From 34d9a16e54becbea4c790b19804d35453018a53f Mon Sep 17 00:00:00 2001 From: Matt Arsenault Date: Sat, 18 Jan 2020 22:41:36 -0500 Subject: [PATCH] AMDGPU: Add option to expand 64-bit integer division in IR I didn't realize we were already expanding 24/32-bit division here already. Use the available IntegerDivision utilities. This uses loops, so produces significantly smaller code than the inline DAG expansion. This now requires width reductions of 64-bit divisions before introducing the expanded loops. This helps work around missing legalization in GlobalISel for division, which are the only remaining core instructions that didn't work at all. I think this is plausibly a better implementation than exists in the DAG, although turning it on by default misses out on the constant value optimizations and also needs benchmarking. --- llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp | 197 ++- .../CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll | 581 +++++---- llvm/test/CodeGen/AMDGPU/sdiv64.ll | 1300 ++++++++++++++++--- llvm/test/CodeGen/AMDGPU/srem64.ll | 1371 +++++++++++++++++--- llvm/test/CodeGen/AMDGPU/udiv64.ll | 1194 +++++++++++++---- llvm/test/CodeGen/AMDGPU/urem64.ll | 1177 +++++++++++++---- 6 files changed, 4712 insertions(+), 1108 deletions(-) diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp index 8840b0a1..a63da45 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp @@ -39,6 +39,7 @@ #include "llvm/IR/Operator.h" #include "llvm/IR/Type.h" #include "llvm/IR/Value.h" +#include "llvm/Transforms/Utils/IntegerDivision.h" #include "llvm/InitializePasses.h" #include "llvm/Pass.h" #include "llvm/Support/Casting.h" @@ -63,6 +64,12 @@ static cl::opt UseMul24Intrin( cl::ReallyHidden, cl::init(true)); +static cl::opt ExpandDiv64InIR( + "amdgpu-codegenprepare-expand-div64", + cl::desc("Expand 64-bit division in AMDGPUCodeGenPrepare"), + cl::ReallyHidden, + cl::init(false)); + class AMDGPUCodeGenPrepare : public FunctionPass, public InstVisitor { const GCNSubtarget *ST = nullptr; @@ -160,16 +167,27 @@ class AMDGPUCodeGenPrepare : public FunctionPass, bool divHasSpecialOptimization(BinaryOperator &I, Value *Num, Value *Den) const; + int getDivNumBits(BinaryOperator &I, + Value *Num, Value *Den, + unsigned AtLeast, bool Signed) const; /// Expands 24 bit div or rem. Value* expandDivRem24(IRBuilder<> &Builder, BinaryOperator &I, Value *Num, Value *Den, bool IsDiv, bool IsSigned) const; + Value *expandDivRem24Impl(IRBuilder<> &Builder, BinaryOperator &I, + Value *Num, Value *Den, unsigned NumBits, + bool IsDiv, bool IsSigned) const; + /// Expands 32 bit div or rem. Value* expandDivRem32(IRBuilder<> &Builder, BinaryOperator &I, Value *Num, Value *Den) const; + Value *shrinkDivRem64(IRBuilder<> &Builder, BinaryOperator &I, + Value *Num, Value *Den) const; + void expandDivRem64(BinaryOperator &I) const; + /// Widen a scalar load. /// /// \details \p Widen scalar load for uniform, small type loads from constant @@ -806,30 +824,49 @@ static Value* getMulHu(IRBuilder<> &Builder, Value *LHS, Value *RHS) { return getMul64(Builder, LHS, RHS).second; } -// The fractional part of a float is enough to accurately represent up to -// a 24-bit signed integer. -Value* AMDGPUCodeGenPrepare::expandDivRem24(IRBuilder<> &Builder, - BinaryOperator &I, - Value *Num, Value *Den, - bool IsDiv, bool IsSigned) const { - assert(Num->getType()->isIntegerTy(32)); - +/// Figure out how many bits are really needed for this ddivision. \p AtLeast is +/// an optimization hint to bypass the second ComputeNumSignBits call if we the +/// first one is insufficient. Returns -1 on failure. +int AMDGPUCodeGenPrepare::getDivNumBits(BinaryOperator &I, + Value *Num, Value *Den, + unsigned AtLeast, bool IsSigned) const { const DataLayout &DL = Mod->getDataLayout(); unsigned LHSSignBits = ComputeNumSignBits(Num, DL, 0, AC, &I); - if (LHSSignBits < 9) - return nullptr; + if (LHSSignBits < AtLeast) + return -1; unsigned RHSSignBits = ComputeNumSignBits(Den, DL, 0, AC, &I); - if (RHSSignBits < 9) - return nullptr; - + if (RHSSignBits < AtLeast) + return -1; unsigned SignBits = std::min(LHSSignBits, RHSSignBits); - unsigned DivBits = 32 - SignBits; + unsigned DivBits = Num->getType()->getScalarSizeInBits() - SignBits; if (IsSigned) ++DivBits; + return DivBits; +} + +// The fractional part of a float is enough to accurately represent up to +// a 24-bit signed integer. +Value *AMDGPUCodeGenPrepare::expandDivRem24(IRBuilder<> &Builder, + BinaryOperator &I, + Value *Num, Value *Den, + bool IsDiv, bool IsSigned) const { + int DivBits = getDivNumBits(I, Num, Den, 9, IsSigned); + if (DivBits == -1) + return nullptr; + return expandDivRem24Impl(Builder, I, Num, Den, DivBits, IsDiv, IsSigned); +} +Value *AMDGPUCodeGenPrepare::expandDivRem24Impl(IRBuilder<> &Builder, + BinaryOperator &I, + Value *Num, Value *Den, + unsigned DivBits, + bool IsDiv, bool IsSigned) const { Type *I32Ty = Builder.getInt32Ty(); + Num = Builder.CreateTrunc(Num, I32Ty); + Den = Builder.CreateTrunc(Den, I32Ty); + Type *F32Ty = Builder.getFloatTy(); ConstantInt *One = Builder.getInt32(1); Value *JQ = One; @@ -901,13 +938,18 @@ Value* AMDGPUCodeGenPrepare::expandDivRem24(IRBuilder<> &Builder, Res = Builder.CreateSub(Num, Rem); } - // Extend in register from the number of bits this divide really is. - if (IsSigned) { - Res = Builder.CreateShl(Res, 32 - DivBits); - Res = Builder.CreateAShr(Res, 32 - DivBits); - } else { - ConstantInt *TruncMask = Builder.getInt32((UINT64_C(1) << DivBits) - 1); - Res = Builder.CreateAnd(Res, TruncMask); + if (DivBits != 0 && DivBits < 32) { + // Extend in register from the number of bits this divide really is. + if (IsSigned) { + int InRegBits = 32 - DivBits; + + Res = Builder.CreateShl(Res, InRegBits); + Res = Builder.CreateAShr(Res, InRegBits); + } else { + ConstantInt *TruncMask + = Builder.getInt32((UINT64_C(1) << DivBits) - 1); + Res = Builder.CreateAnd(Res, TruncMask); + } } return Res; @@ -981,8 +1023,8 @@ Value* AMDGPUCodeGenPrepare::expandDivRem32(IRBuilder<> &Builder, } if (Value *Res = expandDivRem24(Builder, I, Num, Den, IsDiv, IsSigned)) { - Res = Builder.CreateTrunc(Res, Ty); - return Res; + return IsSigned ? Builder.CreateSExtOrTrunc(Res, Ty) : + Builder.CreateZExtOrTrunc(Res, Ty); } ConstantInt *Zero = Builder.getInt32(0); @@ -1093,6 +1135,53 @@ Value* AMDGPUCodeGenPrepare::expandDivRem32(IRBuilder<> &Builder, return Res; } +Value *AMDGPUCodeGenPrepare::shrinkDivRem64(IRBuilder<> &Builder, + BinaryOperator &I, + Value *Num, Value *Den) const { + if (!ExpandDiv64InIR && divHasSpecialOptimization(I, Num, Den)) + return nullptr; // Keep it for later optimization. + + Instruction::BinaryOps Opc = I.getOpcode(); + + bool IsDiv = Opc == Instruction::SDiv || Opc == Instruction::UDiv; + bool IsSigned = Opc == Instruction::SDiv || Opc == Instruction::SRem; + + int NumDivBits = getDivNumBits(I, Num, Den, 32, IsSigned); + if (NumDivBits == -1) + return nullptr; + + Value *Narrowed = nullptr; + if (NumDivBits <= 24) { + Narrowed = expandDivRem24Impl(Builder, I, Num, Den, NumDivBits, + IsDiv, IsSigned); + } else if (NumDivBits <= 32) { + Narrowed = expandDivRem32(Builder, I, Num, Den); + } + + if (Narrowed) { + return IsSigned ? Builder.CreateSExt(Narrowed, Num->getType()) : + Builder.CreateZExt(Narrowed, Num->getType()); + } + + return nullptr; +} + +void AMDGPUCodeGenPrepare::expandDivRem64(BinaryOperator &I) const { + Instruction::BinaryOps Opc = I.getOpcode(); + // Do the general expansion. + if (Opc == Instruction::UDiv || Opc == Instruction::SDiv) { + expandDivisionUpTo64Bits(&I); + return; + } + + if (Opc == Instruction::URem || Opc == Instruction::SRem) { + expandRemainderUpTo64Bits(&I); + return; + } + + llvm_unreachable("not a division"); +} + bool AMDGPUCodeGenPrepare::visitBinaryOperator(BinaryOperator &I) { if (foldBinOpIntoSelect(I)) return true; @@ -1108,9 +1197,13 @@ bool AMDGPUCodeGenPrepare::visitBinaryOperator(BinaryOperator &I) { Instruction::BinaryOps Opc = I.getOpcode(); Type *Ty = I.getType(); Value *NewDiv = nullptr; + unsigned ScalarSize = Ty->getScalarSizeInBits(); + + SmallVector Div64ToExpand; + if ((Opc == Instruction::URem || Opc == Instruction::UDiv || Opc == Instruction::SRem || Opc == Instruction::SDiv) && - Ty->getScalarSizeInBits() <= 32) { + ScalarSize <= 64) { Value *Num = I.getOperand(0); Value *Den = I.getOperand(1); IRBuilder<> Builder(&I); @@ -1122,13 +1215,35 @@ bool AMDGPUCodeGenPrepare::visitBinaryOperator(BinaryOperator &I) { for (unsigned N = 0, E = VT->getNumElements(); N != E; ++N) { Value *NumEltN = Builder.CreateExtractElement(Num, N); Value *DenEltN = Builder.CreateExtractElement(Den, N); - Value *NewElt = expandDivRem32(Builder, I, NumEltN, DenEltN); - if (!NewElt) - NewElt = Builder.CreateBinOp(Opc, NumEltN, DenEltN); + + Value *NewElt; + if (ScalarSize <= 32) { + NewElt = expandDivRem32(Builder, I, NumEltN, DenEltN); + if (!NewElt) + NewElt = Builder.CreateBinOp(Opc, NumEltN, DenEltN); + } else { + // See if this 64-bit division can be shrunk to 32/24-bits before + // producing the general expansion. + NewElt = shrinkDivRem64(Builder, I, NumEltN, DenEltN); + if (!NewElt) { + // The general 64-bit expansion introduces control flow and doesn't + // return the new value. Just insert a scalar copy and defer + // expanding it. + NewElt = Builder.CreateBinOp(Opc, NumEltN, DenEltN); + Div64ToExpand.push_back(cast(NewElt)); + } + } + NewDiv = Builder.CreateInsertElement(NewDiv, NewElt, N); } } else { - NewDiv = expandDivRem32(Builder, I, Num, Den); + if (ScalarSize <= 32) + NewDiv = expandDivRem32(Builder, I, Num, Den); + else { + NewDiv = shrinkDivRem64(Builder, I, Num, Den); + if (!NewDiv) + Div64ToExpand.push_back(&I); + } } if (NewDiv) { @@ -1138,6 +1253,14 @@ bool AMDGPUCodeGenPrepare::visitBinaryOperator(BinaryOperator &I) { } } + if (ExpandDiv64InIR) { + // TODO: We get much worse code in specially handled constant cases. + for (BinaryOperator *Div : Div64ToExpand) { + expandDivRem64(*Div); + Changed = true; + } + } + return Changed; } @@ -1255,11 +1378,25 @@ bool AMDGPUCodeGenPrepare::runOnFunction(Function &F) { bool MadeChange = false; - for (BasicBlock &BB : F) { + Function::iterator NextBB; + for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; FI = NextBB) { + BasicBlock *BB = &*FI; + NextBB = std::next(FI); + BasicBlock::iterator Next; - for (BasicBlock::iterator I = BB.begin(), E = BB.end(); I != E; I = Next) { + for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; I = Next) { Next = std::next(I); + MadeChange |= visit(*I); + + if (Next != E) { // Control flow changed + BasicBlock *NextInstBB = Next->getParent(); + if (NextInstBB != BB) { + BB = NextInstBB; + E = BB->end(); + FE = F.end(); + } + } } } diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll index daf56e4..0f4c094 100644 --- a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll @@ -5505,8 +5505,13 @@ define amdgpu_kernel void @udiv_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 % define amdgpu_kernel void @udiv_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x) { ; CHECK-LABEL: @udiv_v2i64_pow2k_denom( -; CHECK-NEXT: [[R:%.*]] = udiv <2 x i64> [[X:%.*]], -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = udiv i64 [[TMP1]], 4096 +; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x i64> undef, i64 [[TMP2]], i64 0 +; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP5:%.*]] = udiv i64 [[TMP4]], 4096 +; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i64> [[TMP3]], i64 [[TMP5]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP6]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: udiv_v2i64_pow2k_denom: @@ -5516,8 +5521,8 @@ define amdgpu_kernel void @udiv_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_lshr_b64 s[2:3], s[2:3], 12 ; GCN-NEXT: s_lshr_b64 s[0:1], s[0:1], 12 +; GCN-NEXT: s_lshr_b64 s[2:3], s[2:3], 12 ; GCN-NEXT: v_mov_b32_e32 v0, s0 ; GCN-NEXT: v_mov_b32_e32 v1, s1 ; GCN-NEXT: v_mov_b32_e32 v2, s2 @@ -5531,8 +5536,13 @@ define amdgpu_kernel void @udiv_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x) { ; CHECK-LABEL: @udiv_v2i64_mixed_pow2k_denom( -; CHECK-NEXT: [[R:%.*]] = udiv <2 x i64> [[X:%.*]], -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = udiv i64 [[TMP1]], 4096 +; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x i64> undef, i64 [[TMP2]], i64 0 +; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP5:%.*]] = udiv i64 [[TMP4]], 4095 +; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i64> [[TMP3]], i64 [[TMP5]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP6]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: udiv_v2i64_mixed_pow2k_denom: @@ -5540,7 +5550,7 @@ define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* ; GCN-NEXT: v_mov_b32_e32 v0, 0x4f800000 ; GCN-NEXT: v_madak_f32 v0, 0, v0, 0x457ff000 ; GCN-NEXT: v_rcp_f32_e32 v0, v0 -; GCN-NEXT: s_movk_i32 s4, 0xf001 +; GCN-NEXT: s_movk_i32 s6, 0xf001 ; GCN-NEXT: v_mov_b32_e32 v7, 0 ; GCN-NEXT: v_mov_b32_e32 v2, 0 ; GCN-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v0 @@ -5549,11 +5559,13 @@ define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* ; GCN-NEXT: v_mac_f32_e32 v0, 0xcf800000, v1 ; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 ; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 +; GCN-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd +; GCN-NEXT: s_movk_i32 s0, 0xfff +; GCN-NEXT: v_mul_hi_u32 v3, v0, s6 +; GCN-NEXT: v_mul_lo_u32 v5, v1, s6 +; GCN-NEXT: v_mul_lo_u32 v4, v0, s6 ; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, -1 -; GCN-NEXT: v_mul_hi_u32 v3, v0, s4 -; GCN-NEXT: v_mul_lo_u32 v5, v1, s4 -; GCN-NEXT: v_mul_lo_u32 v4, v0, s4 ; GCN-NEXT: v_subrev_i32_e32 v3, vcc, v0, v3 ; GCN-NEXT: v_add_i32_e32 v3, vcc, v5, v3 ; GCN-NEXT: v_mul_hi_u32 v6, v0, v4 @@ -5571,19 +5583,17 @@ define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* ; GCN-NEXT: v_add_i32_e32 v3, vcc, v4, v3 ; GCN-NEXT: v_add_i32_e64 v0, s[2:3], v0, v3 ; GCN-NEXT: v_addc_u32_e32 v4, vcc, v7, v5, vcc -; GCN-NEXT: v_mul_hi_u32 v5, v0, s4 +; GCN-NEXT: v_mul_hi_u32 v5, v0, s6 ; GCN-NEXT: v_addc_u32_e64 v3, vcc, v1, v4, s[2:3] -; GCN-NEXT: v_mul_lo_u32 v6, v3, s4 -; GCN-NEXT: v_mul_lo_u32 v8, v0, s4 +; GCN-NEXT: v_mul_lo_u32 v6, v3, s6 +; GCN-NEXT: v_mul_lo_u32 v8, v0, s6 ; GCN-NEXT: v_subrev_i32_e32 v5, vcc, v0, v5 -; GCN-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 -; GCN-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd +; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: v_add_i32_e32 v5, vcc, v5, v6 ; GCN-NEXT: v_mul_lo_u32 v6, v0, v5 ; GCN-NEXT: v_mul_hi_u32 v9, v0, v8 ; GCN-NEXT: v_mul_hi_u32 v10, v0, v5 ; GCN-NEXT: v_mul_hi_u32 v11, v3, v5 -; GCN-NEXT: s_movk_i32 s0, 0xfff ; GCN-NEXT: v_add_i32_e32 v6, vcc, v9, v6 ; GCN-NEXT: v_addc_u32_e32 v9, vcc, v7, v10, vcc ; GCN-NEXT: v_mul_lo_u32 v10, v3, v8 @@ -5608,6 +5618,7 @@ define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* ; GCN-NEXT: v_addc_u32_e32 v4, vcc, v7, v5, vcc ; GCN-NEXT: v_mul_lo_u32 v5, s11, v0 ; GCN-NEXT: v_mul_hi_u32 v0, s11, v0 +; GCN-NEXT: s_lshr_b64 s[2:3], s[8:9], 12 ; GCN-NEXT: v_add_i32_e32 v3, vcc, v5, v3 ; GCN-NEXT: v_addc_u32_e32 v0, vcc, v4, v0, vcc ; GCN-NEXT: v_addc_u32_e32 v2, vcc, v6, v2, vcc @@ -5641,9 +5652,8 @@ define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* ; GCN-NEXT: v_cndmask_b32_e64 v3, v1, v3, s[0:1] ; GCN-NEXT: v_cndmask_b32_e32 v1, v7, v5, vcc ; GCN-NEXT: v_cndmask_b32_e64 v2, v0, v1, s[0:1] -; GCN-NEXT: s_lshr_b64 s[0:1], s[8:9], 12 -; GCN-NEXT: v_mov_b32_e32 v0, s0 -; GCN-NEXT: v_mov_b32_e32 v1, s1 +; GCN-NEXT: v_mov_b32_e32 v0, s2 +; GCN-NEXT: v_mov_b32_e32 v1, s3 ; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 ; GCN-NEXT: s_endpgm %r = udiv <2 x i64> %x, @@ -5654,8 +5664,15 @@ define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* define amdgpu_kernel void @udiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x, <2 x i64> %y) { ; CHECK-LABEL: @udiv_v2i64_pow2_shl_denom( ; CHECK-NEXT: [[SHL_Y:%.*]] = shl <2 x i64> , [[Y:%.*]] -; CHECK-NEXT: [[R:%.*]] = udiv <2 x i64> [[X:%.*]], [[SHL_Y]] -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[SHL_Y]], i64 0 +; CHECK-NEXT: [[TMP3:%.*]] = udiv i64 [[TMP1]], [[TMP2]] +; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i64> undef, i64 [[TMP3]], i64 0 +; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i64> [[SHL_Y]], i64 1 +; CHECK-NEXT: [[TMP7:%.*]] = udiv i64 [[TMP5]], [[TMP6]] +; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i64> [[TMP4]], i64 [[TMP7]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP8]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: udiv_v2i64_pow2_shl_denom: @@ -5666,10 +5683,10 @@ define amdgpu_kernel void @udiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_add_i32 s1, s2, 12 ; GCN-NEXT: s_add_i32 s0, s0, 12 -; GCN-NEXT: s_lshr_b64 s[2:3], s[10:11], s1 +; GCN-NEXT: s_add_i32 s2, s2, 12 ; GCN-NEXT: s_lshr_b64 s[0:1], s[8:9], s0 +; GCN-NEXT: s_lshr_b64 s[2:3], s[10:11], s2 ; GCN-NEXT: v_mov_b32_e32 v0, s0 ; GCN-NEXT: v_mov_b32_e32 v1, s1 ; GCN-NEXT: v_mov_b32_e32 v2, s2 @@ -5874,8 +5891,13 @@ define amdgpu_kernel void @urem_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 % define amdgpu_kernel void @urem_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x) { ; CHECK-LABEL: @urem_v2i64_pow2k_denom( -; CHECK-NEXT: [[R:%.*]] = urem <2 x i64> [[X:%.*]], -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = urem i64 [[TMP1]], 4096 +; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x i64> undef, i64 [[TMP2]], i64 0 +; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP5:%.*]] = urem i64 [[TMP4]], 4096 +; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i64> [[TMP3]], i64 [[TMP5]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP6]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: urem_v2i64_pow2k_denom: @@ -5887,8 +5909,8 @@ define amdgpu_kernel void @urem_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_and_b32 s1, s2, s8 ; GCN-NEXT: s_and_b32 s0, s0, s8 +; GCN-NEXT: s_and_b32 s1, s2, s8 ; GCN-NEXT: v_mov_b32_e32 v0, s0 ; GCN-NEXT: v_mov_b32_e32 v2, s1 ; GCN-NEXT: v_mov_b32_e32 v3, v1 @@ -5902,8 +5924,15 @@ define amdgpu_kernel void @urem_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, define amdgpu_kernel void @urem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x, <2 x i64> %y) { ; CHECK-LABEL: @urem_v2i64_pow2_shl_denom( ; CHECK-NEXT: [[SHL_Y:%.*]] = shl <2 x i64> , [[Y:%.*]] -; CHECK-NEXT: [[R:%.*]] = urem <2 x i64> [[X:%.*]], [[SHL_Y]] -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[SHL_Y]], i64 0 +; CHECK-NEXT: [[TMP3:%.*]] = urem i64 [[TMP1]], [[TMP2]] +; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i64> undef, i64 [[TMP3]], i64 0 +; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i64> [[SHL_Y]], i64 1 +; CHECK-NEXT: [[TMP7:%.*]] = urem i64 [[TMP5]], [[TMP6]] +; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i64> [[TMP4]], i64 [[TMP7]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP8]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: urem_v2i64_pow2_shl_denom: @@ -5916,14 +5945,14 @@ define amdgpu_kernel void @urem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_lshl_b64 s[0:1], s[12:13], s0 ; GCN-NEXT: s_lshl_b64 s[2:3], s[12:13], s2 -; GCN-NEXT: s_add_u32 s2, s2, -1 -; GCN-NEXT: s_addc_u32 s3, s3, -1 -; GCN-NEXT: s_and_b64 s[2:3], s[10:11], s[2:3] +; GCN-NEXT: s_lshl_b64 s[0:1], s[12:13], s0 ; GCN-NEXT: s_add_u32 s0, s0, -1 ; GCN-NEXT: s_addc_u32 s1, s1, -1 ; GCN-NEXT: s_and_b64 s[0:1], s[8:9], s[0:1] +; GCN-NEXT: s_add_u32 s2, s2, -1 +; GCN-NEXT: s_addc_u32 s3, s3, -1 +; GCN-NEXT: s_and_b64 s[2:3], s[10:11], s[2:3] ; GCN-NEXT: v_mov_b32_e32 v0, s0 ; GCN-NEXT: v_mov_b32_e32 v1, s1 ; GCN-NEXT: v_mov_b32_e32 v2, s2 @@ -6249,8 +6278,13 @@ define amdgpu_kernel void @sdiv_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 % define amdgpu_kernel void @sdiv_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x) { ; CHECK-LABEL: @sdiv_v2i64_pow2k_denom( -; CHECK-NEXT: [[R:%.*]] = sdiv <2 x i64> [[X:%.*]], -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = sdiv i64 [[TMP1]], 4096 +; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x i64> undef, i64 [[TMP2]], i64 0 +; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP5:%.*]] = sdiv i64 [[TMP4]], 4096 +; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i64> [[TMP3]], i64 [[TMP5]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP6]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: sdiv_v2i64_pow2k_denom: @@ -6260,16 +6294,16 @@ define amdgpu_kernel void @sdiv_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_ashr_i32 s8, s3, 31 -; GCN-NEXT: s_lshr_b32 s8, s8, 20 -; GCN-NEXT: s_add_u32 s2, s2, s8 -; GCN-NEXT: s_addc_u32 s3, s3, 0 ; GCN-NEXT: s_ashr_i32 s8, s1, 31 -; GCN-NEXT: s_ashr_i64 s[2:3], s[2:3], 12 ; GCN-NEXT: s_lshr_b32 s8, s8, 20 ; GCN-NEXT: s_add_u32 s0, s0, s8 ; GCN-NEXT: s_addc_u32 s1, s1, 0 +; GCN-NEXT: s_ashr_i32 s8, s3, 31 ; GCN-NEXT: s_ashr_i64 s[0:1], s[0:1], 12 +; GCN-NEXT: s_lshr_b32 s8, s8, 20 +; GCN-NEXT: s_add_u32 s2, s2, s8 +; GCN-NEXT: s_addc_u32 s3, s3, 0 +; GCN-NEXT: s_ashr_i64 s[2:3], s[2:3], 12 ; GCN-NEXT: v_mov_b32_e32 v0, s0 ; GCN-NEXT: v_mov_b32_e32 v1, s1 ; GCN-NEXT: v_mov_b32_e32 v2, s2 @@ -6283,101 +6317,112 @@ define amdgpu_kernel void @sdiv_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, define amdgpu_kernel void @ssdiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x) { ; CHECK-LABEL: @ssdiv_v2i64_mixed_pow2k_denom( -; CHECK-NEXT: [[R:%.*]] = sdiv <2 x i64> [[X:%.*]], -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = sdiv i64 [[TMP1]], 4096 +; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x i64> undef, i64 [[TMP2]], i64 0 +; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP5:%.*]] = sdiv i64 [[TMP4]], 4095 +; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i64> [[TMP3]], i64 [[TMP5]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP6]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: ssdiv_v2i64_mixed_pow2k_denom: ; GCN: ; %bb.0: -; GCN-NEXT: v_mov_b32_e32 v0, 0x4f800000 -; GCN-NEXT: v_madak_f32 v0, 0, v0, 0x457ff000 +; GCN-NEXT: v_mov_b32_e32 v0, 0x457ff000 +; GCN-NEXT: v_mov_b32_e32 v1, 0x4f800000 +; GCN-NEXT: v_mac_f32_e32 v0, 0, v1 ; GCN-NEXT: v_rcp_f32_e32 v0, v0 ; GCN-NEXT: s_movk_i32 s6, 0xf001 -; GCN-NEXT: v_mov_b32_e32 v7, 0 -; GCN-NEXT: v_mov_b32_e32 v5, 0 +; GCN-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 +; GCN-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd +; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v0 ; GCN-NEXT: v_mul_f32_e32 v1, 0x2f800000, v0 ; GCN-NEXT: v_trunc_f32_e32 v1, v1 ; GCN-NEXT: v_mac_f32_e32 v0, 0xcf800000, v1 ; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 ; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 -; GCN-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 -; GCN-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: v_mul_hi_u32 v3, s6, v0 -; GCN-NEXT: v_mul_lo_u32 v2, v1, s6 -; GCN-NEXT: v_mul_lo_u32 v4, v0, s6 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 +; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_ashr_i32 s0, s9, 31 +; GCN-NEXT: s_lshr_b32 s0, s0, 20 +; GCN-NEXT: v_mul_hi_u32 v2, s6, v0 +; GCN-NEXT: v_mul_lo_u32 v3, v1, s6 +; GCN-NEXT: s_add_u32 s2, s8, s0 +; GCN-NEXT: s_addc_u32 s3, s9, 0 +; GCN-NEXT: s_ashr_i32 s8, s11, 31 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v3 +; GCN-NEXT: v_mul_lo_u32 v3, v0, s6 ; GCN-NEXT: v_subrev_i32_e32 v2, vcc, v0, v2 -; GCN-NEXT: v_mul_hi_u32 v3, v0, v4 -; GCN-NEXT: v_mul_lo_u32 v6, v0, v2 -; GCN-NEXT: v_mul_hi_u32 v8, v0, v2 -; GCN-NEXT: v_mul_hi_u32 v9, v1, v2 +; GCN-NEXT: v_mul_lo_u32 v4, v0, v2 +; GCN-NEXT: v_mul_hi_u32 v6, v0, v2 +; GCN-NEXT: v_mul_hi_u32 v5, v0, v3 +; GCN-NEXT: v_mul_hi_u32 v7, v1, v2 ; GCN-NEXT: v_mul_lo_u32 v2, v1, v2 -; GCN-NEXT: v_add_i32_e32 v3, vcc, v3, v6 -; GCN-NEXT: v_addc_u32_e32 v6, vcc, v7, v8, vcc -; GCN-NEXT: v_mul_lo_u32 v8, v1, v4 -; GCN-NEXT: v_mul_hi_u32 v4, v1, v4 -; GCN-NEXT: v_add_i32_e32 v3, vcc, v8, v3 -; GCN-NEXT: v_addc_u32_e32 v3, vcc, v6, v4, vcc -; GCN-NEXT: v_addc_u32_e32 v4, vcc, v9, v5, vcc -; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 -; GCN-NEXT: v_add_i32_e64 v0, s[2:3], v0, v2 -; GCN-NEXT: v_addc_u32_e32 v3, vcc, v7, v4, vcc -; GCN-NEXT: v_addc_u32_e64 v2, vcc, v1, v3, s[2:3] -; GCN-NEXT: v_mul_lo_u32 v4, v2, s6 -; GCN-NEXT: v_mul_hi_u32 v6, s6, v0 +; GCN-NEXT: s_ashr_i64 s[2:3], s[2:3], 12 +; GCN-NEXT: v_add_i32_e32 v4, vcc, v5, v4 +; GCN-NEXT: v_addc_u32_e32 v5, vcc, 0, v6, vcc +; GCN-NEXT: v_mul_lo_u32 v6, v1, v3 +; GCN-NEXT: v_mul_hi_u32 v3, v1, v3 +; GCN-NEXT: s_mov_b32 s9, s8 ; GCN-NEXT: v_add_i32_e32 v4, vcc, v6, v4 -; GCN-NEXT: v_mul_lo_u32 v6, v0, s6 -; GCN-NEXT: v_subrev_i32_e32 v4, vcc, v0, v4 -; GCN-NEXT: v_mul_lo_u32 v10, v0, v4 -; GCN-NEXT: v_mul_hi_u32 v12, v0, v4 -; GCN-NEXT: v_mul_hi_u32 v11, v0, v6 -; GCN-NEXT: v_mul_hi_u32 v9, v2, v6 -; GCN-NEXT: v_mul_lo_u32 v6, v2, v6 -; GCN-NEXT: v_mul_hi_u32 v8, v2, v4 +; GCN-NEXT: v_addc_u32_e32 v3, vcc, v5, v3, vcc +; GCN-NEXT: v_mov_b32_e32 v4, 0 +; GCN-NEXT: v_addc_u32_e32 v5, vcc, v7, v4, vcc +; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 +; GCN-NEXT: v_mov_b32_e32 v6, 0 +; GCN-NEXT: v_add_i32_e64 v0, s[0:1], v0, v2 +; GCN-NEXT: v_addc_u32_e32 v3, vcc, v6, v5, vcc +; GCN-NEXT: v_addc_u32_e64 v2, vcc, v1, v3, s[0:1] +; GCN-NEXT: v_mul_lo_u32 v5, v2, s6 +; GCN-NEXT: v_mul_hi_u32 v7, s6, v0 +; GCN-NEXT: v_add_i32_e32 v5, vcc, v7, v5 +; GCN-NEXT: v_mul_lo_u32 v7, v0, s6 +; GCN-NEXT: v_subrev_i32_e32 v5, vcc, v0, v5 +; GCN-NEXT: v_mul_lo_u32 v10, v0, v5 +; GCN-NEXT: v_mul_hi_u32 v12, v0, v5 +; GCN-NEXT: v_mul_hi_u32 v11, v0, v7 +; GCN-NEXT: v_mul_hi_u32 v9, v2, v7 +; GCN-NEXT: v_mul_lo_u32 v7, v2, v7 +; GCN-NEXT: v_mul_hi_u32 v8, v2, v5 ; GCN-NEXT: v_add_i32_e32 v10, vcc, v11, v10 -; GCN-NEXT: v_addc_u32_e32 v11, vcc, v7, v12, vcc -; GCN-NEXT: v_mul_lo_u32 v2, v2, v4 -; GCN-NEXT: v_add_i32_e32 v6, vcc, v6, v10 -; GCN-NEXT: v_addc_u32_e32 v6, vcc, v11, v9, vcc -; GCN-NEXT: v_addc_u32_e32 v4, vcc, v8, v5, vcc -; GCN-NEXT: v_add_i32_e32 v2, vcc, v6, v2 -; GCN-NEXT: v_addc_u32_e32 v4, vcc, v7, v4, vcc +; GCN-NEXT: v_addc_u32_e32 v11, vcc, 0, v12, vcc +; GCN-NEXT: v_mul_lo_u32 v2, v2, v5 +; GCN-NEXT: v_add_i32_e32 v7, vcc, v7, v10 +; GCN-NEXT: v_addc_u32_e32 v7, vcc, v11, v9, vcc +; GCN-NEXT: v_addc_u32_e32 v5, vcc, v8, v4, vcc +; GCN-NEXT: v_add_i32_e32 v2, vcc, v7, v2 +; GCN-NEXT: v_addc_u32_e32 v5, vcc, v6, v5, vcc ; GCN-NEXT: v_add_i32_e32 v1, vcc, v1, v3 -; GCN-NEXT: v_addc_u32_e64 v1, vcc, v1, v4, s[2:3] -; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_ashr_i32 s2, s11, 31 -; GCN-NEXT: s_add_u32 s0, s10, s2 +; GCN-NEXT: v_addc_u32_e64 v1, vcc, v1, v5, s[0:1] +; GCN-NEXT: s_add_u32 s0, s10, s8 +; GCN-NEXT: s_addc_u32 s1, s11, s8 ; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 -; GCN-NEXT: s_mov_b32 s3, s2 -; GCN-NEXT: s_addc_u32 s1, s11, s2 -; GCN-NEXT: s_xor_b64 s[0:1], s[0:1], s[2:3] +; GCN-NEXT: s_xor_b64 s[0:1], s[0:1], s[8:9] ; GCN-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; GCN-NEXT: v_mul_lo_u32 v2, s0, v1 ; GCN-NEXT: v_mul_hi_u32 v3, s0, v0 -; GCN-NEXT: v_mul_hi_u32 v4, s0, v1 -; GCN-NEXT: v_mul_hi_u32 v6, s1, v1 +; GCN-NEXT: v_mul_hi_u32 v5, s0, v1 +; GCN-NEXT: v_mul_hi_u32 v7, s1, v1 ; GCN-NEXT: v_mul_lo_u32 v1, s1, v1 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 -; GCN-NEXT: v_addc_u32_e32 v3, vcc, v7, v4, vcc -; GCN-NEXT: v_mul_lo_u32 v4, s1, v0 +; GCN-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc +; GCN-NEXT: v_mul_lo_u32 v5, s1, v0 ; GCN-NEXT: v_mul_hi_u32 v0, s1, v0 -; GCN-NEXT: s_movk_i32 s3, 0xfff +; GCN-NEXT: s_movk_i32 s9, 0xfff ; GCN-NEXT: s_mov_b32 s6, -1 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v4, v2 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v5, v2 ; GCN-NEXT: v_addc_u32_e32 v0, vcc, v3, v0, vcc -; GCN-NEXT: v_addc_u32_e32 v2, vcc, v6, v5, vcc +; GCN-NEXT: v_addc_u32_e32 v2, vcc, v7, v4, vcc ; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 -; GCN-NEXT: v_addc_u32_e32 v1, vcc, v7, v2, vcc -; GCN-NEXT: v_mul_lo_u32 v2, v1, s3 -; GCN-NEXT: v_mul_hi_u32 v3, s3, v0 -; GCN-NEXT: v_mul_lo_u32 v4, v0, s3 +; GCN-NEXT: v_addc_u32_e32 v1, vcc, v6, v2, vcc +; GCN-NEXT: v_mul_lo_u32 v2, v1, s9 +; GCN-NEXT: v_mul_hi_u32 v3, s9, v0 +; GCN-NEXT: v_mul_lo_u32 v4, v0, s9 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 ; GCN-NEXT: v_sub_i32_e32 v4, vcc, s0, v4 ; GCN-NEXT: v_mov_b32_e32 v3, s1 ; GCN-NEXT: v_subb_u32_e32 v2, vcc, v3, v2, vcc -; GCN-NEXT: v_subrev_i32_e32 v3, vcc, s3, v4 +; GCN-NEXT: v_subrev_i32_e32 v3, vcc, s9, v4 ; GCN-NEXT: v_subbrev_u32_e32 v5, vcc, 0, v2, vcc ; GCN-NEXT: s_movk_i32 s0, 0xffe ; GCN-NEXT: v_cmp_lt_u32_e32 vcc, s0, v3 @@ -6394,22 +6439,17 @@ define amdgpu_kernel void @ssdiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* ; GCN-NEXT: v_cmp_ne_u32_e32 vcc, 0, v3 ; GCN-NEXT: v_cndmask_b32_e64 v2, -1, v4, s[0:1] ; GCN-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v2 -; GCN-NEXT: v_cndmask_b32_e32 v3, v8, v6, vcc ; GCN-NEXT: v_cndmask_b32_e32 v2, v7, v5, vcc -; GCN-NEXT: v_cndmask_b32_e64 v1, v1, v3, s[0:1] ; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: s_ashr_i32 s0, s9, 31 -; GCN-NEXT: s_lshr_b32 s0, s0, 20 -; GCN-NEXT: s_add_u32 s0, s8, s0 -; GCN-NEXT: s_addc_u32 s1, s9, 0 -; GCN-NEXT: v_xor_b32_e32 v0, s2, v0 -; GCN-NEXT: v_xor_b32_e32 v1, s2, v1 -; GCN-NEXT: v_mov_b32_e32 v3, s2 -; GCN-NEXT: v_subrev_i32_e32 v2, vcc, s2, v0 -; GCN-NEXT: s_ashr_i64 s[0:1], s[0:1], 12 +; GCN-NEXT: v_cndmask_b32_e32 v3, v8, v6, vcc +; GCN-NEXT: v_cndmask_b32_e64 v1, v1, v3, s[0:1] +; GCN-NEXT: v_xor_b32_e32 v0, s8, v0 +; GCN-NEXT: v_subrev_i32_e32 v2, vcc, s8, v0 +; GCN-NEXT: v_xor_b32_e32 v1, s8, v1 +; GCN-NEXT: v_mov_b32_e32 v3, s8 ; GCN-NEXT: v_subb_u32_e32 v3, vcc, v1, v3, vcc -; GCN-NEXT: v_mov_b32_e32 v0, s0 -; GCN-NEXT: v_mov_b32_e32 v1, s1 +; GCN-NEXT: v_mov_b32_e32 v0, s2 +; GCN-NEXT: v_mov_b32_e32 v1, s3 ; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 ; GCN-NEXT: s_endpgm %r = sdiv <2 x i64> %x, @@ -6420,8 +6460,15 @@ define amdgpu_kernel void @ssdiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)* define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x, <2 x i64> %y) { ; CHECK-LABEL: @sdiv_v2i64_pow2_shl_denom( ; CHECK-NEXT: [[SHL_Y:%.*]] = shl <2 x i64> , [[Y:%.*]] -; CHECK-NEXT: [[R:%.*]] = sdiv <2 x i64> [[X:%.*]], [[SHL_Y]] -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[SHL_Y]], i64 0 +; CHECK-NEXT: [[TMP3:%.*]] = sdiv i64 [[TMP1]], [[TMP2]] +; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i64> undef, i64 [[TMP3]], i64 0 +; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i64> [[SHL_Y]], i64 1 +; CHECK-NEXT: [[TMP7:%.*]] = sdiv i64 [[TMP5]], [[TMP6]] +; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i64> [[TMP4]], i64 [[TMP7]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP8]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: sdiv_v2i64_pow2_shl_denom: @@ -6432,8 +6479,8 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: s_mov_b32 s18, 0x4f800000 ; GCN-NEXT: s_mov_b32 s19, 0x5f7ffffc ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_lshl_b64 s[12:13], s[2:3], s4 -; GCN-NEXT: s_lshl_b64 s[2:3], s[2:3], s6 +; GCN-NEXT: s_lshl_b64 s[12:13], s[2:3], s6 +; GCN-NEXT: s_lshl_b64 s[2:3], s[2:3], s4 ; GCN-NEXT: s_ashr_i32 s16, s3, 31 ; GCN-NEXT: s_add_u32 s2, s2, s16 ; GCN-NEXT: s_mov_b32 s17, s16 @@ -6503,22 +6550,22 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_add_i32_e32 v1, vcc, v1, v3 ; GCN-NEXT: v_addc_u32_e64 v1, vcc, v1, v5, s[2:3] ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_ashr_i32 s2, s11, 31 -; GCN-NEXT: s_add_u32 s0, s10, s2 +; GCN-NEXT: s_ashr_i32 s2, s9, 31 +; GCN-NEXT: s_add_u32 s0, s8, s2 ; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: s_mov_b32 s3, s2 -; GCN-NEXT: s_addc_u32 s1, s11, s2 -; GCN-NEXT: s_xor_b64 s[10:11], s[0:1], s[2:3] +; GCN-NEXT: s_addc_u32 s1, s9, s2 +; GCN-NEXT: s_xor_b64 s[8:9], s[0:1], s[2:3] ; GCN-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GCN-NEXT: v_mul_lo_u32 v2, s10, v1 -; GCN-NEXT: v_mul_hi_u32 v3, s10, v0 -; GCN-NEXT: v_mul_hi_u32 v5, s10, v1 -; GCN-NEXT: v_mul_hi_u32 v7, s11, v1 -; GCN-NEXT: v_mul_lo_u32 v1, s11, v1 +; GCN-NEXT: v_mul_lo_u32 v2, s8, v1 +; GCN-NEXT: v_mul_hi_u32 v3, s8, v0 +; GCN-NEXT: v_mul_hi_u32 v5, s8, v1 +; GCN-NEXT: v_mul_hi_u32 v7, s9, v1 +; GCN-NEXT: v_mul_lo_u32 v1, s9, v1 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 ; GCN-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc -; GCN-NEXT: v_mul_lo_u32 v5, s11, v0 -; GCN-NEXT: v_mul_hi_u32 v0, s11, v0 +; GCN-NEXT: v_mul_lo_u32 v5, s9, v0 +; GCN-NEXT: v_mul_hi_u32 v0, s9, v0 ; GCN-NEXT: s_xor_b64 s[2:3], s[2:3], s[16:17] ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v5, v2 @@ -6533,8 +6580,8 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 ; GCN-NEXT: v_mul_lo_u32 v3, s14, v0 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v5 -; GCN-NEXT: v_sub_i32_e32 v5, vcc, s11, v2 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, s10, v3 +; GCN-NEXT: v_sub_i32_e32 v5, vcc, s9, v2 +; GCN-NEXT: v_sub_i32_e32 v3, vcc, s8, v3 ; GCN-NEXT: v_subb_u32_e64 v5, s[0:1], v5, v7, vcc ; GCN-NEXT: v_subrev_i32_e64 v7, s[0:1], s14, v3 ; GCN-NEXT: v_subbrev_u32_e64 v5, s[0:1], 0, v5, s[0:1] @@ -6548,14 +6595,14 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_addc_u32_e64 v8, s[0:1], 0, v1, s[0:1] ; GCN-NEXT: v_add_i32_e64 v9, s[0:1], 1, v0 ; GCN-NEXT: v_addc_u32_e64 v10, s[0:1], 0, v1, s[0:1] -; GCN-NEXT: s_ashr_i32 s10, s13, 31 +; GCN-NEXT: s_ashr_i32 s8, s13, 31 ; GCN-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v5 -; GCN-NEXT: s_add_u32 s12, s12, s10 +; GCN-NEXT: s_add_u32 s12, s12, s8 ; GCN-NEXT: v_cndmask_b32_e64 v5, v10, v8, s[0:1] -; GCN-NEXT: v_mov_b32_e32 v8, s11 -; GCN-NEXT: s_mov_b32 s11, s10 -; GCN-NEXT: s_addc_u32 s13, s13, s10 -; GCN-NEXT: s_xor_b64 s[12:13], s[12:13], s[10:11] +; GCN-NEXT: v_mov_b32_e32 v8, s9 +; GCN-NEXT: s_mov_b32 s9, s8 +; GCN-NEXT: s_addc_u32 s13, s13, s8 +; GCN-NEXT: s_xor_b64 s[12:13], s[12:13], s[8:9] ; GCN-NEXT: v_cvt_f32_u32_e32 v10, s12 ; GCN-NEXT: v_cvt_f32_u32_e32 v11, s13 ; GCN-NEXT: v_subb_u32_e32 v2, vcc, v8, v2, vcc @@ -6624,42 +6671,42 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_add_i32_e32 v3, vcc, v9, v3 ; GCN-NEXT: v_addc_u32_e32 v8, vcc, v6, v8, vcc ; GCN-NEXT: v_add_i32_e32 v5, vcc, v5, v7 -; GCN-NEXT: s_ashr_i32 s14, s9, 31 +; GCN-NEXT: s_ashr_i32 s14, s11, 31 ; GCN-NEXT: v_addc_u32_e64 v5, vcc, v5, v8, s[0:1] -; GCN-NEXT: s_add_u32 s0, s8, s14 +; GCN-NEXT: s_add_u32 s0, s10, s14 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v3 ; GCN-NEXT: s_mov_b32 s15, s14 -; GCN-NEXT: s_addc_u32 s1, s9, s14 -; GCN-NEXT: s_xor_b64 s[8:9], s[0:1], s[14:15] +; GCN-NEXT: s_addc_u32 s1, s11, s14 +; GCN-NEXT: s_xor_b64 s[10:11], s[0:1], s[14:15] ; GCN-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc -; GCN-NEXT: v_mul_lo_u32 v5, s8, v3 -; GCN-NEXT: v_mul_hi_u32 v7, s8, v2 -; GCN-NEXT: v_mul_hi_u32 v9, s8, v3 -; GCN-NEXT: v_mul_hi_u32 v10, s9, v3 -; GCN-NEXT: v_mul_lo_u32 v3, s9, v3 +; GCN-NEXT: v_mul_lo_u32 v5, s10, v3 +; GCN-NEXT: v_mul_hi_u32 v7, s10, v2 +; GCN-NEXT: v_mul_hi_u32 v9, s10, v3 +; GCN-NEXT: v_mul_hi_u32 v10, s11, v3 +; GCN-NEXT: v_mul_lo_u32 v3, s11, v3 ; GCN-NEXT: v_add_i32_e32 v5, vcc, v7, v5 ; GCN-NEXT: v_addc_u32_e32 v7, vcc, 0, v9, vcc -; GCN-NEXT: v_mul_lo_u32 v9, s9, v2 -; GCN-NEXT: v_mul_hi_u32 v2, s9, v2 +; GCN-NEXT: v_mul_lo_u32 v9, s11, v2 +; GCN-NEXT: v_mul_hi_u32 v2, s11, v2 ; GCN-NEXT: v_mov_b32_e32 v8, s3 ; GCN-NEXT: v_add_i32_e32 v5, vcc, v9, v5 ; GCN-NEXT: v_addc_u32_e32 v2, vcc, v7, v2, vcc ; GCN-NEXT: v_addc_u32_e32 v4, vcc, v10, v4, vcc -; GCN-NEXT: v_add_i32_e32 v5, vcc, v2, v3 -; GCN-NEXT: v_addc_u32_e32 v4, vcc, v6, v4, vcc -; GCN-NEXT: v_mul_lo_u32 v6, s12, v4 -; GCN-NEXT: v_mul_hi_u32 v7, s12, v5 -; GCN-NEXT: v_subrev_i32_e32 v2, vcc, s2, v0 -; GCN-NEXT: v_mul_lo_u32 v0, s13, v5 -; GCN-NEXT: v_subb_u32_e32 v3, vcc, v1, v8, vcc -; GCN-NEXT: v_add_i32_e32 v1, vcc, v7, v6 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v3 +; GCN-NEXT: v_addc_u32_e32 v3, vcc, v6, v4, vcc +; GCN-NEXT: v_mul_lo_u32 v4, s12, v3 +; GCN-NEXT: v_mul_hi_u32 v5, s12, v2 +; GCN-NEXT: v_mul_lo_u32 v6, s13, v2 +; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s2, v0 +; GCN-NEXT: v_subb_u32_e32 v1, vcc, v1, v8, vcc +; GCN-NEXT: v_add_i32_e32 v4, vcc, v5, v4 +; GCN-NEXT: v_mul_lo_u32 v5, s12, v2 +; GCN-NEXT: v_add_i32_e32 v4, vcc, v4, v6 +; GCN-NEXT: v_sub_i32_e32 v6, vcc, s11, v4 ; GCN-NEXT: v_mov_b32_e32 v7, s13 -; GCN-NEXT: v_add_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_mul_lo_u32 v1, s12, v5 -; GCN-NEXT: v_sub_i32_e32 v6, vcc, s9, v0 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s8, v1 +; GCN-NEXT: v_sub_i32_e32 v5, vcc, s10, v5 ; GCN-NEXT: v_subb_u32_e64 v6, s[0:1], v6, v7, vcc -; GCN-NEXT: v_subrev_i32_e64 v7, s[0:1], s12, v1 +; GCN-NEXT: v_subrev_i32_e64 v7, s[0:1], s12, v5 ; GCN-NEXT: v_subbrev_u32_e64 v6, s[0:1], 0, v6, s[0:1] ; GCN-NEXT: v_cmp_le_u32_e64 s[0:1], s13, v6 ; GCN-NEXT: v_cndmask_b32_e64 v8, 0, -1, s[0:1] @@ -6667,30 +6714,30 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[0:1] ; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], s13, v6 ; GCN-NEXT: v_cndmask_b32_e64 v6, v8, v7, s[0:1] -; GCN-NEXT: v_add_i32_e64 v7, s[0:1], 2, v5 -; GCN-NEXT: v_addc_u32_e64 v8, s[0:1], 0, v4, s[0:1] -; GCN-NEXT: v_add_i32_e64 v9, s[0:1], 1, v5 -; GCN-NEXT: v_addc_u32_e64 v10, s[0:1], 0, v4, s[0:1] +; GCN-NEXT: v_add_i32_e64 v7, s[0:1], 2, v2 +; GCN-NEXT: v_addc_u32_e64 v8, s[0:1], 0, v3, s[0:1] +; GCN-NEXT: v_add_i32_e64 v9, s[0:1], 1, v2 +; GCN-NEXT: v_addc_u32_e64 v10, s[0:1], 0, v3, s[0:1] ; GCN-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v6 ; GCN-NEXT: v_cndmask_b32_e64 v6, v10, v8, s[0:1] -; GCN-NEXT: v_mov_b32_e32 v8, s9 -; GCN-NEXT: v_subb_u32_e32 v0, vcc, v8, v0, vcc -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s13, v0 +; GCN-NEXT: v_mov_b32_e32 v8, s11 +; GCN-NEXT: v_subb_u32_e32 v4, vcc, v8, v4, vcc +; GCN-NEXT: v_cmp_le_u32_e32 vcc, s13, v4 ; GCN-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s12, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, 0, -1, vcc -; GCN-NEXT: v_cmp_eq_u32_e32 vcc, s13, v0 -; GCN-NEXT: v_cndmask_b32_e32 v0, v8, v1, vcc -; GCN-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 -; GCN-NEXT: v_cndmask_b32_e64 v1, v9, v7, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc -; GCN-NEXT: s_xor_b64 s[0:1], s[14:15], s[10:11] -; GCN-NEXT: v_cndmask_b32_e32 v0, v4, v6, vcc -; GCN-NEXT: v_xor_b32_e32 v1, s0, v1 -; GCN-NEXT: v_xor_b32_e32 v4, s1, v0 -; GCN-NEXT: v_mov_b32_e32 v5, s1 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s0, v1 -; GCN-NEXT: v_subb_u32_e32 v1, vcc, v4, v5, vcc +; GCN-NEXT: v_cmp_le_u32_e32 vcc, s12, v5 +; GCN-NEXT: v_cndmask_b32_e64 v5, 0, -1, vcc +; GCN-NEXT: v_cmp_eq_u32_e32 vcc, s13, v4 +; GCN-NEXT: v_cndmask_b32_e32 v4, v8, v5, vcc +; GCN-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4 +; GCN-NEXT: v_cndmask_b32_e64 v4, v9, v7, s[0:1] +; GCN-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc +; GCN-NEXT: s_xor_b64 s[0:1], s[14:15], s[8:9] +; GCN-NEXT: v_cndmask_b32_e32 v3, v3, v6, vcc +; GCN-NEXT: v_xor_b32_e32 v2, s0, v2 +; GCN-NEXT: v_xor_b32_e32 v3, s1, v3 +; GCN-NEXT: v_mov_b32_e32 v4, s1 +; GCN-NEXT: v_subrev_i32_e32 v2, vcc, s0, v2 +; GCN-NEXT: v_subb_u32_e32 v3, vcc, v3, v4, vcc ; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 ; GCN-NEXT: s_endpgm %shl.y = shl <2 x i64> , %y @@ -7010,8 +7057,13 @@ define amdgpu_kernel void @srem_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 % define amdgpu_kernel void @srem_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x) { ; CHECK-LABEL: @srem_v2i64_pow2k_denom( -; CHECK-NEXT: [[R:%.*]] = srem <2 x i64> [[X:%.*]], -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = srem i64 [[TMP1]], 4096 +; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x i64> undef, i64 [[TMP2]], i64 0 +; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP5:%.*]] = srem i64 [[TMP4]], 4096 +; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i64> [[TMP3]], i64 [[TMP5]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP6]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: srem_v2i64_pow2k_denom: @@ -7022,20 +7074,20 @@ define amdgpu_kernel void @srem_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_ashr_i32 s9, s3, 31 -; GCN-NEXT: s_lshr_b32 s9, s9, 20 -; GCN-NEXT: s_add_u32 s9, s2, s9 -; GCN-NEXT: s_addc_u32 s10, s3, 0 -; GCN-NEXT: s_and_b32 s9, s9, s8 -; GCN-NEXT: s_sub_u32 s2, s2, s9 -; GCN-NEXT: s_subb_u32 s3, s3, s10 ; GCN-NEXT: s_ashr_i32 s9, s1, 31 ; GCN-NEXT: s_lshr_b32 s9, s9, 20 ; GCN-NEXT: s_add_u32 s9, s0, s9 ; GCN-NEXT: s_addc_u32 s10, s1, 0 -; GCN-NEXT: s_and_b32 s8, s9, s8 -; GCN-NEXT: s_sub_u32 s0, s0, s8 +; GCN-NEXT: s_and_b32 s9, s9, s8 +; GCN-NEXT: s_sub_u32 s0, s0, s9 ; GCN-NEXT: s_subb_u32 s1, s1, s10 +; GCN-NEXT: s_ashr_i32 s9, s3, 31 +; GCN-NEXT: s_lshr_b32 s9, s9, 20 +; GCN-NEXT: s_add_u32 s9, s2, s9 +; GCN-NEXT: s_addc_u32 s10, s3, 0 +; GCN-NEXT: s_and_b32 s8, s9, s8 +; GCN-NEXT: s_sub_u32 s2, s2, s8 +; GCN-NEXT: s_subb_u32 s3, s3, s10 ; GCN-NEXT: v_mov_b32_e32 v0, s0 ; GCN-NEXT: v_mov_b32_e32 v1, s1 ; GCN-NEXT: v_mov_b32_e32 v2, s2 @@ -7050,8 +7102,15 @@ define amdgpu_kernel void @srem_v2i64_pow2k_denom(<2 x i64> addrspace(1)* %out, define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x, <2 x i64> %y) { ; CHECK-LABEL: @srem_v2i64_pow2_shl_denom( ; CHECK-NEXT: [[SHL_Y:%.*]] = shl <2 x i64> , [[Y:%.*]] -; CHECK-NEXT: [[R:%.*]] = srem <2 x i64> [[X:%.*]], [[SHL_Y]] -; CHECK-NEXT: store <2 x i64> [[R]], <2 x i64> addrspace(1)* [[OUT:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x i64> [[X:%.*]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[SHL_Y]], i64 0 +; CHECK-NEXT: [[TMP3:%.*]] = srem i64 [[TMP1]], [[TMP2]] +; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i64> undef, i64 [[TMP3]], i64 0 +; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i64> [[X]], i64 1 +; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i64> [[SHL_Y]], i64 1 +; CHECK-NEXT: [[TMP7:%.*]] = srem i64 [[TMP5]], [[TMP6]] +; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i64> [[TMP4]], i64 [[TMP7]], i64 1 +; CHECK-NEXT: store <2 x i64> [[TMP8]], <2 x i64> addrspace(1)* [[OUT:%.*]] ; CHECK-NEXT: ret void ; ; GCN-LABEL: srem_v2i64_pow2_shl_denom: @@ -7062,8 +7121,8 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: s_mov_b32 s18, 0x4f800000 ; GCN-NEXT: s_mov_b32 s19, 0x5f7ffffc ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_lshl_b64 s[14:15], s[2:3], s4 -; GCN-NEXT: s_lshl_b64 s[2:3], s[2:3], s6 +; GCN-NEXT: s_lshl_b64 s[14:15], s[2:3], s6 +; GCN-NEXT: s_lshl_b64 s[2:3], s[2:3], s4 ; GCN-NEXT: s_ashr_i32 s4, s3, 31 ; GCN-NEXT: s_add_u32 s2, s2, s4 ; GCN-NEXT: s_mov_b32 s5, s4 @@ -7086,8 +7145,8 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 ; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_ashr_i32 s12, s11, 31 -; GCN-NEXT: s_add_u32 s0, s10, s12 +; GCN-NEXT: s_ashr_i32 s12, s9, 31 +; GCN-NEXT: s_add_u32 s0, s8, s12 ; GCN-NEXT: v_mul_hi_u32 v3, s6, v0 ; GCN-NEXT: v_mul_lo_u32 v2, s6, v1 ; GCN-NEXT: v_mul_lo_u32 v4, s7, v0 @@ -7104,8 +7163,8 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_addc_u32_e32 v4, vcc, 0, v6, vcc ; GCN-NEXT: v_mul_lo_u32 v6, v1, v5 ; GCN-NEXT: v_mul_hi_u32 v5, v1, v5 -; GCN-NEXT: s_addc_u32 s1, s11, s12 -; GCN-NEXT: s_xor_b64 s[10:11], s[0:1], s[12:13] +; GCN-NEXT: s_addc_u32 s1, s9, s12 +; GCN-NEXT: s_xor_b64 s[8:9], s[0:1], s[12:13] ; GCN-NEXT: v_add_i32_e32 v3, vcc, v6, v3 ; GCN-NEXT: v_addc_u32_e32 v3, vcc, v4, v5, vcc ; GCN-NEXT: v_mov_b32_e32 v4, 0 @@ -7140,15 +7199,15 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_addc_u32_e64 v1, vcc, v1, v5, s[2:3] ; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc -; GCN-NEXT: v_mul_lo_u32 v2, s10, v1 -; GCN-NEXT: v_mul_hi_u32 v3, s10, v0 -; GCN-NEXT: v_mul_hi_u32 v5, s10, v1 -; GCN-NEXT: v_mul_hi_u32 v7, s11, v1 -; GCN-NEXT: v_mul_lo_u32 v1, s11, v1 +; GCN-NEXT: v_mul_lo_u32 v2, s8, v1 +; GCN-NEXT: v_mul_hi_u32 v3, s8, v0 +; GCN-NEXT: v_mul_hi_u32 v5, s8, v1 +; GCN-NEXT: v_mul_hi_u32 v7, s9, v1 +; GCN-NEXT: v_mul_lo_u32 v1, s9, v1 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 ; GCN-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc -; GCN-NEXT: v_mul_lo_u32 v5, s11, v0 -; GCN-NEXT: v_mul_hi_u32 v0, s11, v0 +; GCN-NEXT: v_mul_lo_u32 v5, s9, v0 +; GCN-NEXT: v_mul_hi_u32 v0, s9, v0 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v5, v2 ; GCN-NEXT: v_addc_u32_e32 v0, vcc, v3, v0, vcc @@ -7161,9 +7220,9 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_mul_lo_u32 v0, s16, v0 ; GCN-NEXT: v_add_i32_e32 v1, vcc, v2, v1 ; GCN-NEXT: v_add_i32_e32 v1, vcc, v1, v3 -; GCN-NEXT: v_sub_i32_e32 v2, vcc, s11, v1 +; GCN-NEXT: v_sub_i32_e32 v2, vcc, s9, v1 ; GCN-NEXT: v_mov_b32_e32 v3, s17 -; GCN-NEXT: v_sub_i32_e32 v0, vcc, s10, v0 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s8, v0 ; GCN-NEXT: v_subb_u32_e64 v2, s[0:1], v2, v3, vcc ; GCN-NEXT: v_subrev_i32_e64 v5, s[0:1], s16, v0 ; GCN-NEXT: v_subb_u32_e64 v3, s[2:3], v2, v3, s[0:1] @@ -7178,14 +7237,14 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: s_ashr_i32 s2, s15, 31 ; GCN-NEXT: v_subbrev_u32_e64 v3, s[0:1], 0, v3, s[0:1] ; GCN-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v7 -; GCN-NEXT: s_add_u32 s10, s14, s2 +; GCN-NEXT: s_add_u32 s8, s14, s2 ; GCN-NEXT: v_cndmask_b32_e64 v2, v2, v3, s[0:1] -; GCN-NEXT: v_mov_b32_e32 v3, s11 +; GCN-NEXT: v_mov_b32_e32 v3, s9 ; GCN-NEXT: s_mov_b32 s3, s2 -; GCN-NEXT: s_addc_u32 s11, s15, s2 -; GCN-NEXT: s_xor_b64 s[10:11], s[10:11], s[2:3] -; GCN-NEXT: v_cvt_f32_u32_e32 v7, s10 -; GCN-NEXT: v_cvt_f32_u32_e32 v9, s11 +; GCN-NEXT: s_addc_u32 s9, s15, s2 +; GCN-NEXT: s_xor_b64 s[8:9], s[8:9], s[2:3] +; GCN-NEXT: v_cvt_f32_u32_e32 v7, s8 +; GCN-NEXT: v_cvt_f32_u32_e32 v9, s9 ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v3, v1, vcc ; GCN-NEXT: v_cmp_le_u32_e32 vcc, s17, v1 ; GCN-NEXT: v_cndmask_b32_e64 v3, 0, -1, vcc @@ -7204,13 +7263,13 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_mac_f32_e32 v3, s21, v5 ; GCN-NEXT: v_cvt_u32_f32_e32 v3, v3 ; GCN-NEXT: v_cvt_u32_f32_e32 v5, v5 -; GCN-NEXT: s_sub_u32 s2, 0, s10 +; GCN-NEXT: s_sub_u32 s2, 0, s8 ; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc ; GCN-NEXT: v_mul_hi_u32 v2, s2, v3 ; GCN-NEXT: v_mul_lo_u32 v7, s2, v5 -; GCN-NEXT: s_subb_u32 s3, 0, s11 +; GCN-NEXT: s_subb_u32 s3, 0, s9 ; GCN-NEXT: v_mul_lo_u32 v8, s3, v3 -; GCN-NEXT: s_ashr_i32 s14, s9, 31 +; GCN-NEXT: s_ashr_i32 s14, s11, 31 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v7 ; GCN-NEXT: v_mul_lo_u32 v7, s2, v3 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v8 @@ -7255,68 +7314,68 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou ; GCN-NEXT: v_addc_u32_e32 v8, vcc, v6, v8, vcc ; GCN-NEXT: v_add_i32_e32 v5, vcc, v5, v7 ; GCN-NEXT: v_addc_u32_e64 v5, vcc, v5, v8, s[0:1] -; GCN-NEXT: s_add_u32 s0, s8, s14 +; GCN-NEXT: s_add_u32 s0, s10, s14 ; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v3 -; GCN-NEXT: s_addc_u32 s1, s9, s14 -; GCN-NEXT: s_xor_b64 s[8:9], s[0:1], s[14:15] +; GCN-NEXT: s_addc_u32 s1, s11, s14 +; GCN-NEXT: s_xor_b64 s[10:11], s[0:1], s[14:15] ; GCN-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc -; GCN-NEXT: v_mul_lo_u32 v5, s8, v3 -; GCN-NEXT: v_mul_hi_u32 v7, s8, v2 -; GCN-NEXT: v_mul_hi_u32 v9, s8, v3 -; GCN-NEXT: v_mul_hi_u32 v10, s9, v3 -; GCN-NEXT: v_mul_lo_u32 v3, s9, v3 +; GCN-NEXT: v_mul_lo_u32 v5, s10, v3 +; GCN-NEXT: v_mul_hi_u32 v7, s10, v2 +; GCN-NEXT: v_mul_hi_u32 v9, s10, v3 +; GCN-NEXT: v_mul_hi_u32 v10, s11, v3 +; GCN-NEXT: v_mul_lo_u32 v3, s11, v3 ; GCN-NEXT: v_add_i32_e32 v5, vcc, v7, v5 ; GCN-NEXT: v_addc_u32_e32 v7, vcc, 0, v9, vcc -; GCN-NEXT: v_mul_lo_u32 v9, s9, v2 -; GCN-NEXT: v_mul_hi_u32 v2, s9, v2 +; GCN-NEXT: v_mul_lo_u32 v9, s11, v2 +; GCN-NEXT: v_mul_hi_u32 v2, s11, v2 ; GCN-NEXT: v_mov_b32_e32 v8, s12 ; GCN-NEXT: v_add_i32_e32 v5, vcc, v9, v5 ; GCN-NEXT: v_addc_u32_e32 v2, vcc, v7, v2, vcc ; GCN-NEXT: v_addc_u32_e32 v4, vcc, v10, v4, vcc -; GCN-NEXT: v_add_i32_e32 v5, vcc, v2, v3 -; GCN-NEXT: v_addc_u32_e32 v2, vcc, v6, v4, vcc -; GCN-NEXT: v_mul_lo_u32 v4, s10, v2 -; GCN-NEXT: v_mul_hi_u32 v6, s10, v5 -; GCN-NEXT: v_subrev_i32_e32 v2, vcc, s12, v0 -; GCN-NEXT: v_mul_lo_u32 v0, s11, v5 -; GCN-NEXT: v_subb_u32_e32 v3, vcc, v1, v8, vcc -; GCN-NEXT: v_add_i32_e32 v1, vcc, v6, v4 -; GCN-NEXT: v_add_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_mul_lo_u32 v1, s10, v5 -; GCN-NEXT: v_sub_i32_e32 v4, vcc, s9, v0 -; GCN-NEXT: v_mov_b32_e32 v5, s11 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s8, v1 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v2, v3 +; GCN-NEXT: v_addc_u32_e32 v3, vcc, v6, v4, vcc +; GCN-NEXT: v_mul_lo_u32 v3, s8, v3 +; GCN-NEXT: v_mul_hi_u32 v4, s8, v2 +; GCN-NEXT: v_mul_lo_u32 v5, s9, v2 +; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s12, v0 +; GCN-NEXT: v_mul_lo_u32 v2, s8, v2 +; GCN-NEXT: v_subb_u32_e32 v1, vcc, v1, v8, vcc +; GCN-NEXT: v_add_i32_e32 v3, vcc, v4, v3 +; GCN-NEXT: v_add_i32_e32 v3, vcc, v3, v5 +; GCN-NEXT: v_sub_i32_e32 v4, vcc, s11, v3 +; GCN-NEXT: v_mov_b32_e32 v5, s9 +; GCN-NEXT: v_sub_i32_e32 v2, vcc, s10, v2 ; GCN-NEXT: v_subb_u32_e64 v4, s[0:1], v4, v5, vcc -; GCN-NEXT: v_subrev_i32_e64 v6, s[0:1], s10, v1 +; GCN-NEXT: v_subrev_i32_e64 v6, s[0:1], s8, v2 ; GCN-NEXT: v_subb_u32_e64 v5, s[2:3], v4, v5, s[0:1] ; GCN-NEXT: v_subbrev_u32_e64 v4, s[0:1], 0, v4, s[0:1] -; GCN-NEXT: v_cmp_le_u32_e64 s[0:1], s11, v4 +; GCN-NEXT: v_cmp_le_u32_e64 s[0:1], s9, v4 ; GCN-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[0:1] -; GCN-NEXT: v_cmp_le_u32_e64 s[0:1], s10, v6 +; GCN-NEXT: v_cmp_le_u32_e64 s[0:1], s8, v6 ; GCN-NEXT: v_cndmask_b32_e64 v8, 0, -1, s[0:1] -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], s11, v4 +; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], s9, v4 ; GCN-NEXT: v_cndmask_b32_e64 v7, v7, v8, s[0:1] -; GCN-NEXT: v_subrev_i32_e64 v8, s[0:1], s10, v6 +; GCN-NEXT: v_subrev_i32_e64 v8, s[0:1], s8, v6 ; GCN-NEXT: v_subbrev_u32_e64 v5, s[0:1], 0, v5, s[0:1] ; GCN-NEXT: v_cmp_ne_u32_e64 s[0:1], 0, v7 ; GCN-NEXT: v_cndmask_b32_e64 v4, v4, v5, s[0:1] -; GCN-NEXT: v_mov_b32_e32 v5, s9 -; GCN-NEXT: v_subb_u32_e32 v0, vcc, v5, v0, vcc -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s11, v0 +; GCN-NEXT: v_mov_b32_e32 v5, s11 +; GCN-NEXT: v_subb_u32_e32 v3, vcc, v5, v3, vcc +; GCN-NEXT: v_cmp_le_u32_e32 vcc, s9, v3 ; GCN-NEXT: v_cndmask_b32_e64 v5, 0, -1, vcc -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s10, v1 +; GCN-NEXT: v_cmp_le_u32_e32 vcc, s8, v2 ; GCN-NEXT: v_cndmask_b32_e64 v7, 0, -1, vcc -; GCN-NEXT: v_cmp_eq_u32_e32 vcc, s11, v0 +; GCN-NEXT: v_cmp_eq_u32_e32 vcc, s9, v3 ; GCN-NEXT: v_cndmask_b32_e32 v5, v5, v7, vcc ; GCN-NEXT: v_cmp_ne_u32_e32 vcc, 0, v5 -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v4, vcc +; GCN-NEXT: v_cndmask_b32_e32 v3, v3, v4, vcc ; GCN-NEXT: v_cndmask_b32_e64 v4, v6, v8, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v1, v1, v4, vcc -; GCN-NEXT: v_xor_b32_e32 v1, s14, v1 -; GCN-NEXT: v_xor_b32_e32 v4, s14, v0 -; GCN-NEXT: v_mov_b32_e32 v5, s14 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s14, v1 -; GCN-NEXT: v_subb_u32_e32 v1, vcc, v4, v5, vcc +; GCN-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc +; GCN-NEXT: v_xor_b32_e32 v2, s14, v2 +; GCN-NEXT: v_xor_b32_e32 v3, s14, v3 +; GCN-NEXT: v_mov_b32_e32 v4, s14 +; GCN-NEXT: v_subrev_i32_e32 v2, vcc, s14, v2 +; GCN-NEXT: v_subb_u32_e32 v3, vcc, v3, v4, vcc ; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 ; GCN-NEXT: s_endpgm %shl.y = shl <2 x i64> , %y diff --git a/llvm/test/CodeGen/AMDGPU/sdiv64.ll b/llvm/test/CodeGen/AMDGPU/sdiv64.ll index cd6ceca..29f73d6 100644 --- a/llvm/test/CodeGen/AMDGPU/sdiv64.ll +++ b/llvm/test/CodeGen/AMDGPU/sdiv64.ll @@ -1,5 +1,6 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s +; RUN: llc -march=amdgcn -amdgpu-codegenprepare-expand-div64 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-IR %s define amdgpu_kernel void @s_test_sdiv(i64 addrspace(1)* %out, i64 %x, i64 %y) { ; GCN-LABEL: s_test_sdiv: @@ -139,6 +140,111 @@ define amdgpu_kernel void @s_test_sdiv(i64 addrspace(1)* %out, i64 %x, i64 %y) { ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v2, v5, vcc ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0xd +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_ashr_i32 s2, s7, 31 +; GCN-IR-NEXT: s_ashr_i32 s8, s13, 31 +; GCN-IR-NEXT: s_mov_b32 s3, s2 +; GCN-IR-NEXT: s_mov_b32 s9, s8 +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[2:3], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s10, s0, s2 +; GCN-IR-NEXT: s_subb_u32 s11, s1, s2 +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[8:9], s[12:13] +; GCN-IR-NEXT: s_flbit_i32_b32 s14, s10 +; GCN-IR-NEXT: s_sub_u32 s6, s0, s8 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[10:11], 0 +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s11 +; GCN-IR-NEXT: s_subb_u32 s7, s1, s8 +; GCN-IR-NEXT: s_flbit_i32_b32 s15, s6 +; GCN-IR-NEXT: s_add_i32 s14, s14, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[6:7], 0 +; GCN-IR-NEXT: s_add_i32 s15, s15, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s16, s7 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s14 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s11, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], s[12:13] +; GCN-IR-NEXT: v_mov_b32_e32 v1, s16 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s15 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v1, v0 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[12:13], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[12:13], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[12:13] +; GCN-IR-NEXT: s_cbranch_vccz BB0_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s11 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s10 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB0_7 +; GCN-IR-NEXT: BB0_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[10:11], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB0_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB0_6 +; GCN-IR-NEXT: BB0_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[10:11], v0 +; GCN-IR-NEXT: s_add_u32 s10, s6, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s11, s7, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB0_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s11 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s10, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s6, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s7, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB0_5 +; GCN-IR-NEXT: BB0_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB0_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[8:9], s[2:3] +; GCN-IR-NEXT: v_xor_b32_e32 v0, s0, v0 +; GCN-IR-NEXT: v_xor_b32_e32 v1, s1, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s1 +; GCN-IR-NEXT: v_subrev_i32_e32 v0, vcc, s0, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v2, vcc +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %result = sdiv i64 %x, %y store i64 %result, i64 addrspace(1)* %out ret void @@ -272,6 +378,105 @@ define i64 @v_test_sdiv(i64 %x, i64 %y) { ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v8 ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v2, v8, vcc ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_sdiv: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashrrev_i32_e32 v4, 31, v1 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v5, 31, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v4 +; GCN-IR-NEXT: v_mov_b32_e32 v7, v5 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v4, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v4, v0 +; GCN-IR-NEXT: v_xor_b32_e32 v3, v5, v3 +; GCN-IR-NEXT: v_xor_b32_e32 v2, v5, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, v0, v4 +; GCN-IR-NEXT: v_subb_u32_e32 v13, vcc, v1, v4, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v2, v5 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v3, v5, vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[12:13] +; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 +; GCN-IR-NEXT: v_ffbh_u32_e32 v8, v12 +; GCN-IR-NEXT: v_ffbh_u32_e32 v9, v13 +; GCN-IR-NEXT: s_or_b64 s[6:7], vcc, s[4:5] +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 32, v8 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v13 +; GCN-IR-NEXT: v_cndmask_b32_e32 v3, v9, v8, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, v2, v3 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[8:9] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[4:5], 63, v[8:9] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[6:7], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v13, 0, s[6:7] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[6:7], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v3, v12, 0, s[6:7] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[4:5] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB1_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 1, v8 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v9, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v10, vcc, 63, v8 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[8:9] +; GCN-IR-NEXT: v_lshl_b64 v[10:11], v[12:13], v10 +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB1_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[16:17], v[12:13], v2 +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v14, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v15, 0 +; GCN-IR-NEXT: BB1_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[16:17], v[16:17], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v8, 31, v11 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: v_lshl_b64 v[10:11], v[10:11], 1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_or_b32_e32 v16, v16, v8 +; GCN-IR-NEXT: v_or_b32_e32 v11, v15, v11 +; GCN-IR-NEXT: v_or_b32_e32 v10, v14, v10 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v12, v16 +; GCN-IR-NEXT: v_subb_u32_e64 v8, s[4:5], v13, v17, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v14, 31, v8 +; GCN-IR-NEXT: v_and_b32_e32 v8, 1, v14 +; GCN-IR-NEXT: v_and_b32_e32 v15, v14, v1 +; GCN-IR-NEXT: v_and_b32_e32 v14, v14, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v16, vcc, v16, v14 +; GCN-IR-NEXT: v_subb_u32_e32 v17, vcc, v17, v15, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v15, v9 +; GCN-IR-NEXT: v_mov_b32_e32 v14, v8 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB1_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB1_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[10:11], 1 +; GCN-IR-NEXT: v_or_b32_e32 v2, v9, v1 +; GCN-IR-NEXT: v_or_b32_e32 v3, v8, v0 +; GCN-IR-NEXT: BB1_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_xor_b32_e32 v1, v7, v6 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v5, v4 +; GCN-IR-NEXT: v_xor_b32_e32 v2, v2, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v3, v3, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v3, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v2, v1, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = sdiv i64 %x, %y ret i64 %result } @@ -289,23 +494,53 @@ define amdgpu_kernel void @s_test_sdiv24_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-NEXT: s_ashr_i64 s[4:5], s[6:7], 40 ; GCN-NEXT: s_ashr_i64 s[6:7], s[8:9], 40 ; GCN-NEXT: s_xor_b32 s5, s4, s6 -; GCN-NEXT: v_cvt_f32_i32_e32 v0, s6 -; GCN-NEXT: v_cvt_f32_i32_e32 v1, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s6 ; GCN-NEXT: s_ashr_i32 s4, s5, 30 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 ; GCN-NEXT: s_or_b32 s4, s4, 1 -; GCN-NEXT: v_mul_f32_e32 v2, v1, v2 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 ; GCN-NEXT: v_trunc_f32_e32 v2, v2 ; GCN-NEXT: v_mov_b32_e32 v3, s4 -; GCN-NEXT: v_mad_f32 v1, -v2, v0, v1 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, |v0| +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| ; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v2 -; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv24_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_ashr_i64 s[4:5], s[6:7], 40 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[8:9], 40 +; GCN-IR-NEXT: s_xor_b32 s5, s4, s6 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-IR-NEXT: s_ashr_i32 s4, s5, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s4, s4, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s4 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 40 %2 = ashr i64 %y, 40 %result = sdiv i64 %1, %2 @@ -319,31 +554,36 @@ define i64 @v_test_sdiv24_64(i64 %x, i64 %y) { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_lshrrev_b32_e32 v0, 8, v1 ; GCN-NEXT: v_lshrrev_b32_e32 v1, 8, v3 -; GCN-NEXT: v_cvt_f32_u32_e32 v2, v1 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v2 -; GCN-NEXT: v_mul_f32_e32 v2, 0x4f800000, v2 -; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 -; GCN-NEXT: v_mul_hi_u32 v3, v2, v1 -; GCN-NEXT: v_mul_lo_u32 v4, v2, v1 -; GCN-NEXT: v_sub_i32_e32 v5, vcc, 0, v4 -; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3 -; GCN-NEXT: v_cndmask_b32_e32 v3, v4, v5, vcc -; GCN-NEXT: v_mul_hi_u32 v3, v3, v2 -; GCN-NEXT: v_add_i32_e64 v4, s[4:5], v2, v3 -; GCN-NEXT: v_sub_i32_e64 v2, s[4:5], v2, v3 -; GCN-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc -; GCN-NEXT: v_mul_hi_u32 v2, v2, v0 -; GCN-NEXT: v_mul_lo_u32 v3, v2, v1 -; GCN-NEXT: v_add_i32_e32 v4, vcc, 1, v2 -; GCN-NEXT: v_add_i32_e32 v5, vcc, -1, v2 -; GCN-NEXT: v_cmp_ge_u32_e32 vcc, v0, v3 -; GCN-NEXT: v_sub_i32_e64 v0, s[4:5], v0, v3 -; GCN-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v1 -; GCN-NEXT: s_and_b64 s[4:5], s[4:5], vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v4, s[4:5] -; GCN-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc -; GCN-NEXT: v_mov_b32_e32 v1, 0 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 25 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_sdiv24_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v1, 8, v3 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, v1 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 25 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %1 = lshr i64 %x, 40 %2 = lshr i64 %y, 40 %result = sdiv i64 %1, %2 @@ -354,47 +594,56 @@ define amdgpu_kernel void @s_test_sdiv32_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_sdiv32_64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s2, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_ashr_i32 s3, s2, 31 -; GCN-NEXT: s_ashr_i32 s4, s7, 31 -; GCN-NEXT: s_add_i32 s2, s2, s3 -; GCN-NEXT: s_add_i32 s0, s7, s4 -; GCN-NEXT: s_xor_b32 s5, s2, s3 -; GCN-NEXT: s_xor_b32 s2, s0, s4 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s5 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s5 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v1, v0, s5 -; GCN-NEXT: v_add_i32_e32 v2, vcc, 1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, -1, v0 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s2, v1 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s2, v1 -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s5, v1 -; GCN-NEXT: s_and_b64 vcc, vcc, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] -; GCN-NEXT: s_xor_b32 s0, s4, s3 -; GCN-NEXT: v_xor_b32_e32 v0, s0, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s0, v0 +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_xor_b32 s4, s7, s8 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s7 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s8 +; GCN-NEXT: s_ashr_i32 s4, s4, 30 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: s_or_b32 s4, s4, 1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mov_b32_e32 v3, s4 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv32_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_xor_b32 s4, s7, s8 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s7 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s8 +; GCN-IR-NEXT: s_ashr_i32 s4, s4, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s4, s4, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s4 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 32 %2 = ashr i64 %y, 32 %result = sdiv i64 %1, %2 @@ -406,49 +655,62 @@ define amdgpu_kernel void @s_test_sdiv31_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_sdiv31_64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s3, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_ashr_i64 s[0:1], s[6:7], 33 -; GCN-NEXT: s_ashr_i64 s[2:3], s[2:3], 33 -; GCN-NEXT: s_ashr_i32 s3, s2, 31 -; GCN-NEXT: s_ashr_i32 s4, s0, 31 -; GCN-NEXT: s_add_i32 s1, s2, s3 -; GCN-NEXT: s_add_i32 s0, s0, s4 -; GCN-NEXT: s_xor_b32 s5, s1, s3 -; GCN-NEXT: s_xor_b32 s2, s0, s4 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s5 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s5 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v1, v0, s5 -; GCN-NEXT: v_add_i32_e32 v2, vcc, 1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, -1, v0 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s2, v1 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s2, v1 -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s5, v1 -; GCN-NEXT: s_and_b64 vcc, vcc, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] -; GCN-NEXT: s_xor_b32 s0, s4, s3 -; GCN-NEXT: v_xor_b32_e32 v0, s0, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s0, v0 +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_ashr_i64 s[4:5], s[6:7], 33 +; GCN-NEXT: s_ashr_i64 s[6:7], s[8:9], 33 +; GCN-NEXT: s_xor_b32 s5, s4, s6 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-NEXT: s_ashr_i32 s4, s5, 30 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: s_or_b32 s4, s4, 1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mov_b32_e32 v3, s4 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 31 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv31_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_ashr_i64 s[4:5], s[6:7], 33 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[8:9], 33 +; GCN-IR-NEXT: s_xor_b32 s5, s4, s6 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-IR-NEXT: s_ashr_i32 s4, s5, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s4, s4, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s4 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 31 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 33 %2 = ashr i64 %y, 33 %result = sdiv i64 %1, %2 @@ -469,23 +731,53 @@ define amdgpu_kernel void @s_test_sdiv23_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-NEXT: s_ashr_i64 s[4:5], s[6:7], 41 ; GCN-NEXT: s_ashr_i64 s[6:7], s[8:9], 41 ; GCN-NEXT: s_xor_b32 s5, s4, s6 -; GCN-NEXT: v_cvt_f32_i32_e32 v0, s6 -; GCN-NEXT: v_cvt_f32_i32_e32 v1, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s6 ; GCN-NEXT: s_ashr_i32 s4, s5, 30 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 ; GCN-NEXT: s_or_b32 s4, s4, 1 -; GCN-NEXT: v_mul_f32_e32 v2, v1, v2 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 ; GCN-NEXT: v_trunc_f32_e32 v2, v2 ; GCN-NEXT: v_mov_b32_e32 v3, s4 -; GCN-NEXT: v_mad_f32 v1, -v2, v0, v1 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, |v0| +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| ; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v2 -; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 23 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv23_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_ashr_i64 s[4:5], s[6:7], 41 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[8:9], 41 +; GCN-IR-NEXT: s_xor_b32 s5, s4, s6 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-IR-NEXT: s_ashr_i32 s4, s5, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s4, s4, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s4 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 23 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 41 %2 = ashr i64 %y, 41 %result = sdiv i64 %1, %2 @@ -497,49 +789,62 @@ define amdgpu_kernel void @s_test_sdiv25_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_sdiv25_64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s3, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_ashr_i64 s[0:1], s[6:7], 39 -; GCN-NEXT: s_ashr_i64 s[2:3], s[2:3], 39 -; GCN-NEXT: s_ashr_i32 s3, s2, 31 -; GCN-NEXT: s_ashr_i32 s4, s0, 31 -; GCN-NEXT: s_add_i32 s1, s2, s3 -; GCN-NEXT: s_add_i32 s0, s0, s4 -; GCN-NEXT: s_xor_b32 s5, s1, s3 -; GCN-NEXT: s_xor_b32 s2, s0, s4 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s5 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s5 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v1, v0, s5 -; GCN-NEXT: v_add_i32_e32 v2, vcc, 1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, -1, v0 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s2, v1 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s2, v1 -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s5, v1 -; GCN-NEXT: s_and_b64 vcc, vcc, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] -; GCN-NEXT: s_xor_b32 s0, s4, s3 -; GCN-NEXT: v_xor_b32_e32 v0, s0, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s0, v0 +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_ashr_i64 s[4:5], s[6:7], 39 +; GCN-NEXT: s_ashr_i64 s[6:7], s[8:9], 39 +; GCN-NEXT: s_xor_b32 s5, s4, s6 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-NEXT: s_ashr_i32 s4, s5, 30 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: s_or_b32 s4, s4, 1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mov_b32_e32 v3, s4 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 25 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv25_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_ashr_i64 s[4:5], s[6:7], 39 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[8:9], 39 +; GCN-IR-NEXT: s_xor_b32 s5, s4, s6 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-IR-NEXT: s_ashr_i32 s4, s5, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s4, s4, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s4 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 25 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 39 %2 = ashr i64 %y, 39 %result = sdiv i64 %1, %2 @@ -556,10 +861,10 @@ define amdgpu_kernel void @s_test_sdiv24_v2i64(<2 x i64> addrspace(1)* %out, <2 ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_ashr_i64 s[0:1], s[8:9], 40 -; GCN-NEXT: s_ashr_i64 s[2:3], s[10:11], 40 -; GCN-NEXT: s_ashr_i64 s[8:9], s[12:13], 40 -; GCN-NEXT: s_ashr_i64 s[10:11], s[14:15], 40 +; GCN-NEXT: s_ashr_i64 s[0:1], s[10:11], 40 +; GCN-NEXT: s_ashr_i64 s[2:3], s[8:9], 40 +; GCN-NEXT: s_ashr_i64 s[8:9], s[14:15], 40 +; GCN-NEXT: s_ashr_i64 s[10:11], s[12:13], 40 ; GCN-NEXT: s_xor_b32 s1, s2, s10 ; GCN-NEXT: v_cvt_f32_i32_e32 v0, s2 ; GCN-NEXT: v_cvt_f32_i32_e32 v1, s10 @@ -588,12 +893,59 @@ define amdgpu_kernel void @s_test_sdiv24_v2i64(<2 x i64> addrspace(1)* %out, <2 ; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v7, vcc ; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v4 ; GCN-NEXT: v_add_i32_e32 v1, vcc, v1, v5 -; GCN-NEXT: v_bfe_i32 v2, v0, 0, 24 -; GCN-NEXT: v_bfe_i32 v0, v1, 0, 24 -; GCN-NEXT: v_ashrrev_i32_e32 v3, 31, v2 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-NEXT: v_bfe_i32 v2, v1, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v3, 31, v2 ; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv24_v2i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd +; GCN-IR-NEXT: s_load_dwordx4 s[12:15], s[0:1], 0x11 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_ashr_i64 s[0:1], s[10:11], 40 +; GCN-IR-NEXT: s_ashr_i64 s[2:3], s[8:9], 40 +; GCN-IR-NEXT: s_ashr_i64 s[8:9], s[14:15], 40 +; GCN-IR-NEXT: s_ashr_i64 s[10:11], s[12:13], 40 +; GCN-IR-NEXT: s_xor_b32 s1, s2, s10 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s2 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s10 +; GCN-IR-NEXT: s_xor_b32 s2, s0, s8 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v2, s0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v3, s8 +; GCN-IR-NEXT: s_ashr_i32 s0, s1, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v4, v1 +; GCN-IR-NEXT: s_ashr_i32 s1, s2, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v5, v3 +; GCN-IR-NEXT: s_or_b32 s0, s0, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v4, v0, v4 +; GCN-IR-NEXT: s_or_b32 s1, s1, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v5, v2, v5 +; GCN-IR-NEXT: v_trunc_f32_e32 v4, v4 +; GCN-IR-NEXT: v_mov_b32_e32 v6, s0 +; GCN-IR-NEXT: v_trunc_f32_e32 v5, v5 +; GCN-IR-NEXT: v_mov_b32_e32 v7, s1 +; GCN-IR-NEXT: v_mad_f32 v0, -v4, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v4, v4 +; GCN-IR-NEXT: v_mad_f32 v2, -v5, v3, v2 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v5, v5 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v6, vcc +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, |v3| +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, 0, v7, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v4 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v1, v5 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_bfe_i32 v2, v1, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v3, 31, v2 +; GCN-IR-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr <2 x i64> %x, %2 = ashr <2 x i64> %y, %result = sdiv <2 x i64> %1, %2 @@ -636,6 +988,119 @@ define amdgpu_kernel void @s_test_sdiv24_48(i48 addrspace(1)* %out, i48 %x, i48 ; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 ; GCN-NEXT: buffer_store_short v1, off, s[4:7], 0 offset:4 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv24_48: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dword s6, s[0:1], 0xe +; GCN-IR-NEXT: s_load_dword s3, s[0:1], 0xc +; GCN-IR-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s2, s[0:1], 0xb +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xd +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_sext_i32_i16 s3, s3 +; GCN-IR-NEXT: s_sext_i32_i16 s9, s6 +; GCN-IR-NEXT: s_ashr_i64 s[0:1], s[2:3], 24 +; GCN-IR-NEXT: s_ashr_i32 s2, s3, 31 +; GCN-IR-NEXT: s_ashr_i32 s6, s9, 31 +; GCN-IR-NEXT: s_ashr_i64 s[8:9], s[8:9], 24 +; GCN-IR-NEXT: s_mov_b32 s3, s2 +; GCN-IR-NEXT: s_mov_b32 s7, s6 +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[2:3], s[0:1] +; GCN-IR-NEXT: s_sub_u32 s10, s0, s2 +; GCN-IR-NEXT: s_subb_u32 s11, s1, s2 +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[6:7], s[8:9] +; GCN-IR-NEXT: s_flbit_i32_b32 s14, s10 +; GCN-IR-NEXT: s_sub_u32 s8, s0, s6 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[12:13], s[10:11], 0 +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s11 +; GCN-IR-NEXT: s_subb_u32 s9, s1, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s15, s8 +; GCN-IR-NEXT: s_add_i32 s14, s14, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[8:9], 0 +; GCN-IR-NEXT: s_add_i32 s15, s15, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s16, s9 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s14 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s11, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], s[12:13] +; GCN-IR-NEXT: v_mov_b32_e32 v1, s16 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s15 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s9, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v1, v0 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[12:13], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[12:13], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[12:13] +; GCN-IR-NEXT: s_cbranch_vccz BB9_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s11 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s10 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB9_7 +; GCN-IR-NEXT: BB9_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[10:11], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB9_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB9_6 +; GCN-IR-NEXT: BB9_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[10:11], v0 +; GCN-IR-NEXT: s_add_u32 s10, s8, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s11, s9, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB9_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s11 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s10, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s8, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s9, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB9_5 +; GCN-IR-NEXT: BB9_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB9_7: ; %udiv-end +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[6:7], s[2:3] +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, s0, v0 +; GCN-IR-NEXT: v_xor_b32_e32 v1, s1, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s1 +; GCN-IR-NEXT: v_subrev_i32_e32 v0, vcc, s0, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v2, vcc +; GCN-IR-NEXT: buffer_store_short v1, off, s[4:7], 0 offset:4 +; GCN-IR-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i48 %x, 24 %2 = ashr i48 %y, 24 %result = sdiv i48 %1, %2 @@ -769,6 +1234,93 @@ define amdgpu_kernel void @s_test_sdiv_k_num_i64(i64 addrspace(1)* %out, i64 %x) ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v3, v2, vcc ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_ashr_i32 s2, s7, 31 +; GCN-IR-NEXT: s_mov_b32 s3, s2 +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[2:3], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s8, s0, s2 +; GCN-IR-NEXT: s_subb_u32 s9, s1, s2 +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s8 +; GCN-IR-NEXT: s_add_i32 s0, s0, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s1, s9 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s1 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s0 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s9, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 0xffffffc5, v0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[8:9], 0 +; GCN-IR-NEXT: v_addc_u32_e64 v3, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[6:7] +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_cbranch_vccz BB10_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, 24, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB10_7 +; GCN-IR-NEXT: BB10_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], 24, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB10_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB10_6 +; GCN-IR-NEXT: BB10_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], 24, v0 +; GCN-IR-NEXT: s_add_u32 s7, s8, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s10, s9, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB10_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s10 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s7, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s8, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s9, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB10_5 +; GCN-IR-NEXT: BB10_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB10_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: v_xor_b32_e32 v0, s2, v0 +; GCN-IR-NEXT: v_xor_b32_e32 v1, s3, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s3 +; GCN-IR-NEXT: v_subrev_i32_e32 v0, vcc, s2, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v2, vcc +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %result = sdiv i64 24, %x store i64 %result, i64 addrspace(1)* %out ret void @@ -892,6 +1444,91 @@ define i64 @v_test_sdiv_k_num_i64(i64 %x) { ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v3, v2, vcc ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_sdiv_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashrrev_i32_e32 v2, 31, v1 +; GCN-IR-NEXT: s_movk_i32 s4, 0xffc5 +; GCN-IR-NEXT: v_mov_b32_e32 v3, v2 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v2, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v2, vcc +; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 32, v4 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, s4, v4 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_addc_u32_e64 v7, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[6:7] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[6:7] +; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v4, 24, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], s[4:5], -1 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB11_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 1, v6 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, 63, v6 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[4:5], v[6:7] +; GCN-IR-NEXT: v_lshl_b64 v[8:9], 24, v8 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB11_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[14:15], 24, v4 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v12, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v13, 0 +; GCN-IR-NEXT: BB11_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[14:15], v[14:15], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v9 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, -1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, -1, v5, vcc +; GCN-IR-NEXT: v_or_b32_e32 v14, v14, v6 +; GCN-IR-NEXT: v_or_b32_e32 v9, v13, v9 +; GCN-IR-NEXT: v_or_b32_e32 v8, v12, v8 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v10, v14 +; GCN-IR-NEXT: v_subb_u32_e64 v6, s[4:5], v11, v15, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v12 +; GCN-IR-NEXT: v_and_b32_e32 v13, v12, v1 +; GCN-IR-NEXT: v_and_b32_e32 v12, v12, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v14, vcc, v14, v12 +; GCN-IR-NEXT: v_subb_u32_e32 v15, vcc, v15, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v13, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v12, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB11_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB11_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[8:9], 1 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v1 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v0 +; GCN-IR-NEXT: BB11_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_xor_b32_e32 v1, v5, v3 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v4, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = sdiv i64 24, %x ret i64 %result } @@ -1015,6 +1652,96 @@ define i64 @v_test_sdiv_pow2_k_num_i64(i64 %x) { ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v3, v2, vcc ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_sdiv_pow2_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashrrev_i32_e32 v2, 31, v1 +; GCN-IR-NEXT: s_mov_b32 s11, 0 +; GCN-IR-NEXT: s_movk_i32 s4, 0xffd0 +; GCN-IR-NEXT: s_mov_b32 s10, 0x8000 +; GCN-IR-NEXT: v_mov_b32_e32 v3, v2 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v2, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v4, s10 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v2, vcc +; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v6, v1 +; GCN-IR-NEXT: v_add_i32_e32 v5, vcc, 32, v5 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v5, v6, v5, vcc +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, s4, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_addc_u32_e64 v7, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[6:7] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[6:7] +; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v4, v4, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], s[4:5], -1 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB12_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 1, v6 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, 63, v6 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[4:5], v[6:7] +; GCN-IR-NEXT: v_lshl_b64 v[6:7], s[10:11], v8 +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB12_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: s_mov_b32 s5, 0 +; GCN-IR-NEXT: s_mov_b32 s4, 0x8000 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_lshr_b64 v[14:15], s[4:5], v4 +; GCN-IR-NEXT: v_mov_b32_e32 v12, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v13, 0 +; GCN-IR-NEXT: BB12_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[14:15], v[14:15], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v8, 31, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, -1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, -1, v5, vcc +; GCN-IR-NEXT: v_or_b32_e32 v14, v14, v8 +; GCN-IR-NEXT: v_or_b32_e32 v7, v13, v7 +; GCN-IR-NEXT: v_or_b32_e32 v6, v12, v6 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v10, v14 +; GCN-IR-NEXT: v_subb_u32_e64 v8, s[4:5], v11, v15, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v8 +; GCN-IR-NEXT: v_and_b32_e32 v8, 1, v12 +; GCN-IR-NEXT: v_and_b32_e32 v13, v12, v1 +; GCN-IR-NEXT: v_and_b32_e32 v12, v12, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v14, vcc, v14, v12 +; GCN-IR-NEXT: v_subb_u32_e32 v15, vcc, v15, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v13, v9 +; GCN-IR-NEXT: v_mov_b32_e32 v12, v8 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB12_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB12_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[6:7], 1 +; GCN-IR-NEXT: v_or_b32_e32 v5, v9, v1 +; GCN-IR-NEXT: v_or_b32_e32 v4, v8, v0 +; GCN-IR-NEXT: BB12_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_xor_b32_e32 v1, v5, v3 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v4, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = sdiv i64 32768, %x ret i64 %result } @@ -1029,6 +1756,89 @@ define i64 @v_test_sdiv_pow2_k_den_i64(i64 %x) { ; GCN-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; GCN-NEXT: v_ashr_i64 v[0:1], v[0:1], 15 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_sdiv_pow2_k_den_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashrrev_i32_e32 v2, 31, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, v2 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v2, v1 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v9, vcc, v1, v2, vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[8:9] +; GCN-IR-NEXT: v_ffbh_u32_e32 v0, v8 +; GCN-IR-NEXT: v_ffbh_u32_e32 v1, v9 +; GCN-IR-NEXT: v_add_i32_e64 v0, s[4:5], 32, v0 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v9 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v1, v0, s[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 48, v0 +; GCN-IR-NEXT: v_subb_u32_e64 v5, s[4:5], 0, 0, s[4:5] +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] +; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v9, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[4:5], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v8, 0, s[4:5] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB13_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[0:1], v[4:5] +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[8:9], v6 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB13_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[10:11], v[8:9], v0 +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: s_movk_i32 s12, 0x7fff +; GCN-IR-NEXT: BB13_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[10:11], v[10:11], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v12, 0 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v10, v10, v4 +; GCN-IR-NEXT: v_or_b32_e32 v7, v9, v7 +; GCN-IR-NEXT: v_or_b32_e32 v6, v8, v6 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, s12, v10 +; GCN-IR-NEXT: s_or_b64 s[8:9], s[4:5], s[8:9] +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v11, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v8, 0x8000, v8 +; GCN-IR-NEXT: v_sub_i32_e32 v10, vcc, v10, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v11, vcc, v11, v12, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB13_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB13_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[6:7], 1 +; GCN-IR-NEXT: v_or_b32_e32 v1, v5, v1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v4, v0 +; GCN-IR-NEXT: BB13_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v3 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = sdiv i64 %x, 32768 ret i64 %result } @@ -1044,22 +1854,49 @@ define amdgpu_kernel void @s_test_sdiv24_k_num_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: s_mov_b32 s4, s0 ; GCN-NEXT: s_mov_b32 s5, s1 ; GCN-NEXT: s_ashr_i64 s[0:1], s[2:3], 40 +; GCN-NEXT: s_ashr_i32 s1, s0, 30 ; GCN-NEXT: v_cvt_f32_i32_e32 v0, s0 -; GCN-NEXT: s_ashr_i32 s0, s0, 30 +; GCN-NEXT: s_or_b32 s0, s1, 1 ; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v0 -; GCN-NEXT: s_or_b32 s0, s0, 1 ; GCN-NEXT: v_mul_f32_e32 v1, s2, v1 ; GCN-NEXT: v_mov_b32_e32 v2, s0 ; GCN-NEXT: v_trunc_f32_e32 v1, v1 ; GCN-NEXT: v_mad_f32 v3, -v1, v0, s2 +; GCN-NEXT: v_cvt_i32_f32_e32 v1, v1 ; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v0| ; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v1 ; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv24_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s2, 0x41c00000 +; GCN-IR-NEXT: s_mov_b32 s4, s0 +; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_ashr_i64 s[0:1], s[2:3], 40 +; GCN-IR-NEXT: s_ashr_i32 s1, s0, 30 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s0 +; GCN-IR-NEXT: s_or_b32 s0, s1, 1 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, s2, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s0 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v3, -v1, v0, s2 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v0| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %x.shr = ashr i64 %x, 40 %result = sdiv i64 24, %x.shr store i64 %result, i64 addrspace(1)* %out @@ -1077,21 +1914,47 @@ define amdgpu_kernel void @s_test_sdiv24_k_den_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: s_mov_b32 s4, s0 ; GCN-NEXT: s_mov_b32 s5, s1 ; GCN-NEXT: s_ashr_i64 s[0:1], s[2:3], 40 +; GCN-NEXT: s_ashr_i32 s1, s0, 30 ; GCN-NEXT: v_cvt_f32_i32_e32 v0, s0 -; GCN-NEXT: s_ashr_i32 s0, s0, 30 +; GCN-NEXT: s_or_b32 s0, s1, 1 ; GCN-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 -; GCN-NEXT: s_or_b32 s0, s0, 1 ; GCN-NEXT: v_trunc_f32_e32 v1, v1 ; GCN-NEXT: v_mov_b32_e32 v2, s0 ; GCN-NEXT: v_mad_f32 v0, -v1, s2, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v1, v1 ; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s2 ; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v1 ; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_sdiv24_k_den_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s2, 0x46b6fe00 +; GCN-IR-NEXT: s_mov_b32 s4, s0 +; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_ashr_i64 s[0:1], s[2:3], 40 +; GCN-IR-NEXT: s_ashr_i32 s1, s0, 30 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s0 +; GCN-IR-NEXT: s_or_b32 s0, s1, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s0 +; GCN-IR-NEXT: v_mad_f32 v0, -v1, s2, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s2 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %x.shr = ashr i64 %x, 40 %result = sdiv i64 %x.shr, 23423 store i64 %result, i64 addrspace(1)* %out @@ -1104,20 +1967,40 @@ define i64 @v_test_sdiv24_k_num_i64(i64 %x) { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 ; GCN-NEXT: s_mov_b32 s4, 0x41c00000 -; GCN-NEXT: v_cvt_f32_i32_e32 v1, v0 -; GCN-NEXT: v_ashrrev_i32_e32 v0, 30, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 -; GCN-NEXT: v_or_b32_e32 v0, 1, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0 ; GCN-NEXT: v_mul_f32_e32 v2, s4, v2 ; GCN-NEXT: v_trunc_f32_e32 v2, v2 -; GCN-NEXT: v_mad_f32 v3, -v2, v1, s4 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v1| -; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v2 -; GCN-NEXT: v_add_i32_e32 v0, vcc, v1, v0 +; GCN-NEXT: v_mad_f32 v3, -v2, v0, s4 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v0| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v2, v0 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_sdiv24_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 +; GCN-IR-NEXT: s_mov_b32 s4, 0x41c00000 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v2, s4, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v3, -v2, v0, s4 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v0| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v2, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = ashr i64 %x, 40 %result = sdiv i64 24, %x.shr ret i64 %result @@ -1129,20 +2012,40 @@ define i64 @v_test_sdiv24_pow2_k_num_i64(i64 %x) { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 ; GCN-NEXT: s_mov_b32 s4, 0x47000000 -; GCN-NEXT: v_cvt_f32_i32_e32 v1, v0 -; GCN-NEXT: v_ashrrev_i32_e32 v0, 30, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 -; GCN-NEXT: v_or_b32_e32 v0, 1, v0 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0 ; GCN-NEXT: v_mul_f32_e32 v2, s4, v2 ; GCN-NEXT: v_trunc_f32_e32 v2, v2 -; GCN-NEXT: v_mad_f32 v3, -v2, v1, s4 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v1| -; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v0, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v2 -; GCN-NEXT: v_add_i32_e32 v0, vcc, v1, v0 +; GCN-NEXT: v_mad_f32 v3, -v2, v0, s4 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v0| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v2, v0 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_sdiv24_pow2_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 +; GCN-IR-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v2, s4, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v3, -v2, v0, s4 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v0| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v2, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = ashr i64 %x, 40 %result = sdiv i64 32768, %x.shr ret i64 %result @@ -1158,6 +2061,25 @@ define i64 @v_test_sdiv24_pow2_k_den_i64(i64 %x) { ; GCN-NEXT: v_addc_u32_e32 v1, vcc, 0, v1, vcc ; GCN-NEXT: v_ashr_i64 v[0:1], v[0:1], 15 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_sdiv24_pow2_k_den_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 +; GCN-IR-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, 0x38000000, v0 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, s4, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s4 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v2, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = ashr i64 %x, 40 %result = sdiv i64 %x.shr, 32768 ret i64 %result diff --git a/llvm/test/CodeGen/AMDGPU/srem64.ll b/llvm/test/CodeGen/AMDGPU/srem64.ll index 9113f6c..73da5d4 100644 --- a/llvm/test/CodeGen/AMDGPU/srem64.ll +++ b/llvm/test/CodeGen/AMDGPU/srem64.ll @@ -1,5 +1,6 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s +; RUN: llc -march=amdgcn -amdgpu-codegenprepare-expand-div64 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-IR %s define amdgpu_kernel void @s_test_srem(i64 addrspace(1)* %out, i64 %x, i64 %y) { ; GCN-LABEL: s_test_srem: @@ -122,6 +123,106 @@ define amdgpu_kernel void @s_test_srem(i64 addrspace(1)* %out, i64 %x, i64 %y) { ; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_flbit_i32_b32 s10, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s2 +; GCN-IR-NEXT: s_add_i32 s11, s0, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s12, s3 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[2:3], 0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[6:7], 0 +; GCN-IR-NEXT: s_flbit_i32_b32 s13, s7 +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s12 +; GCN-IR-NEXT: s_add_i32 s8, s10, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s13 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s11 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s3, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v2, s8 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v0, v1 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_vccz BB0_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s7 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s6 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB0_7 +; GCN-IR-NEXT: BB0_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[6:7], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB0_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB0_6 +; GCN-IR-NEXT: BB0_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[6:7], v0 +; GCN-IR-NEXT: s_add_u32 s8, s2, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s9, s3, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB0_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s9 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s8, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s2, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s3, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB0_5 +; GCN-IR-NEXT: BB0_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB0_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s11, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s10, -1 +; GCN-IR-NEXT: s_mov_b32 s8, s4 +; GCN-IR-NEXT: s_mov_b32 s9, s5 +; GCN-IR-NEXT: v_mul_lo_u32 v1, s2, v1 +; GCN-IR-NEXT: v_mul_hi_u32 v2, s2, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v3, s3, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v0, s2, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v4, s7 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v1, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s6, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v4, v1, vcc +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-IR-NEXT: s_endpgm %result = urem i64 %x, %y store i64 %result, i64 addrspace(1)* %out ret void @@ -253,6 +354,110 @@ define i64 @v_test_srem(i64 %x, i64 %y) { ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v7 ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v1, v7, vcc ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_srem: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashrrev_i32_e32 v4, 31, v1 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v5, v4 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v4 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v4 +; GCN-IR-NEXT: v_xor_b32_e32 v3, v3, v6 +; GCN-IR-NEXT: v_xor_b32_e32 v2, v2, v6 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v4 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v4, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v2, v6 +; GCN-IR-NEXT: v_subb_u32_e32 v3, vcc, v3, v6, vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v6, v2 +; GCN-IR-NEXT: v_ffbh_u32_e32 v7, v3 +; GCN-IR-NEXT: v_ffbh_u32_e32 v8, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v9, v1 +; GCN-IR-NEXT: s_or_b64 s[6:7], vcc, s[4:5] +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 32, v6 +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, 32, v8 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3 +; GCN-IR-NEXT: v_cndmask_b32_e32 v6, v7, v6, vcc +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v7, v9, v8, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, v6, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[4:5], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[8:9] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[4:5], 63, v[8:9] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[6:7], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v7, v1, 0, s[6:7] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[6:7], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v6, v0, 0, s[6:7] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[4:5] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB1_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 1, v8 +; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, 0, v9, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v10, vcc, 63, v8 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[6:7], v[8:9] +; GCN-IR-NEXT: v_lshl_b64 v[10:11], v[0:1], v10 +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB1_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[16:17], v[0:1], v6 +; GCN-IR-NEXT: v_add_i32_e32 v12, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v13, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v14, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v15, 0 +; GCN-IR-NEXT: BB1_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[16:17], v[16:17], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v8, 31, v11 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: v_lshl_b64 v[10:11], v[10:11], 1 +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, -1, v6 +; GCN-IR-NEXT: v_addc_u32_e32 v7, vcc, -1, v7, vcc +; GCN-IR-NEXT: v_or_b32_e32 v16, v16, v8 +; GCN-IR-NEXT: v_or_b32_e32 v11, v15, v11 +; GCN-IR-NEXT: v_or_b32_e32 v10, v14, v10 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[6:7] +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v12, v16 +; GCN-IR-NEXT: v_subb_u32_e64 v8, s[4:5], v13, v17, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v14, 31, v8 +; GCN-IR-NEXT: v_and_b32_e32 v8, 1, v14 +; GCN-IR-NEXT: v_and_b32_e32 v15, v14, v3 +; GCN-IR-NEXT: v_and_b32_e32 v14, v14, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v16, vcc, v16, v14 +; GCN-IR-NEXT: v_subb_u32_e32 v17, vcc, v17, v15, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v15, v9 +; GCN-IR-NEXT: v_mov_b32_e32 v14, v8 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB1_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB1_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[10:11], 1 +; GCN-IR-NEXT: v_or_b32_e32 v7, v9, v7 +; GCN-IR-NEXT: v_or_b32_e32 v6, v8, v6 +; GCN-IR-NEXT: BB1_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mul_lo_u32 v7, v2, v7 +; GCN-IR-NEXT: v_mul_hi_u32 v8, v2, v6 +; GCN-IR-NEXT: v_mul_lo_u32 v3, v3, v6 +; GCN-IR-NEXT: v_mul_lo_u32 v2, v2, v6 +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, v8, v7 +; GCN-IR-NEXT: v_add_i32_e32 v3, vcc, v6, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v4 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v5 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v4 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v5, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = srem i64 %x, %y ret i64 %result } @@ -270,25 +475,57 @@ define amdgpu_kernel void @s_test_srem23_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-NEXT: s_ashr_i64 s[4:5], s[6:7], 41 ; GCN-NEXT: s_ashr_i64 s[6:7], s[8:9], 41 ; GCN-NEXT: s_xor_b32 s5, s4, s6 -; GCN-NEXT: v_cvt_f32_i32_e32 v0, s6 -; GCN-NEXT: v_cvt_f32_i32_e32 v1, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s6 ; GCN-NEXT: s_ashr_i32 s5, s5, 30 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 ; GCN-NEXT: s_or_b32 s5, s5, 1 -; GCN-NEXT: v_mul_f32_e32 v2, v1, v2 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 ; GCN-NEXT: v_trunc_f32_e32 v2, v2 ; GCN-NEXT: v_mov_b32_e32 v3, s5 -; GCN-NEXT: v_mad_f32 v1, -v2, v0, v1 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, |v0| +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| ; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v2 -; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_mul_lo_u32 v0, v0, s6 ; GCN-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 23 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem23_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_ashr_i64 s[4:5], s[6:7], 41 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[8:9], 41 +; GCN-IR-NEXT: s_xor_b32 s5, s4, s6 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-IR-NEXT: s_ashr_i32 s5, s5, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s5, s5, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s5 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s6 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 23 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 41 %2 = ashr i64 %y, 41 %result = srem i64 %1, %2 @@ -309,25 +546,57 @@ define amdgpu_kernel void @s_test_srem24_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-NEXT: s_ashr_i64 s[4:5], s[6:7], 40 ; GCN-NEXT: s_ashr_i64 s[6:7], s[8:9], 40 ; GCN-NEXT: s_xor_b32 s5, s4, s6 -; GCN-NEXT: v_cvt_f32_i32_e32 v0, s6 -; GCN-NEXT: v_cvt_f32_i32_e32 v1, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s6 ; GCN-NEXT: s_ashr_i32 s5, s5, 30 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v0 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 ; GCN-NEXT: s_or_b32 s5, s5, 1 -; GCN-NEXT: v_mul_f32_e32 v2, v1, v2 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 ; GCN-NEXT: v_trunc_f32_e32 v2, v2 ; GCN-NEXT: v_mov_b32_e32 v3, s5 -; GCN-NEXT: v_mad_f32 v1, -v2, v0, v1 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, |v0| +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| ; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v2 -; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_mul_lo_u32 v0, v0, s6 ; GCN-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem24_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_ashr_i64 s[4:5], s[6:7], 40 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[8:9], 40 +; GCN-IR-NEXT: s_xor_b32 s5, s4, s6 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-IR-NEXT: s_ashr_i32 s5, s5, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s5, s5, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s5 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s6 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 40 %2 = ashr i64 %y, 40 %result = srem i64 %1, %2 @@ -350,15 +619,39 @@ define i64 @v_test_srem24_64(i64 %x, i64 %y) { ; GCN-NEXT: v_mul_f32_e32 v5, v3, v5 ; GCN-NEXT: v_trunc_f32_e32 v5, v5 ; GCN-NEXT: v_mad_f32 v3, -v5, v4, v3 +; GCN-NEXT: v_cvt_i32_f32_e32 v5, v5 ; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v4| ; GCN-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v3, v5 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v3, v2 +; GCN-NEXT: v_add_i32_e32 v2, vcc, v5, v2 ; GCN-NEXT: v_mul_lo_u32 v1, v2, v1 ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v1 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_srem24_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 +; GCN-IR-NEXT: v_ashr_i64 v[1:2], v[2:3], 40 +; GCN-IR-NEXT: v_xor_b32_e32 v2, v0, v1 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v3, v0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v4, v1 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v2, 30, v2 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v5, v4 +; GCN-IR-NEXT: v_or_b32_e32 v2, 1, v2 +; GCN-IR-NEXT: v_mul_f32_e32 v5, v3, v5 +; GCN-IR-NEXT: v_trunc_f32_e32 v5, v5 +; GCN-IR-NEXT: v_mad_f32 v3, -v5, v4, v3 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v5, v5 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v4| +; GCN-IR-NEXT: v_cndmask_b32_e32 v2, 0, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, v5, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v1, v2, v1 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v1 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %1 = ashr i64 %x, 40 %2 = ashr i64 %y, 40 %result = srem i64 %1, %2 @@ -369,48 +662,66 @@ define amdgpu_kernel void @s_test_srem25_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_srem25_64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s3, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_ashr_i64 s[0:1], s[6:7], 39 -; GCN-NEXT: s_ashr_i64 s[2:3], s[2:3], 39 -; GCN-NEXT: s_ashr_i32 s1, s2, 31 -; GCN-NEXT: s_ashr_i32 s4, s0, 31 -; GCN-NEXT: s_add_i32 s2, s2, s1 -; GCN-NEXT: s_add_i32 s0, s0, s4 -; GCN-NEXT: s_xor_b32 s5, s2, s1 -; GCN-NEXT: s_xor_b32 s2, s0, s4 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s5 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s5 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v0, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s2, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, s5, v1 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s2, v0 -; GCN-NEXT: v_cmp_le_u32_e64 s[2:3], s5, v1 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s5, v1 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] -; GCN-NEXT: v_xor_b32_e32 v0, s4, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_ashr_i64 s[4:5], s[6:7], 39 +; GCN-NEXT: s_ashr_i64 s[6:7], s[8:9], 39 +; GCN-NEXT: s_xor_b32 s5, s4, s6 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-NEXT: s_ashr_i32 s5, s5, 30 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: s_or_b32 s5, s5, 1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mov_b32_e32 v3, s5 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-NEXT: v_mul_lo_u32 v0, v0, s6 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 25 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem25_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_ashr_i64 s[4:5], s[6:7], 39 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[8:9], 39 +; GCN-IR-NEXT: s_xor_b32 s5, s4, s6 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-IR-NEXT: s_ashr_i32 s5, s5, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s5, s5, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s5 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s6 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 25 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 39 %2 = ashr i64 %y, 39 %result = srem i64 %1, %2 @@ -422,48 +733,66 @@ define amdgpu_kernel void @s_test_srem31_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_srem31_64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s3, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_ashr_i64 s[0:1], s[6:7], 33 -; GCN-NEXT: s_ashr_i64 s[2:3], s[2:3], 33 -; GCN-NEXT: s_ashr_i32 s1, s2, 31 -; GCN-NEXT: s_ashr_i32 s4, s0, 31 -; GCN-NEXT: s_add_i32 s2, s2, s1 -; GCN-NEXT: s_add_i32 s0, s0, s4 -; GCN-NEXT: s_xor_b32 s5, s2, s1 -; GCN-NEXT: s_xor_b32 s2, s0, s4 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s5 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s5 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v0, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s2, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, s5, v1 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s2, v0 -; GCN-NEXT: v_cmp_le_u32_e64 s[2:3], s5, v1 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s5, v1 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] -; GCN-NEXT: v_xor_b32_e32 v0, s4, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_ashr_i64 s[4:5], s[6:7], 33 +; GCN-NEXT: s_ashr_i64 s[6:7], s[8:9], 33 +; GCN-NEXT: s_xor_b32 s5, s4, s6 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-NEXT: s_ashr_i32 s5, s5, 30 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: s_or_b32 s5, s5, 1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mov_b32_e32 v3, s5 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-NEXT: v_mul_lo_u32 v0, v0, s6 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: v_bfe_i32 v0, v0, 0, 31 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem31_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s9, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_ashr_i64 s[4:5], s[6:7], 33 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[8:9], 33 +; GCN-IR-NEXT: s_xor_b32 s5, s4, s6 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s6 +; GCN-IR-NEXT: s_ashr_i32 s5, s5, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s5, s5, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s5 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s6 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 31 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 33 %2 = ashr i64 %y, 33 %result = srem i64 %1, %2 @@ -476,46 +805,60 @@ define amdgpu_kernel void @s_test_srem32_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_srem32_64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s2, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_ashr_i32 s0, s2, 31 -; GCN-NEXT: s_ashr_i32 s4, s7, 31 -; GCN-NEXT: s_add_i32 s2, s2, s0 -; GCN-NEXT: s_add_i32 s1, s7, s4 -; GCN-NEXT: s_xor_b32 s5, s2, s0 -; GCN-NEXT: s_xor_b32 s2, s1, s4 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s5 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s5 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v0, v0, s5 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s2, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, s5, v1 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s2, v0 -; GCN-NEXT: v_cmp_le_u32_e64 s[2:3], s5, v1 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s5, v1 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] -; GCN-NEXT: v_xor_b32_e32 v0, s4, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_xor_b32 s4, s7, s8 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s7 +; GCN-NEXT: v_cvt_f32_i32_e32 v1, s8 +; GCN-NEXT: s_ashr_i32 s4, s4, 30 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: s_or_b32 s4, s4, 1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mov_b32_e32 v3, s4 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-NEXT: v_mul_lo_u32 v0, v0, s8 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s7, v0 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem32_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_xor_b32 s4, s7, s8 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s7 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v1, s8 +; GCN-IR-NEXT: s_ashr_i32 s4, s4, 30 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: s_or_b32 s4, s4, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s4 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, |v1| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s8 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s7, v0 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 32 %2 = ashr i64 %y, 32 %result = srem i64 %1, %2 @@ -662,6 +1005,121 @@ define amdgpu_kernel void @s_test_srem33_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v2, v3, vcc ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem33_64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0xd +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_ashr_i64 s[0:1], s[6:7], 31 +; GCN-IR-NEXT: s_ashr_i32 s2, s7, 31 +; GCN-IR-NEXT: s_ashr_i32 s6, s9, 31 +; GCN-IR-NEXT: s_ashr_i64 s[8:9], s[8:9], 31 +; GCN-IR-NEXT: s_mov_b32 s3, s2 +; GCN-IR-NEXT: s_mov_b32 s7, s6 +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[0:1], s[2:3] +; GCN-IR-NEXT: s_xor_b64 s[10:11], s[8:9], s[6:7] +; GCN-IR-NEXT: s_sub_u32 s8, s0, s2 +; GCN-IR-NEXT: s_subb_u32 s9, s1, s2 +; GCN-IR-NEXT: s_flbit_i32_b32 s7, s8 +; GCN-IR-NEXT: s_sub_u32 s10, s10, s6 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[8:9], 0 +; GCN-IR-NEXT: s_flbit_i32_b32 s12, s9 +; GCN-IR-NEXT: s_subb_u32 s11, s11, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s13, s10 +; GCN-IR-NEXT: s_add_i32 s14, s7, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s12 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[6:7], s[10:11], 0 +; GCN-IR-NEXT: s_add_i32 s12, s13, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s13, s11 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s14 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s9, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: s_or_b64 s[0:1], s[6:7], s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v1, s13 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s12 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s11, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v1, v0 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[6:7], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[6:7] +; GCN-IR-NEXT: s_cbranch_vccz BB8_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s9 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s8 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB8_7 +; GCN-IR-NEXT: BB8_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[8:9], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB8_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB8_6 +; GCN-IR-NEXT: BB8_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[8:9], v0 +; GCN-IR-NEXT: s_add_u32 s6, s10, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s7, s11, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB8_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s7 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s6, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s10, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s11, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB8_5 +; GCN-IR-NEXT: BB8_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB8_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: v_mul_lo_u32 v1, s10, v1 +; GCN-IR-NEXT: v_mul_hi_u32 v2, s10, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v3, s11, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v0, s10, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v4, s9 +; GCN-IR-NEXT: v_mov_b32_e32 v5, s3 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v1, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s8, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v4, v1, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, s2, v0 +; GCN-IR-NEXT: v_xor_b32_e32 v1, s3, v1 +; GCN-IR-NEXT: v_subrev_i32_e32 v0, vcc, s2, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v5, vcc +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i64 %x, 31 %2 = ashr i64 %y, 31 %result = srem i64 %1, %2 @@ -706,6 +1164,127 @@ define amdgpu_kernel void @s_test_srem24_48(i48 addrspace(1)* %out, i48 %x, i48 ; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0 ; GCN-NEXT: buffer_store_short v1, off, s[4:7], 0 offset:4 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem24_48: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dword s7, s[0:1], 0xe +; GCN-IR-NEXT: s_load_dword s3, s[0:1], 0xc +; GCN-IR-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s2, s[0:1], 0xb +; GCN-IR-NEXT: s_load_dword s6, s[0:1], 0xd +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_sext_i32_i16 s3, s3 +; GCN-IR-NEXT: s_sext_i32_i16 s7, s7 +; GCN-IR-NEXT: s_ashr_i64 s[0:1], s[2:3], 24 +; GCN-IR-NEXT: s_ashr_i32 s2, s3, 31 +; GCN-IR-NEXT: s_ashr_i32 s10, s7, 31 +; GCN-IR-NEXT: s_ashr_i64 s[6:7], s[6:7], 24 +; GCN-IR-NEXT: s_mov_b32 s3, s2 +; GCN-IR-NEXT: s_mov_b32 s11, s10 +; GCN-IR-NEXT: s_xor_b64 s[0:1], s[0:1], s[2:3] +; GCN-IR-NEXT: s_xor_b64 s[6:7], s[6:7], s[10:11] +; GCN-IR-NEXT: s_sub_u32 s8, s0, s2 +; GCN-IR-NEXT: s_subb_u32 s9, s1, s2 +; GCN-IR-NEXT: s_flbit_i32_b32 s11, s8 +; GCN-IR-NEXT: s_sub_u32 s6, s6, s10 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[8:9], 0 +; GCN-IR-NEXT: s_flbit_i32_b32 s12, s9 +; GCN-IR-NEXT: s_subb_u32 s7, s7, s10 +; GCN-IR-NEXT: s_flbit_i32_b32 s13, s6 +; GCN-IR-NEXT: s_add_i32 s14, s11, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s12 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[10:11], s[6:7], 0 +; GCN-IR-NEXT: s_add_i32 s12, s13, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s13, s7 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s14 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s9, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: s_or_b64 s[0:1], s[10:11], s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v1, s13 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s12 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v1, v0 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[10:11], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[10:11], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[10:11] +; GCN-IR-NEXT: s_cbranch_vccz BB9_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s9 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s8 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB9_7 +; GCN-IR-NEXT: BB9_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[8:9], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB9_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB9_6 +; GCN-IR-NEXT: BB9_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[8:9], v0 +; GCN-IR-NEXT: s_add_u32 s10, s6, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s11, s7, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB9_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s11 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s10, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s6, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s7, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB9_5 +; GCN-IR-NEXT: BB9_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB9_7: ; %udiv-end +; GCN-IR-NEXT: v_mul_lo_u32 v1, s6, v1 +; GCN-IR-NEXT: v_mul_hi_u32 v2, s6, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v3, s7, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v0, s6, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v4, s9 +; GCN-IR-NEXT: v_mov_b32_e32 v5, s3 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v1, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s8, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v4, v1, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v0, s2, v0 +; GCN-IR-NEXT: v_xor_b32_e32 v1, s3, v1 +; GCN-IR-NEXT: v_subrev_i32_e32 v0, vcc, s2, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v5, vcc +; GCN-IR-NEXT: buffer_store_short v1, off, s[4:7], 0 offset:4 +; GCN-IR-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %1 = ashr i48 %x, 24 %2 = ashr i48 %y, 24 %result = srem i48 %1, %2 @@ -833,6 +1412,96 @@ define amdgpu_kernel void @s_test_srem_k_num_i64(i64 addrspace(1)* %out, i64 %x) ; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_ashr_i32 s0, s7, 31 +; GCN-IR-NEXT: s_mov_b32 s1, s0 +; GCN-IR-NEXT: s_xor_b64 s[2:3], s[6:7], s[0:1] +; GCN-IR-NEXT: s_sub_u32 s2, s2, s0 +; GCN-IR-NEXT: s_subb_u32 s3, s3, s0 +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s2 +; GCN-IR-NEXT: s_add_i32 s0, s0, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s1, s3 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s1 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s0 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s3, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 0xffffffc5, v0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[2:3], 0 +; GCN-IR-NEXT: v_addc_u32_e64 v3, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[6:7] +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_cbranch_vccz BB10_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, 24, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB10_7 +; GCN-IR-NEXT: BB10_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], 24, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB10_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB10_6 +; GCN-IR-NEXT: BB10_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], 24, v0 +; GCN-IR-NEXT: s_add_u32 s7, s2, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s8, s3, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB10_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s8 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s7, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s2, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s3, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB10_5 +; GCN-IR-NEXT: BB10_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB10_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: v_mul_lo_u32 v1, s2, v1 +; GCN-IR-NEXT: v_mul_hi_u32 v2, s2, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v3, s3, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v0, s2, v0 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v1, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, 0, v1, vcc +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %result = srem i64 24, %x store i64 %result, i64 addrspace(1)* %out ret void @@ -951,6 +1620,94 @@ define i64 @v_test_srem_k_num_i64(i64 %x) { ; GCN-NEXT: v_cndmask_b32_e32 v1, v4, v7, vcc ; GCN-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5] ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_srem_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashrrev_i32_e32 v2, 31, v1 +; GCN-IR-NEXT: s_movk_i32 s4, 0xffc5 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v2 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v2, vcc +; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, s4, v2 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_addc_u32_e64 v5, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] +; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, 24, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], s[4:5], -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB11_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[4:5] +; GCN-IR-NEXT: v_lshl_b64 v[6:7], 24, v6 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB11_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[12:13], 24, v2 +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: BB11_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[12:13], v[12:13], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_or_b32_e32 v12, v12, v4 +; GCN-IR-NEXT: v_or_b32_e32 v7, v11, v7 +; GCN-IR-NEXT: v_or_b32_e32 v6, v10, v6 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], v8, v12 +; GCN-IR-NEXT: v_subb_u32_e64 v4, s[4:5], v9, v13, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v4 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v10 +; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 +; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, v12, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v13, vcc, v13, v11, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v11, v5 +; GCN-IR-NEXT: v_mov_b32_e32 v10, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB11_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB11_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[6:7], 1 +; GCN-IR-NEXT: v_or_b32_e32 v3, v5, v3 +; GCN-IR-NEXT: v_or_b32_e32 v2, v4, v2 +; GCN-IR-NEXT: BB11_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mul_lo_u32 v3, v0, v3 +; GCN-IR-NEXT: v_mul_hi_u32 v4, v0, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v1, v1, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, v2 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, v4, v3 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, 0, v1, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = srem i64 24, %x ret i64 %result } @@ -1069,6 +1826,99 @@ define i64 @v_test_srem_pow2_k_num_i64(i64 %x) { ; GCN-NEXT: v_cndmask_b32_e32 v1, v4, v7, vcc ; GCN-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5] ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_srem_pow2_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashrrev_i32_e32 v2, 31, v1 +; GCN-IR-NEXT: s_mov_b32 s11, 0 +; GCN-IR-NEXT: s_movk_i32 s4, 0xffd0 +; GCN-IR-NEXT: s_mov_b32 s10, 0x8000 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v2 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s10 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v2, vcc +; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v2, v4, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, s4, v2 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_addc_u32_e64 v5, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] +; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v3, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], s[4:5], -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB12_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[4:5] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[10:11], v6 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB12_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: s_mov_b32 s5, 0 +; GCN-IR-NEXT: s_mov_b32 s4, 0x8000 +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_lshr_b64 v[12:13], s[4:5], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: BB12_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[12:13], v[12:13], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v5 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_or_b32_e32 v12, v12, v6 +; GCN-IR-NEXT: v_or_b32_e32 v5, v11, v5 +; GCN-IR-NEXT: v_or_b32_e32 v4, v10, v4 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v8, v12 +; GCN-IR-NEXT: v_subb_u32_e64 v6, s[4:5], v9, v13, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v10 +; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 +; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, v12, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v13, vcc, v13, v11, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v11, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v10, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB12_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB12_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v3, v7, v3 +; GCN-IR-NEXT: v_or_b32_e32 v2, v6, v2 +; GCN-IR-NEXT: BB12_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mul_lo_u32 v3, v0, v3 +; GCN-IR-NEXT: v_mul_hi_u32 v4, v0, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v1, v1, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, v2 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, v4, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 0x8000, v0 +; GCN-IR-NEXT: v_add_i32_e64 v1, s[4:5], v2, v1 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, 0, v1, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = srem i64 32768, %x ret i64 %result } @@ -1085,6 +1935,92 @@ define i64 @v_test_srem_pow2_k_den_i64(i64 %x) { ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_srem_pow2_k_den_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashrrev_i32_e32 v2, 31, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, v2 +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v2 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v2, vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v1 +; GCN-IR-NEXT: v_add_i32_e64 v4, s[4:5], 32, v4 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v4, v5, v4, s[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], 48, v4 +; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, s[4:5] +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[6:7] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[6:7] +; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] +; GCN-IR-NEXT: v_cndmask_b32_e64 v5, v1, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[4:5], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v4, v0, 0, s[4:5] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB13_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 1, v6 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, 63, v6 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[4:5], v[6:7] +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[0:1], v8 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB13_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[12:13], v[0:1], v4 +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: s_movk_i32 s12, 0x7fff +; GCN-IR-NEXT: BB13_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[12:13], v[12:13], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v9 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v14, 0 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, -1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, -1, v5, vcc +; GCN-IR-NEXT: v_or_b32_e32 v12, v12, v6 +; GCN-IR-NEXT: v_or_b32_e32 v9, v11, v9 +; GCN-IR-NEXT: v_or_b32_e32 v8, v10, v8 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[4:5] +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, s12, v12 +; GCN-IR-NEXT: s_or_b64 s[8:9], s[4:5], s[8:9] +; GCN-IR-NEXT: v_subb_u32_e32 v6, vcc, 0, v13, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v10 +; GCN-IR-NEXT: v_and_b32_e32 v10, 0x8000, v10 +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, v12, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v13, vcc, v13, v14, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v11, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v10, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB13_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB13_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[8:9], 1 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: BB13_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 15 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v4 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v5, vcc +; GCN-IR-NEXT: v_xor_b32_e32 v1, v1, v3 +; GCN-IR-NEXT: v_xor_b32_e32 v0, v0, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = srem i64 %x, 32768 ret i64 %result } @@ -1100,17 +2036,17 @@ define amdgpu_kernel void @s_test_srem24_k_num_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: s_mov_b32 s4, s0 ; GCN-NEXT: s_mov_b32 s5, s1 ; GCN-NEXT: s_ashr_i64 s[0:1], s[2:3], 40 -; GCN-NEXT: v_cvt_f32_i32_e32 v0, s0 ; GCN-NEXT: s_ashr_i32 s1, s0, 30 -; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s0 ; GCN-NEXT: s_or_b32 s1, s1, 1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v0 ; GCN-NEXT: v_mul_f32_e32 v1, s2, v1 ; GCN-NEXT: v_mov_b32_e32 v2, s1 ; GCN-NEXT: v_trunc_f32_e32 v1, v1 ; GCN-NEXT: v_mad_f32 v3, -v1, v0, s2 +; GCN-NEXT: v_cvt_i32_f32_e32 v1, v1 ; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v0| ; GCN-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v1, v1 ; GCN-NEXT: v_add_i32_e32 v0, vcc, v0, v1 ; GCN-NEXT: v_mul_lo_u32 v0, v0, s0 ; GCN-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 @@ -1118,6 +2054,35 @@ define amdgpu_kernel void @s_test_srem24_k_num_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem24_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s2, 0x41c00000 +; GCN-IR-NEXT: s_mov_b32 s4, s0 +; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_ashr_i64 s[0:1], s[2:3], 40 +; GCN-IR-NEXT: s_ashr_i32 s1, s0, 30 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s0 +; GCN-IR-NEXT: s_or_b32 s1, s1, 1 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, s2, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s1 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v3, -v1, v0, s2 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, |v0| +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s0 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %x.shr = ashr i64 %x, 40 %result = srem i64 24, %x.shr store i64 %result, i64 addrspace(1)* %out @@ -1135,10 +2100,10 @@ define amdgpu_kernel void @s_test_srem24_k_den_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: s_mov_b32 s4, s0 ; GCN-NEXT: s_mov_b32 s5, s1 ; GCN-NEXT: s_ashr_i64 s[0:1], s[2:3], 40 -; GCN-NEXT: v_cvt_f32_i32_e32 v0, s0 ; GCN-NEXT: s_ashr_i32 s1, s0, 30 -; GCN-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v0, s0 ; GCN-NEXT: s_or_b32 s1, s1, 1 +; GCN-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 ; GCN-NEXT: v_trunc_f32_e32 v1, v1 ; GCN-NEXT: v_mov_b32_e32 v2, s1 ; GCN-NEXT: v_mad_f32 v0, -v1, s2, v0 @@ -1153,6 +2118,35 @@ define amdgpu_kernel void @s_test_srem24_k_den_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_srem24_k_den_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s2, 0x46b6fe00 +; GCN-IR-NEXT: s_mov_b32 s4, s0 +; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_ashr_i64 s[0:1], s[2:3], 40 +; GCN-IR-NEXT: s_ashr_i32 s1, s0, 30 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v0, s0 +; GCN-IR-NEXT: s_or_b32 s1, s1, 1 +; GCN-IR-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s1 +; GCN-IR-NEXT: v_mad_f32 v0, -v1, s2, v0 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s2 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, 0, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, v0, v1 +; GCN-IR-NEXT: s_movk_i32 s1, 0x5b7f +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s1 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s0, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %x.shr = ashr i64 %x, 40 %result = srem i64 %x.shr, 23423 store i64 %result, i64 addrspace(1)* %out @@ -1165,22 +2159,44 @@ define i64 @v_test_srem24_k_num_i64(i64 %x) { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 ; GCN-NEXT: s_mov_b32 s4, 0x41c00000 -; GCN-NEXT: v_cvt_f32_i32_e32 v1, v0 -; GCN-NEXT: v_ashrrev_i32_e32 v2, 30, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v3, v1 -; GCN-NEXT: v_or_b32_e32 v2, 1, v2 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v2, v0 +; GCN-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v3, v2 ; GCN-NEXT: v_mul_f32_e32 v3, s4, v3 ; GCN-NEXT: v_trunc_f32_e32 v3, v3 -; GCN-NEXT: v_mad_f32 v4, -v3, v1, s4 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, |v1| -; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v2, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v2, v3 -; GCN-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-NEXT: v_mad_f32 v4, -v3, v2, s4 +; GCN-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, |v2| +; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc +; GCN-NEXT: v_add_i32_e32 v1, vcc, v3, v1 ; GCN-NEXT: v_mul_lo_u32 v0, v1, v0 ; GCN-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_srem24_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 +; GCN-IR-NEXT: s_mov_b32 s4, 0x41c00000 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v3, v2 +; GCN-IR-NEXT: v_mul_f32_e32 v3, s4, v3 +; GCN-IR-NEXT: v_trunc_f32_e32 v3, v3 +; GCN-IR-NEXT: v_mad_f32 v4, -v3, v2, s4 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, |v2| +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v3, v1 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v1, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = ashr i64 %x, 40 %result = srem i64 24, %x.shr ret i64 %result @@ -1192,22 +2208,44 @@ define i64 @v_test_srem24_pow2_k_num_i64(i64 %x) { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 ; GCN-NEXT: s_mov_b32 s4, 0x47000000 -; GCN-NEXT: v_cvt_f32_i32_e32 v1, v0 -; GCN-NEXT: v_ashrrev_i32_e32 v2, 30, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v3, v1 -; GCN-NEXT: v_or_b32_e32 v2, 1, v2 +; GCN-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-NEXT: v_cvt_f32_i32_e32 v2, v0 +; GCN-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v3, v2 ; GCN-NEXT: v_mul_f32_e32 v3, s4, v3 ; GCN-NEXT: v_trunc_f32_e32 v3, v3 -; GCN-NEXT: v_mad_f32 v4, -v3, v1, s4 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, |v1| -; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v2, vcc -; GCN-NEXT: v_cvt_i32_f32_e32 v2, v3 -; GCN-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-NEXT: v_mad_f32 v4, -v3, v2, s4 +; GCN-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, |v2| +; GCN-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc +; GCN-NEXT: v_add_i32_e32 v1, vcc, v3, v1 ; GCN-NEXT: v_mul_lo_u32 v0, v1, v0 ; GCN-NEXT: v_sub_i32_e32 v0, vcc, 0x8000, v0 ; GCN-NEXT: v_bfe_i32 v0, v0, 0, 24 ; GCN-NEXT: v_ashrrev_i32_e32 v1, 31, v0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_srem24_pow2_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 +; GCN-IR-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v3, v2 +; GCN-IR-NEXT: v_mul_f32_e32 v3, s4, v3 +; GCN-IR-NEXT: v_trunc_f32_e32 v3, v3 +; GCN-IR-NEXT: v_mad_f32 v4, -v3, v2, s4 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v4|, |v2| +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v3, v1 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v1, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 0x8000, v0 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = ashr i64 %x, 40 %result = srem i64 32768, %x.shr ret i64 %result @@ -1225,6 +2263,27 @@ define i64 @v_test_srem24_pow2_k_den_i64(i64 %x) { ; GCN-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 ; GCN-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_srem24_pow2_k_den_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_ashr_i64 v[0:1], v[0:1], 40 +; GCN-IR-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 30, v0 +; GCN-IR-NEXT: v_cvt_f32_i32_e32 v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, 1, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v3, 0x38000000, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v3, v3 +; GCN-IR-NEXT: v_mad_f32 v2, -v3, s4, v2 +; GCN-IR-NEXT: v_cvt_i32_f32_e32 v3, v3 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, s4 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, 0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v3, v1 +; GCN-IR-NEXT: v_lshlrev_b32_e32 v1, 15, v1 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v1 +; GCN-IR-NEXT: v_bfe_i32 v0, v0, 0, 24 +; GCN-IR-NEXT: v_ashrrev_i32_e32 v1, 31, v0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = ashr i64 %x, 40 %result = srem i64 %x.shr, 32768 ret i64 %result diff --git a/llvm/test/CodeGen/AMDGPU/udiv64.ll b/llvm/test/CodeGen/AMDGPU/udiv64.ll index c4795f1..375fdc6 100644 --- a/llvm/test/CodeGen/AMDGPU/udiv64.ll +++ b/llvm/test/CodeGen/AMDGPU/udiv64.ll @@ -1,5 +1,6 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s +; RUN: llc -march=amdgcn -amdgpu-codegenprepare-expand-div64 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-IR %s define amdgpu_kernel void @s_test_udiv_i64(i64 addrspace(1)* %out, i64 %x, i64 %y) { ; GCN-LABEL: s_test_udiv_i64: @@ -123,6 +124,95 @@ define amdgpu_kernel void @s_test_udiv_i64(i64 addrspace(1)* %out, i64 %x, i64 % ; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_flbit_i32_b32 s10, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s2 +; GCN-IR-NEXT: s_add_i32 s11, s0, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s12, s3 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[2:3], 0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[6:7], 0 +; GCN-IR-NEXT: s_flbit_i32_b32 s13, s7 +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s12 +; GCN-IR-NEXT: s_add_i32 s8, s10, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s13 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s11 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s3, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v2, s8 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v0, v1 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_vccz BB0_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s7 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s6 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB0_7 +; GCN-IR-NEXT: BB0_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[6:7], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB0_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB0_6 +; GCN-IR-NEXT: BB0_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[6:7], v0 +; GCN-IR-NEXT: s_add_u32 s6, s2, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s7, s3, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB0_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s7 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s6, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s2, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s3, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB0_5 +; GCN-IR-NEXT: BB0_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB0_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %result = udiv i64 %x, %y store i64 %result, i64 addrspace(1)* %out ret void @@ -241,6 +331,89 @@ define i64 @v_test_udiv_i64(i64 %x, i64 %y) { ; GCN-NEXT: v_cndmask_b32_e32 v1, v13, v11, vcc ; GCN-NEXT: v_cndmask_b32_e64 v1, v5, v1, s[4:5] ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_udiv_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v2 +; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v3 +; GCN-IR-NEXT: v_ffbh_u32_e32 v6, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v7, v1 +; GCN-IR-NEXT: s_or_b64 s[6:7], vcc, s[4:5] +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 32, v4 +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 32, v6 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3 +; GCN-IR-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v5, v7, v6, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v4, v5 +; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[6:7] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[4:5], 63, v[6:7] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[6:7], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v4, v1, 0, s[6:7] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[6:7], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v5, v0, 0, s[6:7] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[4:5] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB1_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 1, v6 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, 63, v6 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[4:5], v[6:7] +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[0:1], v8 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB1_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[12:13], v[0:1], v4 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: BB1_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[12:13], v[12:13], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v9 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, -1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, -1, v5, vcc +; GCN-IR-NEXT: v_or_b32_e32 v12, v12, v6 +; GCN-IR-NEXT: v_or_b32_e32 v9, v11, v9 +; GCN-IR-NEXT: v_or_b32_e32 v8, v10, v8 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v0, v12 +; GCN-IR-NEXT: v_subb_u32_e64 v6, s[4:5], v1, v13, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v10 +; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v3 +; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, v12, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v13, vcc, v13, v11, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v11, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v10, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB1_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB1_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[8:9], 1 +; GCN-IR-NEXT: v_or_b32_e32 v4, v7, v1 +; GCN-IR-NEXT: v_or_b32_e32 v5, v6, v0 +; GCN-IR-NEXT: BB1_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mov_b32_e32 v0, v5 +; GCN-IR-NEXT: v_mov_b32_e32 v1, v4 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = udiv i64 %x, %y ret i64 %result } @@ -249,40 +422,52 @@ define amdgpu_kernel void @s_test_udiv24_64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_udiv24_64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s2, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_lshr_b32 s3, s7, 8 -; GCN-NEXT: s_lshr_b32 s2, s2, 8 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s2 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s2 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s3 -; GCN-NEXT: v_mul_lo_u32 v1, v0, s2 -; GCN-NEXT: v_add_i32_e32 v2, vcc, 1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, -1, v0 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s3, v1 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s3, v1 -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s2, v1 -; GCN-NEXT: s_and_b64 vcc, vcc, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_lshr_b32 s4, s7, 8 +; GCN-NEXT: s_lshr_b32 s5, s8, 8 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv24_64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_lshr_b32 s4, s7, 8 +; GCN-IR-NEXT: s_lshr_b32 s5, s8, 8 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr i64 %x, 40 %2 = lshr i64 %y, 40 %result = udiv i64 %1, %2 @@ -296,31 +481,36 @@ define i64 @v_test_udiv24_i64(i64 %x, i64 %y) { ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_lshrrev_b32_e32 v0, 8, v1 ; GCN-NEXT: v_lshrrev_b32_e32 v1, 8, v3 -; GCN-NEXT: v_cvt_f32_u32_e32 v2, v1 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v2 -; GCN-NEXT: v_mul_f32_e32 v2, 0x4f800000, v2 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GCN-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 ; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 -; GCN-NEXT: v_mul_hi_u32 v3, v2, v1 -; GCN-NEXT: v_mul_lo_u32 v4, v2, v1 -; GCN-NEXT: v_sub_i32_e32 v5, vcc, 0, v4 -; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3 -; GCN-NEXT: v_cndmask_b32_e32 v3, v4, v5, vcc -; GCN-NEXT: v_mul_hi_u32 v3, v3, v2 -; GCN-NEXT: v_add_i32_e64 v4, s[4:5], v2, v3 -; GCN-NEXT: v_sub_i32_e64 v2, s[4:5], v2, v3 -; GCN-NEXT: v_cndmask_b32_e32 v2, v2, v4, vcc -; GCN-NEXT: v_mul_hi_u32 v2, v2, v0 -; GCN-NEXT: v_mul_lo_u32 v3, v2, v1 -; GCN-NEXT: v_add_i32_e32 v4, vcc, 1, v2 -; GCN-NEXT: v_add_i32_e32 v5, vcc, -1, v2 -; GCN-NEXT: v_cmp_ge_u32_e32 vcc, v0, v3 -; GCN-NEXT: v_sub_i32_e64 v0, s[4:5], v0, v3 -; GCN-NEXT: v_cmp_ge_u32_e64 s[4:5], v0, v1 -; GCN-NEXT: s_and_b64 s[4:5], s[4:5], vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v4, s[4:5] -; GCN-NEXT: v_cndmask_b32_e32 v0, v5, v0, vcc +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_udiv24_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v1, 8, v3 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, v1 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %1 = lshr i64 %x, 40 %2 = lshr i64 %y, 40 %result = udiv i64 %1, %2 @@ -331,38 +521,46 @@ define amdgpu_kernel void @s_test_udiv32_i64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_udiv32_i64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s2, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s2 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s2 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s7 -; GCN-NEXT: v_mul_lo_u32 v1, v0, s2 -; GCN-NEXT: v_add_i32_e32 v2, vcc, 1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, -1, v0 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s7, v1 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s7, v1 -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s2, v1 -; GCN-NEXT: s_and_b64 vcc, vcc, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s7 +; GCN-NEXT: v_cvt_f32_u32_e32 v1, s8 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc ; GCN-NEXT: v_mov_b32_e32 v1, 0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv32_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s7 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, s8 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr i64 %x, 32 %2 = lshr i64 %y, 32 %result = udiv i64 %1, %2 @@ -374,40 +572,52 @@ define amdgpu_kernel void @s_test_udiv31_i64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_udiv31_i64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s2, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_lshr_b32 s3, s7, 1 -; GCN-NEXT: s_lshr_b32 s2, s2, 1 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s2 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s2 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s3 -; GCN-NEXT: v_mul_lo_u32 v1, v0, s2 -; GCN-NEXT: v_add_i32_e32 v2, vcc, 1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, -1, v0 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s3, v1 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s3, v1 -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s2, v1 -; GCN-NEXT: s_and_b64 vcc, vcc, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_lshr_b32 s4, s7, 1 +; GCN-NEXT: s_lshr_b32 s5, s8, 1 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv31_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_lshr_b32 s4, s7, 1 +; GCN-IR-NEXT: s_lshr_b32 s5, s8, 1 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr i64 %x, 33 %2 = lshr i64 %y, 33 %result = udiv i64 %1, %2 @@ -422,24 +632,49 @@ define amdgpu_kernel void @s_test_udiv23_i64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-NEXT: s_load_dword s8, s[0:1], 0xe ; GCN-NEXT: s_mov_b32 s3, 0xf000 ; GCN-NEXT: s_mov_b32 s2, -1 -; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: s_mov_b32 s0, s4 ; GCN-NEXT: s_mov_b32 s1, s5 ; GCN-NEXT: s_lshr_b32 s4, s7, 9 ; GCN-NEXT: s_lshr_b32 s5, s8, 9 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s5 -; GCN-NEXT: v_cvt_f32_u32_e32 v2, s4 -; GCN-NEXT: v_rcp_iflag_f32_e32 v3, v0 -; GCN-NEXT: v_mul_f32_e32 v3, v2, v3 -; GCN-NEXT: v_trunc_f32_e32 v3, v3 -; GCN-NEXT: v_mad_f32 v2, -v3, v0, v2 -; GCN-NEXT: v_cvt_u32_f32_e32 v3, v3 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, |v0| -; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v3, vcc +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s4 +; GCN-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc ; GCN-NEXT: v_and_b32_e32 v0, 0x7fffff, v0 +; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv23_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_lshr_b32 s4, s7, 9 +; GCN-IR-NEXT: s_lshr_b32 s5, s8, 9 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0x7fffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr i64 %x, 41 %2 = lshr i64 %y, 41 %result = udiv i64 %1, %2 @@ -570,6 +805,107 @@ define amdgpu_kernel void @s_test_udiv24_i48(i48 addrspace(1)* %out, i48 %x, i48 ; GCN-NEXT: buffer_store_short v0, off, s[4:7], 0 offset:4 ; GCN-NEXT: buffer_store_dword v1, off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv24_i48: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s2, s[0:1], 0xb +; GCN-IR-NEXT: s_load_dword s3, s[0:1], 0xc +; GCN-IR-NEXT: s_load_dword s6, s[0:1], 0xd +; GCN-IR-NEXT: s_load_dword s7, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s8, 0xffff +; GCN-IR-NEXT: s_mov_b32 s9, 0xff000000 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_and_b32 s1, s3, s8 +; GCN-IR-NEXT: s_and_b32 s0, s2, s9 +; GCN-IR-NEXT: s_and_b32 s3, s7, s8 +; GCN-IR-NEXT: s_and_b32 s2, s6, s9 +; GCN-IR-NEXT: s_lshr_b64 s[6:7], s[0:1], 24 +; GCN-IR-NEXT: s_lshr_b64 s[2:3], s[2:3], 24 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[2:3], 0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[6:7], 0 +; GCN-IR-NEXT: s_flbit_i32_b32 s10, s2 +; GCN-IR-NEXT: s_flbit_i32_b32 s11, s3 +; GCN-IR-NEXT: s_flbit_i32_b32 s12, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s13, s7 +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], s[8:9] +; GCN-IR-NEXT: s_add_i32 s8, s10, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s11 +; GCN-IR-NEXT: s_add_i32 s9, s12, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s13 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s8 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s3, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v2, s9 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v0, v1 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_vccz BB7_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s7 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s6 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB7_7 +; GCN-IR-NEXT: BB7_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[6:7], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB7_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB7_6 +; GCN-IR-NEXT: BB7_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[6:7], v0 +; GCN-IR-NEXT: s_add_u32 s6, s2, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s7, s3, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB7_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s7 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s6, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s2, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s3, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB7_5 +; GCN-IR-NEXT: BB7_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB7_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: buffer_store_short v1, off, s[4:7], 0 offset:4 +; GCN-IR-NEXT: buffer_store_dword v0, off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr i48 %x, 24 %2 = lshr i48 %y, 24 %result = udiv i48 %1, %2 @@ -689,6 +1025,84 @@ define amdgpu_kernel void @s_test_udiv_k_num_i64(i64 addrspace(1)* %out, i64 %x) ; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s1, s7 +; GCN-IR-NEXT: s_add_i32 s0, s0, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s1 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s0 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 0xffffffc5, v0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[6:7], 0 +; GCN-IR-NEXT: v_addc_u32_e64 v3, s[2:3], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[2:3], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[2:3] +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_cbranch_vccz BB8_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, 24, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB8_7 +; GCN-IR-NEXT: BB8_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], 24, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB8_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB8_6 +; GCN-IR-NEXT: BB8_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], 24, v0 +; GCN-IR-NEXT: s_add_u32 s3, s6, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s8, s7, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB8_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s8 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s3, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s6, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s7, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB8_5 +; GCN-IR-NEXT: BB8_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB8_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, s2 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %result = udiv i64 24, %x store i64 %result, i64 addrspace(1)* %out ret void @@ -798,6 +1212,88 @@ define i64 @v_test_udiv_pow2_k_num_i64(i64 %x) { ; GCN-NEXT: v_cndmask_b32_e32 v1, v7, v8, vcc ; GCN-NEXT: v_cndmask_b32_e64 v1, 0, v1, s[4:5] ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_udiv_pow2_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 +; GCN-IR-NEXT: s_movk_i32 s6, 0xffd0 +; GCN-IR-NEXT: s_mov_b32 s10, 0x8000 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, s6, v2 +; GCN-IR-NEXT: s_mov_b32 s11, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s10 +; GCN-IR-NEXT: v_addc_u32_e64 v5, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] +; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v3, v2, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], s[4:5], -1 +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB9_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[4:5] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[10:11], v6 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB9_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: s_mov_b32 s5, 0 +; GCN-IR-NEXT: s_mov_b32 s4, 0x8000 +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_lshr_b64 v[12:13], s[4:5], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: BB9_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[12:13], v[12:13], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v5 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_or_b32_e32 v12, v12, v6 +; GCN-IR-NEXT: v_or_b32_e32 v5, v11, v5 +; GCN-IR-NEXT: v_or_b32_e32 v4, v10, v4 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v8, v12 +; GCN-IR-NEXT: v_subb_u32_e64 v6, s[4:5], v9, v13, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v10 +; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 +; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, v12, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v13, vcc, v13, v11, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v11, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v10, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB9_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB9_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v2, v7, v1 +; GCN-IR-NEXT: v_or_b32_e32 v3, v6, v0 +; GCN-IR-NEXT: BB9_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mov_b32_e32 v0, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v1, v2 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = udiv i64 32768, %x ret i64 %result } @@ -809,6 +1305,81 @@ define i64 @v_test_udiv_pow2_k_den_i64(i64 %x) { ; GCN-NEXT: v_alignbit_b32 v0, v1, v0, 15 ; GCN-NEXT: v_lshrrev_b32_e32 v1, 15, v1 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_udiv_pow2_k_den_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 +; GCN-IR-NEXT: v_add_i32_e64 v2, s[4:5], 32, v2 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v3, v2, s[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 48, v2 +; GCN-IR-NEXT: v_subb_u32_e64 v5, s[4:5], 0, 0, s[4:5] +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] +; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v1, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[4:5], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v3, v0, 0, s[4:5] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB10_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[4:5] +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[0:1], v6 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB10_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], v[0:1], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v0, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_movk_i32 s12, 0x7fff +; GCN-IR-NEXT: BB10_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v7, v1, v7 +; GCN-IR-NEXT: v_or_b32_e32 v6, v0, v6 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s12, v8 +; GCN-IR-NEXT: s_or_b64 s[8:9], s[4:5], s[8:9] +; GCN-IR-NEXT: v_subb_u32_e32 v0, vcc, 0, v9, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v0, 31, v0 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v0 +; GCN-IR-NEXT: v_and_b32_e32 v0, 0x8000, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, v8, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v9, vcc, v9, v10, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v0, v4 +; GCN-IR-NEXT: v_mov_b32_e32 v1, v5 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB10_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB10_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[6:7], 1 +; GCN-IR-NEXT: v_or_b32_e32 v2, v5, v1 +; GCN-IR-NEXT: v_or_b32_e32 v3, v4, v0 +; GCN-IR-NEXT: BB10_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mov_b32_e32 v0, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v1, v2 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = udiv i64 %x, 32768 ret i64 %result } @@ -921,6 +1492,81 @@ define amdgpu_kernel void @s_test_udiv_k_den_i64(i64 addrspace(1)* %out, i64 %x) ; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv_k_den_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s2, s7 +; GCN-IR-NEXT: s_add_i32 s3, s0, 32 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[6:7], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s2 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s3 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 59, v0 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[2:3], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[2:3], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[2:3] +; GCN-IR-NEXT: s_cbranch_vccz BB11_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s7 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s6 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB11_7 +; GCN-IR-NEXT: BB11_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[6:7], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB11_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB11_6 +; GCN-IR-NEXT: BB11_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[6:7], v0 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB11_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], 23, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], 0, v9, s[0:1] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 24, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v6 +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_subbrev_u32_e64 v9, s[0:1], 0, v9, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB11_5 +; GCN-IR-NEXT: BB11_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB11_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %result = udiv i64 %x, 24 store i64 %result, i64 addrspace(1)* %out ret void @@ -1027,6 +1673,79 @@ define i64 @v_test_udiv_k_den_i64(i64 %x) { ; GCN-NEXT: v_cndmask_b32_e32 v1, v10, v8, vcc ; GCN-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5] ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_udiv_k_den_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 +; GCN-IR-NEXT: v_add_i32_e64 v2, s[4:5], 32, v2 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v3, v2, s[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 59, v2 +; GCN-IR-NEXT: v_subb_u32_e64 v5, s[4:5], 0, 0, s[4:5] +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] +; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v1, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[4:5], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v3, v0, 0, s[4:5] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB12_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[4:5] +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[0:1], v6 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB12_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], v[0:1], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v0, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: BB12_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v4 +; GCN-IR-NEXT: v_or_b32_e32 v7, v1, v7 +; GCN-IR-NEXT: v_or_b32_e32 v6, v0, v6 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_sub_i32_e64 v0, s[4:5], 23, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v0, s[4:5], 0, v9, s[4:5] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v0, 31, v0 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v0 +; GCN-IR-NEXT: v_and_b32_e32 v0, 24, v0 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[4:5], v8, v0 +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_subbrev_u32_e64 v9, vcc, 0, v9, s[4:5] +; GCN-IR-NEXT: v_mov_b32_e32 v0, v4 +; GCN-IR-NEXT: v_mov_b32_e32 v1, v5 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB12_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB12_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[6:7], 1 +; GCN-IR-NEXT: v_or_b32_e32 v2, v5, v1 +; GCN-IR-NEXT: v_or_b32_e32 v3, v4, v0 +; GCN-IR-NEXT: BB12_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mov_b32_e32 v0, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v1, v2 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = udiv i64 %x, 24 ret i64 %result } @@ -1038,35 +1757,45 @@ define amdgpu_kernel void @s_test_udiv24_k_num_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s2, 0x41c00000 ; GCN-NEXT: s_mov_b32 s4, s0 ; GCN-NEXT: s_mov_b32 s5, s1 -; GCN-NEXT: s_lshr_b32 s2, s3, 8 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s2 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s2 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, 24 -; GCN-NEXT: v_mul_lo_u32 v1, v0, s2 -; GCN-NEXT: v_add_i32_e32 v2, vcc, 1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, -1, v0 -; GCN-NEXT: v_cmp_gt_u32_e64 s[0:1], 25, v1 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, 24, v1 -; GCN-NEXT: v_cmp_le_u32_e32 vcc, s2, v1 -; GCN-NEXT: s_and_b64 vcc, vcc, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] +; GCN-NEXT: s_lshr_b32 s0, s3, 8 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s0 +; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-NEXT: v_mul_f32_e32 v1, s2, v1 +; GCN-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-NEXT: v_mad_f32 v2, -v1, v0, s2 +; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, v0 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv24_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s2, 0x41c00000 +; GCN-IR-NEXT: s_mov_b32 s4, s0 +; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_lshr_b32 s0, s3, 8 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s0 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, s2, v1 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v2, -v1, v0, s2 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %x.shr = lshr i64 %x, 40 %result = udiv i64 24, %x.shr store i64 %result, i64 addrspace(1)* %out @@ -1080,37 +1809,43 @@ define amdgpu_kernel void @s_test_udiv24_k_den_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_movk_i32 s2, 0x5b7f -; GCN-NEXT: s_movk_i32 s8, 0x5b7e -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s2 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 +; GCN-NEXT: s_mov_b32 s2, 0x46b6fe00 ; GCN-NEXT: s_mov_b32 s4, s0 ; GCN-NEXT: s_mov_b32 s5, s1 -; GCN-NEXT: s_lshr_b32 s3, s3, 8 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s2 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s3 -; GCN-NEXT: v_mul_lo_u32 v1, v0, s2 -; GCN-NEXT: v_add_i32_e32 v2, vcc, 1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, -1, v0 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s3, v1 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s3, v1 -; GCN-NEXT: v_cmp_lt_u32_e32 vcc, s8, v1 -; GCN-NEXT: s_and_b64 vcc, vcc, s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] +; GCN-NEXT: s_lshr_b32 s0, s3, 8 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s0 +; GCN-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 +; GCN-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-NEXT: v_mad_f32 v0, -v1, s2, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s2 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_udiv24_k_den_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s2, 0x46b6fe00 +; GCN-IR-NEXT: s_mov_b32 s4, s0 +; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_lshr_b32 s0, s3, 8 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v0, -v1, s2, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s2 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %x.shr = lshr i64 %x, 40 %result = udiv i64 %x.shr, 23423 store i64 %result, i64 addrspace(1)* %out @@ -1122,31 +1857,35 @@ define i64 @v_test_udiv24_k_num_i64(i64 %x) { ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_lshrrev_b32_e32 v0, 8, v1 -; GCN-NEXT: v_cvt_f32_u32_e32 v1, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v1 -; GCN-NEXT: v_mul_f32_e32 v1, 0x4f800000, v1 +; GCN-NEXT: s_mov_b32 s4, 0x41c00000 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-NEXT: v_mul_f32_e32 v1, s4, v1 +; GCN-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-NEXT: v_mad_f32 v2, -v1, v0, s4 ; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 -; GCN-NEXT: v_mul_hi_u32 v2, v1, v0 -; GCN-NEXT: v_mul_lo_u32 v3, v1, v0 -; GCN-NEXT: v_sub_i32_e32 v4, vcc, 0, v3 -; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v2 -; GCN-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GCN-NEXT: v_mul_hi_u32 v2, v2, v1 -; GCN-NEXT: v_add_i32_e64 v3, s[4:5], v1, v2 -; GCN-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v2 -; GCN-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc -; GCN-NEXT: v_mul_hi_u32 v1, v1, 24 -; GCN-NEXT: v_mul_lo_u32 v2, v1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, 1, v1 -; GCN-NEXT: v_add_i32_e32 v4, vcc, -1, v1 -; GCN-NEXT: v_cmp_gt_u32_e32 vcc, 25, v2 -; GCN-NEXT: v_sub_i32_e64 v2, s[4:5], 24, v2 -; GCN-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v0 -; GCN-NEXT: s_and_b64 s[4:5], s[4:5], vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v1, v3, s[4:5] -; GCN-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, v0 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_udiv24_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-IR-NEXT: s_mov_b32 s4, 0x41c00000 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, s4, v1 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v2, -v1, v0, s4 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = lshr i64 %x, 40 %result = udiv i64 24, %x.shr ret i64 %result @@ -1157,32 +1896,35 @@ define i64 @v_test_udiv24_pow2_k_num_i64(i64 %x) { ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_lshrrev_b32_e32 v0, 8, v1 -; GCN-NEXT: s_mov_b32 s6, 0x8001 -; GCN-NEXT: v_cvt_f32_u32_e32 v1, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v1 -; GCN-NEXT: v_mul_f32_e32 v1, 0x4f800000, v1 +; GCN-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-NEXT: v_mul_f32_e32 v1, s4, v1 +; GCN-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-NEXT: v_mad_f32 v2, -v1, v0, s4 ; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 -; GCN-NEXT: v_mul_hi_u32 v2, v1, v0 -; GCN-NEXT: v_mul_lo_u32 v3, v1, v0 -; GCN-NEXT: v_sub_i32_e32 v4, vcc, 0, v3 -; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v2 -; GCN-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GCN-NEXT: v_mul_hi_u32 v2, v2, v1 -; GCN-NEXT: v_add_i32_e64 v3, s[4:5], v1, v2 -; GCN-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v2 -; GCN-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc -; GCN-NEXT: v_lshrrev_b32_e32 v1, 17, v1 -; GCN-NEXT: v_mul_u32_u24_e32 v2, v1, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, 1, v1 -; GCN-NEXT: v_add_i32_e32 v4, vcc, -1, v1 -; GCN-NEXT: v_cmp_gt_u32_e64 s[4:5], s6, v2 -; GCN-NEXT: v_sub_i32_e32 v2, vcc, 0x8000, v2 -; GCN-NEXT: v_cmp_ge_u32_e32 vcc, v2, v0 -; GCN-NEXT: s_and_b64 vcc, vcc, s[4:5] -; GCN-NEXT: v_cndmask_b32_e32 v0, v1, v3, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v4, v0, s[4:5] +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, v0 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_udiv24_pow2_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-IR-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, s4, v1 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v2, -v1, v0, s4 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = lshr i64 %x, 40 %result = udiv i64 32768, %x.shr ret i64 %result @@ -1195,6 +1937,22 @@ define i64 @v_test_udiv24_pow2_k_den_i64(i64 %x) { ; GCN-NEXT: v_lshrrev_b32_e32 v0, 23, v1 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_udiv24_pow2_k_den_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-IR-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, 0x38000000, v0 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v0, -v1, s4, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s4 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = lshr i64 %x, 40 %result = udiv i64 %x.shr, 32768 ret i64 %result diff --git a/llvm/test/CodeGen/AMDGPU/urem64.ll b/llvm/test/CodeGen/AMDGPU/urem64.ll index 220f6ad..845d862 100644 --- a/llvm/test/CodeGen/AMDGPU/urem64.ll +++ b/llvm/test/CodeGen/AMDGPU/urem64.ll @@ -1,5 +1,6 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s +; RUN: llc -march=amdgcn -amdgpu-codegenprepare-expand-div64 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-IR %s define amdgpu_kernel void @s_test_urem_i64(i64 addrspace(1)* %out, i64 %x, i64 %y) { ; GCN-LABEL: s_test_urem_i64: @@ -122,6 +123,106 @@ define amdgpu_kernel void @s_test_urem_i64(i64 addrspace(1)* %out, i64 %x, i64 % ; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xd +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_flbit_i32_b32 s10, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s2 +; GCN-IR-NEXT: s_add_i32 s11, s0, 32 +; GCN-IR-NEXT: s_flbit_i32_b32 s12, s3 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[2:3], 0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[8:9], s[6:7], 0 +; GCN-IR-NEXT: s_flbit_i32_b32 s13, s7 +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], s[8:9] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s12 +; GCN-IR-NEXT: s_add_i32 s8, s10, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s13 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s11 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s3, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v2, s8 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, v0, v1 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[8:9], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[8:9], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_vccz BB0_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s7 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s6 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB0_7 +; GCN-IR-NEXT: BB0_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[6:7], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB0_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB0_6 +; GCN-IR-NEXT: BB0_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[6:7], v0 +; GCN-IR-NEXT: s_add_u32 s8, s2, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s9, s3, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB0_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s9 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s8, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s2, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s3, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB0_5 +; GCN-IR-NEXT: BB0_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB0_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s11, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s10, -1 +; GCN-IR-NEXT: s_mov_b32 s8, s4 +; GCN-IR-NEXT: s_mov_b32 s9, s5 +; GCN-IR-NEXT: v_mul_lo_u32 v1, s2, v1 +; GCN-IR-NEXT: v_mul_hi_u32 v2, s2, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v3, s3, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v0, s2, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v4, s7 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v1, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s6, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v4, v1, vcc +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-IR-NEXT: s_endpgm %result = urem i64 %x, %y store i64 %result, i64 addrspace(1)* %out ret void @@ -239,6 +340,95 @@ define i64 @v_test_urem_i64(i64 %x, i64 %y) { ; GCN-NEXT: v_cndmask_b32_e32 v2, v4, v7, vcc ; GCN-NEXT: v_cndmask_b32_e64 v1, v1, v2, s[4:5] ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_urem_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v4, v2 +; GCN-IR-NEXT: v_ffbh_u32_e32 v5, v3 +; GCN-IR-NEXT: v_ffbh_u32_e32 v6, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v7, v1 +; GCN-IR-NEXT: s_or_b64 s[6:7], vcc, s[4:5] +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 32, v4 +; GCN-IR-NEXT: v_add_i32_e32 v6, vcc, 32, v6 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v3 +; GCN-IR-NEXT: v_cndmask_b32_e32 v4, v5, v4, vcc +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v5, v7, v6, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, v4, v5 +; GCN-IR-NEXT: v_subb_u32_e64 v7, s[4:5], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[6:7] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[4:5], 63, v[6:7] +; GCN-IR-NEXT: s_or_b64 s[6:7], s[6:7], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v5, v1, 0, s[6:7] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[6:7], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v4, v0, 0, s[6:7] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[4:5] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB1_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, 1, v6 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, 0, v7, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v8, vcc, 63, v6 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[4:5], v[6:7] +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[0:1], v8 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB1_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[14:15], v[0:1], v4 +; GCN-IR-NEXT: v_add_i32_e32 v10, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v11, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v12, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v13, 0 +; GCN-IR-NEXT: BB1_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[14:15], v[14:15], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v9 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, -1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v5, vcc, -1, v5, vcc +; GCN-IR-NEXT: v_or_b32_e32 v14, v14, v6 +; GCN-IR-NEXT: v_or_b32_e32 v9, v13, v9 +; GCN-IR-NEXT: v_or_b32_e32 v8, v12, v8 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v10, v14 +; GCN-IR-NEXT: v_subb_u32_e64 v6, s[4:5], v11, v15, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v12, 31, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v12 +; GCN-IR-NEXT: v_and_b32_e32 v13, v12, v3 +; GCN-IR-NEXT: v_and_b32_e32 v12, v12, v2 +; GCN-IR-NEXT: v_sub_i32_e32 v14, vcc, v14, v12 +; GCN-IR-NEXT: v_subb_u32_e32 v15, vcc, v15, v13, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v13, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v12, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB1_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB1_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[8:9], 1 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: BB1_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mul_lo_u32 v5, v2, v5 +; GCN-IR-NEXT: v_mul_hi_u32 v6, v2, v4 +; GCN-IR-NEXT: v_mul_lo_u32 v3, v3, v4 +; GCN-IR-NEXT: v_mul_lo_u32 v2, v2, v4 +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, v6, v5 +; GCN-IR-NEXT: v_add_i32_e32 v3, vcc, v4, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = urem i64 %x, %y ret i64 %result } @@ -247,40 +437,56 @@ define amdgpu_kernel void @s_test_urem31_i64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_urem31_i64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s2, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_lshr_b32 s3, s7, 1 -; GCN-NEXT: s_lshr_b32 s4, s2, 1 +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_lshr_b32 s4, s7, 1 +; GCN-NEXT: s_lshr_b32 s5, s8, 1 ; GCN-NEXT: v_cvt_f32_u32_e32 v0, s4 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s4 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s4 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s3 -; GCN-NEXT: v_mul_lo_u32 v0, v0, s4 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s3, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, s4, v1 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s3, v0 -; GCN-NEXT: v_cmp_le_u32_e64 s[2:3], s4, v1 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s4, v1 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] +; GCN-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-NEXT: v_mul_lo_u32 v0, v0, s5 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem31_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_lshr_b32 s4, s7, 1 +; GCN-IR-NEXT: s_lshr_b32 s5, s8, 1 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s5 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-IR-NEXT: v_and_b32_e32 v0, 0x7fffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr i64 %x, 33 %2 = lshr i64 %y, 33 %result = urem i64 %1, %2 @@ -291,66 +497,87 @@ define amdgpu_kernel void @s_test_urem31_i64(i64 addrspace(1)* %out, i64 %x, i64 define amdgpu_kernel void @s_test_urem31_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> %x, <2 x i64> %y) { ; GCN-LABEL: s_test_urem31_v2i64: ; GCN: ; %bb.0: -; GCN-NEXT: s_load_dwordx2 s[8:9], s[0:1], 0x9 -; GCN-NEXT: s_load_dwordx4 s[12:15], s[0:1], 0xd -; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x11 -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0x9 +; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd +; GCN-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x11 +; GCN-NEXT: s_mov_b32 s15, 0xf000 +; GCN-NEXT: s_mov_b32 s14, -1 ; GCN-NEXT: v_mov_b32_e32 v1, 0 +; GCN-NEXT: s_brev_b32 s0, -2 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_lshr_b32 s4, s13, 1 -; GCN-NEXT: s_lshr_b32 s6, s15, 1 -; GCN-NEXT: s_lshr_b32 s12, s5, 1 -; GCN-NEXT: s_lshr_b32 s5, s7, 1 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s5 -; GCN-NEXT: v_cvt_f32_u32_e32 v2, s12 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v2 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_mul_f32_e32 v2, 0x4f800000, v2 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 -; GCN-NEXT: v_mul_hi_u32 v3, v0, s5 -; GCN-NEXT: v_mul_lo_u32 v4, v0, s5 -; GCN-NEXT: v_mul_hi_u32 v5, v2, s12 -; GCN-NEXT: v_mul_lo_u32 v6, v2, s12 -; GCN-NEXT: v_sub_i32_e32 v7, vcc, 0, v4 -; GCN-NEXT: v_sub_i32_e32 v8, vcc, 0, v6 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v3 -; GCN-NEXT: v_cndmask_b32_e64 v3, v4, v7, s[0:1] -; GCN-NEXT: v_cmp_eq_u32_e64 s[2:3], 0, v5 -; GCN-NEXT: v_cndmask_b32_e64 v4, v6, v8, s[2:3] -; GCN-NEXT: v_mul_hi_u32 v3, v3, v0 -; GCN-NEXT: v_mul_hi_u32 v4, v4, v2 -; GCN-NEXT: v_add_i32_e32 v5, vcc, v3, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v3, v0 -; GCN-NEXT: v_add_i32_e32 v3, vcc, v4, v2 -; GCN-NEXT: v_subrev_i32_e32 v2, vcc, v4, v2 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v5, s[0:1] -; GCN-NEXT: v_cndmask_b32_e64 v2, v2, v3, s[2:3] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s6 -; GCN-NEXT: v_mul_hi_u32 v2, v2, s4 -; GCN-NEXT: v_mul_lo_u32 v0, v0, s5 -; GCN-NEXT: v_mul_lo_u32 v2, v2, s12 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, s6, v0 -; GCN-NEXT: v_sub_i32_e32 v4, vcc, s4, v2 -; GCN-NEXT: v_add_i32_e32 v5, vcc, s5, v3 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s6, v0 -; GCN-NEXT: v_cmp_le_u32_e64 s[2:3], s5, v3 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s5, v3 -; GCN-NEXT: v_add_i32_e32 v6, vcc, s12, v4 -; GCN-NEXT: v_cmp_ge_u32_e64 s[4:5], s4, v2 -; GCN-NEXT: v_cmp_le_u32_e64 s[6:7], s12, v4 -; GCN-NEXT: v_subrev_i32_e32 v2, vcc, s12, v4 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc -; GCN-NEXT: s_and_b64 vcc, s[6:7], s[4:5] -; GCN-NEXT: v_cndmask_b32_e32 v3, v4, v2, vcc -; GCN-NEXT: v_cndmask_b32_e64 v2, v5, v0, s[0:1] -; GCN-NEXT: v_cndmask_b32_e64 v0, v6, v3, s[4:5] +; GCN-NEXT: s_lshr_b32 s1, s7, 1 +; GCN-NEXT: s_lshr_b32 s2, s5, 1 +; GCN-NEXT: s_lshr_b32 s3, s11, 1 +; GCN-NEXT: s_lshr_b32 s4, s9, 1 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s2 +; GCN-NEXT: v_cvt_f32_u32_e32 v2, s4 +; GCN-NEXT: v_cvt_f32_u32_e32 v3, s1 +; GCN-NEXT: v_cvt_f32_u32_e32 v4, s3 +; GCN-NEXT: v_rcp_iflag_f32_e32 v5, v2 +; GCN-NEXT: v_rcp_iflag_f32_e32 v6, v4 +; GCN-NEXT: v_mul_f32_e32 v5, v0, v5 +; GCN-NEXT: v_mul_f32_e32 v6, v3, v6 +; GCN-NEXT: v_trunc_f32_e32 v5, v5 +; GCN-NEXT: v_trunc_f32_e32 v6, v6 +; GCN-NEXT: v_mad_f32 v0, -v5, v2, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v5, v5 +; GCN-NEXT: v_mad_f32 v3, -v6, v4, v3 +; GCN-NEXT: v_cvt_u32_f32_e32 v6, v6 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v2 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v5, vcc +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v4 +; GCN-NEXT: v_addc_u32_e32 v2, vcc, 0, v6, vcc +; GCN-NEXT: v_mul_lo_u32 v0, v0, s4 +; GCN-NEXT: v_mul_lo_u32 v2, v2, s3 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s2, v0 +; GCN-NEXT: v_sub_i32_e32 v2, vcc, s1, v2 +; GCN-NEXT: v_and_b32_e32 v0, s0, v0 +; GCN-NEXT: v_and_b32_e32 v2, s0, v2 ; GCN-NEXT: v_mov_b32_e32 v3, v1 -; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[12:15], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem31_v2i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd +; GCN-IR-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x11 +; GCN-IR-NEXT: s_mov_b32 s15, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s14, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_brev_b32 s0, -2 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_lshr_b32 s1, s7, 1 +; GCN-IR-NEXT: s_lshr_b32 s2, s5, 1 +; GCN-IR-NEXT: s_lshr_b32 s3, s11, 1 +; GCN-IR-NEXT: s_lshr_b32 s4, s9, 1 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s2 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v2, s4 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v3, s1 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v4, s3 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v5, v2 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v6, v4 +; GCN-IR-NEXT: v_mul_f32_e32 v5, v0, v5 +; GCN-IR-NEXT: v_mul_f32_e32 v6, v3, v6 +; GCN-IR-NEXT: v_trunc_f32_e32 v5, v5 +; GCN-IR-NEXT: v_trunc_f32_e32 v6, v6 +; GCN-IR-NEXT: v_mad_f32 v0, -v5, v2, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v5, v5 +; GCN-IR-NEXT: v_mad_f32 v3, -v6, v4, v3 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v6, v6 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v2, vcc, 0, v6, vcc +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s4 +; GCN-IR-NEXT: v_mul_lo_u32 v2, v2, s3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s2, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, s1, v2 +; GCN-IR-NEXT: v_and_b32_e32 v0, s0, v0 +; GCN-IR-NEXT: v_and_b32_e32 v2, s0, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, v1 +; GCN-IR-NEXT: buffer_store_dwordx4 v[0:3], off, s[12:15], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr <2 x i64> %x, %2 = lshr <2 x i64> %y, %result = urem <2 x i64> %1, %2 @@ -362,40 +589,56 @@ define amdgpu_kernel void @s_test_urem24_i64(i64 addrspace(1)* %out, i64 %x, i64 ; GCN-LABEL: s_test_urem24_i64: ; GCN: ; %bb.0: ; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 -; GCN-NEXT: s_load_dword s2, s[0:1], 0xe -; GCN-NEXT: s_mov_b32 s11, 0xf000 -; GCN-NEXT: s_mov_b32 s10, -1 +; GCN-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-NEXT: s_mov_b32 s3, 0xf000 +; GCN-NEXT: s_mov_b32 s2, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_mov_b32 s8, s4 -; GCN-NEXT: s_mov_b32 s9, s5 -; GCN-NEXT: s_lshr_b32 s3, s7, 8 -; GCN-NEXT: s_lshr_b32 s4, s2, 8 +; GCN-NEXT: s_mov_b32 s0, s4 +; GCN-NEXT: s_mov_b32 s1, s5 +; GCN-NEXT: s_lshr_b32 s4, s7, 8 +; GCN-NEXT: s_lshr_b32 s5, s8, 8 ; GCN-NEXT: v_cvt_f32_u32_e32 v0, s4 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s4 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s4 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s3 -; GCN-NEXT: v_mul_lo_u32 v0, v0, s4 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s3, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, s4, v1 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s3, v0 -; GCN-NEXT: v_cmp_le_u32_e64 s[2:3], s4, v1 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s4, v1 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] +; GCN-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-NEXT: v_mul_lo_u32 v0, v0, s5 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 -; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[8:11], 0 +; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem24_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dword s8, s[0:1], 0xe +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: s_lshr_b32 s4, s7, 8 +; GCN-IR-NEXT: s_lshr_b32 s5, s8, 8 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s4 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, s5 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, v0, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v0, -v2, v1, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s5 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s4, v0 +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr i64 %x, 40 %2 = lshr i64 %y, 40 %result = urem i64 %1, %2 @@ -406,55 +649,87 @@ define amdgpu_kernel void @s_test_urem24_i64(i64 addrspace(1)* %out, i64 %x, i64 define amdgpu_kernel void @s_test_urem23_64_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> %x, <2 x i64> %y) { ; GCN-LABEL: s_test_urem23_64_v2i64: ; GCN: ; %bb.0: -; GCN-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9 -; GCN-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd -; GCN-NEXT: s_load_dwordx4 s[12:15], s[0:1], 0x11 -; GCN-NEXT: s_mov_b32 s7, 0xf000 -; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0x9 +; GCN-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd +; GCN-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x11 +; GCN-NEXT: s_mov_b32 s15, 0xf000 +; GCN-NEXT: s_mov_b32 s14, -1 ; GCN-NEXT: v_mov_b32_e32 v1, 0 +; GCN-NEXT: s_brev_b32 s0, -2 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: s_lshr_b32 s2, s11, 9 -; GCN-NEXT: s_lshr_b32 s3, s9, 1 -; GCN-NEXT: s_lshr_b32 s8, s15, 9 -; GCN-NEXT: s_lshr_b32 s9, s13, 1 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s9 -; GCN-NEXT: v_cvt_f32_u32_e32 v2, s2 -; GCN-NEXT: v_cvt_f32_u32_e32 v3, s8 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v4, v3 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_mul_f32_e32 v4, v2, v4 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_trunc_f32_e32 v4, v4 -; GCN-NEXT: v_mul_hi_u32 v5, v0, s9 -; GCN-NEXT: v_mul_lo_u32 v6, v0, s9 -; GCN-NEXT: v_mad_f32 v2, -v4, v3, v2 -; GCN-NEXT: v_cvt_u32_f32_e32 v4, v4 -; GCN-NEXT: v_sub_i32_e32 v7, vcc, 0, v6 -; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, |v3| -; GCN-NEXT: v_addc_u32_e32 v2, vcc, 0, v4, vcc -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v5 -; GCN-NEXT: v_cndmask_b32_e64 v3, v6, v7, s[0:1] -; GCN-NEXT: v_mul_lo_u32 v2, v2, s8 -; GCN-NEXT: v_mul_hi_u32 v3, v3, v0 -; GCN-NEXT: v_sub_i32_e32 v2, vcc, s2, v2 -; GCN-NEXT: v_add_i32_e32 v4, vcc, v3, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v3, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v4, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s3 -; GCN-NEXT: v_mul_lo_u32 v0, v0, s9 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, s3, v0 -; GCN-NEXT: v_add_i32_e32 v4, vcc, s9, v3 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s3, v0 -; GCN-NEXT: v_cmp_le_u32_e64 s[2:3], s9, v3 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s9, v3 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v4, v0, s[0:1] -; GCN-NEXT: v_and_b32_e32 v2, 0x7fffff, v2 +; GCN-NEXT: s_lshr_b32 s1, s7, 9 +; GCN-NEXT: s_lshr_b32 s2, s5, 1 +; GCN-NEXT: s_lshr_b32 s3, s11, 9 +; GCN-NEXT: s_lshr_b32 s4, s9, 1 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s2 +; GCN-NEXT: v_cvt_f32_u32_e32 v2, s4 +; GCN-NEXT: v_cvt_f32_u32_e32 v3, s1 +; GCN-NEXT: v_cvt_f32_u32_e32 v4, s3 +; GCN-NEXT: v_rcp_iflag_f32_e32 v5, v2 +; GCN-NEXT: v_rcp_iflag_f32_e32 v6, v4 +; GCN-NEXT: v_mul_f32_e32 v5, v0, v5 +; GCN-NEXT: v_mul_f32_e32 v6, v3, v6 +; GCN-NEXT: v_trunc_f32_e32 v5, v5 +; GCN-NEXT: v_trunc_f32_e32 v6, v6 +; GCN-NEXT: v_mad_f32 v0, -v5, v2, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v5, v5 +; GCN-NEXT: v_mad_f32 v3, -v6, v4, v3 +; GCN-NEXT: v_cvt_u32_f32_e32 v6, v6 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v2 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v5, vcc +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v4 +; GCN-NEXT: v_addc_u32_e32 v2, vcc, 0, v6, vcc +; GCN-NEXT: v_mul_lo_u32 v0, v0, s4 +; GCN-NEXT: v_mul_lo_u32 v2, v2, s3 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s2, v0 +; GCN-NEXT: v_sub_i32_e32 v2, vcc, s1, v2 +; GCN-NEXT: v_and_b32_e32 v0, s0, v0 +; GCN-NEXT: v_and_b32_e32 v2, s0, v2 ; GCN-NEXT: v_mov_b32_e32 v3, v1 -; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[4:7], 0 +; GCN-NEXT: buffer_store_dwordx4 v[0:3], off, s[12:15], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem23_64_v2i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0x9 +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd +; GCN-IR-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x11 +; GCN-IR-NEXT: s_mov_b32 s15, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s14, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_brev_b32 s0, -2 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_lshr_b32 s1, s7, 9 +; GCN-IR-NEXT: s_lshr_b32 s2, s5, 1 +; GCN-IR-NEXT: s_lshr_b32 s3, s11, 9 +; GCN-IR-NEXT: s_lshr_b32 s4, s9, 1 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s2 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v2, s4 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v3, s1 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v4, s3 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v5, v2 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v6, v4 +; GCN-IR-NEXT: v_mul_f32_e32 v5, v0, v5 +; GCN-IR-NEXT: v_mul_f32_e32 v6, v3, v6 +; GCN-IR-NEXT: v_trunc_f32_e32 v5, v5 +; GCN-IR-NEXT: v_trunc_f32_e32 v6, v6 +; GCN-IR-NEXT: v_mad_f32 v0, -v5, v2, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v5, v5 +; GCN-IR-NEXT: v_mad_f32 v3, -v6, v4, v3 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v6, v6 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v2, vcc, 0, v6, vcc +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s4 +; GCN-IR-NEXT: v_mul_lo_u32 v2, v2, s3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s2, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, s1, v2 +; GCN-IR-NEXT: v_and_b32_e32 v0, s0, v0 +; GCN-IR-NEXT: v_and_b32_e32 v2, s0, v2 +; GCN-IR-NEXT: v_mov_b32_e32 v3, v1 +; GCN-IR-NEXT: buffer_store_dwordx4 v[0:3], off, s[12:15], 0 +; GCN-IR-NEXT: s_endpgm %1 = lshr <2 x i64> %x, %2 = lshr <2 x i64> %y, %result = urem <2 x i64> %1, %2 @@ -573,6 +848,93 @@ define amdgpu_kernel void @s_test_urem_k_num_i64(i64 addrspace(1)* %out, i64 %x) ; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v0, s[0:1] ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s1, s7 +; GCN-IR-NEXT: s_add_i32 s0, s0, 32 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s1 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s0 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 0xffffffc5, v0 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[6:7], 0 +; GCN-IR-NEXT: v_addc_u32_e64 v3, s[2:3], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[2:3], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[2:3] +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_cbranch_vccz BB6_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, 24, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB6_7 +; GCN-IR-NEXT: BB6_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], 24, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB6_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB6_6 +; GCN-IR-NEXT: BB6_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], 24, v0 +; GCN-IR-NEXT: s_add_u32 s3, s6, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_addc_u32 s8, s7, -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB6_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v10, s8 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], s3, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], v10, v9, s[0:1] +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v7, s6, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, s7, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v7 +; GCN-IR-NEXT: v_subb_u32_e64 v9, s[0:1], v9, v6, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB6_5 +; GCN-IR-NEXT: BB6_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB6_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: v_mul_lo_u32 v1, s6, v1 +; GCN-IR-NEXT: v_mul_hi_u32 v2, s6, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v3, s7, v0 +; GCN-IR-NEXT: v_mul_lo_u32 v0, s6, v0 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v1, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, 0, v1, vcc +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %result = urem i64 24, %x store i64 %result, i64 addrspace(1)* %out ret void @@ -684,6 +1046,90 @@ define amdgpu_kernel void @s_test_urem_k_den_i64(i64 addrspace(1)* %out, i64 %x) ; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem_k_den_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x9 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_flbit_i32_b32 s0, s6 +; GCN-IR-NEXT: s_flbit_i32_b32 s2, s7 +; GCN-IR-NEXT: s_add_i32 s3, s0, 32 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[0:1], s[6:7], 0 +; GCN-IR-NEXT: v_mov_b32_e32 v0, s2 +; GCN-IR-NEXT: v_mov_b32_e32 v1, s3 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 vcc, s7, 0 +; GCN-IR-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 59, v0 +; GCN-IR-NEXT: v_subb_u32_e64 v3, s[2:3], 0, 0, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[0:1], s[0:1], vcc +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 63, v[2:3] +; GCN-IR-NEXT: s_or_b64 s[2:3], s[0:1], vcc +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[2:3] +; GCN-IR-NEXT: s_cbranch_vccz BB7_2 +; GCN-IR-NEXT: ; %bb.1: +; GCN-IR-NEXT: v_mov_b32_e32 v0, s7 +; GCN-IR-NEXT: v_cndmask_b32_e64 v1, v0, 0, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v0, s6 +; GCN-IR-NEXT: v_cndmask_b32_e64 v0, v0, 0, s[0:1] +; GCN-IR-NEXT: s_branch BB7_7 +; GCN-IR-NEXT: BB7_2: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, 1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v3, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[0:1], v[0:1], v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v2, vcc, 63, v2 +; GCN-IR-NEXT: s_and_b64 vcc, exec, s[0:1] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[6:7], v2 +; GCN-IR-NEXT: s_cbranch_vccz BB7_4 +; GCN-IR-NEXT: ; %bb.3: +; GCN-IR-NEXT: v_mov_b32_e32 v2, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_branch BB7_6 +; GCN-IR-NEXT: BB7_4: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[8:9], s[6:7], v0 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: BB7_5: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[8:9], v[8:9], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v2, 31, v5 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_add_i32_e32 v0, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_or_b32_e32 v8, v8, v2 +; GCN-IR-NEXT: v_or_b32_e32 v4, v6, v4 +; GCN-IR-NEXT: v_or_b32_e32 v5, v7, v5 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_sub_i32_e64 v2, s[0:1], 23, v8 +; GCN-IR-NEXT: v_subb_u32_e64 v2, s[0:1], 0, v9, s[0:1] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v6, 31, v2 +; GCN-IR-NEXT: v_and_b32_e32 v2, 1, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 24, v6 +; GCN-IR-NEXT: v_sub_i32_e64 v8, s[0:1], v8, v6 +; GCN-IR-NEXT: s_and_b64 vcc, exec, vcc +; GCN-IR-NEXT: v_subbrev_u32_e64 v9, s[0:1], 0, v9, s[0:1] +; GCN-IR-NEXT: v_mov_b32_e32 v7, v3 +; GCN-IR-NEXT: v_mov_b32_e32 v6, v2 +; GCN-IR-NEXT: s_cbranch_vccz BB7_5 +; GCN-IR-NEXT: BB7_6: ; %udiv-loop-exit +; GCN-IR-NEXT: v_lshl_b64 v[0:1], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v0, v2, v0 +; GCN-IR-NEXT: v_or_b32_e32 v1, v3, v1 +; GCN-IR-NEXT: BB7_7: ; %udiv-end +; GCN-IR-NEXT: s_mov_b32 s3, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s2, -1 +; GCN-IR-NEXT: s_mov_b32 s0, s4 +; GCN-IR-NEXT: s_mov_b32 s1, s5 +; GCN-IR-NEXT: v_mul_lo_u32 v1, v1, 24 +; GCN-IR-NEXT: v_mul_hi_u32 v2, v0, 24 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, 24 +; GCN-IR-NEXT: v_mov_b32_e32 v3, s7 +; GCN-IR-NEXT: v_add_i32_e32 v1, vcc, v2, v1 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s6, v0 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v3, v1, vcc +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0 +; GCN-IR-NEXT: s_endpgm %result = urem i64 %x, 24 store i64 %result, i64 addrspace(1)* %out ret void @@ -793,6 +1239,94 @@ define i64 @v_test_urem_pow2_k_num_i64(i64 %x) { ; GCN-NEXT: v_cndmask_b32_e32 v1, v4, v7, vcc ; GCN-NEXT: v_cndmask_b32_e64 v1, v3, v1, s[4:5] ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_urem_pow2_k_num_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 +; GCN-IR-NEXT: s_movk_i32 s6, 0xffd0 +; GCN-IR-NEXT: s_mov_b32 s10, 0x8000 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 32, v2 +; GCN-IR-NEXT: v_cmp_eq_u32_e32 vcc, 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc +; GCN-IR-NEXT: v_add_i32_e32 v4, vcc, s6, v2 +; GCN-IR-NEXT: s_mov_b32 s11, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v2, s10 +; GCN-IR-NEXT: v_addc_u32_e64 v5, s[6:7], 0, -1, vcc +; GCN-IR-NEXT: v_cmp_lt_u64_e32 vcc, 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] +; GCN-IR-NEXT: s_or_b64 s[4:5], s[4:5], vcc +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v2, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[4:5], s[4:5], -1 +; GCN-IR-NEXT: v_mov_b32_e32 v3, 0 +; GCN-IR-NEXT: s_and_b64 s[4:5], s[4:5], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB8_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[4:5] +; GCN-IR-NEXT: v_lshl_b64 v[4:5], s[10:11], v6 +; GCN-IR-NEXT: v_mov_b32_e32 v6, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB8_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: s_mov_b32 s5, 0 +; GCN-IR-NEXT: s_mov_b32 s4, 0x8000 +; GCN-IR-NEXT: v_add_i32_e32 v8, vcc, -1, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v9, vcc, -1, v1, vcc +; GCN-IR-NEXT: v_lshr_b64 v[12:13], s[4:5], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v10, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v11, 0 +; GCN-IR-NEXT: BB8_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[12:13], v[12:13], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v6, 31, v5 +; GCN-IR-NEXT: v_mov_b32_e32 v7, 0 +; GCN-IR-NEXT: v_lshl_b64 v[4:5], v[4:5], 1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_or_b32_e32 v12, v12, v6 +; GCN-IR-NEXT: v_or_b32_e32 v5, v11, v5 +; GCN-IR-NEXT: v_or_b32_e32 v4, v10, v4 +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[2:3] +; GCN-IR-NEXT: v_sub_i32_e64 v6, s[4:5], v8, v12 +; GCN-IR-NEXT: v_subb_u32_e64 v6, s[4:5], v9, v13, s[4:5] +; GCN-IR-NEXT: s_or_b64 s[8:9], vcc, s[8:9] +; GCN-IR-NEXT: v_ashrrev_i32_e32 v10, 31, v6 +; GCN-IR-NEXT: v_and_b32_e32 v6, 1, v10 +; GCN-IR-NEXT: v_and_b32_e32 v11, v10, v1 +; GCN-IR-NEXT: v_and_b32_e32 v10, v10, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v12, vcc, v12, v10 +; GCN-IR-NEXT: v_subb_u32_e32 v13, vcc, v13, v11, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v11, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v10, v6 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB8_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB8_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[4:5], 1 +; GCN-IR-NEXT: v_or_b32_e32 v3, v7, v3 +; GCN-IR-NEXT: v_or_b32_e32 v2, v6, v2 +; GCN-IR-NEXT: BB8_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_mul_lo_u32 v3, v0, v3 +; GCN-IR-NEXT: v_mul_hi_u32 v4, v0, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v1, v1, v2 +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, v2 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, v4, v3 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 0x8000, v0 +; GCN-IR-NEXT: v_add_i32_e64 v1, s[4:5], v2, v1 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, 0, v1, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = urem i64 32768, %x ret i64 %result } @@ -804,6 +1338,82 @@ define i64 @v_test_urem_pow2_k_den_i64(i64 %x) { ; GCN-NEXT: v_and_b32_e32 v0, 0x7fff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_urem_pow2_k_den_i64: +; GCN-IR: ; %bb.0: ; %_udiv-special-cases +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_cmp_eq_u64_e32 vcc, 0, v[0:1] +; GCN-IR-NEXT: v_ffbh_u32_e32 v2, v0 +; GCN-IR-NEXT: v_ffbh_u32_e32 v3, v1 +; GCN-IR-NEXT: v_add_i32_e64 v2, s[4:5], 32, v2 +; GCN-IR-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v3, v2, s[4:5] +; GCN-IR-NEXT: v_sub_i32_e64 v4, s[4:5], 48, v2 +; GCN-IR-NEXT: v_subb_u32_e64 v5, s[4:5], 0, 0, s[4:5] +; GCN-IR-NEXT: v_cmp_lt_u64_e64 s[4:5], 63, v[4:5] +; GCN-IR-NEXT: v_cmp_ne_u64_e64 s[6:7], 63, v[4:5] +; GCN-IR-NEXT: s_or_b64 s[4:5], vcc, s[4:5] +; GCN-IR-NEXT: v_cndmask_b32_e64 v3, v1, 0, s[4:5] +; GCN-IR-NEXT: s_xor_b64 s[8:9], s[4:5], -1 +; GCN-IR-NEXT: v_cndmask_b32_e64 v2, v0, 0, s[4:5] +; GCN-IR-NEXT: s_and_b64 s[4:5], s[8:9], s[6:7] +; GCN-IR-NEXT: s_and_saveexec_b64 s[6:7], s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB9_6 +; GCN-IR-NEXT: ; %bb.1: ; %udiv-bb1 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, 1, v4 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, 0, v5, vcc +; GCN-IR-NEXT: v_sub_i32_e32 v6, vcc, 63, v4 +; GCN-IR-NEXT: s_mov_b64 s[8:9], 0 +; GCN-IR-NEXT: v_cmp_ge_u64_e32 vcc, v[2:3], v[4:5] +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[0:1], v6 +; GCN-IR-NEXT: v_mov_b32_e32 v4, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: s_and_saveexec_b64 s[4:5], vcc +; GCN-IR-NEXT: s_xor_b64 s[10:11], exec, s[4:5] +; GCN-IR-NEXT: s_cbranch_execz BB9_5 +; GCN-IR-NEXT: ; %bb.2: ; %udiv-preheader +; GCN-IR-NEXT: v_lshr_b64 v[10:11], v[0:1], v2 +; GCN-IR-NEXT: v_mov_b32_e32 v8, 0 +; GCN-IR-NEXT: v_mov_b32_e32 v9, 0 +; GCN-IR-NEXT: s_movk_i32 s12, 0x7fff +; GCN-IR-NEXT: BB9_3: ; %udiv-do-while +; GCN-IR-NEXT: ; =>This Inner Loop Header: Depth=1 +; GCN-IR-NEXT: v_lshl_b64 v[10:11], v[10:11], 1 +; GCN-IR-NEXT: v_lshrrev_b32_e32 v4, 31, v7 +; GCN-IR-NEXT: v_mov_b32_e32 v5, 0 +; GCN-IR-NEXT: v_lshl_b64 v[6:7], v[6:7], 1 +; GCN-IR-NEXT: v_mov_b32_e32 v12, 0 +; GCN-IR-NEXT: v_add_i32_e32 v2, vcc, -1, v2 +; GCN-IR-NEXT: v_addc_u32_e32 v3, vcc, -1, v3, vcc +; GCN-IR-NEXT: v_or_b32_e32 v10, v10, v4 +; GCN-IR-NEXT: v_or_b32_e32 v7, v9, v7 +; GCN-IR-NEXT: v_or_b32_e32 v6, v8, v6 +; GCN-IR-NEXT: v_cmp_eq_u64_e64 s[4:5], 0, v[2:3] +; GCN-IR-NEXT: v_sub_i32_e32 v4, vcc, s12, v10 +; GCN-IR-NEXT: s_or_b64 s[8:9], s[4:5], s[8:9] +; GCN-IR-NEXT: v_subb_u32_e32 v4, vcc, 0, v11, vcc +; GCN-IR-NEXT: v_ashrrev_i32_e32 v8, 31, v4 +; GCN-IR-NEXT: v_and_b32_e32 v4, 1, v8 +; GCN-IR-NEXT: v_and_b32_e32 v8, 0x8000, v8 +; GCN-IR-NEXT: v_sub_i32_e32 v10, vcc, v10, v8 +; GCN-IR-NEXT: v_subb_u32_e32 v11, vcc, v11, v12, vcc +; GCN-IR-NEXT: v_mov_b32_e32 v9, v5 +; GCN-IR-NEXT: v_mov_b32_e32 v8, v4 +; GCN-IR-NEXT: s_andn2_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: s_cbranch_execnz BB9_3 +; GCN-IR-NEXT: ; %bb.4: ; %Flow +; GCN-IR-NEXT: s_or_b64 exec, exec, s[8:9] +; GCN-IR-NEXT: BB9_5: ; %Flow1 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[10:11] +; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[6:7], 1 +; GCN-IR-NEXT: v_or_b32_e32 v3, v5, v3 +; GCN-IR-NEXT: v_or_b32_e32 v2, v4, v2 +; GCN-IR-NEXT: BB9_6: ; %Flow2 +; GCN-IR-NEXT: s_or_b64 exec, exec, s[6:7] +; GCN-IR-NEXT: v_lshl_b64 v[2:3], v[2:3], 15 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v2 +; GCN-IR-NEXT: v_subb_u32_e32 v1, vcc, v1, v3, vcc +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %result = urem i64 %x, 32768 ret i64 %result } @@ -815,35 +1425,49 @@ define amdgpu_kernel void @s_test_urem24_k_num_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) +; GCN-NEXT: s_mov_b32 s2, 0x41c00000 ; GCN-NEXT: s_mov_b32 s4, s0 ; GCN-NEXT: s_mov_b32 s5, s1 -; GCN-NEXT: s_lshr_b32 s8, s3, 8 -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s8 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s8 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s8 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, 24 -; GCN-NEXT: v_mul_lo_u32 v0, v0, s8 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, 24, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, s8, v1 -; GCN-NEXT: v_cmp_gt_u32_e64 s[0:1], 25, v0 -; GCN-NEXT: v_cmp_le_u32_e64 s[2:3], s8, v1 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, s8, v1 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] +; GCN-NEXT: s_lshr_b32 s0, s3, 8 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s0 +; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-NEXT: v_mul_f32_e32 v1, s2, v1 +; GCN-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-NEXT: v_mad_f32 v2, -v1, v0, s2 +; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, v0 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-NEXT: v_mul_lo_u32 v0, v0, s0 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem24_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_mov_b32 s2, 0x41c00000 +; GCN-IR-NEXT: s_mov_b32 s4, s0 +; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_lshr_b32 s0, s3, 8 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s0 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v1, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, s2, v1 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v2, -v1, v0, s2 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v2|, v0 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s0 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %x.shr = lshr i64 %x, 40 %result = urem i64 24, %x.shr store i64 %result, i64 addrspace(1)* %out @@ -856,38 +1480,50 @@ define amdgpu_kernel void @s_test_urem24_k_den_i64(i64 addrspace(1)* %out, i64 % ; GCN-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 ; GCN-NEXT: s_mov_b32 s7, 0xf000 ; GCN-NEXT: s_mov_b32 s6, -1 +; GCN-NEXT: s_mov_b32 s8, 0x46b6fe00 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: s_movk_i32 s2, 0x5b7f -; GCN-NEXT: s_movk_i32 s8, 0x5b7e -; GCN-NEXT: v_cvt_f32_u32_e32 v0, s2 -; GCN-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GCN-NEXT: v_mul_f32_e32 v0, 0x4f800000, v0 ; GCN-NEXT: s_mov_b32 s4, s0 ; GCN-NEXT: s_mov_b32 s5, s1 -; GCN-NEXT: s_lshr_b32 s3, s3, 8 -; GCN-NEXT: v_cvt_u32_f32_e32 v0, v0 -; GCN-NEXT: v_mul_hi_u32 v1, v0, s2 -; GCN-NEXT: v_mul_lo_u32 v2, v0, s2 -; GCN-NEXT: v_sub_i32_e32 v3, vcc, 0, v2 -; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v1 -; GCN-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v1, v1, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, v1, v0 -; GCN-NEXT: v_subrev_i32_e32 v0, vcc, v1, v0 -; GCN-NEXT: v_cndmask_b32_e64 v0, v0, v2, s[0:1] -; GCN-NEXT: v_mul_hi_u32 v0, v0, s3 +; GCN-NEXT: s_lshr_b32 s0, s3, 8 +; GCN-NEXT: v_cvt_f32_u32_e32 v0, s0 +; GCN-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 +; GCN-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-NEXT: v_mad_f32 v0, -v1, s8, v0 +; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s8 +; GCN-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc ; GCN-NEXT: v_mul_lo_u32 v0, v0, s2 -; GCN-NEXT: v_sub_i32_e32 v1, vcc, s3, v0 -; GCN-NEXT: v_add_i32_e32 v2, vcc, s2, v1 -; GCN-NEXT: v_cmp_ge_u32_e64 s[0:1], s3, v0 -; GCN-NEXT: v_cmp_lt_u32_e64 s[2:3], s8, v1 -; GCN-NEXT: v_add_i32_e32 v0, vcc, 0xffffa481, v1 -; GCN-NEXT: s_and_b64 vcc, s[2:3], s[0:1] -; GCN-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[0:1] +; GCN-NEXT: v_sub_i32_e32 v0, vcc, s0, v0 +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 ; GCN-NEXT: s_endpgm +; +; GCN-IR-LABEL: s_test_urem24_k_den_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9 +; GCN-IR-NEXT: s_mov_b32 s7, 0xf000 +; GCN-IR-NEXT: s_mov_b32 s6, -1 +; GCN-IR-NEXT: s_mov_b32 s8, 0x46b6fe00 +; GCN-IR-NEXT: s_waitcnt lgkmcnt(0) +; GCN-IR-NEXT: s_movk_i32 s2, 0x5b7f +; GCN-IR-NEXT: s_mov_b32 s4, s0 +; GCN-IR-NEXT: s_mov_b32 s5, s1 +; GCN-IR-NEXT: s_lshr_b32 s0, s3, 8 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v0, s0 +; GCN-IR-NEXT: v_mul_f32_e32 v1, 0x38331158, v0 +; GCN-IR-NEXT: v_trunc_f32_e32 v1, v1 +; GCN-IR-NEXT: v_mad_f32 v0, -v1, s8, v0 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v1, v1 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v0|, s8 +; GCN-IR-NEXT: v_addc_u32_e32 v0, vcc, 0, v1, vcc +; GCN-IR-NEXT: v_mul_lo_u32 v0, v0, s2 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, s0, v0 +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0 +; GCN-IR-NEXT: s_endpgm %x.shr = lshr i64 %x, 40 %result = urem i64 %x.shr, 23423 store i64 %result, i64 addrspace(1)* %out @@ -899,31 +1535,39 @@ define i64 @v_test_urem24_k_num_i64(i64 %x) { ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-NEXT: s_mov_b32 s4, 0x41c00000 ; GCN-NEXT: v_cvt_f32_u32_e32 v1, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v1 -; GCN-NEXT: v_mul_f32_e32 v1, 0x4f800000, v1 -; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 -; GCN-NEXT: v_mul_hi_u32 v2, v1, v0 -; GCN-NEXT: v_mul_lo_u32 v3, v1, v0 -; GCN-NEXT: v_sub_i32_e32 v4, vcc, 0, v3 -; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v2 -; GCN-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GCN-NEXT: v_mul_hi_u32 v2, v2, v1 -; GCN-NEXT: v_add_i32_e64 v3, s[4:5], v1, v2 -; GCN-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v2 -; GCN-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc -; GCN-NEXT: v_mul_hi_u32 v1, v1, 24 -; GCN-NEXT: v_mul_lo_u32 v1, v1, v0 -; GCN-NEXT: v_sub_i32_e32 v2, vcc, 24, v1 -; GCN-NEXT: v_add_i32_e32 v3, vcc, v2, v0 -; GCN-NEXT: v_cmp_gt_u32_e32 vcc, 25, v1 -; GCN-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v0 -; GCN-NEXT: v_sub_i32_e64 v0, s[6:7], v2, v0 -; GCN-NEXT: s_and_b64 s[4:5], s[4:5], vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[4:5] -; GCN-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, s4, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v3, -v2, v1, s4 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v1 +; GCN-NEXT: v_addc_u32_e32 v1, vcc, 0, v2, vcc +; GCN-NEXT: v_mul_lo_u32 v0, v1, v0 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_urem24_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-IR-NEXT: s_mov_b32 s4, 0x41c00000 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, v0 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, s4, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v3, -v2, v1, s4 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_mul_lo_u32 v0, v1, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 24, v0 +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = lshr i64 %x, 40 %result = urem i64 24, %x.shr ret i64 %result @@ -934,32 +1578,39 @@ define i64 @v_test_urem24_pow2_k_num_i64(i64 %x) { ; GCN: ; %bb.0: ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: v_lshrrev_b32_e32 v0, 8, v1 -; GCN-NEXT: s_mov_b32 s6, 0x8001 +; GCN-NEXT: s_mov_b32 s4, 0x47000000 ; GCN-NEXT: v_cvt_f32_u32_e32 v1, v0 -; GCN-NEXT: v_rcp_iflag_f32_e32 v1, v1 -; GCN-NEXT: v_mul_f32_e32 v1, 0x4f800000, v1 -; GCN-NEXT: v_cvt_u32_f32_e32 v1, v1 -; GCN-NEXT: v_mul_hi_u32 v2, v1, v0 -; GCN-NEXT: v_mul_lo_u32 v3, v1, v0 -; GCN-NEXT: v_sub_i32_e32 v4, vcc, 0, v3 -; GCN-NEXT: v_cmp_eq_u32_e32 vcc, 0, v2 -; GCN-NEXT: v_cndmask_b32_e32 v2, v3, v4, vcc -; GCN-NEXT: v_mul_hi_u32 v2, v2, v1 -; GCN-NEXT: v_add_i32_e64 v3, s[4:5], v1, v2 -; GCN-NEXT: v_sub_i32_e64 v1, s[4:5], v1, v2 -; GCN-NEXT: v_cndmask_b32_e32 v1, v1, v3, vcc -; GCN-NEXT: v_lshrrev_b32_e32 v1, 17, v1 -; GCN-NEXT: v_mul_u32_u24_e32 v1, v1, v0 -; GCN-NEXT: v_sub_i32_e32 v2, vcc, 0x8000, v1 -; GCN-NEXT: v_add_i32_e32 v3, vcc, v2, v0 -; GCN-NEXT: v_cmp_gt_u32_e32 vcc, s6, v1 -; GCN-NEXT: v_cmp_ge_u32_e64 s[4:5], v2, v0 -; GCN-NEXT: v_sub_i32_e64 v0, s[6:7], v2, v0 -; GCN-NEXT: s_and_b64 s[4:5], s[4:5], vcc -; GCN-NEXT: v_cndmask_b32_e64 v0, v2, v0, s[4:5] -; GCN-NEXT: v_cndmask_b32_e32 v0, v3, v0, vcc +; GCN-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-NEXT: v_mul_f32_e32 v2, s4, v2 +; GCN-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-NEXT: v_mad_f32 v3, -v2, v1, s4 +; GCN-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v1 +; GCN-NEXT: v_addc_u32_e32 v1, vcc, 0, v2, vcc +; GCN-NEXT: v_mul_lo_u32 v0, v1, v0 +; GCN-NEXT: v_sub_i32_e32 v0, vcc, 0x8000, v0 +; GCN-NEXT: v_and_b32_e32 v0, 0xffffff, v0 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_urem24_pow2_k_num_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-IR-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, v0 +; GCN-IR-NEXT: v_rcp_iflag_f32_e32 v2, v1 +; GCN-IR-NEXT: v_mul_f32_e32 v2, s4, v2 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v3, -v2, v1, s4 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v3|, v1 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_mul_lo_u32 v0, v1, v0 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, 0x8000, v0 +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = lshr i64 %x, 40 %result = urem i64 32768, %x.shr ret i64 %result @@ -972,6 +1623,24 @@ define i64 @v_test_urem24_pow2_k_den_i64(i64 %x) { ; GCN-NEXT: v_bfe_u32 v0, v1, 8, 15 ; GCN-NEXT: v_mov_b32_e32 v1, 0 ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GCN-IR-LABEL: v_test_urem24_pow2_k_den_i64: +; GCN-IR: ; %bb.0: +; GCN-IR-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GCN-IR-NEXT: v_lshrrev_b32_e32 v0, 8, v1 +; GCN-IR-NEXT: s_mov_b32 s4, 0x47000000 +; GCN-IR-NEXT: v_cvt_f32_u32_e32 v1, v0 +; GCN-IR-NEXT: v_mul_f32_e32 v2, 0x38000000, v1 +; GCN-IR-NEXT: v_trunc_f32_e32 v2, v2 +; GCN-IR-NEXT: v_mad_f32 v1, -v2, s4, v1 +; GCN-IR-NEXT: v_cvt_u32_f32_e32 v2, v2 +; GCN-IR-NEXT: v_cmp_ge_f32_e64 vcc, |v1|, s4 +; GCN-IR-NEXT: v_addc_u32_e32 v1, vcc, 0, v2, vcc +; GCN-IR-NEXT: v_lshlrev_b32_e32 v1, 15, v1 +; GCN-IR-NEXT: v_sub_i32_e32 v0, vcc, v0, v1 +; GCN-IR-NEXT: v_and_b32_e32 v0, 0xffffff, v0 +; GCN-IR-NEXT: v_mov_b32_e32 v1, 0 +; GCN-IR-NEXT: s_setpc_b64 s[30:31] %x.shr = lshr i64 %x, 40 %result = urem i64 %x.shr, 32768 ret i64 %result -- 2.7.4