https://github.com/fhahn updated https://github.com/llvm/llvm-project/pull/164821
>From a08699d6405945ffc8043ff5dd2d4cbcea91898f Mon Sep 17 00:00:00 2001 From: Florian Hahn <[email protected]> Date: Thu, 23 Oct 2025 14:01:19 +0100 Subject: [PATCH 1/3] [Clang] Add test --- .../CodeGen/AArch64/ext-vector-coercion.c | 42 +++++++++++++++++++ 1 file changed, 42 insertions(+) create mode 100644 clang/test/CodeGen/AArch64/ext-vector-coercion.c diff --git a/clang/test/CodeGen/AArch64/ext-vector-coercion.c b/clang/test/CodeGen/AArch64/ext-vector-coercion.c new file mode 100644 index 0000000000000..3680225e1e84a --- /dev/null +++ b/clang/test/CodeGen/AArch64/ext-vector-coercion.c @@ -0,0 +1,42 @@ +// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 6 +// RUN: %clang_cc1 -fenable-matrix -triple arm64-apple-macosx %s -emit-llvm -disable-llvm-passes -o - | FileCheck %s + +typedef float float3 __attribute__((ext_vector_type(3))); +struct Vec3 { + union { + struct { + float x; + float y; + float z; + }; + float vec __attribute__((ext_vector_type(3))); + }; +}; + +// CHECK-LABEL: define i128 @add( +// CHECK-SAME: i128 [[A_COERCE:%.*]]) #[[ATTR0:[0-9]+]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[RETVAL:%.*]] = alloca [[STRUCT_VEC3:%.*]], align 16 +// CHECK-NEXT: [[A:%.*]] = alloca [[STRUCT_VEC3]], align 16 +// CHECK-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[A]], i32 0, i32 0 +// CHECK-NEXT: store i128 [[A_COERCE]], ptr [[COERCE_DIVE]], align 16 +// CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[A]], i32 0, i32 0 +// CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x float>, ptr [[TMP0]], align 16 +// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <4 x float> [[LOADVECN]], <4 x float> poison, <3 x i32> <i32 0, i32 1, i32 2> +// CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[A]], i32 0, i32 0 +// CHECK-NEXT: [[LOADVECN1:%.*]] = load <4 x float>, ptr [[TMP1]], align 16 +// CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <4 x float> [[LOADVECN1]], <4 x float> poison, <3 x i32> <i32 0, i32 1, i32 2> +// CHECK-NEXT: [[ADD:%.*]] = fadd <3 x float> [[EXTRACTVEC]], [[EXTRACTVEC2]] +// CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0 +// CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <3 x float> [[ADD]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> +// CHECK-NEXT: store <4 x float> [[EXTRACTVEC3]], ptr [[TMP2]], align 16 +// CHECK-NEXT: [[COERCE_DIVE4:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0 +// CHECK-NEXT: [[TMP3:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16 +// CHECK-NEXT: ret i128 [[TMP3]] +// +struct Vec3 add(struct Vec3 a) { + struct Vec3 res; + res.vec = a.vec + a.vec; + return res; +} + >From 4195a46bf1d310a7a6083d51949ca78a5f2148e6 Mon Sep 17 00:00:00 2001 From: Florian Hahn <[email protected]> Date: Thu, 23 Oct 2025 14:03:30 +0100 Subject: [PATCH 2/3] [Clang] Freeze padded vectors before storing. Currently Clang usually leaves padding bits uninitialized, which means they are undef at the moment. When expanding stores of vector types to include padding, the padding lanes will be poison, hence the padding bits will be poison. This interacts badly with coercion of arguments and return values, where 3 x float vectors will be loaded as i128 integer; poisoning the padding bits will make the whole value poison. --- clang/lib/CodeGen/CGExpr.cpp | 3 +++ clang/test/CodeGen/AArch64/ext-vector-coercion.c | 7 ++++--- 2 files changed, 7 insertions(+), 3 deletions(-) diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp index 301d5770cf78f..a581cf821092f 100644 --- a/clang/lib/CodeGen/CGExpr.cpp +++ b/clang/lib/CodeGen/CGExpr.cpp @@ -2300,6 +2300,9 @@ void CodeGenFunction::EmitStoreOfScalar(llvm::Value *Value, Address Addr, SmallVector<int, 16> Mask(NewVecTy->getNumElements(), -1); std::iota(Mask.begin(), Mask.begin() + VecTy->getNumElements(), 0); Value = Builder.CreateShuffleVector(Value, Mask, "extractVec"); + // The extra lanes will be poison. Freeze the whole vector to make sure + // the padding memory is not poisoned, which may break coercion. + Value = Builder.CreateFreeze(Value); SrcTy = NewVecTy; } if (Addr.getElementType() != SrcTy) diff --git a/clang/test/CodeGen/AArch64/ext-vector-coercion.c b/clang/test/CodeGen/AArch64/ext-vector-coercion.c index 3680225e1e84a..44638fac1560a 100644 --- a/clang/test/CodeGen/AArch64/ext-vector-coercion.c +++ b/clang/test/CodeGen/AArch64/ext-vector-coercion.c @@ -29,10 +29,11 @@ struct Vec3 { // CHECK-NEXT: [[ADD:%.*]] = fadd <3 x float> [[EXTRACTVEC]], [[EXTRACTVEC2]] // CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0 // CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <3 x float> [[ADD]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> -// CHECK-NEXT: store <4 x float> [[EXTRACTVEC3]], ptr [[TMP2]], align 16 +// CHECK-NEXT: [[TMP3:%.*]] = freeze <4 x float> [[EXTRACTVEC3]] +// CHECK-NEXT: store <4 x float> [[TMP3]], ptr [[TMP2]], align 16 // CHECK-NEXT: [[COERCE_DIVE4:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0 -// CHECK-NEXT: [[TMP3:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16 -// CHECK-NEXT: ret i128 [[TMP3]] +// CHECK-NEXT: [[TMP4:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16 +// CHECK-NEXT: ret i128 [[TMP4]] // struct Vec3 add(struct Vec3 a) { struct Vec3 res; >From ae3e4afa1d1c86d91e62e83ed0660eaa8b285737 Mon Sep 17 00:00:00 2001 From: Florian Hahn <[email protected]> Date: Thu, 23 Oct 2025 19:23:07 +0100 Subject: [PATCH 3/3] !fixup use undef instead of poison for padding. --- clang/lib/CodeGen/CGExpr.cpp | 8 ++++---- clang/test/CodeGen/AArch64/ext-vector-coercion.c | 9 ++++----- clang/test/CodeGenCXX/matrix-vector-bit-int.cpp | 8 ++++---- clang/test/CodeGenOpenCL/preserve_vec3.cl | 8 ++++---- 4 files changed, 16 insertions(+), 17 deletions(-) diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp index a581cf821092f..fbec449c9f608 100644 --- a/clang/lib/CodeGen/CGExpr.cpp +++ b/clang/lib/CodeGen/CGExpr.cpp @@ -2299,10 +2299,10 @@ void CodeGenFunction::EmitStoreOfScalar(llvm::Value *Value, Address Addr, VecTy != NewVecTy) { SmallVector<int, 16> Mask(NewVecTy->getNumElements(), -1); std::iota(Mask.begin(), Mask.begin() + VecTy->getNumElements(), 0); - Value = Builder.CreateShuffleVector(Value, Mask, "extractVec"); - // The extra lanes will be poison. Freeze the whole vector to make sure - // the padding memory is not poisoned, which may break coercion. - Value = Builder.CreateFreeze(Value); + // Use undef instead of poison for the padding lanes, to make sure no + // padding bits are poisoned, which may break coercion. + Value = Builder.CreateShuffleVector(Value, llvm::UndefValue::get(VecTy), + Mask, "extractVec"); SrcTy = NewVecTy; } if (Addr.getElementType() != SrcTy) diff --git a/clang/test/CodeGen/AArch64/ext-vector-coercion.c b/clang/test/CodeGen/AArch64/ext-vector-coercion.c index 44638fac1560a..ba6a2947eb05c 100644 --- a/clang/test/CodeGen/AArch64/ext-vector-coercion.c +++ b/clang/test/CodeGen/AArch64/ext-vector-coercion.c @@ -28,12 +28,11 @@ struct Vec3 { // CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <4 x float> [[LOADVECN1]], <4 x float> poison, <3 x i32> <i32 0, i32 1, i32 2> // CHECK-NEXT: [[ADD:%.*]] = fadd <3 x float> [[EXTRACTVEC]], [[EXTRACTVEC2]] // CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0 -// CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <3 x float> [[ADD]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> -// CHECK-NEXT: [[TMP3:%.*]] = freeze <4 x float> [[EXTRACTVEC3]] -// CHECK-NEXT: store <4 x float> [[TMP3]], ptr [[TMP2]], align 16 +// CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <3 x float> [[ADD]], <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> +// CHECK-NEXT: store <4 x float> [[EXTRACTVEC3]], ptr [[TMP2]], align 16 // CHECK-NEXT: [[COERCE_DIVE4:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0 -// CHECK-NEXT: [[TMP4:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16 -// CHECK-NEXT: ret i128 [[TMP4]] +// CHECK-NEXT: [[TMP3:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16 +// CHECK-NEXT: ret i128 [[TMP3]] // struct Vec3 add(struct Vec3 a) { struct Vec3 res; diff --git a/clang/test/CodeGenCXX/matrix-vector-bit-int.cpp b/clang/test/CodeGenCXX/matrix-vector-bit-int.cpp index 2e7531b334ecb..0decf75b80e6a 100644 --- a/clang/test/CodeGenCXX/matrix-vector-bit-int.cpp +++ b/clang/test/CodeGenCXX/matrix-vector-bit-int.cpp @@ -19,7 +19,7 @@ using i4x3x3 = _BitInt(4) __attribute__((matrix_type(3, 3))); // CHECK-NEXT: store i32 [[A_COERCE]], ptr [[A]], align 4 // CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x i8>, ptr [[A]], align 4 // CHECK-NEXT: [[A1:%.*]] = shufflevector <4 x i8> [[LOADVECN]], <4 x i8> poison, <3 x i32> <i32 0, i32 1, i32 2> -// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i8> [[A1]], <3 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> +// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i8> [[A1]], <3 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> // CHECK-NEXT: store <4 x i8> [[EXTRACTVEC]], ptr [[A_ADDR]], align 4 // CHECK-NEXT: [[LOADVECN2:%.*]] = load <4 x i8>, ptr [[A_ADDR]], align 4 // CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <4 x i8> [[LOADVECN2]], <4 x i8> poison, <3 x i32> <i32 0, i32 1, i32 2> @@ -38,7 +38,7 @@ i8x3 v1(i8x3 a) { // CHECK-SAME: <3 x i32> noundef [[A:%.*]]) #[[ATTR1:[0-9]+]] { // CHECK-NEXT: [[ENTRY:.*:]] // CHECK-NEXT: [[A_ADDR:%.*]] = alloca <3 x i32>, align 16 -// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i32> [[A]], <3 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> +// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i32> [[A]], <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> // CHECK-NEXT: store <4 x i32> [[EXTRACTVEC]], ptr [[A_ADDR]], align 16 // CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x i32>, ptr [[A_ADDR]], align 16 // CHECK-NEXT: [[EXTRACTVEC1:%.*]] = shufflevector <4 x i32> [[LOADVECN]], <4 x i32> poison, <3 x i32> <i32 0, i32 1, i32 2> @@ -57,7 +57,7 @@ i32x3 v2(i32x3 a) { // CHECK-NEXT: [[A_ADDR:%.*]] = alloca <3 x i512>, align 256 // CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x i512>, ptr [[TMP0]], align 256 // CHECK-NEXT: [[A:%.*]] = shufflevector <4 x i512> [[LOADVECN]], <4 x i512> poison, <3 x i32> <i32 0, i32 1, i32 2> -// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i512> [[A]], <3 x i512> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> +// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i512> [[A]], <3 x i512> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> // CHECK-NEXT: store <4 x i512> [[EXTRACTVEC]], ptr [[A_ADDR]], align 256 // CHECK-NEXT: [[LOADVECN1:%.*]] = load <4 x i512>, ptr [[A_ADDR]], align 256 // CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <4 x i512> [[LOADVECN1]], <4 x i512> poison, <3 x i32> <i32 0, i32 1, i32 2> @@ -80,7 +80,7 @@ i512x3 v3(i512x3 a) { // CHECK-NEXT: store i32 [[A_COERCE]], ptr [[A]], align 4 // CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x i4>, ptr [[A]], align 4 // CHECK-NEXT: [[A1:%.*]] = shufflevector <4 x i4> [[LOADVECN]], <4 x i4> poison, <3 x i32> <i32 0, i32 1, i32 2> -// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i4> [[A1]], <3 x i4> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> +// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i4> [[A1]], <3 x i4> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> // CHECK-NEXT: store <4 x i4> [[EXTRACTVEC]], ptr [[A_ADDR]], align 4 // CHECK-NEXT: [[LOADVECN2:%.*]] = load <4 x i4>, ptr [[A_ADDR]], align 4 // CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <4 x i4> [[LOADVECN2]], <4 x i4> poison, <3 x i32> <i32 0, i32 1, i32 2> diff --git a/clang/test/CodeGenOpenCL/preserve_vec3.cl b/clang/test/CodeGenOpenCL/preserve_vec3.cl index e76aa81f918cb..66d955b4af443 100644 --- a/clang/test/CodeGenOpenCL/preserve_vec3.cl +++ b/clang/test/CodeGenOpenCL/preserve_vec3.cl @@ -71,8 +71,8 @@ void kernel char8_to_short3(global short3 *a, global char8 *b) { // CHECK-LABEL: define dso_local spir_func void @from_char3( // CHECK-SAME: <3 x i8> noundef [[A:%.*]], ptr addrspace(1) noundef writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] { // CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i8> [[A]], <3 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> -// CHECK-NEXT: store <4 x i8> [[EXTRACTVEC]], ptr addrspace(1) [[OUT]], align 4, !tbaa [[INT_TBAA3:![0-9]+]] +// CHECK-NEXT: [[TMP0:%.*]] = shufflevector <3 x i8> [[A]], <3 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> +// CHECK-NEXT: store <4 x i8> [[TMP0]], ptr addrspace(1) [[OUT]], align 4, !tbaa [[INT_TBAA3:![0-9]+]] // CHECK-NEXT: ret void // void from_char3(char3 a, global int *out) { @@ -82,8 +82,8 @@ void from_char3(char3 a, global int *out) { // CHECK-LABEL: define dso_local spir_func void @from_short3( // CHECK-SAME: <3 x i16> noundef [[A:%.*]], ptr addrspace(1) noundef writeonly captures(none) initializes((0, 8)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR2]] { // CHECK-NEXT: [[ENTRY:.*:]] -// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i16> [[A]], <3 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> -// CHECK-NEXT: store <4 x i16> [[EXTRACTVEC]], ptr addrspace(1) [[OUT]], align 8, !tbaa [[LONG_TBAA19:![0-9]+]] +// CHECK-NEXT: [[TMP0:%.*]] = shufflevector <3 x i16> [[A]], <3 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison> +// CHECK-NEXT: store <4 x i16> [[TMP0]], ptr addrspace(1) [[OUT]], align 8, !tbaa [[LONG_TBAA19:![0-9]+]] // CHECK-NEXT: ret void // void from_short3(short3 a, global long *out) { _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
