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v13, v4
+; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, -1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1
-; GISEL-NEXT: v_cndmask_b32_e32 v5, v5, v11, vcc
-; GISEL-NEXT: v_sub_i32_e32 v11, vcc, v8, v4
-; GISEL-NEXT: v_subbrev_u32_e32 v14, vcc, 0, v1, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v5, v5, v12, vcc
+; GISEL-NEXT: v_sub_i32_e32 v12, vcc, v11, v4
+; GISEL-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v1, vcc
; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3
-; GISEL-NEXT: v_cndmask_b32_e32 v13, v18, v13, vcc
-; GISEL-NEXT: v_sub_i32_e32 v4, vcc, v12, v4
-; GISEL-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v3, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v10, v10, v14, vcc
+; GISEL-NEXT: v_sub_i32_e32 v4, vcc, v13, v4
+; GISEL-NEXT: v_subbrev_u32_e32 v14, vcc, 0, v3, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
-; GISEL-NEXT: v_cndmask_b32_e32 v5, v8, v11, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v13
-; GISEL-NEXT: v_cndmask_b32_e64 v4, v12, v4, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v14, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
+; GISEL-NEXT: v_cndmask_b32_e32 v5, v11, v12, vcc
+; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v4, v13, v4, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v15, vcc
+; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v7
; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v5, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v3, v3, v15, s[4:5]
-; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v7
+; GISEL-NEXT: v_cndmask_b32_e64 v3, v3, v14, s[4:5]
+; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v6
; GISEL-NEXT: v_cndmask_b32_e64 v2, v2, v4, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v9, v1, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v3, v10, v3, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v8, v1, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v3, v9, v3, s[4:5]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_urem_v2i64_oddk_denom:
; CGP: ; %bb.0:
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CGP-NEXT: s_mov_b32 s8, 0x12d8fb
+; CGP-NEXT: s_mov_b32 s6, 0x12d8fb
; CGP-NEXT: v_mov_b32_e32 v4, 0x12d8fb
-; CGP-NEXT: v_cvt_f32_u32_e32 v5, 0x12d8fb
-; CGP-NEXT: v_cvt_f32_ubyte0_e32 v6, 0
-; CGP-NEXT: s_mov_b32 s6, 0xffed2705
-; CGP-NEXT: s_bfe_i32 s4, -1, 0x10000
-; CGP-NEXT: s_bfe_i32 s5, -1, 0x10000
+; CGP-NEXT: v_cvt_f32_u32_e32 v6, 0x12d8fb
+; CGP-NEXT: v_cvt_f32_ubyte0_e32 v5, 0
+; CGP-NEXT: s_mov_b32 s7, 0xffed2705
+; CGP-NEXT: s_bfe_i32 s4, 1, 0x10000
; CGP-NEXT: v_cvt_f32_u32_e32 v7, 0x12d8fb
; CGP-NEXT: v_cvt_f32_ubyte0_e32 v8, 0
-; CGP-NEXT: s_bfe_i32 s7, -1, 0x10000
-; CGP-NEXT: s_bfe_i32 s9, -1, 0x10000
-; CGP-NEXT: v_mac_f32_e32 v5, 0x4f800000, v6
-; CGP-NEXT: v_mov_b32_e32 v6, s4
-; CGP-NEXT: v_mov_b32_e32 v9, s5
+; CGP-NEXT: v_mac_f32_e32 v6, 0x4f800000, v5
+; CGP-NEXT: v_mov_b32_e32 v5, s4
; CGP-NEXT: v_mac_f32_e32 v7, 0x4f800000, v8
-; CGP-NEXT: v_rcp_iflag_f32_e32 v5, v5
+; CGP-NEXT: v_rcp_iflag_f32_e32 v6, v6
; CGP-NEXT: v_rcp_iflag_f32_e32 v7, v7
-; CGP-NEXT: v_mul_f32_e32 v5, 0x5f7ffffc, v5
+; CGP-NEXT: v_mul_f32_e32 v6, 0x5f7ffffc, v6
; CGP-NEXT: v_mul_f32_e32 v7, 0x5f7ffffc, v7
-; CGP-NEXT: v_mul_f32_e32 v8, 0x2f800000, v5
-; CGP-NEXT: v_mul_f32_e32 v10, 0x2f800000, v7
+; CGP-NEXT: v_mul_f32_e32 v8, 0x2f800000, v6
+; CGP-NEXT: v_mul_f32_e32 v9, 0x2f800000, v7
; CGP-NEXT: v_trunc_f32_e32 v8, v8
-; CGP-NEXT: v_trunc_f32_e32 v10, v10
-; CGP-NEXT: v_mac_f32_e32 v5, 0xcf800000, v8
+; CGP-NEXT: v_trunc_f32_e32 v9, v9
+; CGP-NEXT: v_mac_f32_e32 v6, 0xcf800000, v8
; CGP-NEXT: v_cvt_u32_f32_e32 v8, v8
-; CGP-NEXT: v_mac_f32_e32 v7, 0xcf800000, v10
-; CGP-NEXT: v_cvt_u32_f32_e32 v10, v10
-; CGP-NEXT: v_cvt_u32_f32_e32 v5, v5
-; CGP-NEXT: v_mul_lo_u32 v11, v8, s6
+; CGP-NEXT: v_mac_f32_e32 v7, 0xcf800000, v9
+; CGP-NEXT: v_cvt_u32_f32_e32 v9, v9
+; CGP-NEXT: v_cvt_u32_f32_e32 v6, v6
+; CGP-NEXT: v_mul_lo_u32 v10, v8, s7
; CGP-NEXT: v_cvt_u32_f32_e32 v7, v7
-; CGP-NEXT: v_mul_lo_u32 v12, v10, s6
-; CGP-NEXT: v_mul_lo_u32 v13, v5, s6
-; CGP-NEXT: v_mul_hi_u32 v14, s6, v5
-; CGP-NEXT: v_sub_i32_e32 v11, vcc, v11, v5
-; CGP-NEXT: v_mul_lo_u32 v15, v7, s6
-; CGP-NEXT: v_mul_hi_u32 v16, s6, v7
-; CGP-NEXT: v_sub_i32_e32 v12, vcc, v12, v7
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v14
-; CGP-NEXT: v_mul_lo_u32 v14, v8, v13
-; CGP-NEXT: v_mul_hi_u32 v17, v5, v13
-; CGP-NEXT: v_mul_hi_u32 v13, v8, v13
-; CGP-NEXT: v_add_i32_e32 v12, vcc, v12, v16
-; CGP-NEXT: v_mul_lo_u32 v16, v10, v15
-; CGP-NEXT: v_mul_hi_u32 v18, v7, v15
-; CGP-NEXT: v_mul_hi_u32 v15, v10, v15
-; CGP-NEXT: v_mul_lo_u32 v19, v7, v12
-; CGP-NEXT: v_add_i32_e32 v16, vcc, v16, v19
+; CGP-NEXT: v_mul_lo_u32 v11, v9, s7
+; CGP-NEXT: v_mul_lo_u32 v12, v6, s7
+; CGP-NEXT: v_mul_hi_u32 v13, s7, v6
+; CGP-NEXT: v_sub_i32_e32 v10, vcc, v10, v6
+; CGP-NEXT: v_mul_lo_u32 v14, v7, s7
+; CGP-NEXT: v_mul_hi_u32 v15, s7, v7
+; CGP-NEXT: v_sub_i32_e32 v11, vcc, v11, v7
+; CGP-NEXT: v_add_i32_e32 v10, vcc, v10, v13
+; CGP-NEXT: v_mul_lo_u32 v13, v8, v12
+; CGP-NEXT: v_mul_hi_u32 v16, v6, v12
+; CGP-NEXT: v_mul_hi_u32 v12, v8, v12
+; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v15
+; CGP-NEXT: v_mul_lo_u32 v15, v9, v14
+; CGP-NEXT: v_mul_hi_u32 v17, v7, v14
+; CGP-NEXT: v_mul_hi_u32 v14, v9, v14
+; CGP-NEXT: v_mul_lo_u32 v18, v6, v10
+; CGP-NEXT: v_mul_lo_u32 v19, v7, v11
+; CGP-NEXT: v_add_i32_e32 v15, vcc, v15, v19
; CGP-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v16, vcc, v16, v18
-; CGP-NEXT: v_mul_lo_u32 v16, v5, v11
-; CGP-NEXT: v_mul_lo_u32 v18, v8, v11
-; CGP-NEXT: v_add_i32_e64 v14, s[4:5], v14, v16
-; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, s[4:5]
-; CGP-NEXT: v_add_i32_e64 v14, s[4:5], v14, v17
-; CGP-NEXT: v_mul_hi_u32 v14, v5, v11
-; CGP-NEXT: v_mul_hi_u32 v11, v8, v11
-; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, s[4:5]
-; CGP-NEXT: v_add_i32_e64 v16, s[4:5], v16, v17
-; CGP-NEXT: v_mul_lo_u32 v17, v10, v12
-; CGP-NEXT: v_add_i32_e64 v13, s[4:5], v18, v13
+; CGP-NEXT: v_add_i32_e32 v15, vcc, v15, v17
+; CGP-NEXT: v_mul_lo_u32 v15, v8, v10
+; CGP-NEXT: v_mul_hi_u32 v17, v6, v10
+; CGP-NEXT: v_mul_hi_u32 v10, v8, v10
+; CGP-NEXT: v_add_i32_e64 v13, s[4:5], v13, v18
; CGP-NEXT: v_cndmask_b32_e64 v18, 0, 1, s[4:5]
+; CGP-NEXT: v_add_i32_e64 v13, s[4:5], v13, v16
+; CGP-NEXT: v_mul_lo_u32 v13, v9, v11
+; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, s[4:5]
+; CGP-NEXT: v_add_i32_e64 v16, s[4:5], v18, v16
+; CGP-NEXT: v_mul_hi_u32 v18, v7, v11
+; CGP-NEXT: v_mul_hi_u32 v11, v9, v11
+; CGP-NEXT: v_add_i32_e64 v12, s[4:5], v15, v12
+; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, s[4:5]
; CGP-NEXT: v_add_i32_e64 v13, s[4:5], v13, v14
; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, s[4:5]
-; CGP-NEXT: v_add_i32_e64 v14, s[4:5], v18, v14
-; CGP-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v18, vcc, v19, v18
-; CGP-NEXT: v_mul_hi_u32 v19, v7, v12
-; CGP-NEXT: v_mul_hi_u32 v12, v10, v12
-; CGP-NEXT: v_add_i32_e32 v15, vcc, v17, v15
+; CGP-NEXT: v_add_i32_e64 v12, s[4:5], v12, v17
+; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, s[4:5]
+; CGP-NEXT: v_add_i32_e64 v15, s[4:5], v15, v17
; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v15, vcc, v15, v19
-; CGP-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v17, vcc, v17, v19
-; CGP-NEXT: v_add_i32_e32 v13, vcc, v13, v16
-; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v15, vcc, v15, v18
+; CGP-NEXT: v_add_i32_e32 v13, vcc, v13, v18
; CGP-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v14, vcc, v14, v16
-; CGP-NEXT: v_add_i32_e32 v16, vcc, v17, v18
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v14
+; CGP-NEXT: v_add_i32_e32 v17, vcc, v19, v17
+; CGP-NEXT: v_add_i32_e32 v14, vcc, v14, v18
; CGP-NEXT: v_add_i32_e32 v12, vcc, v12, v16
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v13
-; CGP-NEXT: v_addc_u32_e32 v8, vcc, v8, v11, vcc
-; CGP-NEXT: v_mul_lo_u32 v11, v5, s6
-; CGP-NEXT: v_mul_hi_u32 v13, s6, v5
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v15
-; CGP-NEXT: v_addc_u32_e32 v10, vcc, v10, v12, vcc
-; CGP-NEXT: v_mul_lo_u32 v12, v7, s6
-; CGP-NEXT: v_mul_hi_u32 v14, s6, v7
-; CGP-NEXT: v_mul_lo_u32 v15, v8, s6
-; CGP-NEXT: v_mul_lo_u32 v16, v8, v11
-; CGP-NEXT: v_mul_hi_u32 v17, v5, v11
-; CGP-NEXT: v_mul_hi_u32 v11, v8, v11
-; CGP-NEXT: v_mul_lo_u32 v18, v10, s6
-; CGP-NEXT: v_mul_lo_u32 v19, v10, v12
-; CGP-NEXT: v_sub_i32_e32 v15, vcc, v15, v5
-; CGP-NEXT: v_add_i32_e32 v13, vcc, v15, v13
-; CGP-NEXT: v_mul_hi_u32 v15, v7, v12
-; CGP-NEXT: v_mul_hi_u32 v12, v10, v12
-; CGP-NEXT: v_sub_i32_e32 v18, vcc, v18, v7
-; CGP-NEXT: v_add_i32_e32 v14, vcc, v18, v14
-; CGP-NEXT: v_mul_lo_u32 v18, v7, v14
-; CGP-NEXT: v_add_i32_e32 v18, vcc, v19, v18
-; CGP-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v15, vcc, v18, v15
-; CGP-NEXT: v_mul_lo_u32 v15, v5, v13
-; CGP-NEXT: v_mul_lo_u32 v18, v8, v13
-; CGP-NEXT: v_add_i32_e64 v15, s[4:5], v16, v15
+; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v13, vcc, v13, v17
+; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v15, vcc, v15, v16
+; CGP-NEXT: v_add_i32_e32 v14, vcc, v14, v17
+; CGP-NEXT: v_add_i32_e32 v10, vcc, v10, v15
+; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v14
+; CGP-NEXT: v_add_i32_e32 v6, vcc, v6, v12
+; CGP-NEXT: v_addc_u32_e32 v8, vcc, v8, v10, vcc
+; CGP-NEXT: v_mul_lo_u32 v10, v6, s7
+; CGP-NEXT: v_mul_hi_u32 v12, s7, v6
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v13
+; CGP-NEXT: v_addc_u32_e32 v9, vcc, v9, v11, vcc
+; CGP-NEXT: v_mul_lo_u32 v11, v7, s7
+; CGP-NEXT: v_mul_hi_u32 v13, s7, v7
+; CGP-NEXT: v_mul_lo_u32 v14, v8, s7
+; CGP-NEXT: v_mul_lo_u32 v15, v8, v10
+; CGP-NEXT: v_mul_hi_u32 v16, v6, v10
+; CGP-NEXT: v_mul_hi_u32 v10, v8, v10
+; CGP-NEXT: v_mul_lo_u32 v17, v9, s7
+; CGP-NEXT: v_mul_lo_u32 v18, v9, v11
+; CGP-NEXT: v_mul_hi_u32 v19, v7, v11
+; CGP-NEXT: v_mul_hi_u32 v11, v9, v11
+; CGP-NEXT: v_sub_i32_e32 v14, vcc, v14, v6
+; CGP-NEXT: v_sub_i32_e32 v17, vcc, v17, v7
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v14, v12
+; CGP-NEXT: v_add_i32_e32 v13, vcc, v17, v13
+; CGP-NEXT: v_mul_lo_u32 v14, v6, v12
+; CGP-NEXT: v_mul_lo_u32 v17, v7, v13
+; CGP-NEXT: v_add_i32_e32 v17, vcc, v18, v17
+; CGP-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v17, vcc, v17, v19
+; CGP-NEXT: v_mul_lo_u32 v17, v8, v12
+; CGP-NEXT: v_mul_hi_u32 v19, v6, v12
+; CGP-NEXT: v_mul_hi_u32 v12, v8, v12
+; CGP-NEXT: v_add_i32_e64 v14, s[4:5], v15, v14
+; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, s[4:5]
+; CGP-NEXT: v_add_i32_e64 v14, s[4:5], v14, v16
+; CGP-NEXT: v_mul_lo_u32 v14, v9, v13
; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, s[4:5]
-; CGP-NEXT: v_add_i32_e64 v15, s[4:5], v15, v17
-; CGP-NEXT: v_mul_hi_u32 v15, v5, v13
+; CGP-NEXT: v_add_i32_e64 v15, s[4:5], v15, v16
+; CGP-NEXT: v_mul_hi_u32 v16, v7, v13
+; CGP-NEXT: v_mul_hi_u32 v13, v9, v13
+; CGP-NEXT: v_add_i32_e64 v10, s[4:5], v17, v10
; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, s[4:5]
-; CGP-NEXT: v_add_i32_e64 v16, s[4:5], v16, v17
-; CGP-NEXT: v_mul_lo_u32 v17, v10, v14
-; CGP-NEXT: v_add_i32_e64 v11, s[4:5], v18, v11
-; CGP-NEXT: v_cndmask_b32_e64 v18, 0, 1, s[4:5]
-; CGP-NEXT: v_add_i32_e64 v11, s[4:5], v11, v15
-; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, s[4:5]
-; CGP-NEXT: v_add_i32_e64 v15, s[4:5], v18, v15
-; CGP-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v18, vcc, v19, v18
-; CGP-NEXT: v_mul_hi_u32 v19, v7, v14
-; CGP-NEXT: v_add_i32_e32 v12, vcc, v17, v12
-; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v12, vcc, v12, v19
+; CGP-NEXT: v_add_i32_e64 v11, s[4:5], v14, v11
+; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, s[4:5]
+; CGP-NEXT: v_add_i32_e64 v10, s[4:5], v10, v19
+; CGP-NEXT: v_cndmask_b32_e64 v19, 0, 1, s[4:5]
+; CGP-NEXT: v_add_i32_e64 v17, s[4:5], v17, v19
; CGP-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v17, vcc, v17, v19
-; CGP-NEXT: v_mov_b32_e32 v19, s7
; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v16
; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v15, vcc, v15, v16
-; CGP-NEXT: v_mov_b32_e32 v16, s9
-; CGP-NEXT: v_mul_hi_u32 v13, v8, v13
-; CGP-NEXT: v_mul_hi_u32 v14, v10, v14
-; CGP-NEXT: v_add_i32_e32 v12, vcc, v12, v18
-; CGP-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v17, vcc, v17, v18
-; CGP-NEXT: v_add_i32_e32 v13, vcc, v13, v15
-; CGP-NEXT: v_add_i32_e32 v14, vcc, v14, v17
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v11
-; CGP-NEXT: v_addc_u32_e32 v8, vcc, v8, v13, vcc
-; CGP-NEXT: v_mul_lo_u32 v11, v1, v5
-; CGP-NEXT: v_mul_hi_u32 v13, v0, v5
-; CGP-NEXT: v_mul_hi_u32 v5, v1, v5
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v12
-; CGP-NEXT: v_addc_u32_e32 v10, vcc, v10, v14, vcc
-; CGP-NEXT: v_mul_lo_u32 v12, v3, v7
-; CGP-NEXT: v_mul_hi_u32 v14, v2, v7
+; CGP-NEXT: v_add_i32_e32 v18, vcc, v18, v19
+; CGP-NEXT: v_add_i32_e32 v14, vcc, v14, v16
+; CGP-NEXT: v_add_i32_e32 v10, vcc, v10, v15
+; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v18
+; CGP-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v15, vcc, v17, v15
+; CGP-NEXT: v_add_i32_e32 v14, vcc, v14, v16
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v12, v15
+; CGP-NEXT: v_add_i32_e32 v13, vcc, v13, v14
+; CGP-NEXT: v_add_i32_e32 v6, vcc, v6, v10
+; CGP-NEXT: v_addc_u32_e32 v8, vcc, v8, v12, vcc
+; CGP-NEXT: v_mul_lo_u32 v10, v1, v6
+; CGP-NEXT: v_mul_hi_u32 v12, v0, v6
+; CGP-NEXT: v_mul_hi_u32 v6, v1, v6
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v11
+; CGP-NEXT: v_addc_u32_e32 v9, vcc, v9, v13, vcc
+; CGP-NEXT: v_mul_lo_u32 v11, v3, v7
+; CGP-NEXT: v_mul_hi_u32 v13, v2, v7
; CGP-NEXT: v_mul_hi_u32 v7, v3, v7
-; CGP-NEXT: v_mul_lo_u32 v15, v0, v8
-; CGP-NEXT: v_mul_lo_u32 v17, v1, v8
-; CGP-NEXT: v_mul_hi_u32 v18, v0, v8
+; CGP-NEXT: v_mul_lo_u32 v14, v0, v8
+; CGP-NEXT: v_mul_lo_u32 v15, v1, v8
+; CGP-NEXT: v_mul_hi_u32 v16, v0, v8
; CGP-NEXT: v_mul_hi_u32 v8, v1, v8
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v15
+; CGP-NEXT: v_mul_lo_u32 v17, v2, v9
+; CGP-NEXT: v_mul_lo_u32 v18, v3, v9
+; CGP-NEXT: v_mul_hi_u32 v19, v2, v9
+; CGP-NEXT: v_mul_hi_u32 v9, v3, v9
+; CGP-NEXT: v_add_i32_e32 v10, vcc, v10, v14
+; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v6, vcc, v15, v6
; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v13
-; CGP-NEXT: v_mul_lo_u32 v11, v2, v10
-; CGP-NEXT: v_mul_lo_u32 v13, v3, v10
-; CGP-NEXT: v_add_i32_e64 v11, s[4:5], v12, v11
-; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5]
-; CGP-NEXT: v_add_i32_e64 v11, s[4:5], v11, v14
-; CGP-NEXT: v_mul_hi_u32 v11, v2, v10
-; CGP-NEXT: v_mul_hi_u32 v10, v3, v10
-; CGP-NEXT: v_add_i32_e64 v5, s[6:7], v17, v5
-; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, s[6:7]
-; CGP-NEXT: v_add_i32_e64 v7, s[6:7], v13, v7
-; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[6:7]
+; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v17
; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v18
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v18, v7
; CGP-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v15, vcc, v15, v17
-; CGP-NEXT: v_cndmask_b32_e64 v17, 0, 1, s[4:5]
+; CGP-NEXT: v_add_i32_e32 v10, vcc, v10, v12
+; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v6, vcc, v6, v16
+; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v13
+; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v19
+; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
+; CGP-NEXT: v_add_i32_e32 v10, vcc, v14, v10
+; CGP-NEXT: v_add_i32_e32 v12, vcc, v15, v12
+; CGP-NEXT: v_add_i32_e32 v11, vcc, v17, v11
+; CGP-NEXT: v_add_i32_e32 v13, vcc, v18, v13
+; CGP-NEXT: v_add_i32_e32 v6, vcc, v6, v10
+; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v11
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v14, vcc, v14, v18
-; CGP-NEXT: v_add_i32_e32 v12, vcc, v12, v17
+; CGP-NEXT: v_add_i32_e32 v10, vcc, v12, v10
+; CGP-NEXT: v_mul_lo_u32 v12, v6, s6
+; CGP-NEXT: v_mul_hi_u32 v6, s6, v6
; CGP-NEXT: v_add_i32_e32 v11, vcc, v13, v11
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v15
-; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v12
-; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v13, vcc, v14, v13
-; CGP-NEXT: v_mul_lo_u32 v14, v5, s8
-; CGP-NEXT: v_mul_hi_u32 v5, s8, v5
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v12
-; CGP-NEXT: v_mul_lo_u32 v12, v7, s8
-; CGP-NEXT: v_mul_hi_u32 v7, s8, v7
-; CGP-NEXT: v_add_i32_e32 v8, vcc, v8, v13
-; CGP-NEXT: v_add_i32_e32 v10, vcc, v10, v11
-; CGP-NEXT: v_mul_lo_u32 v8, v8, s8
-; CGP-NEXT: v_mul_lo_u32 v10, v10, s8
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v8, v5
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v10, v7
-; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v14
-; CGP-NEXT: v_subb_u32_e64 v8, s[4:5], v1, v5, vcc
-; CGP-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v5
+; CGP-NEXT: v_mul_lo_u32 v13, v7, s6
+; CGP-NEXT: v_mul_hi_u32 v7, s6, v7
+; CGP-NEXT: v_add_i32_e32 v8, vcc, v8, v10
+; CGP-NEXT: v_add_i32_e32 v9, vcc, v9, v11
+; CGP-NEXT: v_mul_lo_u32 v8, v8, s6
+; CGP-NEXT: v_mul_lo_u32 v9, v9, s6
+; CGP-NEXT: v_add_i32_e32 v6, vcc, v8, v6
+; CGP-NEXT: v_add_i32_e32 v7, vcc, v9, v7
+; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v12
+; CGP-NEXT: v_subb_u32_e64 v8, s[4:5], v1, v6, vcc
+; CGP-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v6
; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v4
-; CGP-NEXT: v_cndmask_b32_e64 v5, 0, -1, s[4:5]
-; CGP-NEXT: v_sub_i32_e64 v2, s[4:5], v2, v12
-; CGP-NEXT: v_subb_u32_e64 v10, s[6:7], v3, v7, s[4:5]
+; CGP-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
+; CGP-NEXT: v_sub_i32_e64 v2, s[4:5], v2, v13
+; CGP-NEXT: v_subb_u32_e64 v9, s[6:7], v3, v7, s[4:5]
; CGP-NEXT: v_sub_i32_e64 v3, s[6:7], v3, v7
; CGP-NEXT: v_cmp_ge_u32_e64 s[6:7], v2, v4
; CGP-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[6:7]
; CGP-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v8
-; CGP-NEXT: v_cndmask_b32_e64 v5, v6, v5, s[6:7]
+; CGP-NEXT: v_cndmask_b32_e64 v6, v5, v6, s[6:7]
; CGP-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v10
-; CGP-NEXT: v_cndmask_b32_e32 v6, v19, v7, vcc
+; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v9
+; CGP-NEXT: v_cndmask_b32_e32 v7, v5, v7, vcc
; CGP-NEXT: v_subbrev_u32_e64 v3, vcc, 0, v3, s[4:5]
-; CGP-NEXT: v_sub_i32_e32 v7, vcc, v0, v4
+; CGP-NEXT: v_sub_i32_e32 v10, vcc, v0, v4
; CGP-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v7, v4
+; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v10, v4
; CGP-NEXT: v_cndmask_b32_e64 v11, 0, -1, vcc
; CGP-NEXT: v_sub_i32_e32 v12, vcc, v2, v4
; CGP-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v12, v4
; CGP-NEXT: v_cndmask_b32_e64 v13, 0, -1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1
-; CGP-NEXT: v_cndmask_b32_e32 v9, v9, v11, vcc
-; CGP-NEXT: v_sub_i32_e32 v11, vcc, v7, v4
-; CGP-NEXT: v_subbrev_u32_e32 v14, vcc, 0, v1, vcc
+; CGP-NEXT: v_cndmask_b32_e32 v11, v5, v11, vcc
+; CGP-NEXT: v_sub_i32_e32 v14, vcc, v10, v4
+; CGP-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v1, vcc
; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3
-; CGP-NEXT: v_cndmask_b32_e32 v13, v16, v13, vcc
+; CGP-NEXT: v_cndmask_b32_e32 v5, v5, v13, vcc
; CGP-NEXT: v_sub_i32_e32 v4, vcc, v12, v4
-; CGP-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v3, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9
-; CGP-NEXT: v_cndmask_b32_e32 v7, v7, v11, vcc
-; CGP-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v13
+; CGP-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v3, vcc
+; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v11
+; CGP-NEXT: v_cndmask_b32_e32 v10, v10, v14, vcc
+; CGP-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v5
; CGP-NEXT: v_cndmask_b32_e64 v4, v12, v4, s[4:5]
-; CGP-NEXT: v_cndmask_b32_e32 v1, v1, v14, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
-; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v7, vcc
-; CGP-NEXT: v_cndmask_b32_e64 v3, v3, v15, s[4:5]
-; CGP-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v6
+; CGP-NEXT: v_cndmask_b32_e32 v1, v1, v15, vcc
+; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
+; CGP-NEXT: v_cndmask_b32_e32 v0, v0, v10, vcc
+; CGP-NEXT: v_cndmask_b32_e64 v3, v3, v13, s[4:5]
+; CGP-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v7
; CGP-NEXT: v_cndmask_b32_e64 v2, v2, v4, s[4:5]
; CGP-NEXT: v_cndmask_b32_e32 v1, v8, v1, vcc
-; CGP-NEXT: v_cndmask_b32_e64 v3, v10, v3, s[4:5]
+; CGP-NEXT: v_cndmask_b32_e64 v3, v9, v3, s[4:5]
; CGP-NEXT: s_setpc_b64 s[30:31]
%result = urem <2 x i64> %num, <i64 1235195, i64 1235195>
ret <2 x i64> %result
@@ -2435,17 +2424,17 @@ define <2 x i64> @v_urem_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_cvt_u32_f32_e32 v8, v8
; GISEL-NEXT: v_mac_f32_e32 v7, 0xcf800000, v11
; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v11
-; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v6
-; GISEL-NEXT: v_mul_lo_u32 v6, v4, v8
+; GISEL-NEXT: v_cvt_u32_f32_e32 v6, v6
+; GISEL-NEXT: v_mul_lo_u32 v12, v4, v8
; GISEL-NEXT: v_cvt_u32_f32_e32 v7, v7
; GISEL-NEXT: v_mul_lo_u32 v13, v9, v11
-; GISEL-NEXT: v_mul_lo_u32 v14, v4, v12
-; GISEL-NEXT: v_mul_lo_u32 v15, v5, v12
-; GISEL-NEXT: v_mul_hi_u32 v16, v4, v12
+; GISEL-NEXT: v_mul_lo_u32 v14, v4, v6
+; GISEL-NEXT: v_mul_lo_u32 v15, v5, v6
+; GISEL-NEXT: v_mul_hi_u32 v16, v4, v6
; GISEL-NEXT: v_mul_lo_u32 v17, v9, v7
; GISEL-NEXT: v_mul_lo_u32 v18, v10, v7
; GISEL-NEXT: v_mul_hi_u32 v19, v9, v7
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v15, v6
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v15, v12
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v18, v13
; GISEL-NEXT: v_mul_lo_u32 v15, v11, v17
; GISEL-NEXT: v_mul_hi_u32 v18, v7, v17
@@ -2455,212 +2444,206 @@ define <2 x i64> @v_urem_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
; GISEL-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
; GISEL-NEXT: v_add_i32_e32 v15, vcc, v15, v18
; GISEL-NEXT: v_mul_lo_u32 v15, v8, v14
-; GISEL-NEXT: v_mul_hi_u32 v18, v12, v14
+; GISEL-NEXT: v_mul_hi_u32 v18, v6, v14
; GISEL-NEXT: v_mul_hi_u32 v14, v8, v14
; GISEL-NEXT: v_mul_hi_u32 v17, v11, v17
-; GISEL-NEXT: v_add_i32_e64 v16, s[4:5], v6, v16
-; GISEL-NEXT: v_mul_lo_u32 v6, v12, v16
-; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v15, v6
+; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v12, v16
+; GISEL-NEXT: v_mul_lo_u32 v16, v6, v12
+; GISEL-NEXT: v_add_i32_e64 v15, s[4:5], v15, v16
+; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v15, s[4:5], v15, v18
+; GISEL-NEXT: v_mul_lo_u32 v15, v8, v12
+; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v16, s[4:5], v16, v18
+; GISEL-NEXT: v_mul_hi_u32 v18, v6, v12
+; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v15, v14
; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v6, v18
-; GISEL-NEXT: v_mul_lo_u32 v6, v8, v16
+; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v14, v18
; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, s[4:5]
; GISEL-NEXT: v_add_i32_e64 v15, s[4:5], v15, v18
-; GISEL-NEXT: v_mul_hi_u32 v18, v12, v16
-; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v6, v14
-; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v18, s[4:5], v6, v18
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v14, s[4:5], v14, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v19, vcc, v19, v6
-; GISEL-NEXT: v_mul_lo_u32 v6, v11, v13
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v17
-; GISEL-NEXT: v_mul_hi_u32 v17, v7, v13
+; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v18, vcc, v19, v18
+; GISEL-NEXT: v_mul_lo_u32 v19, v11, v13
+; GISEL-NEXT: v_add_i32_e32 v17, vcc, v19, v17
+; GISEL-NEXT: v_mul_hi_u32 v19, v7, v13
; GISEL-NEXT: v_cndmask_b32_e64 v20, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v17, vcc, v6, v17
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v20, vcc, v20, v6
-; GISEL-NEXT: v_and_b32_e32 v6, 0xffffff, v0
-; GISEL-NEXT: v_and_b32_e32 v0, 0xffffff, v2
-; GISEL-NEXT: s_bfe_i32 s4, -1, 0x10000
-; GISEL-NEXT: s_bfe_i32 s5, -1, 0x10000
-; GISEL-NEXT: s_bfe_i32 s6, -1, 0x10000
-; GISEL-NEXT: s_bfe_i32 s7, -1, 0x10000
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v18, v15
-; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v14, vcc, v14, v15
-; GISEL-NEXT: v_mov_b32_e32 v15, s4
; GISEL-NEXT: v_add_i32_e32 v17, vcc, v17, v19
-; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v18, vcc, v20, v18
-; GISEL-NEXT: v_mov_b32_e32 v19, s5
-; GISEL-NEXT: v_mul_hi_u32 v16, v8, v16
-; GISEL-NEXT: v_add_i32_e32 v14, vcc, v16, v14
-; GISEL-NEXT: v_mov_b32_e32 v16, s6
+; GISEL-NEXT: v_cndmask_b32_e64 v19, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v19, vcc, v20, v19
+; GISEL-NEXT: v_and_b32_e32 v0, 0xffffff, v0
+; GISEL-NEXT: v_and_b32_e32 v2, 0xffffff, v2
+; GISEL-NEXT: s_bfe_i32 s4, 1, 0x10000
+; GISEL-NEXT: v_add_i32_e32 v14, vcc, v14, v16
+; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v15, vcc, v15, v16
+; GISEL-NEXT: v_mov_b32_e32 v16, s4
+; GISEL-NEXT: v_mul_hi_u32 v12, v8, v12
; GISEL-NEXT: v_mul_hi_u32 v13, v11, v13
+; GISEL-NEXT: v_add_i32_e32 v17, vcc, v17, v18
+; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v18, vcc, v19, v18
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v15
; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v18
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v12, v2
-; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v8, v14, vcc
-; GISEL-NEXT: v_mul_lo_u32 v12, v4, v2
-; GISEL-NEXT: v_mul_lo_u32 v5, v5, v2
-; GISEL-NEXT: v_mul_hi_u32 v14, v4, v2
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v14
+; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v8, v12, vcc
+; GISEL-NEXT: v_mul_lo_u32 v12, v4, v6
+; GISEL-NEXT: v_mul_lo_u32 v5, v5, v6
+; GISEL-NEXT: v_mul_hi_u32 v14, v4, v6
; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v17
; GISEL-NEXT: v_addc_u32_e32 v11, vcc, v11, v13, vcc
; GISEL-NEXT: v_mul_lo_u32 v13, v9, v7
; GISEL-NEXT: v_mul_lo_u32 v10, v10, v7
-; GISEL-NEXT: v_mul_hi_u32 v17, v9, v7
+; GISEL-NEXT: v_mul_hi_u32 v15, v9, v7
; GISEL-NEXT: v_mul_lo_u32 v4, v4, v8
-; GISEL-NEXT: v_mul_lo_u32 v18, v8, v12
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v4
-; GISEL-NEXT: v_mul_hi_u32 v5, v2, v12
+; GISEL-NEXT: v_mul_lo_u32 v17, v8, v12
+; GISEL-NEXT: v_mul_hi_u32 v18, v6, v12
; GISEL-NEXT: v_mul_hi_u32 v12, v8, v12
; GISEL-NEXT: v_mul_lo_u32 v9, v9, v11
+; GISEL-NEXT: v_mul_lo_u32 v19, v11, v13
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v4
+; GISEL-NEXT: v_mul_hi_u32 v5, v7, v13
+; GISEL-NEXT: v_mul_hi_u32 v13, v11, v13
; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v9
-; GISEL-NEXT: v_mul_lo_u32 v10, v11, v13
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v14
-; GISEL-NEXT: v_mul_hi_u32 v14, v7, v13
-; GISEL-NEXT: v_mul_hi_u32 v13, v11, v13
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v17
-; GISEL-NEXT: v_mul_lo_u32 v17, v7, v9
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v17
-; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v14
-; GISEL-NEXT: v_mul_lo_u32 v10, v2, v4
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v15
+; GISEL-NEXT: v_mul_lo_u32 v10, v6, v4
; GISEL-NEXT: v_mul_lo_u32 v14, v8, v4
-; GISEL-NEXT: v_add_i32_e64 v10, s[4:5], v18, v10
-; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v10, v5
-; GISEL-NEXT: v_mul_hi_u32 v5, v2, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v10, s[4:5], v18, v10
-; GISEL-NEXT: v_mul_lo_u32 v18, v11, v9
-; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v14, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v12, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v14, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v14, vcc, v17, v14
-; GISEL-NEXT: v_mul_hi_u32 v17, v7, v9
-; GISEL-NEXT: v_add_i32_e32 v13, vcc, v18, v13
-; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v17
-; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v17, vcc, v18, v17
-; GISEL-NEXT: v_mov_b32_e32 v18, s7
+; GISEL-NEXT: v_mul_hi_u32 v15, v6, v4
; GISEL-NEXT: v_mul_hi_u32 v4, v8, v4
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v17, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v18
+; GISEL-NEXT: v_mul_lo_u32 v10, v7, v9
+; GISEL-NEXT: v_mul_lo_u32 v18, v11, v9
+; GISEL-NEXT: v_add_i32_e64 v10, s[4:5], v19, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v19, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v10, v5
+; GISEL-NEXT: v_mul_hi_u32 v5, v7, v9
; GISEL-NEXT: v_mul_hi_u32 v9, v11, v9
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v14
+; GISEL-NEXT: v_add_i32_e64 v10, s[6:7], v14, v12
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[6:7]
+; GISEL-NEXT: v_add_i32_e64 v13, s[6:7], v18, v13
+; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, s[6:7]
+; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v15
+; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v17, vcc, v17, v18
+; GISEL-NEXT: v_cndmask_b32_e64 v18, 0, 1, s[4:5]
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v15
+; GISEL-NEXT: v_add_i32_e32 v15, vcc, v19, v18
+; GISEL-NEXT: v_add_i32_e32 v13, vcc, v14, v13
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v17
; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v12, v10
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v17, v14
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v10
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v12
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v5
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v15
+; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v14
+; GISEL-NEXT: v_add_i32_e32 v13, vcc, v13, v15
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v13
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10
; GISEL-NEXT: v_addc_u32_e32 v4, vcc, v8, v4, vcc
-; GISEL-NEXT: v_mul_lo_u32 v5, 0, v2
-; GISEL-NEXT: v_mul_hi_u32 v8, v6, v2
-; GISEL-NEXT: v_mul_hi_u32 v2, 0, v2
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v13
-; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v11, v9, vcc
-; GISEL-NEXT: v_mul_lo_u32 v10, 0, v7
-; GISEL-NEXT: v_mul_hi_u32 v11, v0, v7
-; GISEL-NEXT: v_mul_hi_u32 v7, 0, v7
-; GISEL-NEXT: v_mul_lo_u32 v12, v6, v4
+; GISEL-NEXT: v_mul_lo_u32 v8, 0, v6
+; GISEL-NEXT: v_mul_hi_u32 v10, v0, v6
+; GISEL-NEXT: v_mul_hi_u32 v6, 0, v6
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
+; GISEL-NEXT: v_addc_u32_e32 v7, vcc, v11, v9, vcc
+; GISEL-NEXT: v_mul_lo_u32 v9, 0, v5
+; GISEL-NEXT: v_mul_hi_u32 v11, v2, v5
+; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5
+; GISEL-NEXT: v_mul_lo_u32 v12, v0, v4
; GISEL-NEXT: v_mul_lo_u32 v13, 0, v4
-; GISEL-NEXT: v_mul_hi_u32 v14, v6, v4
+; GISEL-NEXT: v_mul_hi_u32 v14, v0, v4
; GISEL-NEXT: v_mul_hi_u32 v4, 0, v4
-; GISEL-NEXT: v_mul_lo_u32 v17, v0, v9
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v17
+; GISEL-NEXT: v_mul_lo_u32 v15, v2, v7
+; GISEL-NEXT: v_mul_lo_u32 v17, 0, v7
+; GISEL-NEXT: v_mul_hi_u32 v18, v2, v7
+; GISEL-NEXT: v_mul_hi_u32 v7, 0, v7
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v12
+; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v13, v6
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v15
+; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v17, v5
; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
-; GISEL-NEXT: v_mul_lo_u32 v10, 0, v9
-; GISEL-NEXT: v_mul_hi_u32 v11, v0, v9
-; GISEL-NEXT: v_mul_hi_u32 v9, 0, v9
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v5, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v2, s[4:5], v13, v2
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v10, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v5, v8
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, s[4:5]
-; GISEL-NEXT: v_add_i32_e64 v2, s[4:5], v2, v14
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v11
-; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v12, v5
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v17, v14
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v12
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v10
+; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v14
+; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v11
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v18
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v8, v5
-; GISEL-NEXT: v_mul_lo_u32 v8, v3, v2
-; GISEL-NEXT: v_mul_lo_u32 v12, 0, v2
-; GISEL-NEXT: v_mul_hi_u32 v2, v3, v2
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11
-; GISEL-NEXT: v_mul_lo_u32 v11, v1, v7
-; GISEL-NEXT: v_mul_lo_u32 v13, 0, v7
-; GISEL-NEXT: v_mul_hi_u32 v7, v1, v7
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v9, v10
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v12, v8
+; GISEL-NEXT: v_add_i32_e32 v10, vcc, v13, v10
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v15, v9
+; GISEL-NEXT: v_add_i32_e32 v11, vcc, v17, v11
+; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v8
+; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v9
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
+; GISEL-NEXT: v_add_i32_e32 v8, vcc, v10, v8
+; GISEL-NEXT: v_mul_lo_u32 v10, v3, v6
+; GISEL-NEXT: v_mul_lo_u32 v12, 0, v6
+; GISEL-NEXT: v_mul_hi_u32 v6, v3, v6
+; GISEL-NEXT: v_add_i32_e32 v9, vcc, v11, v9
+; GISEL-NEXT: v_mul_lo_u32 v11, v1, v5
+; GISEL-NEXT: v_mul_lo_u32 v13, 0, v5
+; GISEL-NEXT: v_mul_hi_u32 v5, v1, v5
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9
; GISEL-NEXT: v_mul_lo_u32 v4, v3, v4
-; GISEL-NEXT: v_mul_lo_u32 v5, v1, v5
+; GISEL-NEXT: v_mul_lo_u32 v7, v1, v7
; GISEL-NEXT: v_add_i32_e32 v4, vcc, v12, v4
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v4, v2
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v7
-; GISEL-NEXT: v_sub_i32_e32 v5, vcc, v6, v8
-; GISEL-NEXT: v_subb_u32_e64 v6, s[4:5], 0, v2, vcc
-; GISEL-NEXT: v_sub_i32_e64 v2, s[4:5], 0, v2
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v5, v3
+; GISEL-NEXT: v_add_i32_e32 v7, vcc, v13, v7
+; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v6
+; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5
+; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v10
+; GISEL-NEXT: v_subb_u32_e64 v6, s[4:5], 0, v4, vcc
+; GISEL-NEXT: v_sub_i32_e64 v4, s[4:5], 0, v4
+; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v3
; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5]
-; GISEL-NEXT: v_sub_i32_e64 v8, s[4:5], v0, v11
-; GISEL-NEXT: v_subb_u32_e64 v9, s[6:7], 0, v4, s[4:5]
-; GISEL-NEXT: v_sub_i32_e64 v0, s[6:7], 0, v4
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v8, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, -1, s[6:7]
+; GISEL-NEXT: v_sub_i32_e64 v2, s[4:5], v2, v11
+; GISEL-NEXT: v_subb_u32_e64 v8, s[6:7], 0, v5, s[4:5]
+; GISEL-NEXT: v_sub_i32_e64 v5, s[6:7], 0, v5
+; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v2, v1
+; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, -1, s[6:7]
; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v7, v15, v7, s[6:7]
-; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v9
-; GISEL-NEXT: v_cndmask_b32_e32 v4, v16, v4, vcc
-; GISEL-NEXT: v_subbrev_u32_e64 v0, vcc, 0, v0, s[4:5]
-; GISEL-NEXT: v_sub_i32_e32 v10, vcc, v5, v3
-; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v7, v16, v7, s[6:7]
+; GISEL-NEXT: v_subbrev_u32_e32 v4, vcc, 0, v4, vcc
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v8
+; GISEL-NEXT: v_cndmask_b32_e32 v9, v16, v9, vcc
+; GISEL-NEXT: v_subbrev_u32_e64 v5, vcc, 0, v5, s[4:5]
+; GISEL-NEXT: v_sub_i32_e32 v10, vcc, v0, v3
+; GISEL-NEXT: v_subbrev_u32_e32 v4, vcc, 0, v4, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v10, v3
; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, -1, vcc
-; GISEL-NEXT: v_sub_i32_e32 v12, vcc, v8, v1
-; GISEL-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v0, vcc
+; GISEL-NEXT: v_sub_i32_e32 v12, vcc, v2, v1
+; GISEL-NEXT: v_subbrev_u32_e32 v5, vcc, 0, v5, vcc
; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v12, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v2
-; GISEL-NEXT: v_cndmask_b32_e32 v11, v19, v11, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, -1, vcc
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
+; GISEL-NEXT: v_cndmask_b32_e32 v11, v16, v11, vcc
; GISEL-NEXT: v_sub_i32_e32 v3, vcc, v10, v3
-; GISEL-NEXT: v_subbrev_u32_e32 v14, vcc, 0, v2, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v13
-; GISEL-NEXT: v_cndmask_b32_e32 v0, v18, v0, vcc
+; GISEL-NEXT: v_subbrev_u32_e32 v14, vcc, 0, v4, vcc
+; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v5
+; GISEL-NEXT: v_cndmask_b32_e32 v13, v16, v13, vcc
; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v12, v1
-; GISEL-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v13, vcc
+; GISEL-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v5, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v11
; GISEL-NEXT: v_cndmask_b32_e32 v3, v10, v3, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v0
+; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v13
; GISEL-NEXT: v_cndmask_b32_e64 v1, v12, v1, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e32 v10, v2, v14, vcc
+; GISEL-NEXT: v_cndmask_b32_e32 v4, v4, v14, vcc
; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v7
-; GISEL-NEXT: v_cndmask_b32_e32 v0, v5, v3, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v3, v13, v15, s[4:5]
-; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v2, v8, v1, s[4:5]
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v6, v10, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v3, v9, v3, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v3, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v3, v5, v15, s[4:5]
+; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v9
+; GISEL-NEXT: v_cndmask_b32_e64 v2, v2, v1, s[4:5]
+; GISEL-NEXT: v_cndmask_b32_e32 v1, v6, v4, vcc
+; GISEL-NEXT: v_cndmask_b32_e64 v3, v8, v3, s[4:5]
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_urem_v2i64_24bit:
More information about the llvm-commits
mailing list