[llvm] [GlobalISel] Handle div-by-pow2 (PR #83155)
Shilei Tian via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 26 10:20:01 PDT 2024
https://github.com/shiltian updated https://github.com/llvm/llvm-project/pull/83155
>From 8c9a37916b51bd3806f22fb8ee8ca0c44ba26f73 Mon Sep 17 00:00:00 2001
From: Shilei Tian <i at tianshilei.me>
Date: Tue, 26 Mar 2024 13:19:47 -0400
Subject: [PATCH] [GlobalISel] Handle div-by-pow2
This patch adds similar handling of div-by-pow2 as in `SelectionDAG`.
---
.../llvm/CodeGen/GlobalISel/CombinerHelper.h | 8 +
llvm/include/llvm/CodeGen/GlobalISel/Utils.h | 5 +
.../include/llvm/Target/GlobalISel/Combine.td | 16 +-
.../lib/CodeGen/GlobalISel/CombinerHelper.cpp | 90 +
llvm/lib/CodeGen/GlobalISel/Utils.cpp | 31 +
.../AMDGPU/GlobalISel/llvm.amdgcn.sbfe.ll | 41 +-
.../CodeGen/AMDGPU/GlobalISel/sdiv.i32.ll | 136 +-
.../CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll | 633 +----
llvm/test/CodeGen/AMDGPU/div_i128.ll | 2298 ++++++++++++++++-
9 files changed, 2517 insertions(+), 741 deletions(-)
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h b/llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h
index 9e8fc5d635c50a..28d9cf6260d620 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h
@@ -673,6 +673,14 @@ class CombinerHelper {
bool matchSDivByConst(MachineInstr &MI);
void applySDivByConst(MachineInstr &MI);
+ /// Given an G_SDIV \p MI expressing a signed divided by a pow2 constant,
+ /// return expressions that implements it by shifting.
+ bool matchDivByPow2(MachineInstr &MI, bool IsSigned);
+ void applySDivByPow2(MachineInstr &MI);
+ /// Given an G_UDIV \p MI expressing an unsigned divided by a pow2 constant,
+ /// return expressions that implements it by shifting.
+ void applyUDivByPow2(MachineInstr &MI);
+
// G_UMULH x, (1 << c)) -> x >> (bitwidth - c)
bool matchUMulHToLShr(MachineInstr &MI);
void applyUMulHToLShr(MachineInstr &MI);
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/Utils.h b/llvm/include/llvm/CodeGen/GlobalISel/Utils.h
index 0241ec4f2111d0..e710183d05db1c 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/Utils.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/Utils.h
@@ -315,6 +315,11 @@ std::optional<SmallVector<unsigned>>
ConstantFoldCountZeros(Register Src, const MachineRegisterInfo &MRI,
std::function<unsigned(APInt)> CB);
+/// Tries to constant fold a G_CTTZ operation on \p Src. If \p Src is a vector
+/// then it tries to do an element-wise constant fold.
+std::optional<SmallVector<unsigned>>
+ConstantFoldCTTZ(Register Src, const MachineRegisterInfo &MRI);
+
/// Test if the given value is known to have exactly one bit set. This differs
/// from computeKnownBits in that it doesn't necessarily determine which bit is
/// set.
diff --git a/llvm/include/llvm/Target/GlobalISel/Combine.td b/llvm/include/llvm/Target/GlobalISel/Combine.td
index 6980cbd04aeb1c..72d3c0ea69bcd2 100644
--- a/llvm/include/llvm/Target/GlobalISel/Combine.td
+++ b/llvm/include/llvm/Target/GlobalISel/Combine.td
@@ -179,6 +179,7 @@ def FmArcp : MIFlagEnum<"FmArcp">;
def FmContract : MIFlagEnum<"FmContract">;
def FmAfn : MIFlagEnum<"FmAfn">;
def FmReassoc : MIFlagEnum<"FmReassoc">;
+def IsExact : MIFlagEnum<"IsExact">;
def MIFlags;
// def not; -> Already defined as a SDNode
@@ -1036,7 +1037,20 @@ def sdiv_by_const : GICombineRule<
[{ return Helper.matchSDivByConst(*${root}); }]),
(apply [{ Helper.applySDivByConst(*${root}); }])>;
-def intdiv_combines : GICombineGroup<[udiv_by_const, sdiv_by_const]>;
+def sdiv_by_pow2 : GICombineRule<
+ (defs root:$root),
+ (match (G_SDIV $dst, $x, $y, (MIFlags (not IsExact))):$root,
+ [{ return Helper.matchDivByPow2(*${root}, /*IsSigned=*/true); }]),
+ (apply [{ Helper.applySDivByPow2(*${root}); }])>;
+
+def udiv_by_pow2 : GICombineRule<
+ (defs root:$root),
+ (match (G_UDIV $dst, $x, $y, (MIFlags (not IsExact))):$root,
+ [{ return Helper.matchDivByPow2(*${root}, /*IsSigned=*/false); }]),
+ (apply [{ Helper.applyUDivByPow2(*${root}); }])>;
+
+def intdiv_combines : GICombineGroup<[udiv_by_const, sdiv_by_const,
+ sdiv_by_pow2, udiv_by_pow2]>;
def reassoc_ptradd : GICombineRule<
(defs root:$root, build_fn_matchinfo:$matchinfo),
diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
index 2a521b6b068af7..1396a312c7a750 100644
--- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp
@@ -5270,6 +5270,96 @@ MachineInstr *CombinerHelper::buildSDivUsingMul(MachineInstr &MI) {
return MIB.buildMul(Ty, Res, Factor);
}
+bool CombinerHelper::matchDivByPow2(MachineInstr &MI, bool IsSigned) {
+ assert((MI.getOpcode() == TargetOpcode::G_SDIV ||
+ MI.getOpcode() == TargetOpcode::G_UDIV) &&
+ "Expected SDIV or UDIV");
+ auto &Div = cast<GenericMachineInstr>(MI);
+ Register RHS = Div.getReg(2);
+ auto MatchPow2 = [&](const Constant *C) {
+ auto *CI = dyn_cast<ConstantInt>(C);
+ return CI && (CI->getValue().isPowerOf2() ||
+ (IsSigned && CI->getValue().isNegatedPowerOf2()));
+ };
+ return matchUnaryPredicate(MRI, RHS, MatchPow2, /*AllowUndefs=*/false);
+}
+
+void CombinerHelper::applySDivByPow2(MachineInstr &MI) {
+ assert(MI.getOpcode() == TargetOpcode::G_SDIV && "Expected SDIV");
+ auto &SDiv = cast<GenericMachineInstr>(MI);
+ Register Dst = SDiv.getReg(0);
+ Register LHS = SDiv.getReg(1);
+ Register RHS = SDiv.getReg(2);
+ LLT Ty = MRI.getType(Dst);
+ LLT ShiftAmtTy = getTargetLowering().getPreferredShiftAmountTy(Ty);
+
+ Builder.setInstrAndDebugLoc(MI);
+
+ // Effectively we want to lower G_SDIV %lhs, %rhs, where %rhs is a power of 2,
+ // to the following version:
+ //
+ // %c1 = G_CTTZ %rhs
+ // %inexact = G_SUB $bitwidth, %c1
+ // %sign = %G_ASHR %lhs, $(bitwidth - 1)
+ // %srl = G_SHR %sign, %inexact
+ // %add = G_ADD %lhs, %srl
+ // %sra = G_ASHR %add, %c1
+ // %sra = G_SELECT, %isoneorallones, %lhs, %sra
+ // %zero = G_CONSTANT $0
+ // %neg = G_NEG %sra
+ // %isneg = G_ICMP SLT %lhs, %zero
+ // %res = G_SELECT %isneg, %neg, %sra
+
+ unsigned Bitwidth = Ty.getScalarSizeInBits();
+ auto Zero = Builder.buildConstant(Ty, 0);
+
+ auto Bits = Builder.buildConstant(ShiftAmtTy, Bitwidth);
+ auto C1 = Builder.buildCTTZ(ShiftAmtTy, RHS);
+ auto Inexact = Builder.buildSub(ShiftAmtTy, Bits, C1);
+ auto Sign = Builder.buildAShr(
+ Ty, LHS, Builder.buildConstant(ShiftAmtTy, Bitwidth - 1));
+
+ // Add (LHS < 0) ? abs2 - 1 : 0;
+ auto Lshr = Builder.buildLShr(Ty, Sign, Inexact);
+ auto Add = Builder.buildAdd(Ty, LHS, Lshr);
+ auto Shr = Builder.buildAShr(Ty, Add, C1);
+
+ LLT CCVT =
+ Ty.isVector() ? LLT::vector(Ty.getElementCount(), 1) : LLT::scalar(1);
+
+ auto One = Builder.buildConstant(Ty, 1);
+ auto AllOnes =
+ Builder.buildConstant(Ty, APInt::getAllOnes(Ty.getScalarSizeInBits()));
+ auto IsOne = Builder.buildICmp(CmpInst::Predicate::ICMP_EQ, CCVT, RHS, One);
+ auto IsAllOnes =
+ Builder.buildICmp(CmpInst::Predicate::ICMP_EQ, CCVT, RHS, AllOnes);
+ auto IsOneOrAllOnes = Builder.buildOr(CCVT, IsOne, IsAllOnes);
+ Shr = Builder.buildSelect(Ty, IsOneOrAllOnes, LHS, Shr);
+
+ // If dividing by a positive value, we're done. Otherwise, the result must
+ // be negated.
+ auto Neg = Builder.buildNeg(Ty, Shr);
+ auto IsNeg = Builder.buildICmp(CmpInst::Predicate::ICMP_SLT, CCVT, RHS, Zero);
+ Builder.buildSelect(MI.getOperand(0).getReg(), IsNeg, Neg, Shr);
+ MI.eraseFromParent();
+}
+
+void CombinerHelper::applyUDivByPow2(MachineInstr &MI) {
+ assert(MI.getOpcode() == TargetOpcode::G_UDIV && "Expected UDIV");
+ auto &UDiv = cast<GenericMachineInstr>(MI);
+ Register Dst = UDiv.getReg(0);
+ Register LHS = UDiv.getReg(1);
+ Register RHS = UDiv.getReg(2);
+ LLT Ty = MRI.getType(Dst);
+ LLT ShiftAmtTy = getTargetLowering().getPreferredShiftAmountTy(Ty);
+
+ Builder.setInstrAndDebugLoc(MI);
+
+ auto C1 = Builder.buildCTTZ(ShiftAmtTy, RHS);
+ Builder.buildLShr(MI.getOperand(0).getReg(), LHS, C1);
+ MI.eraseFromParent();
+}
+
bool CombinerHelper::matchUMulHToLShr(MachineInstr &MI) {
assert(MI.getOpcode() == TargetOpcode::G_UMULH);
Register RHS = MI.getOperand(2).getReg();
diff --git a/llvm/lib/CodeGen/GlobalISel/Utils.cpp b/llvm/lib/CodeGen/GlobalISel/Utils.cpp
index 8c41f8b1bdcdbc..427c73d7455294 100644
--- a/llvm/lib/CodeGen/GlobalISel/Utils.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/Utils.cpp
@@ -997,6 +997,37 @@ llvm::ConstantFoldCountZeros(Register Src, const MachineRegisterInfo &MRI,
return std::nullopt;
}
+std::optional<SmallVector<unsigned>>
+llvm::ConstantFoldCTTZ(Register Src, const MachineRegisterInfo &MRI) {
+ LLT Ty = MRI.getType(Src);
+ SmallVector<unsigned> FoldedCTTZs;
+ auto tryFoldScalar = [&](Register R) -> std::optional<unsigned> {
+ auto MaybeCst = getIConstantVRegVal(R, MRI);
+ if (!MaybeCst)
+ return std::nullopt;
+ return MaybeCst->countTrailingZeros();
+ };
+ if (Ty.isVector()) {
+ // Try to constant fold each element.
+ auto *BV = getOpcodeDef<GBuildVector>(Src, MRI);
+ if (!BV)
+ return std::nullopt;
+ for (unsigned SrcIdx = 0; SrcIdx < BV->getNumSources(); ++SrcIdx) {
+ if (auto MaybeFold = tryFoldScalar(BV->getSourceReg(SrcIdx))) {
+ FoldedCTTZs.emplace_back(*MaybeFold);
+ continue;
+ }
+ return std::nullopt;
+ }
+ return FoldedCTTZs;
+ }
+ if (auto MaybeCst = tryFoldScalar(Src)) {
+ FoldedCTTZs.emplace_back(*MaybeCst);
+ return FoldedCTTZs;
+ }
+ return std::nullopt;
+}
+
bool llvm::isKnownToBeAPowerOfTwo(Register Reg, const MachineRegisterInfo &MRI,
GISelKnownBits *KB) {
std::optional<DefinitionAndSourceRegister> DefSrcReg =
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.sbfe.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.sbfe.ll
index 6eed92ba1d71cc..6d4aa3b04d7612 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.sbfe.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.sbfe.ll
@@ -670,36 +670,19 @@ define amdgpu_kernel void @bfe_sext_in_reg_i24(ptr addrspace(1) %out, ptr addrsp
define amdgpu_kernel void @simplify_demanded_bfe_sdiv(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 {
; GFX6-LABEL: simplify_demanded_bfe_sdiv:
; GFX6: ; %bb.0:
-; GFX6-NEXT: v_rcp_iflag_f32_e32 v0, 2.0
-; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x0
-; GFX6-NEXT: v_mul_f32_e32 v0, 0x4f7ffffe, v0
-; GFX6-NEXT: v_cvt_u32_f32_e32 v0, v0
+; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
-; GFX6-NEXT: s_load_dword s0, s[6:7], 0x0
-; GFX6-NEXT: s_mov_b32 s6, -1
-; GFX6-NEXT: s_mov_b32 s7, 0xf000
-; GFX6-NEXT: v_mul_lo_u32 v1, v0, -2
-; GFX6-NEXT: s_waitcnt lgkmcnt(0)
-; GFX6-NEXT: s_bfe_i32 s0, s0, 0x100001
-; GFX6-NEXT: s_ashr_i32 s2, s0, 31
-; GFX6-NEXT: v_mul_hi_u32 v1, v0, v1
-; GFX6-NEXT: s_add_i32 s0, s0, s2
-; GFX6-NEXT: s_xor_b32 s0, s0, s2
-; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v1
-; GFX6-NEXT: v_mul_hi_u32 v0, s0, v0
-; GFX6-NEXT: v_lshlrev_b32_e32 v1, 1, v0
-; GFX6-NEXT: v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT: v_sub_i32_e32 v1, vcc, s0, v1
-; GFX6-NEXT: v_cmp_le_u32_e32 vcc, 2, v1
-; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT: v_subrev_i32_e64 v2, s[0:1], 2, v1
-; GFX6-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX6-NEXT: v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT: v_cmp_le_u32_e32 vcc, 2, v1
-; GFX6-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT: v_xor_b32_e32 v0, s2, v0
-; GFX6-NEXT: v_subrev_i32_e32 v0, vcc, s2, v0
-; GFX6-NEXT: buffer_store_dword v0, off, s[4:7], 0
+; GFX6-NEXT: s_load_dword s3, s[2:3], 0x0
+; GFX6-NEXT: s_mov_b32 s2, -1
+; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: s_bfe_i32 s3, s3, 0x100001
+; GFX6-NEXT: s_ashr_i32 s4, s3, 31
+; GFX6-NEXT: s_lshr_b32 s4, s4, 31
+; GFX6-NEXT: s_add_i32 s3, s3, s4
+; GFX6-NEXT: s_ashr_i32 s3, s3, 1
+; GFX6-NEXT: v_mov_b32_e32 v0, s3
+; GFX6-NEXT: s_mov_b32 s3, 0xf000
+; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
%src = load i32, ptr addrspace(1) %in, align 4
%bfe = call i32 @llvm.amdgcn.sbfe.i32(i32 %src, i32 1, i32 16)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i32.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i32.ll
index 1061f0003bd489..43b8c0cfa5bf3f 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i32.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i32.ll
@@ -279,125 +279,27 @@ define i32 @v_sdiv_i32_pow2k_denom(i32 %num) {
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT: v_ashrrev_i32_e32 v1, 31, v0
-; CHECK-NEXT: v_rcp_iflag_f32_e32 v2, 0x45800000
-; CHECK-NEXT: v_mov_b32_e32 v3, 0xfffff000
-; CHECK-NEXT: v_mov_b32_e32 v4, 0x1000
+; CHECK-NEXT: v_lshrrev_b32_e32 v1, 20, v1
; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v1
-; CHECK-NEXT: v_mul_f32_e32 v2, 0x4f7ffffe, v2
-; CHECK-NEXT: v_xor_b32_e32 v0, v0, v1
-; CHECK-NEXT: v_cvt_u32_f32_e32 v2, v2
-; CHECK-NEXT: v_mul_lo_u32 v3, v2, v3
-; CHECK-NEXT: v_mul_hi_u32 v3, v2, v3
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
-; CHECK-NEXT: v_mul_hi_u32 v2, v0, v2
-; CHECK-NEXT: v_lshlrev_b32_e32 v3, 12, v2
-; CHECK-NEXT: v_add_i32_e32 v5, vcc, 1, v2
-; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
-; CHECK-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v4
-; CHECK-NEXT: v_cndmask_b32_e64 v2, v2, v5, s[4:5]
-; CHECK-NEXT: v_subrev_i32_e32 v3, vcc, 0x1000, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v0, v0, v3, s[4:5]
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, 1, v2
-; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v0, v4
-; CHECK-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
-; CHECK-NEXT: v_xor_b32_e32 v0, v0, v1
-; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
+; CHECK-NEXT: v_ashrrev_i32_e32 v0, 12, v0
; CHECK-NEXT: s_setpc_b64 s[30:31]
%result = sdiv i32 %num, 4096
ret i32 %result
}
define <2 x i32> @v_sdiv_v2i32_pow2k_denom(<2 x i32> %num) {
-; GISEL-LABEL: v_sdiv_v2i32_pow2k_denom:
-; GISEL: ; %bb.0:
-; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_ashrrev_i32_e32 v2, 31, v0
-; GISEL-NEXT: v_mov_b32_e32 v3, 0x1000
-; GISEL-NEXT: v_cvt_f32_u32_e32 v4, 0x1000
-; GISEL-NEXT: v_mov_b32_e32 v5, 0xfffff000
-; GISEL-NEXT: v_ashrrev_i32_e32 v6, 31, v1
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v2
-; GISEL-NEXT: v_rcp_iflag_f32_e32 v4, v4
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v6
-; GISEL-NEXT: v_xor_b32_e32 v0, v0, v2
-; GISEL-NEXT: v_mul_f32_e32 v4, 0x4f7ffffe, v4
-; GISEL-NEXT: v_xor_b32_e32 v1, v1, v6
-; GISEL-NEXT: v_cvt_u32_f32_e32 v4, v4
-; GISEL-NEXT: v_mul_lo_u32 v5, v4, v5
-; GISEL-NEXT: v_mul_hi_u32 v5, v4, v5
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
-; GISEL-NEXT: v_mul_hi_u32 v5, v0, v4
-; GISEL-NEXT: v_mul_hi_u32 v4, v1, v4
-; GISEL-NEXT: v_lshlrev_b32_e32 v7, 12, v5
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, 1, v5
-; GISEL-NEXT: v_lshlrev_b32_e32 v9, 12, v4
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, 1, v4
-; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v7
-; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v1, v9
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v5, v5, v8, s[4:5]
-; GISEL-NEXT: v_sub_i32_e32 v7, vcc, v0, v3
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v1, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v4, v4, v10, s[6:7]
-; GISEL-NEXT: v_subrev_i32_e32 v8, vcc, 0x1000, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v0, v0, v7, s[4:5]
-; GISEL-NEXT: v_add_i32_e32 v7, vcc, 1, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v1, v1, v8, s[6:7]
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, 1, v4
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v3
-; GISEL-NEXT: v_cndmask_b32_e32 v0, v5, v7, vcc
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v1, v3
-; GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v8, vcc
-; GISEL-NEXT: v_xor_b32_e32 v0, v0, v2
-; GISEL-NEXT: v_xor_b32_e32 v1, v1, v6
-; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v2
-; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v1, v6
-; GISEL-NEXT: s_setpc_b64 s[30:31]
-;
-; CGP-LABEL: v_sdiv_v2i32_pow2k_denom:
-; CGP: ; %bb.0:
-; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CGP-NEXT: v_ashrrev_i32_e32 v2, 31, v0
-; CGP-NEXT: v_rcp_iflag_f32_e32 v3, 0x45800000
-; CGP-NEXT: v_mov_b32_e32 v4, 0xfffff000
-; CGP-NEXT: v_mov_b32_e32 v5, 0x1000
-; CGP-NEXT: v_ashrrev_i32_e32 v6, 31, v1
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v2
-; CGP-NEXT: v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v6
-; CGP-NEXT: v_xor_b32_e32 v0, v0, v2
-; CGP-NEXT: v_cvt_u32_f32_e32 v3, v3
-; CGP-NEXT: v_xor_b32_e32 v1, v1, v6
-; CGP-NEXT: v_mul_lo_u32 v4, v3, v4
-; CGP-NEXT: v_mul_hi_u32 v4, v3, v4
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v4
-; CGP-NEXT: v_mul_hi_u32 v4, v0, v3
-; CGP-NEXT: v_mul_hi_u32 v3, v1, v3
-; CGP-NEXT: v_lshlrev_b32_e32 v7, 12, v4
-; CGP-NEXT: v_add_i32_e32 v8, vcc, 1, v4
-; CGP-NEXT: v_lshlrev_b32_e32 v9, 12, v3
-; CGP-NEXT: v_add_i32_e32 v10, vcc, 1, v3
-; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v7
-; CGP-NEXT: v_sub_i32_e32 v1, vcc, v1, v9
-; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v5
-; CGP-NEXT: v_cndmask_b32_e64 v4, v4, v8, s[4:5]
-; CGP-NEXT: v_sub_i32_e32 v7, vcc, v0, v5
-; CGP-NEXT: v_cmp_ge_u32_e64 s[6:7], v1, v5
-; CGP-NEXT: v_cndmask_b32_e64 v3, v3, v10, s[6:7]
-; CGP-NEXT: v_subrev_i32_e32 v8, vcc, 0x1000, v1
-; CGP-NEXT: v_cndmask_b32_e64 v0, v0, v7, s[4:5]
-; CGP-NEXT: v_add_i32_e32 v7, vcc, 1, v4
-; CGP-NEXT: v_cndmask_b32_e64 v1, v1, v8, s[6:7]
-; CGP-NEXT: v_add_i32_e32 v8, vcc, 1, v3
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v5
-; CGP-NEXT: v_cndmask_b32_e32 v0, v4, v7, vcc
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v1, v5
-; CGP-NEXT: v_cndmask_b32_e32 v1, v3, v8, vcc
-; CGP-NEXT: v_xor_b32_e32 v0, v0, v2
-; CGP-NEXT: v_xor_b32_e32 v1, v1, v6
-; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v2
-; CGP-NEXT: v_sub_i32_e32 v1, vcc, v1, v6
-; CGP-NEXT: s_setpc_b64 s[30:31]
+; CHECK-LABEL: v_sdiv_v2i32_pow2k_denom:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: v_ashrrev_i32_e32 v2, 31, v0
+; CHECK-NEXT: v_ashrrev_i32_e32 v3, 31, v1
+; CHECK-NEXT: v_lshrrev_b32_e32 v2, 20, v2
+; CHECK-NEXT: v_lshrrev_b32_e32 v3, 20, v3
+; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v2
+; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v3
+; CHECK-NEXT: v_ashrrev_i32_e32 v0, 12, v0
+; CHECK-NEXT: v_ashrrev_i32_e32 v1, 12, v1
+; CHECK-NEXT: s_setpc_b64 s[30:31]
%result = sdiv <2 x i32> %num, <i32 4096, i32 4096>
ret <2 x i32> %result
}
@@ -884,3 +786,13 @@ define <2 x i32> @v_sdiv_v2i32_24bit(<2 x i32> %num, <2 x i32> %den) {
%result = sdiv <2 x i32> %num.mask, %den.mask
ret <2 x i32> %result
}
+
+define i32 @v_sdiv_i32_exact(i32 %num) {
+; CHECK-LABEL: v_sdiv_i32_exact:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: v_ashrrev_i32_e32 v0, 12, v0
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+ %result = sdiv exact i32 %num, 4096
+ ret i32 %result
+}
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
index 84906c01a4698a..98a34f9d3711a7 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll
@@ -999,602 +999,45 @@ define i64 @v_sdiv_i64_pow2k_denom(i64 %num) {
; CHECK-LABEL: v_sdiv_i64_pow2k_denom:
; CHECK: ; %bb.0:
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: v_cvt_f32_u32_e32 v2, 0x1000
-; CHECK-NEXT: v_cvt_f32_ubyte0_e32 v3, 0
-; CHECK-NEXT: v_mov_b32_e32 v6, 0xfffff000
-; CHECK-NEXT: v_mac_f32_e32 v2, 0x4f800000, v3
-; CHECK-NEXT: v_rcp_iflag_f32_e32 v2, v2
-; CHECK-NEXT: v_mul_f32_e32 v2, 0x5f7ffffc, v2
-; CHECK-NEXT: v_mul_f32_e32 v3, 0x2f800000, v2
-; CHECK-NEXT: v_trunc_f32_e32 v4, v3
-; CHECK-NEXT: v_mac_f32_e32 v2, 0xcf800000, v4
-; CHECK-NEXT: v_cvt_u32_f32_e32 v5, v2
-; CHECK-NEXT: v_cvt_u32_f32_e32 v7, v4
-; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v6, v5, 0
-; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v6, v7, v[3:4]
-; CHECK-NEXT: v_mul_hi_u32 v8, v5, v2
-; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], -1, v5, v[3:4]
-; CHECK-NEXT: v_mul_lo_u32 v4, v7, v2
-; CHECK-NEXT: v_mul_hi_u32 v2, v7, v2
-; CHECK-NEXT: v_mul_lo_u32 v9, v5, v3
-; CHECK-NEXT: v_mul_lo_u32 v10, v7, v3
-; CHECK-NEXT: v_mul_hi_u32 v11, v5, v3
-; CHECK-NEXT: v_mul_hi_u32 v3, v7, v3
-; CHECK-NEXT: v_add_i32_e32 v4, vcc, v4, v9
-; CHECK-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v10, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v4, vcc, v4, v8
-; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v4, vcc, v9, v4
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v11
-; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v8, vcc, v10, v8
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v4
-; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v4, vcc, v8, v4
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, v3, v4
-; CHECK-NEXT: v_add_i32_e32 v5, vcc, v5, v2
-; CHECK-NEXT: v_addc_u32_e32 v7, vcc, v7, v3, vcc
-; CHECK-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v6, v5, 0
-; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v6, v7, v[3:4]
-; CHECK-NEXT: v_ashrrev_i32_e32 v6, 31, v1
-; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v6
-; CHECK-NEXT: v_mad_u64_u32 v[3:4], s[4:5], -1, v5, v[3:4]
-; CHECK-NEXT: v_addc_u32_e32 v1, vcc, v1, v6, vcc
-; CHECK-NEXT: v_xor_b32_e32 v4, v0, v6
-; CHECK-NEXT: v_mul_lo_u32 v0, v7, v2
-; CHECK-NEXT: v_mul_lo_u32 v8, v5, v3
-; CHECK-NEXT: v_xor_b32_e32 v9, v1, v6
-; CHECK-NEXT: v_mul_hi_u32 v1, v5, v2
-; CHECK-NEXT: v_mul_hi_u32 v2, v7, v2
-; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v8
-; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v1
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; CHECK-NEXT: v_mul_lo_u32 v1, v7, v3
-; CHECK-NEXT: v_add_i32_e32 v0, vcc, v8, v0
-; CHECK-NEXT: v_mul_hi_u32 v8, v5, v3
-; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v2
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v1, vcc, v1, v8
-; CHECK-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v8
-; CHECK-NEXT: v_mul_hi_u32 v3, v7, v3
-; CHECK-NEXT: v_add_i32_e32 v0, vcc, v1, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v1, vcc, v2, v1
-; CHECK-NEXT: v_add_i32_e32 v1, vcc, v3, v1
-; CHECK-NEXT: v_add_i32_e32 v0, vcc, v5, v0
-; CHECK-NEXT: v_addc_u32_e32 v1, vcc, v7, v1, vcc
-; CHECK-NEXT: v_mul_lo_u32 v2, v9, v0
-; CHECK-NEXT: v_mul_lo_u32 v3, v4, v1
-; CHECK-NEXT: v_mul_hi_u32 v7, v4, v0
-; CHECK-NEXT: v_mul_hi_u32 v0, v9, v0
-; CHECK-NEXT: v_mov_b32_e32 v5, 0x1000
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v3
-; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v2, v7
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; CHECK-NEXT: v_mul_lo_u32 v7, v9, v1
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
-; CHECK-NEXT: v_mul_hi_u32 v3, v4, v1
-; CHECK-NEXT: v_add_i32_e32 v0, vcc, v7, v0
-; CHECK-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v3
-; CHECK-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, v7, v3
-; CHECK-NEXT: v_add_i32_e32 v7, vcc, v0, v2
-; CHECK-NEXT: v_mul_hi_u32 v8, v9, v1
-; CHECK-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v5, v7, 0
-; CHECK-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; CHECK-NEXT: v_add_i32_e32 v2, vcc, v3, v2
-; CHECK-NEXT: v_add_i32_e32 v3, vcc, v8, v2
-; CHECK-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v5, v3, v[1:2]
-; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v4, v0
-; 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_cmp_ge_u32_e64 s[4:5], v0, v5
-; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v5
-; CHECK-NEXT: v_cndmask_b32_e64 v4, 0, -1, s[4:5]
-; 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 v2, -1, v4, s[4:5]
-; CHECK-NEXT: v_add_i32_e32 v4, vcc, 1, v7
-; CHECK-NEXT: v_addc_u32_e32 v8, vcc, 0, v3, vcc
-; CHECK-NEXT: v_cmp_ge_u32_e32 vcc, v0, v5
-; CHECK-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc
-; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1
-; CHECK-NEXT: v_cndmask_b32_e32 v0, -1, v0, vcc
-; CHECK-NEXT: v_add_i32_e32 v1, vcc, 1, v4
-; CHECK-NEXT: v_addc_u32_e32 v5, vcc, 0, v8, vcc
-; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0
-; CHECK-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc
-; CHECK-NEXT: v_cndmask_b32_e32 v1, v8, v5, vcc
-; CHECK-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
-; CHECK-NEXT: v_cndmask_b32_e32 v0, v7, v0, vcc
-; CHECK-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc
-; CHECK-NEXT: v_xor_b32_e32 v0, v0, v6
-; CHECK-NEXT: v_xor_b32_e32 v1, v1, v6
-; CHECK-NEXT: v_sub_i32_e32 v0, vcc, v0, v6
-; CHECK-NEXT: v_subb_u32_e32 v1, vcc, v1, v6, vcc
-; CHECK-NEXT: s_setpc_b64 s[30:31]
- %result = sdiv i64 %num, 4096
- ret i64 %result
-}
-
-define <2 x i64> @v_sdiv_v2i64_pow2k_denom(<2 x i64> %num) {
-; GISEL-LABEL: v_sdiv_v2i64_pow2k_denom:
-; GISEL: ; %bb.0:
-; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GISEL-NEXT: v_cvt_f32_u32_e32 v4, 0x1000
-; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v5, 0
-; GISEL-NEXT: s_sub_u32 s6, 0, 0x1000
-; GISEL-NEXT: s_subb_u32 s7, 0, 0
-; GISEL-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
-; GISEL-NEXT: v_rcp_iflag_f32_e32 v4, v4
-; GISEL-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
-; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
-; GISEL-NEXT: v_trunc_f32_e32 v7, v5
-; GISEL-NEXT: v_mac_f32_e32 v4, 0xcf800000, v7
-; GISEL-NEXT: v_cvt_u32_f32_e32 v6, v4
-; GISEL-NEXT: v_cvt_u32_f32_e32 v7, v7
-; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], s6, v6, 0
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], s6, v7, v[5:6]
-; GISEL-NEXT: v_mul_lo_u32 v5, v7, v4
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], s7, v6, v[8:9]
-; GISEL-NEXT: v_mul_hi_u32 v9, v6, v4
-; GISEL-NEXT: v_mul_hi_u32 v4, v7, v4
-; GISEL-NEXT: v_mul_lo_u32 v10, v6, v8
-; GISEL-NEXT: v_mul_lo_u32 v11, v7, v8
-; GISEL-NEXT: v_mul_hi_u32 v12, v6, v8
-; GISEL-NEXT: v_mul_hi_u32 v8, v7, v8
-; 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 v5, vcc, v5, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v10, v5
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v11, v4
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v10
-; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v9, v5
-; GISEL-NEXT: v_add_i32_e32 v5, vcc, v8, v5
-; GISEL-NEXT: v_add_i32_e32 v11, vcc, v6, v4
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], s6, v11, 0
-; GISEL-NEXT: v_addc_u32_e32 v5, vcc, v7, v5, vcc
-; GISEL-NEXT: v_mov_b32_e32 v4, v9
-; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[4:5], s6, v5, v[4:5]
-; GISEL-NEXT: v_ashrrev_i32_e32 v4, 31, v1
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4
-; GISEL-NEXT: v_mad_u64_u32 v[9:10], s[4:5], s7, v11, v[9:10]
-; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v1, v4, vcc
-; GISEL-NEXT: v_xor_b32_e32 v10, v0, v4
-; GISEL-NEXT: v_mul_lo_u32 v0, v5, v8
-; GISEL-NEXT: v_mul_lo_u32 v12, v11, v9
-; GISEL-NEXT: v_xor_b32_e32 v13, v1, v4
-; GISEL-NEXT: v_mul_hi_u32 v1, v11, v8
-; GISEL-NEXT: v_mul_hi_u32 v8, v5, v8
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v1, v5, v9
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v12, v0
-; GISEL-NEXT: v_mul_hi_u32 v12, v11, v9
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v8
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v12
-; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v12
-; GISEL-NEXT: v_mul_hi_u32 v9, v5, v9
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v1, v0
-; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v8, v1
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v9, v1
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v11, v0
-; GISEL-NEXT: v_addc_u32_e32 v1, vcc, v5, v1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v8, v13, v0
-; GISEL-NEXT: v_mul_lo_u32 v9, v10, v1
-; GISEL-NEXT: v_mul_hi_u32 v11, v10, v0
-; GISEL-NEXT: v_mul_hi_u32 v0, v13, v0
-; GISEL-NEXT: v_mov_b32_e32 v5, 0x1000
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v11
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v11, v13, v1
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8
-; GISEL-NEXT: v_mul_hi_u32 v9, v10, v1
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v11, v0
-; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v9
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v9, vcc, v11, v9
-; GISEL-NEXT: v_add_i32_e32 v11, vcc, v0, v8
-; GISEL-NEXT: v_mul_hi_u32 v12, v13, v1
-; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v5, v11, 0
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8
-; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v8
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v5, v12, v[1:2]
-; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v10, v0
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], 0, v11, v[8:9]
-; GISEL-NEXT: s_sub_u32 s6, 0, 0x1000
-; GISEL-NEXT: s_subb_u32 s7, 0, 0
-; GISEL-NEXT: v_subb_u32_e64 v1, s[4:5], v13, v8, vcc
-; GISEL-NEXT: v_sub_i32_e64 v8, s[4:5], v13, v8
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, -1, s[4:5]
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1
-; GISEL-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v8, vcc
-; GISEL-NEXT: v_sub_i32_e32 v8, vcc, v0, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v10, -1, v9, s[4:5]
-; GISEL-NEXT: v_subbrev_u32_e32 v9, vcc, 0, v1, vcc
-; GISEL-NEXT: v_add_i32_e32 v13, vcc, 1, v11
-; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v6, 0
-; GISEL-NEXT: v_addc_u32_e32 v14, vcc, 0, v12, vcc
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v8, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v9
-; GISEL-NEXT: v_cndmask_b32_e32 v15, -1, v8, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], s6, v7, v[1:2]
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, 1, v13
-; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], s7, v6, v[8:9]
-; GISEL-NEXT: v_addc_u32_e32 v16, vcc, 0, v14, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v15
-; GISEL-NEXT: v_cndmask_b32_e32 v9, v13, v1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v1, v7, v0
-; GISEL-NEXT: v_mul_lo_u32 v13, v6, v8
-; GISEL-NEXT: v_mul_hi_u32 v15, v6, v0
-; GISEL-NEXT: v_cndmask_b32_e32 v14, v14, v16, vcc
-; GISEL-NEXT: v_mul_hi_u32 v0, v7, v0
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v13
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v1, v15
-; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v15, v7, v8
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v13, v1
-; GISEL-NEXT: v_mul_hi_u32 v13, v6, v8
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v15, v0
-; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v13
-; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v13, vcc, v15, v13
-; GISEL-NEXT: v_mul_hi_u32 v8, v7, v8
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v1
-; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v13, v1
-; GISEL-NEXT: v_add_i32_e32 v1, vcc, v8, v1
-; GISEL-NEXT: v_add_i32_e32 v8, vcc, v6, v0
-; GISEL-NEXT: v_addc_u32_e32 v13, vcc, v7, v1, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[0:1], s[4:5], s6, v8, 0
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10
-; GISEL-NEXT: v_cndmask_b32_e32 v9, v11, v9, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s6, v13, v[1:2]
-; GISEL-NEXT: v_xor_b32_e32 v1, v9, v4
-; GISEL-NEXT: v_ashrrev_i32_e32 v9, 31, v3
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], s7, v8, v[6:7]
-; GISEL-NEXT: v_cndmask_b32_e32 v10, v12, v14, vcc
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v9
-; GISEL-NEXT: v_addc_u32_e32 v3, vcc, v3, v9, vcc
-; GISEL-NEXT: v_xor_b32_e32 v11, v2, v9
-; GISEL-NEXT: v_mul_lo_u32 v2, v13, v0
-; GISEL-NEXT: v_mul_lo_u32 v7, v8, v6
-; GISEL-NEXT: v_xor_b32_e32 v12, v3, v9
-; GISEL-NEXT: v_mul_hi_u32 v3, v8, v0
-; GISEL-NEXT: v_mul_hi_u32 v0, v13, v0
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v3
-; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v3, v13, v6
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v7, v2
-; GISEL-NEXT: v_mul_hi_u32 v7, v8, v6
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0
-; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v7
-; GISEL-NEXT: v_mul_hi_u32 v6, v13, v6
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v2
-; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v3, v2
-; GISEL-NEXT: v_add_i32_e32 v2, vcc, v6, v2
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0
-; GISEL-NEXT: v_addc_u32_e32 v2, vcc, v13, v2, vcc
-; GISEL-NEXT: v_mul_lo_u32 v3, v12, v0
-; GISEL-NEXT: v_mul_lo_u32 v6, v11, v2
-; GISEL-NEXT: v_mul_hi_u32 v7, v11, v0
-; GISEL-NEXT: v_mul_hi_u32 v0, v12, v0
-; GISEL-NEXT: v_xor_b32_e32 v8, v10, v4
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v7
-; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc
-; GISEL-NEXT: v_mul_lo_u32 v7, v12, v2
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, v6, v3
-; GISEL-NEXT: v_mul_hi_u32 v6, v11, v2
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v7, v0
-; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, v7, v6
-; GISEL-NEXT: v_add_i32_e32 v10, vcc, v0, v3
-; GISEL-NEXT: v_mul_hi_u32 v7, v12, v2
-; GISEL-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v5, v10, 0
-; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0
-; GISEL-NEXT: v_add_i32_e32 v13, vcc, v7, v0
-; GISEL-NEXT: v_mov_b32_e32 v0, v3
-; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v5, v13, v[0:1]
-; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v1, v4
-; GISEL-NEXT: v_subb_u32_e32 v1, vcc, v8, v4, vcc
-; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], 0, v10, v[6:7]
-; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v11, v2
-; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], v12, v3, vcc
-; GISEL-NEXT: v_sub_i32_e64 v3, s[4:5], v12, v3
-; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v5
-; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
-; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4
-; GISEL-NEXT: v_subbrev_u32_e32 v3, vcc, 0, v3, vcc
-; GISEL-NEXT: v_cndmask_b32_e64 v4, -1, v6, s[4:5]
-; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v10
-; GISEL-NEXT: v_addc_u32_e32 v7, vcc, 0, v13, vcc
-; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v2, v5
-; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
-; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3
-; GISEL-NEXT: v_cndmask_b32_e32 v2, -1, v2, vcc
-; GISEL-NEXT: v_add_i32_e32 v3, vcc, 1, v6
-; GISEL-NEXT: v_addc_u32_e32 v5, vcc, 0, v7, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
-; GISEL-NEXT: v_cndmask_b32_e32 v2, v6, v3, vcc
-; GISEL-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc
-; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4
-; GISEL-NEXT: v_cndmask_b32_e32 v2, v10, v2, vcc
-; GISEL-NEXT: v_cndmask_b32_e32 v3, v13, v3, vcc
-; GISEL-NEXT: v_xor_b32_e32 v2, v2, v9
-; GISEL-NEXT: v_xor_b32_e32 v3, v3, v9
-; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v9
-; GISEL-NEXT: v_subb_u32_e32 v3, vcc, v3, v9, vcc
+; CHECK-NEXT: v_ashrrev_i32_e32 v2, 31, v1
+; CHECK-NEXT: v_lshrrev_b32_e32 v2, 20, v2
+; CHECK-NEXT: v_add_i32_e32 v0, vcc, v0, v2
+; CHECK-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; CHECK-NEXT: v_ashr_i64 v[0:1], v[0:1], 12
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+ %result = sdiv i64 %num, 4096
+ ret i64 %result
+}
+
+define <2 x i64> @v_sdiv_v2i64_pow2k_denom(<2 x i64> %num) {
+; GISEL-LABEL: v_sdiv_v2i64_pow2k_denom:
+; GISEL: ; %bb.0:
+; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT: v_ashrrev_i32_e32 v4, 31, v1
+; GISEL-NEXT: v_lshrrev_b32_e32 v4, 20, v4
+; GISEL-NEXT: v_ashrrev_i32_e32 v5, 31, v3
+; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4
+; GISEL-NEXT: v_lshrrev_b32_e32 v5, 20, v5
+; GISEL-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v5
+; GISEL-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; GISEL-NEXT: v_ashr_i64 v[0:1], v[0:1], 12
+; GISEL-NEXT: v_ashr_i64 v[2:3], v[2:3], 12
; GISEL-NEXT: s_setpc_b64 s[30:31]
;
; CGP-LABEL: v_sdiv_v2i64_pow2k_denom:
; CGP: ; %bb.0:
; CGP-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CGP-NEXT: v_cvt_f32_u32_e32 v4, 0x1000
-; CGP-NEXT: v_cvt_f32_ubyte0_e32 v5, 0
-; CGP-NEXT: v_mov_b32_e32 v6, 0xfffff000
-; CGP-NEXT: v_mac_f32_e32 v4, 0x4f800000, v5
-; CGP-NEXT: v_rcp_iflag_f32_e32 v4, v4
-; CGP-NEXT: v_mul_f32_e32 v4, 0x5f7ffffc, v4
-; CGP-NEXT: v_mul_f32_e32 v5, 0x2f800000, v4
-; CGP-NEXT: v_trunc_f32_e32 v7, v5
-; CGP-NEXT: v_mac_f32_e32 v4, 0xcf800000, v7
-; CGP-NEXT: v_cvt_u32_f32_e32 v8, v4
-; CGP-NEXT: v_cvt_u32_f32_e32 v9, v7
-; CGP-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v6, v8, 0
-; CGP-NEXT: v_mov_b32_e32 v7, v5
-; CGP-NEXT: v_mad_u64_u32 v[10:11], s[4:5], v6, v9, v[7:8]
-; CGP-NEXT: v_mul_hi_u32 v12, v9, v4
-; CGP-NEXT: v_mad_u64_u32 v[13:14], s[4:5], -1, v8, v[10:11]
-; CGP-NEXT: v_mul_lo_u32 v10, v9, v4
-; CGP-NEXT: v_mul_hi_u32 v11, v8, v4
-; CGP-NEXT: v_mul_lo_u32 v4, v8, v13
-; CGP-NEXT: v_mul_lo_u32 v7, v9, v13
-; CGP-NEXT: v_mul_hi_u32 v14, v8, v13
-; CGP-NEXT: v_mul_hi_u32 v13, v9, v13
-; CGP-NEXT: v_add_i32_e32 v4, vcc, v10, v4
-; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v4, vcc, v4, v11
-; CGP-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v4, vcc, v15, v4
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v12
-; CGP-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v7, v14
-; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v14, vcc, v15, v14
-; CGP-NEXT: v_add_i32_e32 v4, vcc, v7, v4
-; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v14, v7
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v13, v7
-; CGP-NEXT: v_add_i32_e32 v16, vcc, v8, v4
-; CGP-NEXT: v_mad_u64_u32 v[13:14], s[4:5], v6, v16, 0
-; CGP-NEXT: v_addc_u32_e32 v17, vcc, v9, v7, vcc
-; CGP-NEXT: v_mov_b32_e32 v4, v14
-; CGP-NEXT: v_mad_u64_u32 v[14:15], s[4:5], v6, v17, v[4:5]
-; CGP-NEXT: v_ashrrev_i32_e32 v7, 31, v1
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v7
-; CGP-NEXT: v_mad_u64_u32 v[14:15], s[4:5], -1, v16, v[14:15]
-; CGP-NEXT: v_addc_u32_e32 v1, vcc, v1, v7, vcc
-; CGP-NEXT: v_xor_b32_e32 v15, v0, v7
-; CGP-NEXT: v_mul_lo_u32 v0, v17, v13
-; CGP-NEXT: v_mul_lo_u32 v4, v16, v14
-; CGP-NEXT: v_xor_b32_e32 v18, v1, v7
-; CGP-NEXT: v_mul_hi_u32 v1, v16, v13
-; CGP-NEXT: v_mul_hi_u32 v13, v17, v13
+; CGP-NEXT: v_ashrrev_i32_e32 v4, 31, v1
+; CGP-NEXT: v_lshrrev_b32_e32 v4, 20, v4
; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v4
-; CGP-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v1
-; CGP-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v1, v17, v14
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v4, v0
-; CGP-NEXT: v_mul_hi_u32 v4, v16, v14
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v13
-; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v4
-; CGP-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v4, vcc, v13, v4
-; CGP-NEXT: v_mul_hi_u32 v13, v17, v14
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v1, v0
-; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v4, v1
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v13, v1
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v16, v0
-; CGP-NEXT: v_addc_u32_e32 v1, vcc, v17, v1, vcc
-; CGP-NEXT: v_mul_lo_u32 v13, v18, v0
-; CGP-NEXT: v_mul_lo_u32 v14, v15, v1
-; CGP-NEXT: v_mul_hi_u32 v16, v15, v0
-; CGP-NEXT: v_mul_hi_u32 v0, v18, v0
-; CGP-NEXT: v_mov_b32_e32 v4, 0x1000
-; CGP-NEXT: v_add_i32_e32 v13, vcc, v13, v14
-; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v13, vcc, v13, v16
-; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v16, v18, v1
-; CGP-NEXT: v_add_i32_e32 v13, vcc, v14, v13
-; CGP-NEXT: v_mul_hi_u32 v14, v15, v1
-; 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, v14
-; CGP-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v14, vcc, v16, v14
-; CGP-NEXT: v_add_i32_e32 v16, vcc, v0, v13
-; CGP-NEXT: v_mul_hi_u32 v17, v18, v1
-; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v4, v16, 0
-; CGP-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v13, vcc, v14, v13
-; CGP-NEXT: v_add_i32_e32 v17, vcc, v17, v13
-; CGP-NEXT: v_mad_u64_u32 v[13:14], s[4:5], v4, v17, v[1:2]
-; CGP-NEXT: v_sub_i32_e32 v0, vcc, v15, v0
-; CGP-NEXT: v_subb_u32_e64 v1, s[4:5], v18, v13, vcc
-; CGP-NEXT: v_sub_i32_e64 v13, s[4:5], v18, v13
-; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v4
-; CGP-NEXT: v_cndmask_b32_e64 v14, 0, -1, s[4:5]
-; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1
-; CGP-NEXT: v_subbrev_u32_e32 v1, vcc, 0, v13, vcc
-; CGP-NEXT: v_sub_i32_e32 v0, vcc, v0, v4
-; CGP-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v1, vcc
-; CGP-NEXT: v_add_i32_e32 v15, vcc, 1, v16
-; CGP-NEXT: v_addc_u32_e32 v18, vcc, 0, v17, vcc
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v0, v4
-; CGP-NEXT: v_mov_b32_e32 v0, v5
-; CGP-NEXT: v_cndmask_b32_e64 v14, -1, v14, s[4:5]
-; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v6, v9, v[0:1]
-; CGP-NEXT: v_cndmask_b32_e64 v19, 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], -1, v8, v[0:1]
-; CGP-NEXT: v_cndmask_b32_e32 v5, -1, v19, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, 1, v15
-; CGP-NEXT: v_mul_lo_u32 v19, v8, v0
-; CGP-NEXT: v_addc_u32_e32 v13, vcc, 0, v18, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5
-; CGP-NEXT: v_cndmask_b32_e32 v5, v15, v1, vcc
-; CGP-NEXT: v_cndmask_b32_e32 v13, v18, v13, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v10, v19
-; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v1, v11
-; CGP-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v11, v9, v0
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v10, v1
-; CGP-NEXT: v_mul_hi_u32 v10, v8, v0
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v11, v12
-; CGP-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v10, vcc, v11, v10
-; CGP-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v11, vcc, v12, v11
-; CGP-NEXT: v_mul_hi_u32 v0, v9, v0
-; CGP-NEXT: v_add_i32_e32 v1, vcc, v10, v1
-; CGP-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v10, vcc, v11, v10
-; CGP-NEXT: v_add_i32_e32 v0, vcc, v0, v10
-; CGP-NEXT: v_add_i32_e32 v8, vcc, v8, v1
-; CGP-NEXT: v_addc_u32_e32 v9, vcc, v9, v0, vcc
-; CGP-NEXT: v_mad_u64_u32 v[0:1], s[4:5], v6, v8, 0
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14
-; CGP-NEXT: v_cndmask_b32_e32 v5, v16, v5, vcc
-; CGP-NEXT: v_xor_b32_e32 v11, v5, v7
-; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v6, v9, v[1:2]
-; CGP-NEXT: v_cndmask_b32_e32 v10, v17, v13, vcc
-; CGP-NEXT: v_xor_b32_e32 v1, v10, v7
-; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], -1, v8, v[5:6]
-; CGP-NEXT: v_ashrrev_i32_e32 v10, 31, v3
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v10
-; CGP-NEXT: v_addc_u32_e32 v3, vcc, v3, v10, vcc
-; CGP-NEXT: v_xor_b32_e32 v12, v2, v10
-; CGP-NEXT: v_mul_lo_u32 v2, v9, v0
-; CGP-NEXT: v_mul_lo_u32 v6, v8, v5
-; CGP-NEXT: v_xor_b32_e32 v13, v3, v10
-; CGP-NEXT: v_mul_hi_u32 v3, v8, v0
-; CGP-NEXT: v_mul_hi_u32 v0, v9, v0
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v6
-; CGP-NEXT: v_cndmask_b32_e64 v6, 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, v9, v5
-; CGP-NEXT: v_add_i32_e32 v2, vcc, v6, v2
-; CGP-NEXT: v_mul_hi_u32 v6, v8, v5
-; 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, v6
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v3, v6
-; CGP-NEXT: v_mul_hi_u32 v5, v9, v5
-; 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, v5, v2
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v8, v0
-; CGP-NEXT: v_addc_u32_e32 v2, vcc, v9, v2, vcc
-; CGP-NEXT: v_mul_lo_u32 v5, v13, v3
-; CGP-NEXT: v_mul_lo_u32 v6, v12, v2
-; CGP-NEXT: v_sub_i32_e32 v0, vcc, v11, v7
-; CGP-NEXT: v_subb_u32_e32 v1, vcc, v1, v7, vcc
-; CGP-NEXT: v_mul_hi_u32 v7, v12, v3
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v6
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v5, v7
-; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; CGP-NEXT: v_mul_lo_u32 v7, v13, v2
-; CGP-NEXT: v_mul_hi_u32 v3, v13, v3
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v5
-; CGP-NEXT: v_mul_hi_u32 v6, v12, v2
-; CGP-NEXT: v_add_i32_e32 v3, vcc, v7, v3
-; CGP-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc
-; 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 v6, vcc, v7, v6
-; CGP-NEXT: v_add_i32_e32 v7, vcc, v3, v5
-; CGP-NEXT: v_mul_hi_u32 v8, v13, v2
-; CGP-NEXT: v_mad_u64_u32 v[2:3], s[4:5], v4, v7, 0
-; CGP-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc
-; CGP-NEXT: v_add_i32_e32 v5, vcc, v6, v5
-; CGP-NEXT: v_add_i32_e32 v8, vcc, v8, v5
-; CGP-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v4, v8, v[3:4]
-; CGP-NEXT: v_sub_i32_e32 v2, vcc, v12, v2
-; CGP-NEXT: v_subb_u32_e64 v3, s[4:5], v13, v5, vcc
-; CGP-NEXT: v_sub_i32_e64 v5, s[4:5], v13, v5
-; CGP-NEXT: v_subbrev_u32_e32 v5, vcc, 0, v5, vcc
-; CGP-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v4
-; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v4
-; CGP-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5]
-; CGP-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v3
-; CGP-NEXT: v_subbrev_u32_e32 v5, vcc, 0, v5, vcc
-; CGP-NEXT: v_cndmask_b32_e64 v3, -1, v6, s[4:5]
-; CGP-NEXT: v_add_i32_e32 v6, vcc, 1, v7
-; CGP-NEXT: v_addc_u32_e32 v9, vcc, 0, v8, vcc
-; CGP-NEXT: v_cmp_ge_u32_e32 vcc, v2, v4
-; CGP-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc
-; CGP-NEXT: v_cmp_eq_u32_e32 vcc, 0, v5
-; CGP-NEXT: v_cndmask_b32_e32 v2, -1, v2, vcc
-; CGP-NEXT: v_add_i32_e32 v4, vcc, 1, v6
-; CGP-NEXT: v_addc_u32_e32 v5, vcc, 0, v9, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v2
-; CGP-NEXT: v_cndmask_b32_e32 v2, v6, v4, vcc
-; CGP-NEXT: v_cndmask_b32_e32 v4, v9, v5, vcc
-; CGP-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3
-; CGP-NEXT: v_cndmask_b32_e32 v2, v7, v2, vcc
-; CGP-NEXT: v_cndmask_b32_e32 v3, v8, v4, vcc
-; CGP-NEXT: v_xor_b32_e32 v2, v2, v10
-; CGP-NEXT: v_xor_b32_e32 v3, v3, v10
-; CGP-NEXT: v_sub_i32_e32 v2, vcc, v2, v10
-; CGP-NEXT: v_subb_u32_e32 v3, vcc, v3, v10, vcc
+; CGP-NEXT: v_ashrrev_i32_e32 v4, 31, v3
+; CGP-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc
+; CGP-NEXT: v_lshrrev_b32_e32 v4, 20, v4
+; CGP-NEXT: v_add_i32_e32 v2, vcc, v2, v4
+; CGP-NEXT: v_addc_u32_e32 v3, vcc, 0, v3, vcc
+; CGP-NEXT: v_ashr_i64 v[0:1], v[0:1], 12
+; CGP-NEXT: v_ashr_i64 v[2:3], v[2:3], 12
; CGP-NEXT: s_setpc_b64 s[30:31]
%result = sdiv <2 x i64> %num, <i64 4096, i64 4096>
ret <2 x i64> %result
@@ -3398,3 +2841,13 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) {
%result = sdiv <2 x i64> %num.mask, %den.mask
ret <2 x i64> %result
}
+
+define i64 @v_sdiv_i64_exact(i64 %num) {
+; CHECK-LABEL: v_sdiv_i64_exact:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; CHECK-NEXT: v_ashr_i64 v[0:1], v[0:1], 12
+; CHECK-NEXT: s_setpc_b64 s[30:31]
+ %result = sdiv exact i64 %num, 4096
+ ret i64 %result
+}
diff --git a/llvm/test/CodeGen/AMDGPU/div_i128.ll b/llvm/test/CodeGen/AMDGPU/div_i128.ll
index 2f3d5d9d140c2c..cf99b5d80e13a8 100644
--- a/llvm/test/CodeGen/AMDGPU/div_i128.ll
+++ b/llvm/test/CodeGen/AMDGPU/div_i128.ll
@@ -1,10 +1,9 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
-; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s
-; RUN: llc -O0 -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9-O0,GFX9-SDAG-O0 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9 %s
+; RUN: llc -O0 -global-isel=0 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9-O0 %s
-; FIXME: GlobalISel missing the power-of-2 cases in legalization. https://github.com/llvm/llvm-project/issues/80671
-; xUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9,GFX9 %s
-; xUN: llc -O0 -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9-O0,GFX9-O0 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9-G %s
+; RUN: llc -O0 -global-isel=1 -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -o - %s | FileCheck -check-prefixes=GFX9-G-O0 %s
define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
; GFX9-LABEL: v_sdiv_i128_vv:
@@ -1223,6 +1222,1158 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
; GFX9-O0-NEXT: s_mov_b64 exec, s[4:5]
; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
; GFX9-O0-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-G-LABEL: v_sdiv_i128_vv:
+; GFX9-G: ; %bb.0: ; %_udiv-special-cases
+; GFX9-G-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-G-NEXT: v_ashrrev_i32_e32 v16, 31, v3
+; GFX9-G-NEXT: v_xor_b32_e32 v0, v16, v0
+; GFX9-G-NEXT: v_xor_b32_e32 v1, v16, v1
+; GFX9-G-NEXT: v_sub_co_u32_e32 v10, vcc, v0, v16
+; GFX9-G-NEXT: v_xor_b32_e32 v2, v16, v2
+; GFX9-G-NEXT: v_subb_co_u32_e32 v11, vcc, v1, v16, vcc
+; GFX9-G-NEXT: v_ashrrev_i32_e32 v17, 31, v7
+; GFX9-G-NEXT: v_xor_b32_e32 v3, v16, v3
+; GFX9-G-NEXT: v_subb_co_u32_e32 v12, vcc, v2, v16, vcc
+; GFX9-G-NEXT: v_subb_co_u32_e32 v13, vcc, v3, v16, vcc
+; GFX9-G-NEXT: v_xor_b32_e32 v0, v17, v4
+; GFX9-G-NEXT: v_xor_b32_e32 v1, v17, v5
+; GFX9-G-NEXT: v_sub_co_u32_e32 v18, vcc, v0, v17
+; GFX9-G-NEXT: v_xor_b32_e32 v2, v17, v6
+; GFX9-G-NEXT: v_subb_co_u32_e32 v19, vcc, v1, v17, vcc
+; GFX9-G-NEXT: v_xor_b32_e32 v3, v17, v7
+; GFX9-G-NEXT: v_subb_co_u32_e32 v4, vcc, v2, v17, vcc
+; GFX9-G-NEXT: v_subb_co_u32_e32 v5, vcc, v3, v17, vcc
+; GFX9-G-NEXT: v_or_b32_e32 v0, v18, v4
+; GFX9-G-NEXT: v_or_b32_e32 v1, v19, v5
+; GFX9-G-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1]
+; GFX9-G-NEXT: v_or_b32_e32 v0, v10, v12
+; GFX9-G-NEXT: v_or_b32_e32 v1, v11, v13
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1]
+; GFX9-G-NEXT: v_ffbh_u32_e32 v1, v18
+; GFX9-G-NEXT: v_ffbh_u32_e32 v0, v19
+; GFX9-G-NEXT: v_add_u32_e32 v1, 32, v1
+; GFX9-G-NEXT: v_ffbh_u32_e32 v2, v4
+; GFX9-G-NEXT: v_min_u32_e32 v0, v0, v1
+; GFX9-G-NEXT: v_ffbh_u32_e32 v1, v5
+; GFX9-G-NEXT: v_add_u32_e32 v2, 32, v2
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[6:7], 0, v[4:5]
+; GFX9-G-NEXT: v_add_u32_e32 v0, 64, v0
+; GFX9-G-NEXT: v_min_u32_e32 v1, v1, v2
+; GFX9-G-NEXT: v_ffbh_u32_e32 v2, v10
+; GFX9-G-NEXT: v_cndmask_b32_e64 v0, v1, v0, s[6:7]
+; GFX9-G-NEXT: v_ffbh_u32_e32 v1, v11
+; GFX9-G-NEXT: v_add_u32_e32 v2, 32, v2
+; GFX9-G-NEXT: v_ffbh_u32_e32 v3, v12
+; GFX9-G-NEXT: v_min_u32_e32 v1, v1, v2
+; GFX9-G-NEXT: v_ffbh_u32_e32 v2, v13
+; GFX9-G-NEXT: v_add_u32_e32 v3, 32, v3
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[6:7], 0, v[12:13]
+; GFX9-G-NEXT: v_add_u32_e32 v1, 64, v1
+; GFX9-G-NEXT: v_min_u32_e32 v2, v2, v3
+; GFX9-G-NEXT: v_cndmask_b32_e64 v1, v2, v1, s[6:7]
+; GFX9-G-NEXT: v_sub_co_u32_e64 v0, s[6:7], v0, v1
+; GFX9-G-NEXT: v_subb_co_u32_e64 v1, s[6:7], 0, 0, s[6:7]
+; GFX9-G-NEXT: v_mov_b32_e32 v6, 0x7f
+; GFX9-G-NEXT: v_subb_co_u32_e64 v2, s[6:7], 0, 0, s[6:7]
+; GFX9-G-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-G-NEXT: v_subb_co_u32_e64 v3, s[6:7], 0, 0, s[6:7]
+; GFX9-G-NEXT: v_cmp_gt_u64_e64 s[6:7], v[0:1], v[6:7]
+; GFX9-G-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX9-G-NEXT: v_cndmask_b32_e64 v6, 0, 1, s[6:7]
+; GFX9-G-NEXT: v_cmp_lt_u64_e64 s[6:7], 0, v[2:3]
+; GFX9-G-NEXT: v_or_b32_e32 v15, v1, v3
+; GFX9-G-NEXT: v_cndmask_b32_e64 v7, 0, 1, s[6:7]
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[6:7], 0, v[2:3]
+; GFX9-G-NEXT: s_mov_b64 s[8:9], 0
+; GFX9-G-NEXT: v_cndmask_b32_e64 v6, v7, v6, s[6:7]
+; GFX9-G-NEXT: v_cndmask_b32_e64 v7, 0, 1, s[4:5]
+; GFX9-G-NEXT: v_or_b32_e32 v20, v7, v6
+; GFX9-G-NEXT: v_xor_b32_e32 v6, 0x7f, v0
+; GFX9-G-NEXT: v_or_b32_e32 v14, v6, v2
+; GFX9-G-NEXT: v_and_b32_e32 v6, 1, v20
+; GFX9-G-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6
+; GFX9-G-NEXT: v_cndmask_b32_e64 v6, v10, 0, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e64 v7, v11, 0, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e64 v8, v12, 0, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e64 v9, v13, 0, vcc
+; GFX9-G-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[14:15]
+; GFX9-G-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc
+; GFX9-G-NEXT: v_or_b32_e32 v14, v20, v14
+; GFX9-G-NEXT: v_and_b32_e32 v14, 1, v14
+; GFX9-G-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14
+; GFX9-G-NEXT: s_xor_b64 s[4:5], vcc, -1
+; GFX9-G-NEXT: s_and_saveexec_b64 s[6:7], s[4:5]
+; GFX9-G-NEXT: s_cbranch_execz .LBB0_6
+; GFX9-G-NEXT: ; %bb.1: ; %udiv-bb1
+; GFX9-G-NEXT: v_add_co_u32_e32 v20, vcc, 1, v0
+; GFX9-G-NEXT: v_addc_co_u32_e32 v21, vcc, 0, v1, vcc
+; GFX9-G-NEXT: v_addc_co_u32_e32 v22, vcc, 0, v2, vcc
+; GFX9-G-NEXT: v_addc_co_u32_e32 v23, vcc, 0, v3, vcc
+; GFX9-G-NEXT: s_xor_b64 s[4:5], vcc, -1
+; GFX9-G-NEXT: v_sub_co_u32_e32 v8, vcc, 0x7f, v0
+; GFX9-G-NEXT: v_sub_u32_e32 v0, 64, v8
+; GFX9-G-NEXT: v_lshrrev_b64 v[0:1], v0, v[10:11]
+; GFX9-G-NEXT: v_lshlrev_b64 v[2:3], v8, v[12:13]
+; GFX9-G-NEXT: v_subrev_u32_e32 v9, 64, v8
+; GFX9-G-NEXT: v_lshlrev_b64 v[6:7], v8, v[10:11]
+; GFX9-G-NEXT: v_or_b32_e32 v2, v0, v2
+; GFX9-G-NEXT: v_or_b32_e32 v3, v1, v3
+; GFX9-G-NEXT: v_lshlrev_b64 v[0:1], v9, v[10:11]
+; GFX9-G-NEXT: v_cmp_gt_u32_e32 vcc, 64, v8
+; GFX9-G-NEXT: v_cndmask_b32_e32 v6, 0, v6, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v7, 0, v7, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-G-NEXT: v_cmp_eq_u32_e32 vcc, 0, v8
+; GFX9-G-NEXT: v_cndmask_b32_e32 v8, v0, v12, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v9, v1, v13, vcc
+; GFX9-G-NEXT: s_mov_b64 s[10:11], s[8:9]
+; GFX9-G-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-G-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-G-NEXT: v_mov_b32_e32 v2, s10
+; GFX9-G-NEXT: v_mov_b32_e32 v3, s11
+; GFX9-G-NEXT: s_and_saveexec_b64 s[8:9], s[4:5]
+; GFX9-G-NEXT: s_xor_b64 s[12:13], exec, s[8:9]
+; GFX9-G-NEXT: s_cbranch_execz .LBB0_5
+; GFX9-G-NEXT: ; %bb.2: ; %udiv-preheader
+; GFX9-G-NEXT: v_sub_u32_e32 v2, 64, v20
+; GFX9-G-NEXT: v_lshrrev_b64 v[0:1], v20, v[10:11]
+; GFX9-G-NEXT: v_lshlrev_b64 v[2:3], v2, v[12:13]
+; GFX9-G-NEXT: v_subrev_u32_e32 v24, 64, v20
+; GFX9-G-NEXT: v_lshrrev_b64 v[14:15], v20, v[12:13]
+; GFX9-G-NEXT: v_or_b32_e32 v2, v0, v2
+; GFX9-G-NEXT: v_or_b32_e32 v3, v1, v3
+; GFX9-G-NEXT: v_lshrrev_b64 v[0:1], v24, v[12:13]
+; GFX9-G-NEXT: v_cmp_gt_u32_e32 vcc, 64, v20
+; GFX9-G-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v14, 0, v14, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v15, 0, v15, vcc
+; GFX9-G-NEXT: v_add_co_u32_e32 v24, vcc, -1, v18
+; GFX9-G-NEXT: s_mov_b64 s[8:9], 0
+; GFX9-G-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v20
+; GFX9-G-NEXT: v_addc_co_u32_e32 v25, vcc, -1, v19, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e64 v12, v0, v10, s[4:5]
+; GFX9-G-NEXT: v_cndmask_b32_e64 v13, v1, v11, s[4:5]
+; GFX9-G-NEXT: v_addc_co_u32_e32 v26, vcc, -1, v4, vcc
+; GFX9-G-NEXT: s_mov_b64 s[10:11], s[8:9]
+; GFX9-G-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-G-NEXT: v_addc_co_u32_e32 v27, vcc, -1, v5, vcc
+; GFX9-G-NEXT: v_mov_b32_e32 v11, 0
+; GFX9-G-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-G-NEXT: v_mov_b32_e32 v2, s10
+; GFX9-G-NEXT: v_mov_b32_e32 v3, s11
+; GFX9-G-NEXT: .LBB0_3: ; %udiv-do-while
+; GFX9-G-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-G-NEXT: v_lshlrev_b64 v[2:3], 1, v[6:7]
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v10, 31, v7
+; GFX9-G-NEXT: v_or_b32_e32 v6, v0, v2
+; GFX9-G-NEXT: v_or_b32_e32 v7, v1, v3
+; GFX9-G-NEXT: v_lshlrev_b64 v[2:3], 1, v[12:13]
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v12, 31, v9
+; GFX9-G-NEXT: v_lshlrev_b64 v[0:1], 1, v[14:15]
+; GFX9-G-NEXT: v_or_b32_e32 v2, v2, v12
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v14, 31, v13
+; GFX9-G-NEXT: v_sub_co_u32_e32 v12, vcc, v24, v2
+; GFX9-G-NEXT: v_or_b32_e32 v0, v0, v14
+; GFX9-G-NEXT: v_subb_co_u32_e32 v12, vcc, v25, v3, vcc
+; GFX9-G-NEXT: v_subb_co_u32_e32 v12, vcc, v26, v0, vcc
+; GFX9-G-NEXT: v_subb_co_u32_e32 v12, vcc, v27, v1, vcc
+; GFX9-G-NEXT: v_ashrrev_i32_e32 v28, 31, v12
+; GFX9-G-NEXT: v_and_b32_e32 v12, v28, v18
+; GFX9-G-NEXT: v_sub_co_u32_e32 v12, vcc, v2, v12
+; GFX9-G-NEXT: v_and_b32_e32 v2, v28, v19
+; GFX9-G-NEXT: v_subb_co_u32_e32 v13, vcc, v3, v2, vcc
+; GFX9-G-NEXT: v_and_b32_e32 v2, v28, v4
+; GFX9-G-NEXT: v_subb_co_u32_e32 v14, vcc, v0, v2, vcc
+; GFX9-G-NEXT: v_and_b32_e32 v0, v28, v5
+; GFX9-G-NEXT: v_subb_co_u32_e32 v15, vcc, v1, v0, vcc
+; GFX9-G-NEXT: v_add_co_u32_e32 v20, vcc, -1, v20
+; GFX9-G-NEXT: v_addc_co_u32_e32 v21, vcc, -1, v21, vcc
+; GFX9-G-NEXT: v_addc_co_u32_e32 v22, vcc, -1, v22, vcc
+; GFX9-G-NEXT: v_addc_co_u32_e32 v23, vcc, -1, v23, vcc
+; GFX9-G-NEXT: v_lshlrev_b64 v[8:9], 1, v[8:9]
+; GFX9-G-NEXT: v_or_b32_e32 v0, v20, v22
+; GFX9-G-NEXT: v_or_b32_e32 v1, v21, v23
+; GFX9-G-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1]
+; GFX9-G-NEXT: v_or_b32_e32 v8, v8, v10
+; GFX9-G-NEXT: v_and_b32_e32 v10, 1, v28
+; GFX9-G-NEXT: v_mov_b32_e32 v0, v10
+; GFX9-G-NEXT: s_or_b64 s[8:9], vcc, s[8:9]
+; GFX9-G-NEXT: v_mov_b32_e32 v1, v11
+; GFX9-G-NEXT: s_andn2_b64 exec, exec, s[8:9]
+; GFX9-G-NEXT: s_cbranch_execnz .LBB0_3
+; GFX9-G-NEXT: ; %bb.4: ; %Flow
+; GFX9-G-NEXT: s_or_b64 exec, exec, s[8:9]
+; GFX9-G-NEXT: .LBB0_5: ; %Flow2
+; GFX9-G-NEXT: s_or_b64 exec, exec, s[12:13]
+; GFX9-G-NEXT: v_lshlrev_b64 v[2:3], 1, v[6:7]
+; GFX9-G-NEXT: v_lshlrev_b64 v[8:9], 1, v[8:9]
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v4, 31, v7
+; GFX9-G-NEXT: v_or_b32_e32 v8, v8, v4
+; GFX9-G-NEXT: v_or_b32_e32 v6, v0, v2
+; GFX9-G-NEXT: v_or_b32_e32 v7, v1, v3
+; GFX9-G-NEXT: .LBB0_6: ; %Flow3
+; GFX9-G-NEXT: s_or_b64 exec, exec, s[6:7]
+; GFX9-G-NEXT: v_xor_b32_e32 v3, v17, v16
+; GFX9-G-NEXT: v_xor_b32_e32 v0, v6, v3
+; GFX9-G-NEXT: v_xor_b32_e32 v1, v7, v3
+; GFX9-G-NEXT: v_sub_co_u32_e32 v0, vcc, v0, v3
+; GFX9-G-NEXT: v_xor_b32_e32 v2, v8, v3
+; GFX9-G-NEXT: v_subb_co_u32_e32 v1, vcc, v1, v3, vcc
+; GFX9-G-NEXT: v_xor_b32_e32 v4, v9, v3
+; GFX9-G-NEXT: v_subb_co_u32_e32 v2, vcc, v2, v3, vcc
+; GFX9-G-NEXT: v_subb_co_u32_e32 v3, vcc, v4, v3, vcc
+; GFX9-G-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-G-O0-LABEL: v_sdiv_i128_vv:
+; GFX9-G-O0: ; %bb.0: ; %_udiv-special-cases
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-G-O0-NEXT: s_xor_saveexec_b64 s[4:5], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:328 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:332 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:336 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 offset:340 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:344 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[4:5]
+; GFX9-G-O0-NEXT: ; implicit-def: $vgpr8 : SGPR spill to VGPR lane
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v0
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr8 killed $vgpr8 def $vgpr8_vgpr9_vgpr10_vgpr11 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v3
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v10, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:96 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v7
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:88 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:92 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:96 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: ; kill: def $vgpr13 killed $vgpr13 def $vgpr13_vgpr14_vgpr15_vgpr16 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, v1
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], 0
+; GFX9-G-O0-NEXT: s_mov_b64 s[12:13], 0x7f
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1_vgpr2 killed $vgpr4_vgpr5 killed $exec
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v7
+; GFX9-G-O0-NEXT: ; kill: def $vgpr3 killed $vgpr1 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v2
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s6
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v12, v3, v8
+; GFX9-G-O0-NEXT: ; kill: def $vgpr3 killed $vgpr1 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr2 killed $vgpr2 killed $vgpr1_vgpr2 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s6
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v10, v1, v2
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1_vgpr2 killed $vgpr13_vgpr14 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v16
+; GFX9-G-O0-NEXT: ; kill: def $vgpr3 killed $vgpr1 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v2
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s6
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v11, v3, v8
+; GFX9-G-O0-NEXT: ; kill: def $vgpr3 killed $vgpr1 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr2 killed $vgpr2 killed $vgpr1_vgpr2 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s6
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v9, v1, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v3
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v1, v12, v1
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v4, v12, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v6
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v3, v10, v3
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v2, v10, v2
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v1, s[6:7], v1, v12
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v7, s[6:7], v4, v12, s[6:7]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v6, s[6:7], v3, v10, s[6:7]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v5, s[6:7], v2, v10, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2_vgpr3_vgpr4 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v5
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v16
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v7
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v5, v11, v5
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v8, v11, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v14
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v7, v9, v7
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v6, v9, v6
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v5, s[6:7], v5, v11
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v15, s[6:7], v8, v11, s[6:7]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v14, s[6:7], v7, v9, s[6:7]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v13, s[6:7], v6, v9, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr5 killed $vgpr5 def $vgpr5_vgpr6_vgpr7_vgpr8 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v13
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v13, v11, v12
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v11, v11, v12
+; GFX9-G-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v11, v9, v10
+; GFX9-G-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v9, v9, v10
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v14
+; GFX9-G-O0-NEXT: v_or_b32_e64 v9, v9, v12
+; GFX9-G-O0-NEXT: v_or_b32_e64 v11, v10, v11
+; GFX9-G-O0-NEXT: ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[6:7], v[9:10], v[11:12]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v14
+; GFX9-G-O0-NEXT: v_or_b32_e64 v9, v9, v12
+; GFX9-G-O0-NEXT: v_or_b32_e64 v11, v10, v11
+; GFX9-G-O0-NEXT: ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[8:9], v[9:10], v[11:12]
+; GFX9-G-O0-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[8:9], v[8:9], v[5:6]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v11
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v5, v5
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v6, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, 32
+; GFX9-G-O0-NEXT: v_add_u32_e64 v6, v6, v7
+; GFX9-G-O0-NEXT: v_min_u32_e64 v5, v5, v6
+; GFX9-G-O0-NEXT: s_mov_b32 s10, 64
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s10
+; GFX9-G-O0-NEXT: v_add_u32_e64 v6, v5, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v9
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v5, v5
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v7, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, 32
+; GFX9-G-O0-NEXT: v_add_u32_e64 v7, v7, v8
+; GFX9-G-O0-NEXT: v_min_u32_e64 v5, v5, v7
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v5, v5, v6, s[8:9]
+; GFX9-G-O0-NEXT: s_mov_b32 s16, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[8:9], v[9:10], v[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v12
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v6, v6
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v7, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, 32
+; GFX9-G-O0-NEXT: v_add_u32_e64 v7, v7, v8
+; GFX9-G-O0-NEXT: v_min_u32_e64 v6, v6, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, s10
+; GFX9-G-O0-NEXT: v_add_u32_e64 v7, v6, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v10
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v6, v6
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v8, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, 32
+; GFX9-G-O0-NEXT: v_add_u32_e64 v8, v8, v9
+; GFX9-G-O0-NEXT: v_min_u32_e64 v6, v6, v8
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[8:9]
+; GFX9-G-O0-NEXT: s_mov_b32 s15, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s11, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s14, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s10, 0
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v6, s[8:9], v5, v6
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s16
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, s16
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v7, s[8:9], v5, v7, s[8:9]
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, s14
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v9, s[8:9], v5, v8, s[8:9]
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, s10
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v8, s[8:9], v5, v8, s[8:9]
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, s4
+; GFX9-G-O0-NEXT: v_cmp_gt_u64_e64 s[10:11], v[12:13], v[14:15]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[8:9], v[12:13], v[14:15]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, s12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, s13
+; GFX9-G-O0-NEXT: v_cmp_gt_u64_e64 s[12:13], v[10:11], v[12:13]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v10, v5, v10, s[12:13]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v5, v5, v11, s[10:11]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v10, v5, v10, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v5, v5, v11, s[6:7]
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v5, v10
+; GFX9-G-O0-NEXT: s_mov_b32 s7, 0x7f
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 0
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v6, v6, s7
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v7, v7, s6
+; GFX9-G-O0-NEXT: v_or_b32_e64 v6, v6, v9
+; GFX9-G-O0-NEXT: v_or_b32_e64 v8, v7, v8
+; GFX9-G-O0-NEXT: ; kill: def $vgpr6 killed $vgpr6 def $vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[4:5], v[6:7], v[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v3
+; GFX9-G-O0-NEXT: v_and_b32_e32 v1, 1, v5
+; GFX9-G-O0-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v7
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v1, v1, v4, s[6:7]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v2, v3, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v3
+; GFX9-G-O0-NEXT: v_and_b32_e32 v3, 1, v5
+; GFX9-G-O0-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v9
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[6:7]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v3, v4, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr6 killed $vgpr6 def $vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v3
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1_vgpr2 killed $vgpr1_vgpr2 def $vgpr1_vgpr2_vgpr3_vgpr4 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[4:5]
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v5, v6
+; GFX9-G-O0-NEXT: v_and_b32_e32 v5, 1, v5
+; GFX9-G-O0-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v5
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], -1
+; GFX9-G-O0-NEXT: s_xor_b64 s[6:7], s[4:5], s[6:7]
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], exec
+; GFX9-G-O0-NEXT: v_writelane_b32 v0, s4, 0
+; GFX9-G-O0-NEXT: v_writelane_b32 v0, s5, 1
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7]
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[4:5]
+; GFX9-G-O0-NEXT: s_cbranch_execz .LBB0_3
+; GFX9-G-O0-NEXT: s_branch .LBB0_8
+; GFX9-G-O0-NEXT: .LBB0_1: ; %Flow
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_readlane_b32 s4, v0, 2
+; GFX9-G-O0-NEXT: v_readlane_b32 s5, v0, 3
+; GFX9-G-O0-NEXT: s_or_b64 exec, exec, s[4:5]
+; GFX9-G-O0-NEXT: ; %bb.2: ; %Flow
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:132 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:136 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:140 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:144 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:148 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:152 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:156 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:160 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(4)
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:116 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:120 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:124 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:128 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:104 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:108 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:112 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB0_5
+; GFX9-G-O0-NEXT: .LBB0_3: ; %Flow2
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_readlane_b32 s4, v4, 0
+; GFX9-G-O0-NEXT: v_readlane_b32 s5, v4, 1
+; GFX9-G-O0-NEXT: s_or_b64 exec, exec, s[4:5]
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:164 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:168 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:172 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:176 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB0_9
+; GFX9-G-O0-NEXT: .LBB0_4: ; %udiv-loop-exit
+; GFX9-G-O0-NEXT: buffer_load_dword v14, off, s[0:3], s32 offset:180 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v15, off, s[0:3], s32 offset:184 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v16, off, s[0:3], s32 offset:188 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v17, off, s[0:3], s32 offset:192 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:196 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:200 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:204 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:208 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v7
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[10:11], v0, v[2:3]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[0:1], v0, v[4:5]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr2 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr3 killed $vgpr3 killed $vgpr2_vgpr3 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v6, v2, v3
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v16
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v11
+; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v7
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v1, v5
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v9
+; GFX9-G-O0-NEXT: v_or3_b32 v4, v4, v6, v7
+; GFX9-G-O0-NEXT: v_or3_b32 v2, v2, v3, v5
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 def $vgpr4_vgpr5 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v2
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0_vgpr1 killed $vgpr0_vgpr1 def $vgpr0_vgpr1_vgpr2_vgpr3 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB0_3
+; GFX9-G-O0-NEXT: .LBB0_5: ; %Flow1
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_readlane_b32 s4, v8, 4
+; GFX9-G-O0-NEXT: v_readlane_b32 s5, v8, 5
+; GFX9-G-O0-NEXT: s_or_b64 exec, exec, s[4:5]
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:116 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:120 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:124 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:128 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:100 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:104 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:108 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:112 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:180 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:184 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:188 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:192 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:196 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:200 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:204 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:208 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB0_4
+; GFX9-G-O0-NEXT: .LBB0_6: ; %udiv-do-while
+; GFX9-G-O0-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v16, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_readlane_b32 s6, v16, 6
+; GFX9-G-O0-NEXT: v_readlane_b32 s7, v16, 7
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:212 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v13, off, s[0:3], s32 offset:216 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v14, off, s[0:3], s32 offset:220 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v15, off, s[0:3], s32 offset:224 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:228 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:232 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:236 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:240 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v17, off, s[0:3], s32 offset:244 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v18, off, s[0:3], s32 offset:248 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v19, off, s[0:3], s32 offset:252 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v20, off, s[0:3], s32 offset:256 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v31, off, s[0:3], s32 offset:260 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:264 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v33, off, s[0:3], s32 offset:268 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v34, off, s[0:3], s32 offset:272 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v25, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v28, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:276 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:280 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v10, off, s[0:3], s32 offset:284 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:288 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], 0
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(16)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v5
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s8
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[21:22], v2, v[0:1]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s8
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[4:5], v2, v[3:4]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr2 killed $vgpr0 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 killed $vgpr0_vgpr1 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s9
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v3, v0, v1
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v5
+; GFX9-G-O0-NEXT: v_or_b32_e64 v7, v2, v3
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v0, v1
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0_vgpr1 killed $vgpr12_vgpr13 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v15
+; GFX9-G-O0-NEXT: ; kill: def $vgpr2 killed $vgpr0 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 killed $vgpr0_vgpr1 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s9
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v3, v0, v1
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v21
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v22
+; GFX9-G-O0-NEXT: v_or_b32_e64 v4, v2, v3
+; GFX9-G-O0-NEXT: v_or_b32_e64 v9, v0, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[23:24], v0, v[2:3]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[0:1], v0, v[12:13]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr12 killed $vgpr2 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr3 killed $vgpr3 killed $vgpr2_vgpr3 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s8
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v14, v2, v3
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v1
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(8)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v29, v31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v30, v32
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v21, v33
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v22, v34
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v29
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v30
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v23
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v24
+; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v15
+; GFX9-G-O0-NEXT: v_or_b32_e64 v13, v1, v13
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v21
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v22
+; GFX9-G-O0-NEXT: v_or3_b32 v12, v12, v14, v15
+; GFX9-G-O0-NEXT: v_or3_b32 v2, v2, v3, v13
+; GFX9-G-O0-NEXT: ; kill: def $vgpr12 killed $vgpr12 def $vgpr12_vgpr13 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v2
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0_vgpr1 killed $vgpr0_vgpr1 def $vgpr0_vgpr1_vgpr2_vgpr3 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v13
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v11, s[8:9], v11, v4
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v10, s[8:9], v10, v9, s[8:9]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v8, s[8:9], v8, v7, s[8:9]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v10, s[8:9], v6, v5, s[8:9]
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s8
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v8, v6, v10
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s8
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v6, v6, v10
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 1
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 0
+; GFX9-G-O0-NEXT: v_and_b32_e64 v12, v8, s9
+; GFX9-G-O0-NEXT: v_and_b32_e64 v10, v8, s8
+; GFX9-G-O0-NEXT: ; kill: def $vgpr12 killed $vgpr12 def $vgpr12_vgpr13 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, s4
+; GFX9-G-O0-NEXT: ; kill: def $vgpr12_vgpr13 killed $vgpr12_vgpr13 def $vgpr12_vgpr13_vgpr14_vgpr15 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v23, v25
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v24, v26
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v21, v27
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v22, v28
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v23
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v24
+; GFX9-G-O0-NEXT: v_and_b32_e64 v11, v8, v11
+; GFX9-G-O0-NEXT: v_and_b32_e64 v10, v8, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v21
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v21, v22
+; GFX9-G-O0-NEXT: v_and_b32_e64 v8, v6, v8
+; GFX9-G-O0-NEXT: v_and_b32_e64 v6, v6, v21
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v4, s[8:9], v4, v11
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v10, s[8:9], v9, v10, s[8:9]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v9, s[8:9], v7, v8, s[8:9]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v8, s[8:9], v5, v6, s[8:9]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 def $vgpr4_vgpr5_vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v18
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v19
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v20
+; GFX9-G-O0-NEXT: s_mov_b32 s8, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s12, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s11, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s10, -1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v17, s8
+; GFX9-G-O0-NEXT: v_add_co_u32_e64 v17, s[8:9], v11, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, s12
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v18, s[8:9], v10, v11, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, s11
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v20, s[8:9], v9, v10, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, s10
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v19, s[8:9], v8, v9, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v18
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v20
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v19
+; GFX9-G-O0-NEXT: v_or_b32_e64 v17, v17, v20
+; GFX9-G-O0-NEXT: v_or_b32_e64 v19, v18, v19
+; GFX9-G-O0-NEXT: ; kill: def $vgpr17 killed $vgpr17 def $vgpr17_vgpr18 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v18, v19
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v20, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v19, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[4:5], v[17:18], v[19:20]
+; GFX9-G-O0-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v20, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v19, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v18, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v17, v0
+; GFX9-G-O0-NEXT: buffer_store_dword v17, off, s[0:3], s32 offset:132 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v18, off, s[0:3], s32 offset:136 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v19, off, s[0:3], s32 offset:140 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v20, off, s[0:3], s32 offset:144 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v20, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v19, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v18, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v17, v12
+; GFX9-G-O0-NEXT: buffer_store_dword v17, off, s[0:3], s32 offset:148 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v18, off, s[0:3], s32 offset:152 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v19, off, s[0:3], s32 offset:156 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v20, off, s[0:3], s32 offset:160 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], s[4:5]
+; GFX9-G-O0-NEXT: v_writelane_b32 v16, s6, 2
+; GFX9-G-O0-NEXT: v_writelane_b32 v16, s7, 3
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], s[4:5]
+; GFX9-G-O0-NEXT: v_writelane_b32 v16, s6, 6
+; GFX9-G-O0-NEXT: v_writelane_b32 v16, s7, 7
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:260 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:264 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:268 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v15, off, s[0:3], s32 offset:272 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:244 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:248 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v10, off, s[0:3], s32 offset:252 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:256 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:228 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:232 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:236 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:240 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:212 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:216 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:220 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:224 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_andn2_b64 exec, exec, s[4:5]
+; GFX9-G-O0-NEXT: s_cbranch_execnz .LBB0_6
+; GFX9-G-O0-NEXT: s_branch .LBB0_1
+; GFX9-G-O0-NEXT: .LBB0_7: ; %udiv-preheader
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:292 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:296 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:300 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:304 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:308 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v9, off, s[0:3], s32 offset:312 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v10, off, s[0:3], s32 offset:316 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:320 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: buffer_load_dword v17, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v18, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v19, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v20, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v13, off, s[0:3], s32 offset:324 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 64
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v22, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v21, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, s4
+; GFX9-G-O0-NEXT: v_sub_u32_e64 v4, v13, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s4
+; GFX9-G-O0-NEXT: v_sub_u32_e64 v5, v5, v13
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s4
+; GFX9-G-O0-NEXT: v_cmp_lt_u32_e64 s[4:5], v13, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s6
+; GFX9-G-O0-NEXT: v_cmp_eq_u32_e64 s[6:7], v13, v6
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[6:7], v13, v[21:22]
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[26:27], v13, v[15:16]
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[24:25], v5, v[21:22]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v26
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v27
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v23, v24
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v25
+; GFX9-G-O0-NEXT: v_or_b32_e64 v14, v14, v23
+; GFX9-G-O0-NEXT: v_or_b32_e64 v13, v5, v13
+; GFX9-G-O0-NEXT: s_mov_b64 s[8:9], 0
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[21:22], v4, v[21:22]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v21
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v22
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v4, v4, v14, s[4:5]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v5, v5, v13, s[4:5]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v16
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v4, v4, v14, s[6:7]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v13, v5, v13, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 def $vgpr4_vgpr5 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v6
+; GFX9-G-O0-NEXT: ; kill: def $vgpr7 killed $vgpr7 killed $vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v13, v13, v14, s[4:5]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[4:5]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr13 killed $vgpr13 def $vgpr13_vgpr14 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v6
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4_vgpr5 killed $vgpr4_vgpr5 def $vgpr4_vgpr5_vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v18
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v19
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v20
+; GFX9-G-O0-NEXT: s_mov_b32 s4, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s10, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s7, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s6, -1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v17, s4
+; GFX9-G-O0-NEXT: v_add_co_u32_e64 v16, s[4:5], v16, v17
+; GFX9-G-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 offset:288 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, s10
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v15, s[4:5], v15, v16, s[4:5]
+; GFX9-G-O0-NEXT: buffer_store_dword v15, off, s[0:3], s32 offset:284 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, s7
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v14, s[4:5], v14, v15, s[4:5]
+; GFX9-G-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:280 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, s6
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v13, s[4:5], v13, v14, s[4:5]
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:276 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], s[8:9]
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], s[8:9]
+; GFX9-G-O0-NEXT: v_writelane_b32 v12, s8, 6
+; GFX9-G-O0-NEXT: v_writelane_b32 v12, s9, 7
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, s7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, s6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, s4
+; GFX9-G-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:260 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:264 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:268 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v15, off, s[0:3], s32 offset:272 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:244 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:248 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v10, off, s[0:3], s32 offset:252 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:256 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:228 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:232 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:236 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:240 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:212 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:216 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:220 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:224 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB0_6
+; GFX9-G-O0-NEXT: .LBB0_8: ; %udiv-bb1
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: buffer_load_dword v9, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v10, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], 0
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 1
+; GFX9-G-O0-NEXT: s_mov_b32 s10, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s6
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(3)
+; GFX9-G-O0-NEXT: v_add_co_u32_e64 v5, s[6:7], v2, v5
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:324 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s10
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(1)
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v6, s[6:7], v4, v6, s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, s9
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v8, s[6:7], v3, v4, s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s8
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v7, s[6:7], v1, v3, s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, v7
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:308 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:312 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v15, off, s[0:3], s32 offset:316 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 offset:320 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 0x7f
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s6
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v4, s[6:7], v1, v2
+; GFX9-G-O0-NEXT: s_mov_b32 s7, 64
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-G-O0-NEXT: v_sub_u32_e64 v3, v4, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-G-O0-NEXT: v_sub_u32_e64 v9, v1, v4
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-G-O0-NEXT: v_cmp_lt_u32_e64 s[8:9], v4, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s6
+; GFX9-G-O0-NEXT: v_cmp_eq_u32_e64 s[6:7], v4, v1
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[1:2], v4, v[13:14]
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[18:19], v9, v[13:14]
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[16:17], v4, v[11:12]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v18
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v19
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v16
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v17
+; GFX9-G-O0-NEXT: v_or_b32_e64 v10, v10, v15
+; GFX9-G-O0-NEXT: v_or_b32_e64 v4, v4, v9
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[13:14], v3, v[13:14]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v1, v1, v9, s[8:9]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v2, v3, s[8:9]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v14
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v9, v9, v10, s[8:9]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v3, v4, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v12
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v9, v9, v10, s[6:7]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v3, v4, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v3
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1_vgpr2 killed $vgpr1_vgpr2 def $vgpr1_vgpr2_vgpr3_vgpr4 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v10
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:292 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:296 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:300 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:304 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[8:9], s[4:5]
+; GFX9-G-O0-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v5, v8
+; GFX9-G-O0-NEXT: v_or_b32_e64 v7, v6, v7
+; GFX9-G-O0-NEXT: ; kill: def $vgpr5 killed $vgpr5 def $vgpr5_vgpr6 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, s4
+; GFX9-G-O0-NEXT: v_cmp_ne_u64_e64 s[4:5], v[5:6], v[7:8]
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:116 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:120 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:124 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:128 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, s11
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:104 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:108 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:112 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], exec
+; GFX9-G-O0-NEXT: s_and_b64 s[4:5], s[6:7], s[4:5]
+; GFX9-G-O0-NEXT: s_xor_b64 s[6:7], s[4:5], s[6:7]
+; GFX9-G-O0-NEXT: v_writelane_b32 v0, s6, 4
+; GFX9-G-O0-NEXT: v_writelane_b32 v0, s7, 5
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[4:5]
+; GFX9-G-O0-NEXT: s_cbranch_execz .LBB0_5
+; GFX9-G-O0-NEXT: s_branch .LBB0_7
+; GFX9-G-O0-NEXT: .LBB0_9: ; %udiv-end
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[20:21]
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v9, off, s[0:3], s32 offset:164 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v10, off, s[0:3], s32 offset:168 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:172 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:176 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v2
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v0, v0, v8
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v1, v1, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v10
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v2, v2, v6
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v3, v3, v5
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v0, s[4:5], v0, v8
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v1, s[4:5], v1, v7, s[4:5]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v2, s[4:5], v2, v6, s[4:5]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v3, s[4:5], v3, v5, s[4:5]
+; GFX9-G-O0-NEXT: ; kill: killed $vgpr4
+; GFX9-G-O0-NEXT: s_xor_saveexec_b64 s[4:5], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:328 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_nop 0
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:332 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:336 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v16, off, s[0:3], s32 offset:340 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:344 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[4:5]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: s_setpc_b64 s[30:31]
%div = sdiv i128 %lhs, %rhs
ret i128 %div
}
@@ -2306,6 +3457,1043 @@ define i128 @v_udiv_i128_vv(i128 %lhs, i128 %rhs) {
; GFX9-O0-NEXT: s_mov_b64 exec, s[4:5]
; GFX9-O0-NEXT: s_waitcnt vmcnt(0)
; GFX9-O0-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-G-LABEL: v_udiv_i128_vv:
+; GFX9-G: ; %bb.0: ; %_udiv-special-cases
+; GFX9-G-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-G-NEXT: v_or_b32_e32 v8, v4, v6
+; GFX9-G-NEXT: v_or_b32_e32 v9, v5, v7
+; GFX9-G-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[8:9]
+; GFX9-G-NEXT: v_or_b32_e32 v8, v0, v2
+; GFX9-G-NEXT: v_or_b32_e32 v9, v1, v3
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[8:9]
+; GFX9-G-NEXT: v_ffbh_u32_e32 v9, v4
+; GFX9-G-NEXT: v_ffbh_u32_e32 v8, v5
+; GFX9-G-NEXT: v_add_u32_e32 v9, 32, v9
+; GFX9-G-NEXT: v_ffbh_u32_e32 v10, v6
+; GFX9-G-NEXT: v_min_u32_e32 v8, v8, v9
+; GFX9-G-NEXT: v_ffbh_u32_e32 v9, v7
+; GFX9-G-NEXT: v_add_u32_e32 v10, 32, v10
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[6:7], 0, v[6:7]
+; GFX9-G-NEXT: v_add_u32_e32 v8, 64, v8
+; GFX9-G-NEXT: v_min_u32_e32 v9, v9, v10
+; GFX9-G-NEXT: v_ffbh_u32_e32 v10, v0
+; GFX9-G-NEXT: v_cndmask_b32_e64 v8, v9, v8, s[6:7]
+; GFX9-G-NEXT: v_ffbh_u32_e32 v9, v1
+; GFX9-G-NEXT: v_add_u32_e32 v10, 32, v10
+; GFX9-G-NEXT: v_ffbh_u32_e32 v11, v2
+; GFX9-G-NEXT: v_min_u32_e32 v9, v9, v10
+; GFX9-G-NEXT: v_ffbh_u32_e32 v10, v3
+; GFX9-G-NEXT: v_add_u32_e32 v11, 32, v11
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[6:7], 0, v[2:3]
+; GFX9-G-NEXT: v_add_u32_e32 v9, 64, v9
+; GFX9-G-NEXT: v_min_u32_e32 v10, v10, v11
+; GFX9-G-NEXT: v_cndmask_b32_e64 v9, v10, v9, s[6:7]
+; GFX9-G-NEXT: v_sub_co_u32_e64 v12, s[6:7], v8, v9
+; GFX9-G-NEXT: v_subb_co_u32_e64 v13, s[6:7], 0, 0, s[6:7]
+; GFX9-G-NEXT: v_mov_b32_e32 v8, 0x7f
+; GFX9-G-NEXT: v_subb_co_u32_e64 v14, s[6:7], 0, 0, s[6:7]
+; GFX9-G-NEXT: v_mov_b32_e32 v9, 0
+; GFX9-G-NEXT: v_subb_co_u32_e64 v15, s[6:7], 0, 0, s[6:7]
+; GFX9-G-NEXT: v_cmp_gt_u64_e64 s[6:7], v[12:13], v[8:9]
+; GFX9-G-NEXT: s_or_b64 s[4:5], vcc, s[4:5]
+; GFX9-G-NEXT: v_cndmask_b32_e64 v8, 0, 1, s[6:7]
+; GFX9-G-NEXT: v_cmp_lt_u64_e64 s[6:7], 0, v[14:15]
+; GFX9-G-NEXT: v_or_b32_e32 v17, v13, v15
+; GFX9-G-NEXT: v_cndmask_b32_e64 v9, 0, 1, s[6:7]
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[6:7], 0, v[14:15]
+; GFX9-G-NEXT: s_mov_b64 s[8:9], 0
+; GFX9-G-NEXT: v_cndmask_b32_e64 v8, v9, v8, s[6:7]
+; GFX9-G-NEXT: v_cndmask_b32_e64 v9, 0, 1, s[4:5]
+; GFX9-G-NEXT: v_or_b32_e32 v18, v9, v8
+; GFX9-G-NEXT: v_xor_b32_e32 v8, 0x7f, v12
+; GFX9-G-NEXT: v_or_b32_e32 v16, v8, v14
+; GFX9-G-NEXT: v_and_b32_e32 v8, 1, v18
+; GFX9-G-NEXT: v_cmp_ne_u32_e32 vcc, 0, v8
+; GFX9-G-NEXT: v_cndmask_b32_e64 v10, v0, 0, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e64 v11, v1, 0, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e64 v8, v2, 0, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e64 v9, v3, 0, vcc
+; GFX9-G-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[16:17]
+; GFX9-G-NEXT: v_cndmask_b32_e64 v16, 0, 1, vcc
+; GFX9-G-NEXT: v_or_b32_e32 v16, v18, v16
+; GFX9-G-NEXT: v_and_b32_e32 v16, 1, v16
+; GFX9-G-NEXT: v_cmp_ne_u32_e32 vcc, 0, v16
+; GFX9-G-NEXT: s_xor_b64 s[4:5], vcc, -1
+; GFX9-G-NEXT: s_and_saveexec_b64 s[6:7], s[4:5]
+; GFX9-G-NEXT: s_cbranch_execz .LBB1_6
+; GFX9-G-NEXT: ; %bb.1: ; %udiv-bb1
+; GFX9-G-NEXT: v_add_co_u32_e32 v18, vcc, 1, v12
+; GFX9-G-NEXT: v_addc_co_u32_e32 v19, vcc, 0, v13, vcc
+; GFX9-G-NEXT: v_addc_co_u32_e32 v20, vcc, 0, v14, vcc
+; GFX9-G-NEXT: v_addc_co_u32_e32 v21, vcc, 0, v15, vcc
+; GFX9-G-NEXT: s_xor_b64 s[4:5], vcc, -1
+; GFX9-G-NEXT: v_sub_co_u32_e32 v16, vcc, 0x7f, v12
+; GFX9-G-NEXT: v_sub_u32_e32 v8, 64, v16
+; GFX9-G-NEXT: v_lshrrev_b64 v[8:9], v8, v[0:1]
+; GFX9-G-NEXT: v_lshlrev_b64 v[10:11], v16, v[2:3]
+; GFX9-G-NEXT: v_subrev_u32_e32 v14, 64, v16
+; GFX9-G-NEXT: v_lshlrev_b64 v[12:13], v16, v[0:1]
+; GFX9-G-NEXT: v_or_b32_e32 v10, v8, v10
+; GFX9-G-NEXT: v_or_b32_e32 v11, v9, v11
+; GFX9-G-NEXT: v_lshlrev_b64 v[8:9], v14, v[0:1]
+; GFX9-G-NEXT: v_cmp_gt_u32_e32 vcc, 64, v16
+; GFX9-G-NEXT: s_mov_b64 s[10:11], s[8:9]
+; GFX9-G-NEXT: v_cndmask_b32_e32 v14, 0, v12, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v15, 0, v13, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v8, v8, v10, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v9, v9, v11, vcc
+; GFX9-G-NEXT: v_cmp_eq_u32_e32 vcc, 0, v16
+; GFX9-G-NEXT: v_mov_b32_e32 v13, s11
+; GFX9-G-NEXT: v_cndmask_b32_e32 v8, v8, v2, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v9, v9, v3, vcc
+; GFX9-G-NEXT: v_mov_b32_e32 v11, s9
+; GFX9-G-NEXT: v_mov_b32_e32 v10, s8
+; GFX9-G-NEXT: v_mov_b32_e32 v12, s10
+; GFX9-G-NEXT: s_and_saveexec_b64 s[8:9], s[4:5]
+; GFX9-G-NEXT: s_xor_b64 s[12:13], exec, s[8:9]
+; GFX9-G-NEXT: s_cbranch_execz .LBB1_5
+; GFX9-G-NEXT: ; %bb.2: ; %udiv-preheader
+; GFX9-G-NEXT: v_sub_u32_e32 v12, 64, v18
+; GFX9-G-NEXT: v_subrev_u32_e32 v22, 64, v18
+; GFX9-G-NEXT: v_lshrrev_b64 v[10:11], v18, v[0:1]
+; GFX9-G-NEXT: v_lshlrev_b64 v[12:13], v12, v[2:3]
+; GFX9-G-NEXT: v_lshrrev_b64 v[16:17], v18, v[2:3]
+; GFX9-G-NEXT: v_lshrrev_b64 v[2:3], v22, v[2:3]
+; GFX9-G-NEXT: v_or_b32_e32 v10, v10, v12
+; GFX9-G-NEXT: v_or_b32_e32 v11, v11, v13
+; GFX9-G-NEXT: v_cmp_gt_u32_e32 vcc, 64, v18
+; GFX9-G-NEXT: s_mov_b64 s[8:9], 0
+; GFX9-G-NEXT: v_cndmask_b32_e32 v2, v2, v10, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v3, v3, v11, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v16, 0, v16, vcc
+; GFX9-G-NEXT: v_cndmask_b32_e32 v17, 0, v17, vcc
+; GFX9-G-NEXT: v_add_co_u32_e32 v22, vcc, -1, v4
+; GFX9-G-NEXT: v_addc_co_u32_e32 v23, vcc, -1, v5, vcc
+; GFX9-G-NEXT: s_mov_b64 s[10:11], s[8:9]
+; GFX9-G-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v18
+; GFX9-G-NEXT: v_addc_co_u32_e32 v24, vcc, -1, v6, vcc
+; GFX9-G-NEXT: v_mov_b32_e32 v13, s11
+; GFX9-G-NEXT: v_cndmask_b32_e64 v2, v2, v0, s[4:5]
+; GFX9-G-NEXT: v_cndmask_b32_e64 v3, v3, v1, s[4:5]
+; GFX9-G-NEXT: v_addc_co_u32_e32 v25, vcc, -1, v7, vcc
+; GFX9-G-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-G-NEXT: v_mov_b32_e32 v11, s9
+; GFX9-G-NEXT: v_mov_b32_e32 v10, s8
+; GFX9-G-NEXT: v_mov_b32_e32 v12, s10
+; GFX9-G-NEXT: .LBB1_3: ; %udiv-do-while
+; GFX9-G-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-G-NEXT: v_lshlrev_b64 v[12:13], 1, v[14:15]
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v0, 31, v15
+; GFX9-G-NEXT: v_or_b32_e32 v14, v10, v12
+; GFX9-G-NEXT: v_or_b32_e32 v15, v11, v13
+; GFX9-G-NEXT: v_lshlrev_b64 v[12:13], 1, v[16:17]
+; GFX9-G-NEXT: v_lshlrev_b64 v[10:11], 1, v[2:3]
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v2, 31, v3
+; GFX9-G-NEXT: v_or_b32_e32 v12, v12, v2
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v2, 31, v9
+; GFX9-G-NEXT: v_lshlrev_b64 v[8:9], 1, v[8:9]
+; GFX9-G-NEXT: v_or_b32_e32 v2, v10, v2
+; GFX9-G-NEXT: v_or_b32_e32 v8, v8, v0
+; GFX9-G-NEXT: v_sub_co_u32_e32 v0, vcc, v22, v2
+; GFX9-G-NEXT: v_subb_co_u32_e32 v0, vcc, v23, v11, vcc
+; GFX9-G-NEXT: v_subb_co_u32_e32 v0, vcc, v24, v12, vcc
+; GFX9-G-NEXT: v_subb_co_u32_e32 v0, vcc, v25, v13, vcc
+; GFX9-G-NEXT: v_add_co_u32_e64 v18, s[4:5], -1, v18
+; GFX9-G-NEXT: v_ashrrev_i32_e32 v3, 31, v0
+; GFX9-G-NEXT: v_addc_co_u32_e64 v19, s[4:5], -1, v19, s[4:5]
+; GFX9-G-NEXT: v_and_b32_e32 v10, v3, v4
+; GFX9-G-NEXT: v_addc_co_u32_e64 v20, s[4:5], -1, v20, s[4:5]
+; GFX9-G-NEXT: v_and_b32_e32 v16, v3, v5
+; GFX9-G-NEXT: v_sub_co_u32_e32 v2, vcc, v2, v10
+; GFX9-G-NEXT: v_addc_co_u32_e64 v21, s[4:5], -1, v21, s[4:5]
+; GFX9-G-NEXT: v_and_b32_e32 v0, 1, v3
+; GFX9-G-NEXT: v_and_b32_e32 v17, v3, v6
+; GFX9-G-NEXT: v_and_b32_e32 v26, v3, v7
+; GFX9-G-NEXT: v_subb_co_u32_e32 v3, vcc, v11, v16, vcc
+; GFX9-G-NEXT: v_or_b32_e32 v10, v18, v20
+; GFX9-G-NEXT: v_or_b32_e32 v11, v19, v21
+; GFX9-G-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[10:11]
+; GFX9-G-NEXT: v_subb_co_u32_e32 v16, vcc, v12, v17, vcc
+; GFX9-G-NEXT: v_mov_b32_e32 v11, v1
+; GFX9-G-NEXT: v_subb_co_u32_e32 v17, vcc, v13, v26, vcc
+; GFX9-G-NEXT: s_or_b64 s[8:9], s[4:5], s[8:9]
+; GFX9-G-NEXT: v_mov_b32_e32 v10, v0
+; GFX9-G-NEXT: s_andn2_b64 exec, exec, s[8:9]
+; GFX9-G-NEXT: s_cbranch_execnz .LBB1_3
+; GFX9-G-NEXT: ; %bb.4: ; %Flow
+; GFX9-G-NEXT: s_or_b64 exec, exec, s[8:9]
+; GFX9-G-NEXT: .LBB1_5: ; %Flow2
+; GFX9-G-NEXT: s_or_b64 exec, exec, s[12:13]
+; GFX9-G-NEXT: v_lshlrev_b64 v[0:1], 1, v[14:15]
+; GFX9-G-NEXT: v_lshlrev_b64 v[8:9], 1, v[8:9]
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v2, 31, v15
+; GFX9-G-NEXT: v_or_b32_e32 v8, v8, v2
+; GFX9-G-NEXT: v_or_b32_e32 v10, v10, v0
+; GFX9-G-NEXT: v_or_b32_e32 v11, v11, v1
+; GFX9-G-NEXT: .LBB1_6: ; %Flow3
+; GFX9-G-NEXT: s_or_b64 exec, exec, s[6:7]
+; GFX9-G-NEXT: v_mov_b32_e32 v0, v10
+; GFX9-G-NEXT: v_mov_b32_e32 v1, v11
+; GFX9-G-NEXT: v_mov_b32_e32 v2, v8
+; GFX9-G-NEXT: v_mov_b32_e32 v3, v9
+; GFX9-G-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-G-O0-LABEL: v_udiv_i128_vv:
+; GFX9-G-O0: ; %bb.0: ; %_udiv-special-cases
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-G-O0-NEXT: s_xor_saveexec_b64 s[4:5], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:300 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:304 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:308 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 offset:312 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:316 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[4:5]
+; GFX9-G-O0-NEXT: ; implicit-def: $vgpr8 : SGPR spill to VGPR lane
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v0
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr8 killed $vgpr8 def $vgpr8_vgpr9_vgpr10_vgpr11 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v3
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v10, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v4
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_nop 0
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v5
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v7
+; GFX9-G-O0-NEXT: ; kill: def $vgpr5 killed $vgpr5 def $vgpr5_vgpr6_vgpr7_vgpr8 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v9
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v14
+; GFX9-G-O0-NEXT: v_or_b32_e64 v9, v9, v12
+; GFX9-G-O0-NEXT: v_or_b32_e64 v11, v10, v11
+; GFX9-G-O0-NEXT: ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[6:7], v[9:10], v[11:12]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v14
+; GFX9-G-O0-NEXT: v_or_b32_e64 v9, v9, v12
+; GFX9-G-O0-NEXT: v_or_b32_e64 v11, v10, v11
+; GFX9-G-O0-NEXT: ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[8:9], v[9:10], v[11:12]
+; GFX9-G-O0-NEXT: s_or_b64 s[6:7], s[6:7], s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[8:9], v[8:9], v[5:6]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v11
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v5, v5
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v6, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, 32
+; GFX9-G-O0-NEXT: v_add_u32_e64 v6, v6, v7
+; GFX9-G-O0-NEXT: v_min_u32_e64 v5, v5, v6
+; GFX9-G-O0-NEXT: s_mov_b32 s10, 64
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s10
+; GFX9-G-O0-NEXT: v_add_u32_e64 v6, v5, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v9
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v5, v5
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v7, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, 32
+; GFX9-G-O0-NEXT: v_add_u32_e64 v7, v7, v8
+; GFX9-G-O0-NEXT: v_min_u32_e64 v5, v5, v7
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v5, v5, v6, s[8:9]
+; GFX9-G-O0-NEXT: s_mov_b32 s14, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[8:9], v[9:10], v[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v12
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v6, v6
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v7, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, 32
+; GFX9-G-O0-NEXT: v_add_u32_e64 v7, v7, v8
+; GFX9-G-O0-NEXT: v_min_u32_e64 v6, v6, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, s10
+; GFX9-G-O0-NEXT: v_add_u32_e64 v7, v6, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v10
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v6, v6
+; GFX9-G-O0-NEXT: v_ffbh_u32_e64 v8, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, 32
+; GFX9-G-O0-NEXT: v_add_u32_e64 v8, v8, v9
+; GFX9-G-O0-NEXT: v_min_u32_e64 v6, v6, v8
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[8:9]
+; GFX9-G-O0-NEXT: s_mov_b32 s13, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s11, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s12, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s10, 0
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v6, s[8:9], v5, v6
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, s14
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v7, s[8:9], v5, v7, s[8:9]
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, s12
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v9, s[8:9], v5, v8, s[8:9]
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, s10
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v8, s[8:9], v5, v8, s[8:9]
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[12:13], 0x7f
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, s4
+; GFX9-G-O0-NEXT: v_cmp_gt_u64_e64 s[10:11], v[12:13], v[14:15]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[8:9], v[12:13], v[14:15]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, s12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, s13
+; GFX9-G-O0-NEXT: v_cmp_gt_u64_e64 s[12:13], v[10:11], v[12:13]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v10, v5, v10, s[12:13]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v5, v5, v11, s[10:11]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v10, v5, v10, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v5, v5, v11, s[6:7]
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v5, v10
+; GFX9-G-O0-NEXT: s_mov_b32 s7, 0x7f
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 0
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v6, v6, s7
+; GFX9-G-O0-NEXT: v_xor_b32_e64 v7, v7, s6
+; GFX9-G-O0-NEXT: v_or_b32_e64 v6, v6, v9
+; GFX9-G-O0-NEXT: v_or_b32_e64 v8, v7, v8
+; GFX9-G-O0-NEXT: ; kill: def $vgpr6 killed $vgpr6 def $vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[4:5], v[6:7], v[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v3
+; GFX9-G-O0-NEXT: v_and_b32_e32 v1, 1, v5
+; GFX9-G-O0-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v7
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v1, v1, v4, s[6:7]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v2, v3, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v3
+; GFX9-G-O0-NEXT: v_and_b32_e32 v3, 1, v5
+; GFX9-G-O0-NEXT: v_cmp_ne_u32_e64 s[6:7], 0, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v9
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[6:7]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v3, v4, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr6 killed $vgpr6 def $vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v3
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1_vgpr2 killed $vgpr1_vgpr2 def $vgpr1_vgpr2_vgpr3_vgpr4 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[4:5]
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v5, v6
+; GFX9-G-O0-NEXT: v_and_b32_e32 v5, 1, v5
+; GFX9-G-O0-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v5
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], -1
+; GFX9-G-O0-NEXT: s_xor_b64 s[6:7], s[4:5], s[6:7]
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], exec
+; GFX9-G-O0-NEXT: v_writelane_b32 v0, s4, 0
+; GFX9-G-O0-NEXT: v_writelane_b32 v0, s5, 1
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7]
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[4:5]
+; GFX9-G-O0-NEXT: s_cbranch_execz .LBB1_3
+; GFX9-G-O0-NEXT: s_branch .LBB1_8
+; GFX9-G-O0-NEXT: .LBB1_1: ; %Flow
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_readlane_b32 s4, v0, 2
+; GFX9-G-O0-NEXT: v_readlane_b32 s5, v0, 3
+; GFX9-G-O0-NEXT: s_or_b64 exec, exec, s[4:5]
+; GFX9-G-O0-NEXT: ; %bb.2: ; %Flow
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:104 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:108 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:112 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:116 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:120 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:124 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:128 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:132 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(4)
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:96 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB1_5
+; GFX9-G-O0-NEXT: .LBB1_3: ; %Flow2
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_readlane_b32 s4, v4, 0
+; GFX9-G-O0-NEXT: v_readlane_b32 s5, v4, 1
+; GFX9-G-O0-NEXT: s_or_b64 exec, exec, s[4:5]
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:136 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:140 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:144 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:148 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB1_9
+; GFX9-G-O0-NEXT: .LBB1_4: ; %udiv-loop-exit
+; GFX9-G-O0-NEXT: buffer_load_dword v14, off, s[0:3], s32 offset:152 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v15, off, s[0:3], s32 offset:156 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v16, off, s[0:3], s32 offset:160 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v17, off, s[0:3], s32 offset:164 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:168 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:172 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:176 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:180 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v7
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[10:11], v0, v[2:3]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[0:1], v0, v[4:5]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr2 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr3 killed $vgpr3 killed $vgpr2_vgpr3 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v6, v2, v3
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v16
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v11
+; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v7
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v1, v5
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v9
+; GFX9-G-O0-NEXT: v_or3_b32 v4, v4, v6, v7
+; GFX9-G-O0-NEXT: v_or3_b32 v2, v2, v3, v5
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 def $vgpr4_vgpr5 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v2
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0_vgpr1 killed $vgpr0_vgpr1 def $vgpr0_vgpr1_vgpr2_vgpr3 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB1_3
+; GFX9-G-O0-NEXT: .LBB1_5: ; %Flow1
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_readlane_b32 s4, v8, 4
+; GFX9-G-O0-NEXT: v_readlane_b32 s5, v8, 5
+; GFX9-G-O0-NEXT: s_or_b64 exec, exec, s[4:5]
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:88 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:92 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:96 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:100 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:152 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:156 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:160 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:164 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:168 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:172 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:176 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:180 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB1_4
+; GFX9-G-O0-NEXT: .LBB1_6: ; %udiv-do-while
+; GFX9-G-O0-NEXT: ; =>This Inner Loop Header: Depth=1
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v16, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_readlane_b32 s6, v16, 6
+; GFX9-G-O0-NEXT: v_readlane_b32 s7, v16, 7
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:184 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v13, off, s[0:3], s32 offset:188 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v14, off, s[0:3], s32 offset:192 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v15, off, s[0:3], s32 offset:196 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:200 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:204 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:208 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:212 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v17, off, s[0:3], s32 offset:216 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v18, off, s[0:3], s32 offset:220 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v19, off, s[0:3], s32 offset:224 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v20, off, s[0:3], s32 offset:228 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v31, off, s[0:3], s32 offset:232 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v32, off, s[0:3], s32 offset:236 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v33, off, s[0:3], s32 offset:240 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v34, off, s[0:3], s32 offset:244 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v25, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v26, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v27, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v28, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:248 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:252 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v10, off, s[0:3], s32 offset:256 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:260 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], 0
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(16)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v5
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s8
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[21:22], v2, v[0:1]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s8
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[4:5], v2, v[3:4]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr2 killed $vgpr0 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 killed $vgpr0_vgpr1 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s9
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v3, v0, v1
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v5
+; GFX9-G-O0-NEXT: v_or_b32_e64 v7, v2, v3
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v0, v1
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0_vgpr1 killed $vgpr12_vgpr13 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v15
+; GFX9-G-O0-NEXT: ; kill: def $vgpr2 killed $vgpr0 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 killed $vgpr0_vgpr1 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s9
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v3, v0, v1
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v21
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v22
+; GFX9-G-O0-NEXT: v_or_b32_e64 v4, v2, v3
+; GFX9-G-O0-NEXT: v_or_b32_e64 v9, v0, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[23:24], v0, v[2:3]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[0:1], v0, v[12:13]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr12 killed $vgpr2 killed $exec
+; GFX9-G-O0-NEXT: ; kill: def $vgpr3 killed $vgpr3 killed $vgpr2_vgpr3 killed $exec
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s8
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v14, v2, v3
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, v0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v1
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(8)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v29, v31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v30, v32
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v21, v33
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v22, v34
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v29
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v30
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v23
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v24
+; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v15
+; GFX9-G-O0-NEXT: v_or_b32_e64 v13, v1, v13
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0 killed $vgpr0 def $vgpr0_vgpr1 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v21
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v22
+; GFX9-G-O0-NEXT: v_or3_b32 v12, v12, v14, v15
+; GFX9-G-O0-NEXT: v_or3_b32 v2, v2, v3, v13
+; GFX9-G-O0-NEXT: ; kill: def $vgpr12 killed $vgpr12 def $vgpr12_vgpr13 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v2
+; GFX9-G-O0-NEXT: ; kill: def $vgpr0_vgpr1 killed $vgpr0_vgpr1 def $vgpr0_vgpr1_vgpr2_vgpr3 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v12
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v13
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v11, s[8:9], v11, v4
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v10, s[8:9], v10, v9, s[8:9]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v8, s[8:9], v8, v7, s[8:9]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v10, s[8:9], v6, v5, s[8:9]
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s8
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v8, v6, v10
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s8
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v6, v6, v10
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 1
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 0
+; GFX9-G-O0-NEXT: v_and_b32_e64 v12, v8, s9
+; GFX9-G-O0-NEXT: v_and_b32_e64 v10, v8, s8
+; GFX9-G-O0-NEXT: ; kill: def $vgpr12 killed $vgpr12 def $vgpr12_vgpr13 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, s4
+; GFX9-G-O0-NEXT: ; kill: def $vgpr12_vgpr13 killed $vgpr12_vgpr13 def $vgpr12_vgpr13_vgpr14_vgpr15 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v23, v25
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v24, v26
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v21, v27
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v22, v28
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v23
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v24
+; GFX9-G-O0-NEXT: v_and_b32_e64 v11, v8, v11
+; GFX9-G-O0-NEXT: v_and_b32_e64 v10, v8, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v21
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v21, v22
+; GFX9-G-O0-NEXT: v_and_b32_e64 v8, v6, v8
+; GFX9-G-O0-NEXT: v_and_b32_e64 v6, v6, v21
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v4, s[8:9], v4, v11
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v10, s[8:9], v9, v10, s[8:9]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v9, s[8:9], v7, v8, s[8:9]
+; GFX9-G-O0-NEXT: v_subb_co_u32_e64 v8, s[8:9], v5, v6, s[8:9]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 def $vgpr4_vgpr5_vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v18
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v19
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v20
+; GFX9-G-O0-NEXT: s_mov_b32 s8, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s12, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s11, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s10, -1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v17, s8
+; GFX9-G-O0-NEXT: v_add_co_u32_e64 v17, s[8:9], v11, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, s12
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v18, s[8:9], v10, v11, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, s11
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v20, s[8:9], v9, v10, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, s10
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v19, s[8:9], v8, v9, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v18
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v20
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v11, v19
+; GFX9-G-O0-NEXT: v_or_b32_e64 v17, v17, v20
+; GFX9-G-O0-NEXT: v_or_b32_e64 v19, v18, v19
+; GFX9-G-O0-NEXT: ; kill: def $vgpr17 killed $vgpr17 def $vgpr17_vgpr18 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v18, v19
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v20, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v19, s4
+; GFX9-G-O0-NEXT: v_cmp_eq_u64_e64 s[4:5], v[17:18], v[19:20]
+; GFX9-G-O0-NEXT: s_or_b64 s[4:5], s[4:5], s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v20, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v19, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v18, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v17, v0
+; GFX9-G-O0-NEXT: buffer_store_dword v17, off, s[0:3], s32 offset:104 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v18, off, s[0:3], s32 offset:108 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v19, off, s[0:3], s32 offset:112 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v20, off, s[0:3], s32 offset:116 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v20, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v19, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v18, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v17, v12
+; GFX9-G-O0-NEXT: buffer_store_dword v17, off, s[0:3], s32 offset:120 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v18, off, s[0:3], s32 offset:124 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v19, off, s[0:3], s32 offset:128 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v20, off, s[0:3], s32 offset:132 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], s[4:5]
+; GFX9-G-O0-NEXT: v_writelane_b32 v16, s6, 2
+; GFX9-G-O0-NEXT: v_writelane_b32 v16, s7, 3
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], s[4:5]
+; GFX9-G-O0-NEXT: v_writelane_b32 v16, s6, 6
+; GFX9-G-O0-NEXT: v_writelane_b32 v16, s7, 7
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:232 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:236 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:240 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v15, off, s[0:3], s32 offset:244 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:216 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:220 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v10, off, s[0:3], s32 offset:224 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:228 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:200 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:204 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:208 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:212 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:184 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:188 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:192 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:196 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_andn2_b64 exec, exec, s[4:5]
+; GFX9-G-O0-NEXT: s_cbranch_execnz .LBB1_6
+; GFX9-G-O0-NEXT: s_branch .LBB1_1
+; GFX9-G-O0-NEXT: .LBB1_7: ; %udiv-preheader
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:264 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:268 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:272 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:276 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:280 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v9, off, s[0:3], s32 offset:284 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v10, off, s[0:3], s32 offset:288 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:292 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: buffer_load_dword v17, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v18, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v19, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v20, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v13, off, s[0:3], s32 offset:296 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 64
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v22, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v21, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, s4
+; GFX9-G-O0-NEXT: v_sub_u32_e64 v4, v13, v4
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s4
+; GFX9-G-O0-NEXT: v_sub_u32_e64 v5, v5, v13
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s4
+; GFX9-G-O0-NEXT: v_cmp_lt_u32_e64 s[4:5], v13, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s6
+; GFX9-G-O0-NEXT: v_cmp_eq_u32_e64 s[6:7], v13, v6
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[6:7], v13, v[21:22]
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[26:27], v13, v[15:16]
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[24:25], v5, v[21:22]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v26
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v27
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v23, v24
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v25
+; GFX9-G-O0-NEXT: v_or_b32_e64 v14, v14, v23
+; GFX9-G-O0-NEXT: v_or_b32_e64 v13, v5, v13
+; GFX9-G-O0-NEXT: s_mov_b64 s[8:9], 0
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[21:22], v4, v[21:22]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v21
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v22
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v4, v4, v14, s[4:5]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v5, v5, v13, s[4:5]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v15
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v16
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v4, v4, v14, s[6:7]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v13, v5, v13, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 def $vgpr4_vgpr5 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v6
+; GFX9-G-O0-NEXT: ; kill: def $vgpr7 killed $vgpr7 killed $vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v13, v13, v14, s[4:5]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v6, v6, v7, s[4:5]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr13 killed $vgpr13 def $vgpr13_vgpr14 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v6
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4_vgpr5 killed $vgpr4_vgpr5 def $vgpr4_vgpr5_vgpr6_vgpr7 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, v14
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, v17
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v18
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v19
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v20
+; GFX9-G-O0-NEXT: s_mov_b32 s4, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s10, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s7, -1
+; GFX9-G-O0-NEXT: s_mov_b32 s6, -1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v17, s4
+; GFX9-G-O0-NEXT: v_add_co_u32_e64 v16, s[4:5], v16, v17
+; GFX9-G-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 offset:260 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, s10
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v15, s[4:5], v15, v16, s[4:5]
+; GFX9-G-O0-NEXT: buffer_store_dword v15, off, s[0:3], s32 offset:256 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, s7
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v14, s[4:5], v14, v15, s[4:5]
+; GFX9-G-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:252 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, s6
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v13, s[4:5], v13, v14, s[4:5]
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:248 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], s[8:9]
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], s[8:9]
+; GFX9-G-O0-NEXT: v_writelane_b32 v12, s8, 6
+; GFX9-G-O0-NEXT: v_writelane_b32 v12, s9, 7
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, s7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, s6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v12, s4
+; GFX9-G-O0-NEXT: buffer_store_dword v12, off, s[0:3], s32 offset:232 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:236 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:240 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v15, off, s[0:3], s32 offset:244 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v8, off, s[0:3], s32 offset:216 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v9, off, s[0:3], s32 offset:220 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v10, off, s[0:3], s32 offset:224 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v11, off, s[0:3], s32 offset:228 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:200 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:204 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v6, off, s[0:3], s32 offset:208 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v7, off, s[0:3], s32 offset:212 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 offset:184 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:188 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:192 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:196 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_branch .LBB1_6
+; GFX9-G-O0-NEXT: .LBB1_8: ; %udiv-bb1
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: buffer_load_dword v9, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v10, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v11, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v2, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v1, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v3, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 s[4:5], 0
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 1
+; GFX9-G-O0-NEXT: s_mov_b32 s10, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s9, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, s6
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(3)
+; GFX9-G-O0-NEXT: v_add_co_u32_e64 v5, s[6:7], v2, v5
+; GFX9-G-O0-NEXT: buffer_store_dword v5, off, s[0:3], s32 offset:296 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, s10
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(1)
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v6, s[6:7], v4, v6, s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, s9
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v8, s[6:7], v3, v4, s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s8
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v7, s[6:7], v1, v3, s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v16, v7
+; GFX9-G-O0-NEXT: buffer_store_dword v13, off, s[0:3], s32 offset:280 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v14, off, s[0:3], s32 offset:284 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v15, off, s[0:3], s32 offset:288 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v16, off, s[0:3], s32 offset:292 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 0x7f
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s6
+; GFX9-G-O0-NEXT: v_sub_co_u32_e64 v4, s[6:7], v1, v2
+; GFX9-G-O0-NEXT: s_mov_b32 s7, 64
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v14, v10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v13, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-G-O0-NEXT: v_sub_u32_e64 v3, v4, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-G-O0-NEXT: v_sub_u32_e64 v9, v1, v4
+; GFX9-G-O0-NEXT: s_mov_b32 s6, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s7
+; GFX9-G-O0-NEXT: v_cmp_lt_u32_e64 s[8:9], v4, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s6
+; GFX9-G-O0-NEXT: v_cmp_eq_u32_e64 s[6:7], v4, v1
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[1:2], v4, v[13:14]
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[18:19], v9, v[13:14]
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[16:17], v4, v[11:12]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v18
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v19
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v15, v16
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v17
+; GFX9-G-O0-NEXT: v_or_b32_e64 v10, v10, v15
+; GFX9-G-O0-NEXT: v_or_b32_e64 v4, v4, v9
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[13:14], v3, v[13:14]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v2
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, 0
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v1, v1, v9, s[8:9]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v2, v3, s[8:9]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v9, v13
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v14
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v9, v9, v10, s[8:9]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v3, v4, s[8:9]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v11
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v12
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v9, v9, v10, s[6:7]
+; GFX9-G-O0-NEXT: v_cndmask_b32_e64 v3, v3, v4, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v10, v3
+; GFX9-G-O0-NEXT: ; kill: def $vgpr1_vgpr2 killed $vgpr1_vgpr2 def $vgpr1_vgpr2_vgpr3_vgpr4 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v10
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:264 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:268 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:272 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:276 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[8:9], s[4:5]
+; GFX9-G-O0-NEXT: s_mov_b64 s[10:11], s[4:5]
+; GFX9-G-O0-NEXT: v_or_b32_e64 v5, v5, v8
+; GFX9-G-O0-NEXT: v_or_b32_e64 v7, v6, v7
+; GFX9-G-O0-NEXT: ; kill: def $vgpr5 killed $vgpr5 def $vgpr5_vgpr6 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v8, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v7, s4
+; GFX9-G-O0-NEXT: v_cmp_ne_u64_e64 s[4:5], v[5:6], v[7:8]
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:96 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:100 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s8
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s9
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s10
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, s11
+; GFX9-G-O0-NEXT: buffer_store_dword v1, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: buffer_store_dword v2, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v3, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: buffer_store_dword v4, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 s[6:7], exec
+; GFX9-G-O0-NEXT: s_and_b64 s[4:5], s[6:7], s[4:5]
+; GFX9-G-O0-NEXT: s_xor_b64 s[6:7], s[4:5], s[6:7]
+; GFX9-G-O0-NEXT: v_writelane_b32 v0, s6, 4
+; GFX9-G-O0-NEXT: v_writelane_b32 v0, s7, 5
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_store_dword v0, off, s[0:3], s32 ; 4-byte Folded Spill
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[4:5]
+; GFX9-G-O0-NEXT: s_cbranch_execz .LBB1_5
+; GFX9-G-O0-NEXT: s_branch .LBB1_7
+; GFX9-G-O0-NEXT: .LBB1_9: ; %udiv-end
+; GFX9-G-O0-NEXT: s_or_saveexec_b64 s[18:19], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[18:19]
+; GFX9-G-O0-NEXT: buffer_load_dword v5, off, s[0:3], s32 offset:136 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v6, off, s[0:3], s32 offset:140 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v7, off, s[0:3], s32 offset:144 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:148 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v7
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v8
+; GFX9-G-O0-NEXT: ; kill: killed $vgpr4
+; GFX9-G-O0-NEXT: s_xor_saveexec_b64 s[4:5], -1
+; GFX9-G-O0-NEXT: buffer_load_dword v0, off, s[0:3], s32 offset:300 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_nop 0
+; GFX9-G-O0-NEXT: buffer_load_dword v4, off, s[0:3], s32 offset:304 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v8, off, s[0:3], s32 offset:308 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v16, off, s[0:3], s32 offset:312 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: buffer_load_dword v12, off, s[0:3], s32 offset:316 ; 4-byte Folded Reload
+; GFX9-G-O0-NEXT: s_mov_b64 exec, s[4:5]
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0)
+; GFX9-G-O0-NEXT: s_setpc_b64 s[30:31]
%div = udiv i128 %lhs, %rhs
ret i128 %div
}
@@ -2388,6 +4576,66 @@ define i128 @v_sdiv_i128_v_pow2k(i128 %lhs) {
; GFX9-O0-NEXT: v_lshrrev_b64 v[3:4], s4, v[3:4]
; GFX9-O0-NEXT: ; kill: def $vgpr3 killed $vgpr3 killed $vgpr3_vgpr4 killed $exec
; GFX9-O0-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-G-LABEL: v_sdiv_i128_v_pow2k:
+; GFX9-G: ; %bb.0:
+; GFX9-G-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-G-NEXT: v_ashrrev_i32_e32 v4, 31, v3
+; GFX9-G-NEXT: v_mov_b32_e32 v5, v4
+; GFX9-G-NEXT: v_lshrrev_b64 v[4:5], 31, v[4:5]
+; GFX9-G-NEXT: v_add_co_u32_e32 v0, vcc, v0, v4
+; GFX9-G-NEXT: v_addc_co_u32_e32 v4, vcc, v1, v5, vcc
+; GFX9-G-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v2, vcc
+; GFX9-G-NEXT: v_addc_co_u32_e32 v2, vcc, 0, v3, vcc
+; GFX9-G-NEXT: v_lshlrev_b64 v[0:1], 31, v[1:2]
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v3, 1, v4
+; GFX9-G-NEXT: v_or_b32_e32 v0, v3, v0
+; GFX9-G-NEXT: v_ashrrev_i32_e32 v3, 31, v2
+; GFX9-G-NEXT: v_ashrrev_i32_e32 v2, 1, v2
+; GFX9-G-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-G-O0-LABEL: v_sdiv_i128_v_pow2k:
+; GFX9-G-O0: ; %bb.0:
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v0
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v0, v0, v3
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v0
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-G-O0-NEXT: v_lshrrev_b64 v[6:7], v0, v[5:6]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v6
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, v7
+; GFX9-G-O0-NEXT: s_mov_b32 s8, 0
+; GFX9-G-O0-NEXT: s_mov_b32 s5, 0
+; GFX9-G-O0-NEXT: v_add_co_u32_e64 v4, s[6:7], v4, v5
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v1, s[6:7], v1, v0, s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s8
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v5, s[6:7], v2, v0, s[6:7]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s5
+; GFX9-G-O0-NEXT: v_addc_co_u32_e64 v4, s[6:7], v3, v0, s[6:7]
+; GFX9-G-O0-NEXT: ; kill: def $vgpr5 killed $vgpr5 def $vgpr5_vgpr6 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v6, v4
+; GFX9-G-O0-NEXT: s_mov_b32 s5, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s5
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v0, v0, v1
+; GFX9-G-O0-NEXT: s_mov_b32 s5, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[5:6], v2, v[5:6]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v6
+; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v3
+; GFX9-G-O0-NEXT: v_or_b32_e64 v1, v1, v2
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v3, v2, v4
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-G-O0-NEXT: v_ashrrev_i32_e64 v2, v2, v4
+; GFX9-G-O0-NEXT: s_setpc_b64 s[30:31]
%div = sdiv i128 %lhs, 8589934592
ret i128 %div
}
@@ -2434,10 +4682,42 @@ define i128 @v_udiv_i128_v_pow2k(i128 %lhs) {
; GFX9-O0-NEXT: ; kill: def $vgpr2 killed $vgpr2 killed $vgpr2_vgpr3 killed $exec
; GFX9-O0-NEXT: v_mov_b32_e32 v3, 0
; GFX9-O0-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-G-LABEL: v_udiv_i128_v_pow2k:
+; GFX9-G: ; %bb.0:
+; GFX9-G-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-G-NEXT: v_mov_b32_e32 v4, v1
+; GFX9-G-NEXT: v_lshlrev_b64 v[0:1], 31, v[2:3]
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v2, 1, v4
+; GFX9-G-NEXT: v_or_b32_e32 v0, v2, v0
+; GFX9-G-NEXT: v_lshrrev_b32_e32 v2, 1, v3
+; GFX9-G-NEXT: v_mov_b32_e32 v3, 0
+; GFX9-G-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-G-O0-LABEL: v_udiv_i128_v_pow2k:
+; GFX9-G-O0: ; %bb.0:
+; GFX9-G-O0-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v2
+; GFX9-G-O0-NEXT: ; kill: def $vgpr4 killed $vgpr4 def $vgpr4_vgpr5 killed $exec
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v5, v3
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v0, s4
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v0, v0, v1
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v1, s4
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 31
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-G-O0-NEXT: v_lshlrev_b64 v[5:6], v2, v[4:5]
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v4, v5
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, v6
+; GFX9-G-O0-NEXT: v_or_b32_e64 v0, v0, v4
+; GFX9-G-O0-NEXT: v_or_b32_e64 v1, v1, v2
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 1
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v2, s4
+; GFX9-G-O0-NEXT: v_lshrrev_b32_e64 v2, v2, v3
+; GFX9-G-O0-NEXT: s_mov_b32 s4, 0
+; GFX9-G-O0-NEXT: v_mov_b32_e32 v3, s4
+; GFX9-G-O0-NEXT: s_setpc_b64 s[30:31]
%div = udiv i128 %lhs, 8589934592
ret i128 %div
}
-
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; GFX9-SDAG: {{.*}}
-; GFX9-SDAG-O0: {{.*}}
More information about the llvm-commits
mailing list