Skip to content

[Clang] Use private address space for builtin_alloca return type for OpenCL #95750

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 9 commits into from
Jul 26, 2024
Merged

[Clang] Use private address space for builtin_alloca return type for OpenCL #95750

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
ac967e4
[Clang] Use private address space for builtin_alloca for OpenCL
vg0204 Jun 17, 2024
8a5efde
updated return pointer to always point to stack/private address space…
vg0204 Jun 17, 2024
8a4a143
Moved the patch to dedicated builtin function in SemaChecking.cpp &
vg0204 Jul 3, 2024
a1b8505
Made some changes in comments in OpenCLBuiltinAllocaAddrSpace().
vg0204 Jul 3, 2024
e2ac64a
Removed non-OpenCL specific code snippet from OpenCLBuiltinAllocaAddr…
vg0204 Jul 10, 2024
3e2e427
Made suggested changes passing opencl_private directly.
vg0204 Jul 11, 2024
9cae2ff
Made the suggested changes around the builtinAllocaAddrSpace function.
vg0204 Jul 22, 2024
4811263
Modified return type of builtinAllocaAddreSpace function.
vg0204 Jul 23, 2024
c4145d2
Removed unused version specific check-prefixes from builtins-alloca.c…
vg0204 Jul 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 15 additions & 0 deletions clang/lib/Sema/SemaChecking.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1477,6 +1477,18 @@ static bool BuiltinSEHScopeCheck(Sema &SemaRef, CallExpr *TheCall,
return false;
}

// In OpenCL, __builtin_alloca_* should return a pointer to address space
// that corresponds to the stack address space i.e private address space.
static void builtinAllocaAddrSpace(Sema &S, CallExpr *TheCall) {
QualType RT = TheCall->getType();
assert((RT->isPointerType() && !(RT->getPointeeType().hasAddressSpace())) &&
"__builtin_alloca has invalid address space");

RT = RT->getPointeeType();
RT = S.Context.getAddrSpaceQualType(RT, LangAS::opencl_private);
TheCall->setType(S.Context.getPointerType(RT));
}

namespace {
enum PointerAuthOpKind {
PAO_Strip,
Expand Down Expand Up @@ -2210,6 +2222,9 @@ Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
case Builtin::BI__builtin_alloca_uninitialized:
Diag(TheCall->getBeginLoc(), diag::warn_alloca)
<< TheCall->getDirectCallee();
if (getLangOpts().OpenCL) {
builtinAllocaAddrSpace(*this, TheCall);
}
Comment on lines +2225 to +2227
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A bit late tot this review, but right now we have getASTAllocaAddressSpace() in CodeGen/, could this be used moved so we can make use of it in Sema instead?

break;
case Builtin::BI__arithmetic_fence:
if (BuiltinArithmeticFence(TheCall))
Expand Down
141 changes: 141 additions & 0 deletions clang/test/CodeGenOpenCL/builtins-alloca.cl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,141 @@
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
// RUN: %clang_cc1 %s -O0 -triple amdgcn-amd-amdhsa -cl-std=CL1.2 \
// RUN: -emit-llvm -o - | FileCheck --check-prefixes=OPENCL %s
// RUN: %clang_cc1 %s -O0 -triple amdgcn-amd-amdhsa -cl-std=CL2.0 \
// RUN: -emit-llvm -o - | FileCheck --check-prefixes=OPENCL %s
// RUN: %clang_cc1 %s -O0 -triple amdgcn-amd-amdhsa -cl-std=CL3.0 \
// RUN: -emit-llvm -o - | FileCheck --check-prefixes=OPENCL %s
// RUN: %clang_cc1 %s -O0 -triple amdgcn-amd-amdhsa -cl-std=CL3.0 -cl-ext=+__opencl_c_generic_address_space \
// RUN: -emit-llvm -o - | FileCheck --check-prefixes=OPENCL %s

// OPENCL-LABEL: define dso_local void @test1_builtin_alloca(
// OPENCL-SAME: i32 noundef [[N:%.*]]) #[[ATTR0:[0-9]+]] {
// OPENCL-NEXT: [[ENTRY:.*:]]
// OPENCL-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
// OPENCL-NEXT: [[ALLOC_PTR:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
// OPENCL-NEXT: store i32 [[N]], ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[CONV:%.*]] = zext i32 [[TMP0]] to i64
// OPENCL-NEXT: [[MUL:%.*]] = mul i64 [[CONV]], 4
// OPENCL-NEXT: [[TMP1:%.*]] = alloca i8, i64 [[MUL]], align 8, addrspace(5)
// OPENCL-NEXT: store ptr addrspace(5) [[TMP1]], ptr addrspace(5) [[ALLOC_PTR]], align 4
// OPENCL-NEXT: ret void
//
void test1_builtin_alloca(unsigned n) {
__private float* alloc_ptr = (__private float*)__builtin_alloca(n*sizeof(int));
}

// OPENCL-LABEL: define dso_local void @test1_builtin_alloca_uninitialized(
// OPENCL-SAME: i32 noundef [[N:%.*]]) #[[ATTR0]] {
// OPENCL-NEXT: [[ENTRY:.*:]]
// OPENCL-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
// OPENCL-NEXT: [[ALLOC_PTR_UNINITIALIZED:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
// OPENCL-NEXT: store i32 [[N]], ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[CONV:%.*]] = zext i32 [[TMP0]] to i64
// OPENCL-NEXT: [[MUL:%.*]] = mul i64 [[CONV]], 4
// OPENCL-NEXT: [[TMP1:%.*]] = alloca i8, i64 [[MUL]], align 8, addrspace(5)
// OPENCL-NEXT: store ptr addrspace(5) [[TMP1]], ptr addrspace(5) [[ALLOC_PTR_UNINITIALIZED]], align 4
// OPENCL-NEXT: ret void
//
void test1_builtin_alloca_uninitialized(unsigned n) {
__private float* alloc_ptr_uninitialized = (__private float*)__builtin_alloca_uninitialized(n*sizeof(int));
}

// OPENCL-LABEL: define dso_local void @test1_builtin_alloca_with_align(
// OPENCL-SAME: i32 noundef [[N:%.*]]) #[[ATTR0]] {
// OPENCL-NEXT: [[ENTRY:.*:]]
// OPENCL-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
// OPENCL-NEXT: [[ALLOC_PTR_ALIGN:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
// OPENCL-NEXT: store i32 [[N]], ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[CONV:%.*]] = zext i32 [[TMP0]] to i64
// OPENCL-NEXT: [[MUL:%.*]] = mul i64 [[CONV]], 4
// OPENCL-NEXT: [[TMP1:%.*]] = alloca i8, i64 [[MUL]], align 1, addrspace(5)
// OPENCL-NEXT: store ptr addrspace(5) [[TMP1]], ptr addrspace(5) [[ALLOC_PTR_ALIGN]], align 4
// OPENCL-NEXT: ret void
//
void test1_builtin_alloca_with_align(unsigned n) {
__private float* alloc_ptr_align = (__private float*)__builtin_alloca_with_align((n*sizeof(int)), 8);
}

// OPENCL-LABEL: define dso_local void @test1_builtin_alloca_with_align_uninitialized(
// OPENCL-SAME: i32 noundef [[N:%.*]]) #[[ATTR0]] {
// OPENCL-NEXT: [[ENTRY:.*:]]
// OPENCL-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
// OPENCL-NEXT: [[ALLOC_PTR_ALIGN_UNINITIALIZED:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
// OPENCL-NEXT: store i32 [[N]], ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[CONV:%.*]] = zext i32 [[TMP0]] to i64
// OPENCL-NEXT: [[MUL:%.*]] = mul i64 [[CONV]], 4
// OPENCL-NEXT: [[TMP1:%.*]] = alloca i8, i64 [[MUL]], align 1, addrspace(5)
// OPENCL-NEXT: store ptr addrspace(5) [[TMP1]], ptr addrspace(5) [[ALLOC_PTR_ALIGN_UNINITIALIZED]], align 4
// OPENCL-NEXT: ret void
//
void test1_builtin_alloca_with_align_uninitialized(unsigned n) {
__private float* alloc_ptr_align_uninitialized = (__private float*)__builtin_alloca_with_align_uninitialized((n*sizeof(int)), 8);
}

// OPENCL-LABEL: define dso_local void @test2_builtin_alloca(
// OPENCL-SAME: i32 noundef [[N:%.*]]) #[[ATTR0]] {
// OPENCL-NEXT: [[ENTRY:.*:]]
// OPENCL-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
// OPENCL-NEXT: [[ALLOC_PTR:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
// OPENCL-NEXT: store i32 [[N]], ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[CONV:%.*]] = zext i32 [[TMP0]] to i64
// OPENCL-NEXT: [[TMP1:%.*]] = alloca i8, i64 [[CONV]], align 8, addrspace(5)
// OPENCL-NEXT: store ptr addrspace(5) [[TMP1]], ptr addrspace(5) [[ALLOC_PTR]], align 4
// OPENCL-NEXT: ret void
//
void test2_builtin_alloca(unsigned n) {
__private void *alloc_ptr = __builtin_alloca(n);
}

// OPENCL-LABEL: define dso_local void @test2_builtin_alloca_uninitialized(
// OPENCL-SAME: i32 noundef [[N:%.*]]) #[[ATTR0]] {
// OPENCL-NEXT: [[ENTRY:.*:]]
// OPENCL-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
// OPENCL-NEXT: [[ALLOC_PTR_UNINITIALIZED:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
// OPENCL-NEXT: store i32 [[N]], ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[CONV:%.*]] = zext i32 [[TMP0]] to i64
// OPENCL-NEXT: [[TMP1:%.*]] = alloca i8, i64 [[CONV]], align 8, addrspace(5)
// OPENCL-NEXT: store ptr addrspace(5) [[TMP1]], ptr addrspace(5) [[ALLOC_PTR_UNINITIALIZED]], align 4
// OPENCL-NEXT: ret void
//
void test2_builtin_alloca_uninitialized(unsigned n) {
__private void *alloc_ptr_uninitialized = __builtin_alloca_uninitialized(n);
}

// OPENCL-LABEL: define dso_local void @test2_builtin_alloca_with_align(
// OPENCL-SAME: i32 noundef [[N:%.*]]) #[[ATTR0]] {
// OPENCL-NEXT: [[ENTRY:.*:]]
// OPENCL-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
// OPENCL-NEXT: [[ALLOC_PTR_ALIGN:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
// OPENCL-NEXT: store i32 [[N]], ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[CONV:%.*]] = zext i32 [[TMP0]] to i64
// OPENCL-NEXT: [[TMP1:%.*]] = alloca i8, i64 [[CONV]], align 1, addrspace(5)
// OPENCL-NEXT: store ptr addrspace(5) [[TMP1]], ptr addrspace(5) [[ALLOC_PTR_ALIGN]], align 4
// OPENCL-NEXT: ret void
//
void test2_builtin_alloca_with_align(unsigned n) {
__private void *alloc_ptr_align = __builtin_alloca_with_align(n, 8);
}

// OPENCL-LABEL: define dso_local void @test2_builtin_alloca_with_align_uninitialized(
// OPENCL-SAME: i32 noundef [[N:%.*]]) #[[ATTR0]] {
// OPENCL-NEXT: [[ENTRY:.*:]]
// OPENCL-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4, addrspace(5)
// OPENCL-NEXT: [[ALLOC_PTR_ALIGN_UNINITIALIZED:%.*]] = alloca ptr addrspace(5), align 4, addrspace(5)
// OPENCL-NEXT: store i32 [[N]], ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(5) [[N_ADDR]], align 4
// OPENCL-NEXT: [[CONV:%.*]] = zext i32 [[TMP0]] to i64
// OPENCL-NEXT: [[TMP1:%.*]] = alloca i8, i64 [[CONV]], align 1, addrspace(5)
// OPENCL-NEXT: store ptr addrspace(5) [[TMP1]], ptr addrspace(5) [[ALLOC_PTR_ALIGN_UNINITIALIZED]], align 4
// OPENCL-NEXT: ret void
//
void test2_builtin_alloca_with_align_uninitialized(unsigned n) {
__private void *alloc_ptr_align_uninitialized = __builtin_alloca_with_align_uninitialized(n, 8);
}