intel/llvm
1
0
Fork 0
mirror of https://github.com/intel/llvm.git synced 2026-01-17 23:45:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Clang][AMDGPU] Enable type-checking on __builtin_amdgcn_raw_ptr_buffer_atomic_{{add|fadd|fmin|fmax}} (#164824)

Browse Source 
The "t" flag is used to mark the builtin signature as meaningless.
This is done on several builtins taking pointers since otherwise HIP
code would not compile
during compilation for the host (even if the builtin is only used in
device code, compilation would fail).

The builtins changed by this patch are not affected by this issue, so
they do not need the "t" flag in the first place.
This commit is contained in:
Juan Manuel Martinez Caamaño
2025-10-27 09:59:54 +01:00
committed by GitHub
parent aa550cdc5f
commit efcc6135da
3 changed files with 42 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
clang/include/clang/Basic/BuiltinsAMDGPU.def
View File

@@ -177,15 +177,15 @@ BUILTIN(__builtin_amdgcn_raw_buffer_load_b64, "V2UiQbiiIi", "n")
BUILTIN(__builtin_amdgcn_raw_buffer_load_b96, "V3UiQbiiIi", "n")
BUILTIN(__builtin_amdgcn_raw_buffer_load_b128, "V4UiQbiiIi", "n")
BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_add_i32, "iiQbiiIi", "t")
BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_add_i32, "iiQbiiIi", "")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fadd_f32, "ffQbiiIi", "t", "atomic-fadd-rtn-insts")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fadd_f32, "ffQbiiIi", "", "atomic-fadd-rtn-insts")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fadd_v2f16, "V2hV2hQbiiIi", "t", "atomic-buffer-global-pk-add-f16-insts")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f32, "ffQbiiIi", "t", "atomic-fmin-fmax-global-f32")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f32, "ffQbiiIi", "t", "atomic-fmin-fmax-global-f32")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f64, "ddQbiiIi", "t", "atomic-fmin-fmax-global-f64")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f64, "ddQbiiIi", "t", "atomic-fmin-fmax-global-f64")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f32, "ffQbiiIi", "", "atomic-fmin-fmax-global-f32")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f32, "ffQbiiIi", "", "atomic-fmin-fmax-global-f32")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f64, "ddQbiiIi", "", "atomic-fmin-fmax-global-f64")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f64, "ddQbiiIi", "", "atomic-fmin-fmax-global-f64")
TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_load_lds, "vQbv*3IUiiiIiIi", "t", "vmem-to-lds-load-insts")
TARGET_BUILTIN(__builtin_amdgcn_struct_ptr_buffer_load_lds, "vQbv*3IUiiiiIiIi", "t", "vmem-to-lds-load-insts")

18
clang/test/SemaHIP/builtins-amdgcn-raw-buffer-atomic-add.hip Normal file
View File

@@ -0,0 +1,18 @@
// RUN: %clang_cc1 -fsyntax-only -triple amdgcn -target-cpu gfx90a -verify %s -fcuda-is-device
// RUN: %clang_cc1 -fsyntax-only -triple x86_64 -aux-triple amdgcn -verify %s
typedef _Float16 __attribute__((ext_vector_type(2))) float16x2_t;
#define __device__ __attribute__((device))
__device__ void test_raw_ptr_atomics(__amdgpu_buffer_rsrc_t rsrc, int i32, float f32, float16x2_t v2f16, int offset, int soffset) {
i32 = __builtin_amdgcn_raw_ptr_buffer_atomic_add_i32(i32, rsrc, offset, soffset, 0);
f32 = __builtin_amdgcn_raw_ptr_buffer_atomic_fadd_f32(f32, rsrc, offset, soffset, 0);
v2f16 = __builtin_amdgcn_raw_ptr_buffer_atomic_fadd_v2f16(v2f16, rsrc, offset, soffset, 0);
}
__device__ void test_raw_ptr_atomics_err(__amdgpu_buffer_rsrc_t rsrc, int i32, float f32, float16x2_t v2f16, int offset, int soffset) {
i32 = __builtin_amdgcn_raw_ptr_buffer_atomic_add_i32(i32, rsrc, offset, soffset, 0, 4); // expected-error{{too many arguments to function call}}
f32 = __builtin_amdgcn_raw_ptr_buffer_atomic_fadd_f32(f32, rsrc, offset, soffset, 0, 4); // expected-error{{too many arguments to function call}}
v2f16 = __builtin_amdgcn_raw_ptr_buffer_atomic_fadd_v2f16(v2f16, rsrc, offset, soffset, 0, 4);
}

18
clang/test/SemaHIP/builtins-amdgcn-raw-buffer-atomic-fmin-max.hip Normal file
View File

@@ -0,0 +1,18 @@
// RUN: %clang_cc1 -fsyntax-only -triple amdgcn -target-cpu gfx90a -verify %s -fcuda-is-device
// RUN: %clang_cc1 -fsyntax-only -triple x86_64 -aux-triple amdgcn -verify %s
#define __device__ __attribute__((device))
__device__ void test_raw_ptr_atomics(__amdgpu_buffer_rsrc_t rsrc, float f32, double f64, int offset, int soffset) {
f32 = __builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f32(f32, rsrc, offset, soffset, 0);
f64 = __builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f64(f64, rsrc, offset, soffset, 0);
f32 = __builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f32(f32, rsrc, offset, soffset, 0);
f64 = __builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f64(f64, rsrc, offset, soffset, 0);
}
__device__ void test_raw_ptr_atomics_err(__amdgpu_buffer_rsrc_t rsrc, float f32, double f64, int offset, int soffset) {
f32 = __builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f32(f32, rsrc, offset, soffset, 0, 4); // expected-error{{too many arguments to function call}}
f64 = __builtin_amdgcn_raw_ptr_buffer_atomic_fmin_f64(f64, rsrc, offset, soffset, 0, 4); // expected-error{{too many arguments to function call}}
f32 = __builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f32(f32, rsrc, offset, soffset, 0, 4); // expected-error{{too many arguments to function call}}
f64 = __builtin_amdgcn_raw_ptr_buffer_atomic_fmax_f64(f64, rsrc, offset, soffset, 0, 4); // expected-error{{too many arguments to function call}}
}