From 89a68b5c210419f014e58c9c89084b168863d838 Mon Sep 17 00:00:00 2001 From: Deric Cheung Date: Tue, 15 Jul 2025 16:36:47 +0000 Subject: [PATCH] Emit array GEPs in memcpy/memset legalization --- llvm/lib/Target/DirectX/DXILLegalizePass.cpp | 22 ++++---- llvm/test/CodeGen/DirectX/legalize-memcpy.ll | 56 ++++++++++---------- llvm/test/CodeGen/DirectX/legalize-memset.ll | 20 +++---- 3 files changed, 49 insertions(+), 49 deletions(-) diff --git a/llvm/lib/Target/DirectX/DXILLegalizePass.cpp b/llvm/lib/Target/DirectX/DXILLegalizePass.cpp index 76a46c7a2b760..6eff77f8d51e5 100644 --- a/llvm/lib/Target/DirectX/DXILLegalizePass.cpp +++ b/llvm/lib/Target/DirectX/DXILLegalizePass.cpp @@ -347,7 +347,6 @@ static void emitMemcpyExpansion(IRBuilder<> &Builder, Value *Dst, Value *Src, if (ByteLength == 0) return; - LLVMContext &Ctx = Builder.getContext(); const DataLayout &DL = Builder.GetInsertBlock()->getModule()->getDataLayout(); auto GetArrTyFromVal = [](Value *Val) -> ArrayType * { @@ -392,10 +391,11 @@ static void emitMemcpyExpansion(IRBuilder<> &Builder, Value *Dst, Value *Src, assert(ByteLength % DstElemByteSize == 0 && "memcpy length must be divisible by array element type"); for (uint64_t I = 0; I < NumElemsToCopy; ++I) { - Value *Offset = ConstantInt::get(Type::getInt32Ty(Ctx), I); - Value *SrcPtr = Builder.CreateInBoundsGEP(SrcElemTy, Src, Offset, "gep"); + SmallVector Indices = {Builder.getInt32(0), + Builder.getInt32(I)}; + Value *SrcPtr = Builder.CreateInBoundsGEP(SrcArrTy, Src, Indices, "gep"); Value *SrcVal = Builder.CreateLoad(SrcElemTy, SrcPtr); - Value *DstPtr = Builder.CreateInBoundsGEP(DstElemTy, Dst, Offset, "gep"); + Value *DstPtr = Builder.CreateInBoundsGEP(DstArrTy, Dst, Indices, "gep"); Builder.CreateStore(SrcVal, DstPtr); } } @@ -403,7 +403,6 @@ static void emitMemcpyExpansion(IRBuilder<> &Builder, Value *Dst, Value *Src, static void emitMemsetExpansion(IRBuilder<> &Builder, Value *Dst, Value *Val, ConstantInt *SizeCI, DenseMap &ReplacedValues) { - LLVMContext &Ctx = Builder.getContext(); [[maybe_unused]] const DataLayout &DL = Builder.GetInsertBlock()->getModule()->getDataLayout(); [[maybe_unused]] uint64_t OrigSize = SizeCI->getZExtValue(); @@ -444,8 +443,9 @@ static void emitMemsetExpansion(IRBuilder<> &Builder, Value *Dst, Value *Val, } for (uint64_t I = 0; I < Size; ++I) { - Value *Offset = ConstantInt::get(Type::getInt32Ty(Ctx), I); - Value *Ptr = Builder.CreateGEP(ElemTy, Dst, Offset, "gep"); + Value *Zero = Builder.getInt32(0); + Value *Offset = Builder.getInt32(I); + Value *Ptr = Builder.CreateGEP(ArrTy, Dst, {Zero, Offset}, "gep"); Builder.CreateStore(TypedVal, Ptr); } } @@ -478,9 +478,9 @@ static void legalizeMemCpy(Instruction &I, ToRemove.push_back(CI); } -static void removeMemSet(Instruction &I, - SmallVectorImpl &ToRemove, - DenseMap &ReplacedValues) { +static void legalizeMemSet(Instruction &I, + SmallVectorImpl &ToRemove, + DenseMap &ReplacedValues) { CallInst *CI = dyn_cast(&I); if (!CI) @@ -603,7 +603,7 @@ class DXILLegalizationPipeline { LegalizationPipeline[Stage1].push_back(legalizeGetHighLowi64Bytes); LegalizationPipeline[Stage1].push_back(legalizeFreeze); LegalizationPipeline[Stage1].push_back(legalizeMemCpy); - LegalizationPipeline[Stage1].push_back(removeMemSet); + LegalizationPipeline[Stage1].push_back(legalizeMemSet); LegalizationPipeline[Stage1].push_back(updateFnegToFsub); // Note: legalizeGetHighLowi64Bytes and // downcastI64toI32InsertExtractElements both modify extractelement, so they diff --git a/llvm/test/CodeGen/DirectX/legalize-memcpy.ll b/llvm/test/CodeGen/DirectX/legalize-memcpy.ll index 0cf9d83b3af25..55bdefc12aa77 100644 --- a/llvm/test/CodeGen/DirectX/legalize-memcpy.ll +++ b/llvm/test/CodeGen/DirectX/legalize-memcpy.ll @@ -6,9 +6,9 @@ define void @replace_int_memcpy_test() #0 { ; CHECK-SAME: ) #[[ATTR0:[0-9]+]] { ; CHECK-NEXT: [[TMP1:%.*]] = alloca [1 x i32], align 4 ; CHECK-NEXT: [[TMP2:%.*]] = alloca [1 x i32], align 4 -; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds [1 x i32], ptr [[TMP1]], i32 0, i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[GEP]], align 4 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 0 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [1 x i32], ptr [[TMP2]], i32 0, i32 0 ; CHECK-NEXT: store i32 [[TMP3]], ptr [[GEP1]], align 4 ; CHECK-NEXT: ret void ; @@ -23,17 +23,17 @@ define void @replace_3int_memcpy_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = alloca [3 x i32], align 4 ; CHECK-NEXT: [[TMP2:%.*]] = alloca [3 x i32], align 4 -; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds [3 x i32], ptr [[TMP1]], i32 0, i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[GEP]], align 4 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 0 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [3 x i32], ptr [[TMP2]], i32 0, i32 0 ; CHECK-NEXT: store i32 [[TMP3]], ptr [[GEP1]], align 4 -; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 1 +; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [3 x i32], ptr [[TMP1]], i32 0, i32 1 ; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[GEP2]], align 4 -; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 1 +; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds [3 x i32], ptr [[TMP2]], i32 0, i32 1 ; CHECK-NEXT: store i32 [[TMP4]], ptr [[GEP3]], align 4 -; CHECK-NEXT: [[GEP4:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 2 +; CHECK-NEXT: [[GEP4:%.*]] = getelementptr inbounds [3 x i32], ptr [[TMP1]], i32 0, i32 2 ; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[GEP4]], align 4 -; CHECK-NEXT: [[GEP5:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 2 +; CHECK-NEXT: [[GEP5:%.*]] = getelementptr inbounds [3 x i32], ptr [[TMP2]], i32 0, i32 2 ; CHECK-NEXT: store i32 [[TMP5]], ptr [[GEP5]], align 4 ; CHECK-NEXT: ret void ; @@ -48,13 +48,13 @@ define void @replace_mismatched_size_int_memcpy_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = alloca [2 x i32], align 4 ; CHECK-NEXT: [[TMP2:%.*]] = alloca [3 x i32], align 4 -; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds [2 x i32], ptr [[TMP1]], i32 0, i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr [[GEP]], align 4 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 0 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [3 x i32], ptr [[TMP2]], i32 0, i32 0 ; CHECK-NEXT: store i32 [[TMP3]], ptr [[GEP1]], align 4 -; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds i32, ptr [[TMP1]], i32 1 +; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [2 x i32], ptr [[TMP1]], i32 0, i32 1 ; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[GEP2]], align 4 -; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 1 +; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds [3 x i32], ptr [[TMP2]], i32 0, i32 1 ; CHECK-NEXT: store i32 [[TMP4]], ptr [[GEP3]], align 4 ; CHECK-NEXT: ret void ; @@ -69,13 +69,13 @@ define void @replace_int16_memcpy_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = alloca [2 x i16], align 2 ; CHECK-NEXT: [[TMP2:%.*]] = alloca [2 x i16], align 2 -; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds [2 x i16], ptr [[TMP1]], i32 0, i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = load i16, ptr [[GEP]], align 2 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds i16, ptr [[TMP2]], i32 0 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [2 x i16], ptr [[TMP2]], i32 0, i32 0 ; CHECK-NEXT: store i16 [[TMP3]], ptr [[GEP1]], align 2 -; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i32 1 +; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [2 x i16], ptr [[TMP1]], i32 0, i32 1 ; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr [[GEP2]], align 2 -; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds i16, ptr [[TMP2]], i32 1 +; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds [2 x i16], ptr [[TMP2]], i32 0, i32 1 ; CHECK-NEXT: store i16 [[TMP4]], ptr [[GEP3]], align 2 ; CHECK-NEXT: ret void ; @@ -90,13 +90,13 @@ define void @replace_float_memcpy_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = alloca [2 x float], align 4 ; CHECK-NEXT: [[TMP2:%.*]] = alloca [2 x float], align 4 -; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds [2 x float], ptr [[TMP1]], i32 0, i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = load float, ptr [[GEP]], align 4 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds float, ptr [[TMP2]], i32 0 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [2 x float], ptr [[TMP2]], i32 0, i32 0 ; CHECK-NEXT: store float [[TMP3]], ptr [[GEP1]], align 4 -; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i32 1 +; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [2 x float], ptr [[TMP1]], i32 0, i32 1 ; CHECK-NEXT: [[TMP4:%.*]] = load float, ptr [[GEP2]], align 4 -; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds float, ptr [[TMP2]], i32 1 +; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds [2 x float], ptr [[TMP2]], i32 0, i32 1 ; CHECK-NEXT: store float [[TMP4]], ptr [[GEP3]], align 4 ; CHECK-NEXT: ret void ; @@ -111,13 +111,13 @@ define void @replace_double_memcpy_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = alloca [2 x double], align 4 ; CHECK-NEXT: [[TMP2:%.*]] = alloca [2 x double], align 4 -; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds double, ptr [[TMP1]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds [2 x double], ptr [[TMP1]], i32 0, i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = load double, ptr [[GEP]], align 8 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds double, ptr [[TMP2]], i32 0 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [2 x double], ptr [[TMP2]], i32 0, i32 0 ; CHECK-NEXT: store double [[TMP3]], ptr [[GEP1]], align 8 -; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds double, ptr [[TMP1]], i32 1 +; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [2 x double], ptr [[TMP1]], i32 0, i32 1 ; CHECK-NEXT: [[TMP4:%.*]] = load double, ptr [[GEP2]], align 8 -; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds double, ptr [[TMP2]], i32 1 +; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds [2 x double], ptr [[TMP2]], i32 0, i32 1 ; CHECK-NEXT: store double [[TMP4]], ptr [[GEP3]], align 8 ; CHECK-NEXT: ret void ; @@ -132,13 +132,13 @@ define void @replace_half_memcpy_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = alloca [2 x half], align 2 ; CHECK-NEXT: [[TMP2:%.*]] = alloca [2 x half], align 2 -; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds half, ptr [[TMP1]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds [2 x half], ptr [[TMP1]], i32 0, i32 0 ; CHECK-NEXT: [[TMP3:%.*]] = load half, ptr [[GEP]], align 2 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds half, ptr [[TMP2]], i32 0 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [2 x half], ptr [[TMP2]], i32 0, i32 0 ; CHECK-NEXT: store half [[TMP3]], ptr [[GEP1]], align 2 -; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds half, ptr [[TMP1]], i32 1 +; CHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [2 x half], ptr [[TMP1]], i32 0, i32 1 ; CHECK-NEXT: [[TMP4:%.*]] = load half, ptr [[GEP2]], align 2 -; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds half, ptr [[TMP2]], i32 1 +; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds [2 x half], ptr [[TMP2]], i32 0, i32 1 ; CHECK-NEXT: store half [[TMP4]], ptr [[GEP3]], align 2 ; CHECK-NEXT: ret void ; diff --git a/llvm/test/CodeGen/DirectX/legalize-memset.ll b/llvm/test/CodeGen/DirectX/legalize-memset.ll index e97817ba824ed..a73e7378cfb9f 100644 --- a/llvm/test/CodeGen/DirectX/legalize-memset.ll +++ b/llvm/test/CodeGen/DirectX/legalize-memset.ll @@ -6,9 +6,9 @@ define void @replace_float_memset_test() #0 { ; CHECK-SAME: ) #[[ATTR0:[0-9]+]] { ; CHECK-NEXT: [[ACCUM_I_FLAT:%.*]] = alloca [2 x float], align 4 ; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 8, ptr nonnull [[ACCUM_I_FLAT]]) -; CHECK-NEXT: [[GEP:%.*]] = getelementptr float, ptr [[ACCUM_I_FLAT]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr [2 x float], ptr [[ACCUM_I_FLAT]], i32 0, i32 0 ; CHECK-NEXT: store float 0.000000e+00, ptr [[GEP]], align 4 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr float, ptr [[ACCUM_I_FLAT]], i32 1 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr [2 x float], ptr [[ACCUM_I_FLAT]], i32 0, i32 1 ; CHECK-NEXT: store float 0.000000e+00, ptr [[GEP1]], align 4 ; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 8, ptr nonnull [[ACCUM_I_FLAT]]) ; CHECK-NEXT: ret void @@ -25,9 +25,9 @@ define void @replace_half_memset_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[ACCUM_I_FLAT:%.*]] = alloca [2 x half], align 4 ; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 4, ptr nonnull [[ACCUM_I_FLAT]]) -; CHECK-NEXT: [[GEP:%.*]] = getelementptr half, ptr [[ACCUM_I_FLAT]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr [2 x half], ptr [[ACCUM_I_FLAT]], i32 0, i32 0 ; CHECK-NEXT: store half 0xH0000, ptr [[GEP]], align 2 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr half, ptr [[ACCUM_I_FLAT]], i32 1 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr [2 x half], ptr [[ACCUM_I_FLAT]], i32 0, i32 1 ; CHECK-NEXT: store half 0xH0000, ptr [[GEP1]], align 2 ; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 4, ptr nonnull [[ACCUM_I_FLAT]]) ; CHECK-NEXT: ret void @@ -44,9 +44,9 @@ define void @replace_double_memset_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[ACCUM_I_FLAT:%.*]] = alloca [2 x double], align 4 ; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 16, ptr nonnull [[ACCUM_I_FLAT]]) -; CHECK-NEXT: [[GEP:%.*]] = getelementptr double, ptr [[ACCUM_I_FLAT]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr [2 x double], ptr [[ACCUM_I_FLAT]], i32 0, i32 0 ; CHECK-NEXT: store double 0.000000e+00, ptr [[GEP]], align 8 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr double, ptr [[ACCUM_I_FLAT]], i32 1 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr [2 x double], ptr [[ACCUM_I_FLAT]], i32 0, i32 1 ; CHECK-NEXT: store double 0.000000e+00, ptr [[GEP1]], align 8 ; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 16, ptr nonnull [[ACCUM_I_FLAT]]) ; CHECK-NEXT: ret void @@ -63,9 +63,9 @@ define void @replace_int16_memset_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[CACHE_I:%.*]] = alloca [2 x i16], align 2 ; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 4, ptr nonnull [[CACHE_I]]) -; CHECK-NEXT: [[GEP:%.*]] = getelementptr i16, ptr [[CACHE_I]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr [2 x i16], ptr [[CACHE_I]], i32 0, i32 0 ; CHECK-NEXT: store i16 0, ptr [[GEP]], align 2 -; CHECK-NEXT: [[GEP1:%.*]] = getelementptr i16, ptr [[CACHE_I]], i32 1 +; CHECK-NEXT: [[GEP1:%.*]] = getelementptr [2 x i16], ptr [[CACHE_I]], i32 0, i32 1 ; CHECK-NEXT: store i16 0, ptr [[GEP1]], align 2 ; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 4, ptr nonnull [[CACHE_I]]) ; CHECK-NEXT: ret void @@ -82,7 +82,7 @@ define void @replace_int_memset_test() #0 { ; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-NEXT: [[ACCUM_I_FLAT:%.*]] = alloca [1 x i32], align 4 ; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 4, ptr nonnull [[ACCUM_I_FLAT]]) -; CHECK-NEXT: [[GEP:%.*]] = getelementptr i32, ptr [[ACCUM_I_FLAT]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr [1 x i32], ptr [[ACCUM_I_FLAT]], i32 0, i32 0 ; CHECK-NEXT: store i32 0, ptr [[GEP]], align 4 ; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 4, ptr nonnull [[ACCUM_I_FLAT]]) ; CHECK-NEXT: ret void @@ -102,7 +102,7 @@ define void @replace_int_memset_to_var_test() #0 { ; CHECK-NEXT: store i32 1, ptr [[I]], align 4 ; CHECK-NEXT: [[I8_LOAD:%.*]] = load i32, ptr [[I]], align 4 ; CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 4, ptr nonnull [[ACCUM_I_FLAT]]) -; CHECK-NEXT: [[GEP:%.*]] = getelementptr i32, ptr [[ACCUM_I_FLAT]], i32 0 +; CHECK-NEXT: [[GEP:%.*]] = getelementptr [1 x i32], ptr [[ACCUM_I_FLAT]], i32 0, i32 0 ; CHECK-NEXT: store i32 [[I8_LOAD]], ptr [[GEP]], align 4 ; CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 4, ptr nonnull [[ACCUM_I_FLAT]]) ; CHECK-NEXT: ret void