Skip to content

Commit ae3e4af

Browse files
fhahnfhahn
committed
!fixup use undef instead of poison for padding.
1 parent 4195a46 commit ae3e4af

File tree

4 files changed

+16
-17
lines changed

4 files changed

+16
-17
lines changed

‎clang/lib/CodeGen/CGExpr.cpp‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -2299,10 +2299,10 @@ void CodeGenFunction::EmitStoreOfScalar(llvm::Value *Value, Address Addr,
22992299
VecTy != NewVecTy) {
23002300
SmallVector<int, 16> Mask(NewVecTy->getNumElements(), -1);
23012301
std::iota(Mask.begin(), Mask.begin() + VecTy->getNumElements(), 0);
2302-
Value = Builder.CreateShuffleVector(Value, Mask, "extractVec");
2303-
// The extra lanes will be poison. Freeze the whole vector to make sure
2304-
// the padding memory is not poisoned, which may break coercion.
2305-
Value = Builder.CreateFreeze(Value);
2302+
// Use undef instead of poison for the padding lanes, to make sure no
2303+
// padding bits are poisoned, which may break coercion.
2304+
Value = Builder.CreateShuffleVector(Value, llvm::UndefValue::get(VecTy),
2305+
Mask, "extractVec");
23062306
SrcTy = NewVecTy;
23072307
}
23082308
if (Addr.getElementType() != SrcTy)

‎clang/test/CodeGen/AArch64/ext-vector-coercion.c‎

Lines changed: 4 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -28,12 +28,11 @@ struct Vec3 {
2828
// CHECK-NEXT: [[EXTRACTVEC2:%.*]] = shufflevector <4 x float> [[LOADVECN1]], <4 x float> poison, <3 x i32> <i32 0, i32 1, i32 2>
2929
// CHECK-NEXT: [[ADD:%.*]] = fadd <3 x float> [[EXTRACTVEC]], [[EXTRACTVEC2]]
3030
// CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0
31-
// CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <3 x float> [[ADD]], <3 x float> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
32-
// CHECK-NEXT: [[TMP3:%.*]] = freeze <4 x float> [[EXTRACTVEC3]]
33-
// CHECK-NEXT: store <4 x float> [[TMP3]], ptr [[TMP2]], align 16
31+
// CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <3 x float> [[ADD]], <3 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
32+
// CHECK-NEXT: store <4 x float> [[EXTRACTVEC3]], ptr [[TMP2]], align 16
3433
// CHECK-NEXT: [[COERCE_DIVE4:%.*]] = getelementptr inbounds nuw [[STRUCT_VEC3]], ptr [[RETVAL]], i32 0, i32 0
35-
// CHECK-NEXT: [[TMP4:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16
36-
// CHECK-NEXT: ret i128 [[TMP4]]
34+
// CHECK-NEXT: [[TMP3:%.*]] = load i128, ptr [[COERCE_DIVE4]], align 16
35+
// CHECK-NEXT: ret i128 [[TMP3]]
3736
//
3837
struct Vec3 add(struct Vec3 a) {
3938
struct Vec3 res;

‎clang/test/CodeGenCXX/matrix-vector-bit-int.cpp‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -19,7 +19,7 @@ using i4x3x3 = _BitInt(4) __attribute__((matrix_type(3, 3)));
1919
// CHECK-NEXT: store i32 [[A_COERCE]], ptr [[A]], align 4
2020
// CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x i8>, ptr [[A]], align 4
2121
// CHECK-NEXT: [[A1:%.*]] = shufflevector <4 x i8> [[LOADVECN]], <4 x i8> poison, <3 x i32> <i32 0, i32 1, i32 2>
22-
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i8> [[A1]], <3 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
22+
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i8> [[A1]], <3 x i8> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
2323
// CHECK-NEXT: store <4 x i8> [[EXTRACTVEC]], ptr [[A_ADDR]], align 4
2424
// CHECK-NEXT: [[LOADVECN2:%.*]] = load <4 x i8>, ptr [[A_ADDR]], align 4
2525
// 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) {
3838
// CHECK-SAME: <3 x i32> noundef [[A:%.*]]) #[[ATTR1:[0-9]+]] {
3939
// CHECK-NEXT: [[ENTRY:.*:]]
4040
// CHECK-NEXT: [[A_ADDR:%.*]] = alloca <3 x i32>, align 16
41-
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i32> [[A]], <3 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
41+
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i32> [[A]], <3 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
4242
// CHECK-NEXT: store <4 x i32> [[EXTRACTVEC]], ptr [[A_ADDR]], align 16
4343
// CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x i32>, ptr [[A_ADDR]], align 16
4444
// 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) {
5757
// CHECK-NEXT: [[A_ADDR:%.*]] = alloca <3 x i512>, align 256
5858
// CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x i512>, ptr [[TMP0]], align 256
5959
// CHECK-NEXT: [[A:%.*]] = shufflevector <4 x i512> [[LOADVECN]], <4 x i512> poison, <3 x i32> <i32 0, i32 1, i32 2>
60-
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i512> [[A]], <3 x i512> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
60+
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i512> [[A]], <3 x i512> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
6161
// CHECK-NEXT: store <4 x i512> [[EXTRACTVEC]], ptr [[A_ADDR]], align 256
6262
// CHECK-NEXT: [[LOADVECN1:%.*]] = load <4 x i512>, ptr [[A_ADDR]], align 256
6363
// 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) {
8080
// CHECK-NEXT: store i32 [[A_COERCE]], ptr [[A]], align 4
8181
// CHECK-NEXT: [[LOADVECN:%.*]] = load <4 x i4>, ptr [[A]], align 4
8282
// CHECK-NEXT: [[A1:%.*]] = shufflevector <4 x i4> [[LOADVECN]], <4 x i4> poison, <3 x i32> <i32 0, i32 1, i32 2>
83-
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i4> [[A1]], <3 x i4> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
83+
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i4> [[A1]], <3 x i4> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
8484
// CHECK-NEXT: store <4 x i4> [[EXTRACTVEC]], ptr [[A_ADDR]], align 4
8585
// CHECK-NEXT: [[LOADVECN2:%.*]] = load <4 x i4>, ptr [[A_ADDR]], align 4
8686
// CHECK-NEXT: [[EXTRACTVEC3:%.*]] = shufflevector <4 x i4> [[LOADVECN2]], <4 x i4> poison, <3 x i32> <i32 0, i32 1, i32 2>

‎clang/test/CodeGenOpenCL/preserve_vec3.cl‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -71,8 +71,8 @@ void kernel char8_to_short3(global short3 *a, global char8 *b) {
7171
// CHECK-LABEL: define dso_local spir_func void @from_char3(
7272
// CHECK-SAME: <3 x i8> noundef [[A:%.*]], ptr addrspace(1) noundef writeonly captures(none) initializes((0, 4)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] {
7373
// CHECK-NEXT: [[ENTRY:.*:]]
74-
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i8> [[A]], <3 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
75-
// CHECK-NEXT: store <4 x i8> [[EXTRACTVEC]], ptr addrspace(1) [[OUT]], align 4, !tbaa [[INT_TBAA3:![0-9]+]]
74+
// CHECK-NEXT: [[TMP0:%.*]] = shufflevector <3 x i8> [[A]], <3 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
75+
// CHECK-NEXT: store <4 x i8> [[TMP0]], ptr addrspace(1) [[OUT]], align 4, !tbaa [[INT_TBAA3:![0-9]+]]
7676
// CHECK-NEXT: ret void
7777
//
7878
void from_char3(char3 a, global int *out) {
@@ -82,8 +82,8 @@ void from_char3(char3 a, global int *out) {
8282
// CHECK-LABEL: define dso_local spir_func void @from_short3(
8383
// CHECK-SAME: <3 x i16> noundef [[A:%.*]], ptr addrspace(1) noundef writeonly captures(none) initializes((0, 8)) [[OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
8484
// CHECK-NEXT: [[ENTRY:.*:]]
85-
// CHECK-NEXT: [[EXTRACTVEC:%.*]] = shufflevector <3 x i16> [[A]], <3 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
86-
// CHECK-NEXT: store <4 x i16> [[EXTRACTVEC]], ptr addrspace(1) [[OUT]], align 8, !tbaa [[LONG_TBAA19:![0-9]+]]
85+
// CHECK-NEXT: [[TMP0:%.*]] = shufflevector <3 x i16> [[A]], <3 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 poison>
86+
// CHECK-NEXT: store <4 x i16> [[TMP0]], ptr addrspace(1) [[OUT]], align 8, !tbaa [[LONG_TBAA19:![0-9]+]]
8787
// CHECK-NEXT: ret void
8888
//
8989
void from_short3(short3 a, global long *out) {

0 commit comments

Comments
 (0)