pratlucas created this revision.
Herald added subscribers: cfe-commits, danielkiss, kristof.beyls.
Herald added a project: clang.
pratlucas added reviewers: t.p.northover, ostannard, pcc.
During CodeGen for AArch64 Neon intrinsics, Clang was incorrectly
assuming all the pointers from which loads were being generated for vld1
intrinsics were aligned according to the intrinsics result type, causing
alignment faults on the code generated by the backend.
This patch updates vld1 intrinsics' CodeGen to properly capture the
correct load alignment based on the type of the pointer provided as
input for the intrinsic.
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D79721
Files:
clang/lib/CodeGen/CGBuiltin.cpp
clang/test/CodeGen/aarch64-neon-intrinsics.c
Index: clang/test/CodeGen/aarch64-neon-intrinsics.c
===================================================================
--- clang/test/CodeGen/aarch64-neon-intrinsics.c
+++ clang/test/CodeGen/aarch64-neon-intrinsics.c
@@ -8956,7 +8956,7 @@
// CHECK-LABEL: @test_vld1q_u8(
// CHECK: [[TMP0:%.*]] = bitcast i8* %a to <16 x i8>*
-// CHECK: [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* [[TMP0]]
+// CHECK: [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* [[TMP0]], align 1
// CHECK: ret <16 x i8> [[TMP1]]
uint8x16_t test_vld1q_u8(uint8_t const *a) {
return vld1q_u8(a);
@@ -8965,7 +8965,7 @@
// CHECK-LABEL: @test_vld1q_u16(
// CHECK: [[TMP0:%.*]] = bitcast i16* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <8 x i16>*
-// CHECK: [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* [[TMP1]], align 2
// CHECK: ret <8 x i16> [[TMP2]]
uint16x8_t test_vld1q_u16(uint16_t const *a) {
return vld1q_u16(a);
@@ -8974,7 +8974,7 @@
// CHECK-LABEL: @test_vld1q_u32(
// CHECK: [[TMP0:%.*]] = bitcast i32* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x i32>*
-// CHECK: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* [[TMP1]], align 4
// CHECK: ret <4 x i32> [[TMP2]]
uint32x4_t test_vld1q_u32(uint32_t const *a) {
return vld1q_u32(a);
@@ -8983,7 +8983,7 @@
// CHECK-LABEL: @test_vld1q_u64(
// CHECK: [[TMP0:%.*]] = bitcast i64* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <2 x i64>*
-// CHECK: [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 8
// CHECK: ret <2 x i64> [[TMP2]]
uint64x2_t test_vld1q_u64(uint64_t const *a) {
return vld1q_u64(a);
@@ -8991,7 +8991,7 @@
// CHECK-LABEL: @test_vld1q_s8(
// CHECK: [[TMP0:%.*]] = bitcast i8* %a to <16 x i8>*
-// CHECK: [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* [[TMP0]]
+// CHECK: [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* [[TMP0]], align 1
// CHECK: ret <16 x i8> [[TMP1]]
int8x16_t test_vld1q_s8(int8_t const *a) {
return vld1q_s8(a);
@@ -9000,7 +9000,7 @@
// CHECK-LABEL: @test_vld1q_s16(
// CHECK: [[TMP0:%.*]] = bitcast i16* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <8 x i16>*
-// CHECK: [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* [[TMP1]], align 2
// CHECK: ret <8 x i16> [[TMP2]]
int16x8_t test_vld1q_s16(int16_t const *a) {
return vld1q_s16(a);
@@ -9009,7 +9009,7 @@
// CHECK-LABEL: @test_vld1q_s32(
// CHECK: [[TMP0:%.*]] = bitcast i32* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x i32>*
-// CHECK: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <4 x i32>, <4 x i32>* [[TMP1]], align 4
// CHECK: ret <4 x i32> [[TMP2]]
int32x4_t test_vld1q_s32(int32_t const *a) {
return vld1q_s32(a);
@@ -9018,7 +9018,7 @@
// CHECK-LABEL: @test_vld1q_s64(
// CHECK: [[TMP0:%.*]] = bitcast i64* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <2 x i64>*
-// CHECK: [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 8
// CHECK: ret <2 x i64> [[TMP2]]
int64x2_t test_vld1q_s64(int64_t const *a) {
return vld1q_s64(a);
@@ -9027,7 +9027,7 @@
// CHECK-LABEL: @test_vld1q_f16(
// CHECK: [[TMP0:%.*]] = bitcast half* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <8 x half>*
-// CHECK: [[TMP2:%.*]] = load <8 x half>, <8 x half>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <8 x half>, <8 x half>* [[TMP1]], align 2
// CHECK: ret <8 x half> [[TMP2]]
float16x8_t test_vld1q_f16(float16_t const *a) {
return vld1q_f16(a);
@@ -9036,7 +9036,7 @@
// CHECK-LABEL: @test_vld1q_f32(
// CHECK: [[TMP0:%.*]] = bitcast float* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x float>*
-// CHECK: [[TMP2:%.*]] = load <4 x float>, <4 x float>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <4 x float>, <4 x float>* [[TMP1]], align 4
// CHECK: ret <4 x float> [[TMP2]]
float32x4_t test_vld1q_f32(float32_t const *a) {
return vld1q_f32(a);
@@ -9045,7 +9045,7 @@
// CHECK-LABEL: @test_vld1q_f64(
// CHECK: [[TMP0:%.*]] = bitcast double* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <2 x double>*
-// CHECK: [[TMP2:%.*]] = load <2 x double>, <2 x double>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <2 x double>, <2 x double>* [[TMP1]], align 8
// CHECK: ret <2 x double> [[TMP2]]
float64x2_t test_vld1q_f64(float64_t const *a) {
return vld1q_f64(a);
@@ -9053,7 +9053,7 @@
// CHECK-LABEL: @test_vld1q_p8(
// CHECK: [[TMP0:%.*]] = bitcast i8* %a to <16 x i8>*
-// CHECK: [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* [[TMP0]]
+// CHECK: [[TMP1:%.*]] = load <16 x i8>, <16 x i8>* [[TMP0]], align 1
// CHECK: ret <16 x i8> [[TMP1]]
poly8x16_t test_vld1q_p8(poly8_t const *a) {
return vld1q_p8(a);
@@ -9062,7 +9062,7 @@
// CHECK-LABEL: @test_vld1q_p16(
// CHECK: [[TMP0:%.*]] = bitcast i16* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <8 x i16>*
-// CHECK: [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* [[TMP1]], align 2
// CHECK: ret <8 x i16> [[TMP2]]
poly16x8_t test_vld1q_p16(poly16_t const *a) {
return vld1q_p16(a);
@@ -9070,7 +9070,7 @@
// CHECK-LABEL: @test_vld1_u8(
// CHECK: [[TMP0:%.*]] = bitcast i8* %a to <8 x i8>*
-// CHECK: [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]]
+// CHECK: [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]], align 1
// CHECK: ret <8 x i8> [[TMP1]]
uint8x8_t test_vld1_u8(uint8_t const *a) {
return vld1_u8(a);
@@ -9079,7 +9079,7 @@
// CHECK-LABEL: @test_vld1_u16(
// CHECK: [[TMP0:%.*]] = bitcast i16* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x i16>*
-// CHECK: [[TMP2:%.*]] = load <4 x i16>, <4 x i16>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <4 x i16>, <4 x i16>* [[TMP1]], align 2
// CHECK: ret <4 x i16> [[TMP2]]
uint16x4_t test_vld1_u16(uint16_t const *a) {
return vld1_u16(a);
@@ -9088,7 +9088,7 @@
// CHECK-LABEL: @test_vld1_u32(
// CHECK: [[TMP0:%.*]] = bitcast i32* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <2 x i32>*
-// CHECK: [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* [[TMP1]], align 4
// CHECK: ret <2 x i32> [[TMP2]]
uint32x2_t test_vld1_u32(uint32_t const *a) {
return vld1_u32(a);
@@ -9097,7 +9097,7 @@
// CHECK-LABEL: @test_vld1_u64(
// CHECK: [[TMP0:%.*]] = bitcast i64* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <1 x i64>*
-// CHECK: [[TMP2:%.*]] = load <1 x i64>, <1 x i64>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <1 x i64>, <1 x i64>* [[TMP1]], align 8
// CHECK: ret <1 x i64> [[TMP2]]
uint64x1_t test_vld1_u64(uint64_t const *a) {
return vld1_u64(a);
@@ -9105,7 +9105,7 @@
// CHECK-LABEL: @test_vld1_s8(
// CHECK: [[TMP0:%.*]] = bitcast i8* %a to <8 x i8>*
-// CHECK: [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]]
+// CHECK: [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]], align 1
// CHECK: ret <8 x i8> [[TMP1]]
int8x8_t test_vld1_s8(int8_t const *a) {
return vld1_s8(a);
@@ -9114,7 +9114,7 @@
// CHECK-LABEL: @test_vld1_s16(
// CHECK: [[TMP0:%.*]] = bitcast i16* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x i16>*
-// CHECK: [[TMP2:%.*]] = load <4 x i16>, <4 x i16>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <4 x i16>, <4 x i16>* [[TMP1]], align 2
// CHECK: ret <4 x i16> [[TMP2]]
int16x4_t test_vld1_s16(int16_t const *a) {
return vld1_s16(a);
@@ -9123,7 +9123,7 @@
// CHECK-LABEL: @test_vld1_s32(
// CHECK: [[TMP0:%.*]] = bitcast i32* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <2 x i32>*
-// CHECK: [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <2 x i32>, <2 x i32>* [[TMP1]], align 4
// CHECK: ret <2 x i32> [[TMP2]]
int32x2_t test_vld1_s32(int32_t const *a) {
return vld1_s32(a);
@@ -9132,7 +9132,7 @@
// CHECK-LABEL: @test_vld1_s64(
// CHECK: [[TMP0:%.*]] = bitcast i64* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <1 x i64>*
-// CHECK: [[TMP2:%.*]] = load <1 x i64>, <1 x i64>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <1 x i64>, <1 x i64>* [[TMP1]], align 8
// CHECK: ret <1 x i64> [[TMP2]]
int64x1_t test_vld1_s64(int64_t const *a) {
return vld1_s64(a);
@@ -9141,7 +9141,7 @@
// CHECK-LABEL: @test_vld1_f16(
// CHECK: [[TMP0:%.*]] = bitcast half* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x half>*
-// CHECK: [[TMP2:%.*]] = load <4 x half>, <4 x half>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <4 x half>, <4 x half>* [[TMP1]], align 2
// CHECK: ret <4 x half> [[TMP2]]
float16x4_t test_vld1_f16(float16_t const *a) {
return vld1_f16(a);
@@ -9150,7 +9150,7 @@
// CHECK-LABEL: @test_vld1_f32(
// CHECK: [[TMP0:%.*]] = bitcast float* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <2 x float>*
-// CHECK: [[TMP2:%.*]] = load <2 x float>, <2 x float>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <2 x float>, <2 x float>* [[TMP1]], align 4
// CHECK: ret <2 x float> [[TMP2]]
float32x2_t test_vld1_f32(float32_t const *a) {
return vld1_f32(a);
@@ -9159,7 +9159,7 @@
// CHECK-LABEL: @test_vld1_f64(
// CHECK: [[TMP0:%.*]] = bitcast double* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <1 x double>*
-// CHECK: [[TMP2:%.*]] = load <1 x double>, <1 x double>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <1 x double>, <1 x double>* [[TMP1]], align 8
// CHECK: ret <1 x double> [[TMP2]]
float64x1_t test_vld1_f64(float64_t const *a) {
return vld1_f64(a);
@@ -9167,7 +9167,7 @@
// CHECK-LABEL: @test_vld1_p8(
// CHECK: [[TMP0:%.*]] = bitcast i8* %a to <8 x i8>*
-// CHECK: [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]]
+// CHECK: [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]], align 1
// CHECK: ret <8 x i8> [[TMP1]]
poly8x8_t test_vld1_p8(poly8_t const *a) {
return vld1_p8(a);
@@ -9176,7 +9176,7 @@
// CHECK-LABEL: @test_vld1_p16(
// CHECK: [[TMP0:%.*]] = bitcast i16* %a to i8*
// CHECK: [[TMP1:%.*]] = bitcast i8* [[TMP0]] to <4 x i16>*
-// CHECK: [[TMP2:%.*]] = load <4 x i16>, <4 x i16>* [[TMP1]]
+// CHECK: [[TMP2:%.*]] = load <4 x i16>, <4 x i16>* [[TMP1]], align 2
// CHECK: ret <4 x i16> [[TMP2]]
poly16x4_t test_vld1_p16(poly16_t const *a) {
return vld1_p16(a);
Index: clang/lib/CodeGen/CGBuiltin.cpp
===================================================================
--- clang/lib/CodeGen/CGBuiltin.cpp
+++ clang/lib/CodeGen/CGBuiltin.cpp
@@ -10283,9 +10283,9 @@
}
case NEON::BI__builtin_neon_vld1_v:
case NEON::BI__builtin_neon_vld1q_v: {
+ auto Alignment = getNaturalPointeeTypeAlignment(
+ E->getArg(0)->IgnoreParenCasts()->getType());
Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy));
- auto Alignment = CharUnits::fromQuantity(
- BuiltinID == NEON::BI__builtin_neon_vld1_v ? 8 : 16);
return Builder.CreateAlignedLoad(VTy, Ops[0], Alignment);
}
case NEON::BI__builtin_neon_vst1_v:
@@ -10298,8 +10298,8 @@
Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
Ty = llvm::PointerType::getUnqual(VTy->getElementType());
Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
- auto Alignment = CharUnits::fromQuantity(
- BuiltinID == NEON::BI__builtin_neon_vld1_lane_v ? 8 : 16);
+ auto Alignment = getNaturalPointeeTypeAlignment(
+ E->getArg(0)->IgnoreParenCasts()->getType());
Ops[0] =
Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment);
return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vld1_lane");
@@ -10309,8 +10309,8 @@
Value *V = UndefValue::get(Ty);
Ty = llvm::PointerType::getUnqual(VTy->getElementType());
Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
- auto Alignment = CharUnits::fromQuantity(
- BuiltinID == NEON::BI__builtin_neon_vld1_dup_v ? 8 : 16);
+ auto Alignment = getNaturalPointeeTypeAlignment(
+ E->getArg(0)->IgnoreParenCasts()->getType());
Ops[0] =
Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment);
llvm::Constant *CI = ConstantInt::get(Int32Ty, 0);
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
