https://github.com/Lukacma updated
https://github.com/llvm/llvm-project/pull/90849
>From 18c489682d38837d7b0abc7b9ecf829a9df4bd0c Mon Sep 17 00:00:00 2001
From: Marian Lukac <marian.lu...@arm.com>
Date: Thu, 2 May 2024 11:33:13 +0000
Subject: [PATCH 1/2] [Clang] Added lifetime markers for temp. allocated
Aggregate types
---
clang/lib/CodeGen/CGCall.cpp | 10 ++++
clang/test/CodeGen/aarch64-byval-temp.c | 46 +++++++++++++++++--
...-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c | 2 +
...cle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.cpp | 2 +
clang/test/CodeGen/nofpclass.c | 22 +++++----
5 files changed, 70 insertions(+), 12 deletions(-)
diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 1b4ca2a8b2fe8..8fdde5326b8ed 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -5202,6 +5202,16 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo
&CallInfo,
Val = Builder.CreateFreeze(Val);
IRCallArgs[FirstIRArg] = Val;
+ // Emit lifetime markers for the temporary alloca.
+ llvm::TypeSize ByvalTempElementSize =
+ CGM.getDataLayout().getTypeAllocSize(Addr.getElementType());
+ llvm::Value *LifetimeSize =
+ EmitLifetimeStart(ByvalTempElementSize, Addr.getPointer());
+
+ // Add cleanup code to emit the end lifetime marker after the call.
+ if (LifetimeSize) // In case we disabled lifetime markers.
+ CallLifetimeEndAfterCall.emplace_back(Addr, LifetimeSize);
+
I->copyInto(*this, Addr);
} else {
// We want to avoid creating an unnecessary temporary+copy here;
diff --git a/clang/test/CodeGen/aarch64-byval-temp.c
b/clang/test/CodeGen/aarch64-byval-temp.c
index e9e2586406e5c..765a298f83ed0 100644
--- a/clang/test/CodeGen/aarch64-byval-temp.c
+++ b/clang/test/CodeGen/aarch64-byval-temp.c
@@ -1,13 +1,15 @@
-// RUN: %clang_cc1 -emit-llvm -triple arm64-- -o - %s -O0 | FileCheck %s
--check-prefix=CHECK-O0
-// RUN: %clang_cc1 -emit-llvm -disable-llvm-optzns -triple arm64-- -o - %s -O3
| FileCheck %s --check-prefix=CHECK-O3
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
UTC_ARGS: --version 4
+// RUN: %clang_cc1 -emit-llvm -triple arm64--
-fexperimental-max-bitint-width=1024 -o - %s -O0 | FileCheck %s
--check-prefix=CHECK-O0
+// RUN: %clang_cc1 -emit-llvm -disable-llvm-optzns
-fexperimental-max-bitint-width=1024 -triple arm64-- -o - %s -O3 | FileCheck
%s --check-prefix=CHECK-O3
struct large {
void* pointers[8];
};
void pass_large(struct large);
+void pass_large_BitInt(_BitInt(129));
-// For arm64, we don't use byval to pass structs but instead we create
+// For arm64, we don't use byval to pass structs and _BitInt(>128) type, but
instead we create
// temporary allocas.
//
// Make sure we generate the appropriate lifetime markers for the temporary
@@ -71,3 +73,41 @@ void example(void) {
// Mark the end of the lifetime of `l`.
// CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 64, ptr %l)
// CHECK-O3-NEXT: ret void
+
+void example_BitInt(void) {
+ _BitInt(129) l = {0};
+ pass_large_BitInt(l);
+ pass_large_BitInt(l);
+}
+// CHECK-O0-LABEL: define dso_local void @example_BitInt(
+// CHECK-O0-NEXT: entry:
+// CHECK-O0-NEXT: [[L:%.*]] = alloca i129, align 16
+// CHECK-O0-NEXT: [[INDIRECT_ARG_TEMP:%.*]] = alloca i129, align 16
+// CHECK-O0-NEXT: [[INDIRECT_ARG_TEMP1:%.*]] = alloca i129, align 16
+// CHECK-O0-NEXT: store i129 0, ptr [[L]], align 16
+// CHECK-O0-NEXT: [[TMP0:%.*]] = load i129, ptr [[L]], align 16
+// CHECK-O0-NEXT: store i129 [[TMP0]], ptr [[INDIRECT_ARG_TEMP]], align 16
+// CHECK-O0-NEXT: call void @pass_large_BitInt(ptr noundef
[[INDIRECT_ARG_TEMP]])
+// CHECK-O0-NEXT: [[TMP1:%.*]] = load i129, ptr [[L]], align 16
+// CHECK-O0-NEXT: store i129 [[TMP1]], ptr [[INDIRECT_ARG_TEMP1]], align 16
+// CHECK-O0-NEXT: call void @pass_large_BitInt(ptr noundef
[[INDIRECT_ARG_TEMP1]])
+// CHECK-O0-NEXT: ret void
+//
+// CHECK-O3-LABEL: define dso_local void @example_BitInt(
+// CHECK-O3-NEXT: entry:
+// CHECK-O3-NEXT: [[L:%.*]] = alloca i129, align 16
+// CHECK-O3-NEXT: [[INDIRECT_ARG_TEMP:%.*]] = alloca i129, align 16
+// CHECK-O3-NEXT: [[INDIRECT_ARG_TEMP1:%.*]] = alloca i129, align 16
+// CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 32, ptr [[L]])
+// CHECK-O3-NEXT: store i129 0, ptr [[L]], align 16, !tbaa [[TBAA6:![0-9]+]]
+// CHECK-O3-NEXT: [[TMP0:%.*]] = load i129, ptr [[L]], align 16, !tbaa
[[TBAA6]]
+// CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 32, ptr
[[INDIRECT_ARG_TEMP]])
+// CHECK-O3-NEXT: store i129 [[TMP0]], ptr [[INDIRECT_ARG_TEMP]], align 16,
!tbaa [[TBAA6]]
+// CHECK-O3-NEXT: call void @pass_large_BitInt(ptr noundef
[[INDIRECT_ARG_TEMP]])
+// CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 32, ptr
[[INDIRECT_ARG_TEMP]])
+// CHECK-O3-NEXT: [[TMP1:%.*]] = load i129, ptr [[L]], align 16, !tbaa
[[TBAA6]]
+// CHECK-O3-NEXT: call void @llvm.lifetime.start.p0(i64 32, ptr
[[INDIRECT_ARG_TEMP1]])
+// CHECK-O3-NEXT: store i129 [[TMP1]], ptr [[INDIRECT_ARG_TEMP1]], align
16, !tbaa [[TBAA6]]
+// CHECK-O3-NEXT: call void @pass_large_BitInt(ptr noundef
[[INDIRECT_ARG_TEMP1]])
+// CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 32, ptr
[[INDIRECT_ARG_TEMP1]])
+// CHECK-O3-NEXT: call void @llvm.lifetime.end.p0(i64 32, ptr [[L]])
diff --git
a/clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c
b/clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c
index a4abe96cc08a6..55e1ed393d848 100644
--- a/clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c
+++ b/clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.c
@@ -88,8 +88,10 @@ typedef svint8_t vec2
__attribute__((arm_sve_vector_bits(N)));
// CHECK-NEXT: entry:
// CHECK-NEXT: [[INDIRECT_ARG_TEMP:%.*]] = alloca <[[#div(VBITS,8)]] x i8>,
align 16
// CHECK-NEXT: [[X:%.*]] = tail call <[[#div(VBITS,8)]] x i8>
@llvm.vector.extract.v[[#div(VBITS,8)]]i8.nxv16i8(<vscale x 16 x i8>
[[X_COERCE:%.*]], i64 0)
+// CHECK-NEXT: call void @llvm.lifetime.start.p0(i64 [[SIZE:[0-9]+]], ptr
nonnull [[INDIRECT_ARG_TEMP]]) #[[ATTR6:[0-9]+]]
// CHECK-NEXT: store <[[#div(VBITS,8)]] x i8> [[X]], ptr
[[INDIRECT_ARG_TEMP]], align 16, [[TBAA6]]
// CHECK-NEXT: call void @f3(ptr noundef nonnull [[INDIRECT_ARG_TEMP]])
[[ATTR5:#.*]]
+// CHECK-NEXT: call void @llvm.lifetime.end.p0(i64 [[SIZE]], ptr nonnull
[[INDIRECT_ARG_TEMP]]) #[[ATTR6:[0-9]+]]
// CHECK-NEXT: ret void
// CHECK128-LABEL: declare void @f3(<16 x i8> noundef)
diff --git
a/clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.cpp
b/clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.cpp
index 05587fd9e7fe2..30ea73b63bce1 100644
--- a/clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.cpp
+++ b/clang/test/CodeGen/aarch64-sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.cpp
@@ -73,8 +73,10 @@ typedef svint16_t vec2
__attribute__((arm_sve_vector_bits(N)));
// CHECK128-NEXT: ret void
// CHECKWIDE-NEXT: [[INDIRECT_ARG_TEMP:%.*]] = alloca <[[#div(VBITS, 16)]] x
i16>, align 16
// CHECKWIDE-NEXT: [[X:%.*]] = tail call <[[#div(VBITS, 16)]] x i16>
@llvm.vector.extract.v[[#div(VBITS, 16)]]i16.nxv8i16(<vscale x 8 x i16>
[[X_COERCE:%.*]], i64 0)
+// CHECKWIDE-NEXT: call void @llvm.lifetime.start.p0(i64 [[SIZE:[0-9]+]],
ptr nonnull [[INDIRECT_ARG_TEMP]]) #[[ATTR6:[0-9]+]]
// CHECKWIDE-NEXT: store <[[#div(VBITS, 16)]] x i16> [[X]], ptr
[[INDIRECT_ARG_TEMP]], align 16, [[TBAA6:!tbaa !.*]]
// CHECKWIDE-NEXT: call void @_Z1fDv[[#div(VBITS, 16)]]_s(ptr noundef
nonnull [[INDIRECT_ARG_TEMP]]) [[ATTR5:#.*]]
+// CHECKWIDE-NEXT: call void @llvm.lifetime.end.p0(i64 [[SIZE]], ptr nonnull
[[INDIRECT_ARG_TEMP]]) #[[ATTR6:[0-9]+]]
// CHECKWIDE-NEXT: ret void
void g(vec2 x) { f(x); } // OK
#endif
diff --git a/clang/test/CodeGen/nofpclass.c b/clang/test/CodeGen/nofpclass.c
index dd90d02f7759b..fc4c64f9b921b 100644
--- a/clang/test/CodeGen/nofpclass.c
+++ b/clang/test/CodeGen/nofpclass.c
@@ -172,7 +172,7 @@ double2 defined_func_v2f64(double2 a, double2 b, double2 c)
{
// CLFINITEONLY-LABEL: define dso_local nofpclass(nan inf) float
@call_extern_func
// CLFINITEONLY-SAME: (float noundef nofpclass(nan inf) [[A:%.*]], double
noundef nofpclass(nan inf) [[B:%.*]], half noundef nofpclass(nan inf)
[[C:%.*]]) local_unnamed_addr #[[ATTR3:[0-9]+]] {
// CLFINITEONLY-NEXT: entry:
-// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
float @extern_func(float noundef nofpclass(nan inf) [[A]], double noundef
nofpclass(nan inf) [[B]], half noundef nofpclass(nan inf) [[C]])
#[[ATTR10:[0-9]+]]
+// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
float @extern_func(float noundef nofpclass(nan inf) [[A]], double noundef
nofpclass(nan inf) [[B]], half noundef nofpclass(nan inf) [[C]])
#[[ATTR11:[0-9]+]]
// CLFINITEONLY-NEXT: ret float [[CALL]]
//
// NONANS: Function Attrs: noinline nounwind optnone
@@ -249,7 +249,7 @@ float call_extern_func(float a, double b, _Float16 c) {
// CLFINITEONLY-LABEL: define dso_local nofpclass(nan inf) double
@call_extern_func_vec
// CLFINITEONLY-SAME: (double noundef nofpclass(nan inf) [[A_COERCE:%.*]], <2
x double> noundef nofpclass(nan inf) [[B:%.*]], i32 noundef [[C_COERCE:%.*]])
local_unnamed_addr #[[ATTR5:[0-9]+]] {
// CLFINITEONLY-NEXT: entry:
-// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
double @extern_func_vec(double noundef nofpclass(nan inf) [[A_COERCE]], <2 x
double> noundef nofpclass(nan inf) [[B]], i32 noundef [[C_COERCE]]) #[[ATTR10]]
+// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
double @extern_func_vec(double noundef nofpclass(nan inf) [[A_COERCE]], <2 x
double> noundef nofpclass(nan inf) [[B]], i32 noundef [[C_COERCE]]) #[[ATTR11]]
// CLFINITEONLY-NEXT: ret double [[CALL]]
//
// NONANS: Function Attrs: noinline nounwind optnone
@@ -389,7 +389,7 @@ float2 call_extern_func_vec(float2 a, double2 b, half2 c) {
// CLFINITEONLY-LABEL: define dso_local nofpclass(nan inf) <2 x float>
@defined_complex_func
// CLFINITEONLY-SAME: (<2 x float> noundef nofpclass(nan inf)
[[A_COERCE:%.*]], double noundef nofpclass(nan inf) [[B_COERCE0:%.*]], double
noundef nofpclass(nan inf) [[B_COERCE1:%.*]], <2 x half> noundef nofpclass(nan
inf) [[C_COERCE:%.*]]) local_unnamed_addr #[[ATTR6:[0-9]+]] {
// CLFINITEONLY-NEXT: entry:
-// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
<2 x float> @extern_complex(<2 x float> noundef nofpclass(nan inf)
[[A_COERCE]], double noundef nofpclass(nan inf) [[B_COERCE0]], double noundef
nofpclass(nan inf) [[B_COERCE1]], <2 x half> noundef nofpclass(nan inf)
[[C_COERCE]]) #[[ATTR10]]
+// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
<2 x float> @extern_complex(<2 x float> noundef nofpclass(nan inf)
[[A_COERCE]], double noundef nofpclass(nan inf) [[B_COERCE0]], double noundef
nofpclass(nan inf) [[B_COERCE1]], <2 x half> noundef nofpclass(nan inf)
[[C_COERCE]]) #[[ATTR11]]
// CLFINITEONLY-NEXT: ret <2 x float> [[CALL]]
//
// NONANS: Function Attrs: noinline nounwind optnone
@@ -927,12 +927,14 @@ _Complex _Float16 defined_complex_func_f16_ret(_Complex
_Float16 c) {
// CLFINITEONLY-NEXT: [[CF16_REAL:%.*]] = load half, ptr [[CF16]], align 8
// CLFINITEONLY-NEXT: [[CF16_IMAGP:%.*]] = getelementptr inbounds i8, ptr
[[CF16]], i64 2
// CLFINITEONLY-NEXT: [[CF16_IMAG:%.*]] = load half, ptr [[CF16_IMAGP]],
align 2
+// CLFINITEONLY-NEXT: call void @llvm.lifetime.start.p0(i64 16, ptr nonnull
[[INDIRECT_ARG_TEMP]]) #[[ATTR12:[0-9]+]]
// CLFINITEONLY-NEXT: [[INDIRECT_ARG_TEMP_IMAGP:%.*]] = getelementptr
inbounds i8, ptr [[INDIRECT_ARG_TEMP]], i64 8
// CLFINITEONLY-NEXT: store double [[CF64_COERCE0]], ptr
[[INDIRECT_ARG_TEMP]], align 8
// CLFINITEONLY-NEXT: store double [[CF64_COERCE1]], ptr
[[INDIRECT_ARG_TEMP_IMAGP]], align 8
// CLFINITEONLY-NEXT: [[COERCE5_SROA_0_0_VEC_INSERT:%.*]] = insertelement
<2 x half> poison, half [[CF16_REAL]], i64 0
// CLFINITEONLY-NEXT: [[COERCE5_SROA_0_2_VEC_INSERT:%.*]] = insertelement
<2 x half> [[COERCE5_SROA_0_0_VEC_INSERT]], half [[CF16_IMAG]], i64 1
-// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
float (float, ...) @variadic(float noundef nofpclass(nan inf) [[F32]], double
noundef nofpclass(nan inf) [[CONV]], double noundef nofpclass(nan inf) [[F64]],
half noundef nofpclass(nan inf) [[F16]], double noundef nofpclass(nan inf)
[[V2F32_COERCE]], <2 x double> noundef nofpclass(nan inf) [[V2F64]], i32
noundef [[V2F16_COERCE]], <2 x float> noundef nofpclass(nan inf)
[[CF32_COERCE]], ptr noundef nonnull byval({ double, double }) align 8
[[INDIRECT_ARG_TEMP]], <2 x half> noundef nofpclass(nan inf)
[[COERCE5_SROA_0_2_VEC_INSERT]]) #[[ATTR10]]
+// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
float (float, ...) @variadic(float noundef nofpclass(nan inf) [[F32]], double
noundef nofpclass(nan inf) [[CONV]], double noundef nofpclass(nan inf) [[F64]],
half noundef nofpclass(nan inf) [[F16]], double noundef nofpclass(nan inf)
[[V2F32_COERCE]], <2 x double> noundef nofpclass(nan inf) [[V2F64]], i32
noundef [[V2F16_COERCE]], <2 x float> noundef nofpclass(nan inf)
[[CF32_COERCE]], ptr noundef nonnull byval({ double, double }) align 8
[[INDIRECT_ARG_TEMP]], <2 x half> noundef nofpclass(nan inf)
[[COERCE5_SROA_0_2_VEC_INSERT]]) #[[ATTR11]]
+// CLFINITEONLY-NEXT: call void @llvm.lifetime.end.p0(i64 16, ptr nonnull
[[INDIRECT_ARG_TEMP]]) #[[ATTR12]]
// CLFINITEONLY-NEXT: ret float [[CALL]]
//
// NONANS: Function Attrs: noinline nounwind optnone
@@ -1178,12 +1180,14 @@ float call_variadic(float f32, double f64, _Float16 f16,
// CLFINITEONLY-NEXT: [[CF16_REAL:%.*]] = load half, ptr [[CF16]], align 8
// CLFINITEONLY-NEXT: [[CF16_IMAGP:%.*]] = getelementptr inbounds i8, ptr
[[CF16]], i64 2
// CLFINITEONLY-NEXT: [[CF16_IMAG:%.*]] = load half, ptr [[CF16_IMAGP]],
align 2
+// CLFINITEONLY-NEXT: call void @llvm.lifetime.start.p0(i64 16, ptr nonnull
[[INDIRECT_ARG_TEMP]]) #[[ATTR12]]
// CLFINITEONLY-NEXT: [[INDIRECT_ARG_TEMP_IMAGP:%.*]] = getelementptr
inbounds i8, ptr [[INDIRECT_ARG_TEMP]], i64 8
// CLFINITEONLY-NEXT: store double [[CF64_COERCE0]], ptr
[[INDIRECT_ARG_TEMP]], align 8
// CLFINITEONLY-NEXT: store double [[CF64_COERCE1]], ptr
[[INDIRECT_ARG_TEMP_IMAGP]], align 8
// CLFINITEONLY-NEXT: [[COERCE5_SROA_0_0_VEC_INSERT:%.*]] = insertelement
<2 x half> poison, half [[CF16_REAL]], i64 0
// CLFINITEONLY-NEXT: [[COERCE5_SROA_0_2_VEC_INSERT:%.*]] = insertelement
<2 x half> [[COERCE5_SROA_0_0_VEC_INSERT]], half [[CF16_IMAG]], i64 1
-// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
float (float, ...) [[FPTR]](float noundef nofpclass(nan inf) [[F32]], double
noundef nofpclass(nan inf) [[CONV]], double noundef nofpclass(nan inf) [[F64]],
half noundef nofpclass(nan inf) [[F16]], double noundef nofpclass(nan inf)
[[V2F32_COERCE]], <2 x double> noundef nofpclass(nan inf) [[V2F64]], i32
noundef [[V2F16_COERCE]], <2 x float> noundef nofpclass(nan inf)
[[CF32_COERCE]], ptr noundef nonnull byval({ double, double }) align 8
[[INDIRECT_ARG_TEMP]], <2 x half> noundef nofpclass(nan inf)
[[COERCE5_SROA_0_2_VEC_INSERT]]) #[[ATTR10]]
+// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
float (float, ...) [[FPTR]](float noundef nofpclass(nan inf) [[F32]], double
noundef nofpclass(nan inf) [[CONV]], double noundef nofpclass(nan inf) [[F64]],
half noundef nofpclass(nan inf) [[F16]], double noundef nofpclass(nan inf)
[[V2F32_COERCE]], <2 x double> noundef nofpclass(nan inf) [[V2F64]], i32
noundef [[V2F16_COERCE]], <2 x float> noundef nofpclass(nan inf)
[[CF32_COERCE]], ptr noundef nonnull byval({ double, double }) align 8
[[INDIRECT_ARG_TEMP]], <2 x half> noundef nofpclass(nan inf)
[[COERCE5_SROA_0_2_VEC_INSERT]]) #[[ATTR11]]
+// CLFINITEONLY-NEXT: call void @llvm.lifetime.end.p0(i64 16, ptr nonnull
[[INDIRECT_ARG_TEMP]]) #[[ATTR12]]
// CLFINITEONLY-NEXT: ret float [[CALL]]
//
// NONANS: Function Attrs: noinline nounwind optnone
@@ -1364,9 +1368,9 @@ extern __m256d extern_m256d(__m256d, ...);
//
// CLFINITEONLY: Function Attrs: convergent norecurse nounwind
// CLFINITEONLY-LABEL: define dso_local nofpclass(nan inf) <4 x double>
@call_m256d
-// CLFINITEONLY-SAME: (<4 x double> noundef nofpclass(nan inf) [[X:%.*]])
local_unnamed_addr #[[ATTR8:[0-9]+]] {
+// CLFINITEONLY-SAME: (<4 x double> noundef nofpclass(nan inf) [[X:%.*]])
local_unnamed_addr #[[ATTR9:[0-9]+]] {
// CLFINITEONLY-NEXT: entry:
-// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
<4 x double> (<4 x double>, ...) @extern_m256d(<4 x double> noundef
nofpclass(nan inf) [[X]], <4 x double> noundef nofpclass(nan inf) [[X]])
#[[ATTR10]]
+// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
<4 x double> (<4 x double>, ...) @extern_m256d(<4 x double> noundef
nofpclass(nan inf) [[X]], <4 x double> noundef nofpclass(nan inf) [[X]])
#[[ATTR11]]
// CLFINITEONLY-NEXT: ret <4 x double> [[CALL]]
//
// NONANS: Function Attrs: noinline nounwind optnone
@@ -1407,9 +1411,9 @@ __m256d call_m256d(__m256d x) {
//
// CLFINITEONLY: Function Attrs: convergent norecurse nounwind
// CLFINITEONLY-LABEL: define dso_local nofpclass(nan inf) <25 x double>
@call_matrix
-// CLFINITEONLY-SAME: (<25 x double> noundef nofpclass(nan inf) [[X:%.*]])
local_unnamed_addr #[[ATTR9:[0-9]+]] {
+// CLFINITEONLY-SAME: (<25 x double> noundef nofpclass(nan inf) [[X:%.*]])
local_unnamed_addr #[[ATTR10:[0-9]+]] {
// CLFINITEONLY-NEXT: entry:
-// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
<25 x double> @extern_matrix(<25 x double> noundef nofpclass(nan inf) [[X]])
#[[ATTR10]]
+// CLFINITEONLY-NEXT: [[CALL:%.*]] = tail call nnan ninf nofpclass(nan inf)
<25 x double> @extern_matrix(<25 x double> noundef nofpclass(nan inf) [[X]])
#[[ATTR11]]
// CLFINITEONLY-NEXT: ret <25 x double> [[CALL]]
//
// NONANS: Function Attrs: noinline nounwind optnone
>From c9dad42ae9017405e5dbd323596eac4be59e4498 Mon Sep 17 00:00:00 2001
From: Marian Lukac <marian.lu...@arm.com>
Date: Wed, 15 May 2024 11:14:38 +0000
Subject: [PATCH 2/2] Restructured the whole case statement
---
clang/lib/CodeGen/CGCall.cpp | 69 +++++++++++++-----------------------
1 file changed, 25 insertions(+), 44 deletions(-)
diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 8fdde5326b8ed..19d62a74cfbc4 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -5192,28 +5192,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo
&CallInfo,
case ABIArgInfo::Indirect:
case ABIArgInfo::IndirectAliased: {
assert(NumIRArgs == 1);
- if (!I->isAggregate()) {
- // Make a temporary alloca to pass the argument.
- RawAddress Addr = CreateMemTempWithoutCast(
- I->Ty, ArgInfo.getIndirectAlign(), "indirect-arg-temp");
-
- llvm::Value *Val = getAsNaturalPointerTo(Addr, I->Ty);
- if (ArgHasMaybeUndefAttr)
- Val = Builder.CreateFreeze(Val);
- IRCallArgs[FirstIRArg] = Val;
-
- // Emit lifetime markers for the temporary alloca.
- llvm::TypeSize ByvalTempElementSize =
- CGM.getDataLayout().getTypeAllocSize(Addr.getElementType());
- llvm::Value *LifetimeSize =
- EmitLifetimeStart(ByvalTempElementSize, Addr.getPointer());
-
- // Add cleanup code to emit the end lifetime marker after the call.
- if (LifetimeSize) // In case we disabled lifetime markers.
- CallLifetimeEndAfterCall.emplace_back(Addr, LifetimeSize);
-
- I->copyInto(*this, Addr);
- } else {
+ if (I->isAggregate()) {
// We want to avoid creating an unnecessary temporary+copy here;
// however, we need one in three cases:
// 1. If the argument is not byval, and we are required to copy the
@@ -5266,28 +5245,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo
&CallInfo,
}
}
- if (NeedCopy) {
- // Create an aligned temporary, and copy to it.
- RawAddress AI = CreateMemTempWithoutCast(
- I->Ty, ArgInfo.getIndirectAlign(), "byval-temp");
- llvm::Value *Val = getAsNaturalPointerTo(AI, I->Ty);
- if (ArgHasMaybeUndefAttr)
- Val = Builder.CreateFreeze(Val);
- IRCallArgs[FirstIRArg] = Val;
-
- // Emit lifetime markers for the temporary alloca.
- llvm::TypeSize ByvalTempElementSize =
- CGM.getDataLayout().getTypeAllocSize(AI.getElementType());
- llvm::Value *LifetimeSize =
- EmitLifetimeStart(ByvalTempElementSize, AI.getPointer());
-
- // Add cleanup code to emit the end lifetime marker after the call.
- if (LifetimeSize) // In case we disabled lifetime markers.
- CallLifetimeEndAfterCall.emplace_back(AI, LifetimeSize);
-
- // Generate the copy.
- I->copyInto(*this, AI);
- } else {
+ if (!NeedCopy) {
// Skip the extra memcpy call.
llvm::Value *V = getAsNaturalPointerTo(Addr, I->Ty);
auto *T = llvm::PointerType::get(
@@ -5299,8 +5257,31 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo
&CallInfo,
if (ArgHasMaybeUndefAttr)
Val = Builder.CreateFreeze(Val);
IRCallArgs[FirstIRArg] = Val;
+ break;
}
}
+
+ // For non-aggregate args and aggregate args meeting conditions above
+ // we need to create an aligned temporary, and copy to it.
+ RawAddress AI = CreateMemTempWithoutCast(
+ I->Ty, ArgInfo.getIndirectAlign(), "byval-temp");
+ llvm::Value *Val = getAsNaturalPointerTo(AI, I->Ty);
+ if (ArgHasMaybeUndefAttr)
+ Val = Builder.CreateFreeze(Val);
+ IRCallArgs[FirstIRArg] = Val;
+
+ // Emit lifetime markers for the temporary alloca.
+ llvm::TypeSize ByvalTempElementSize =
+ CGM.getDataLayout().getTypeAllocSize(AI.getElementType());
+ llvm::Value *LifetimeSize =
+ EmitLifetimeStart(ByvalTempElementSize, AI.getPointer());
+
+ // Add cleanup code to emit the end lifetime marker after the call.
+ if (LifetimeSize) // In case we disabled lifetime markers.
+ CallLifetimeEndAfterCall.emplace_back(AI, LifetimeSize);
+
+ // Generate the copy.
+ I->copyInto(*this, AI);
break;
}
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
