OSSForks/pytorch
1
0
Fork 0
mirror of https://github.com/zebrajr/pytorch.git synced 2025-12-06 12:20:52 +01:00
Code Issues Actions 38 Packages Projects Releases Wiki Activity

[Easy][Building] Fix the warning of int4mm.cu when building (#151427)

Browse Source 
As the title stated.

**Changes Before:**

```C++
[999/1526] Building CUDA object caffe2/CMakeFiles/torch_cuda.dir/__/aten/src/ATen/native/cuda/int4mm.cu.o
/root/Git.d/pytorch/pytorch/aten/src/ATen/native/cuda/int4mm.cu(142): warning #177-D: variable "at::native::kWarpSize" was declared but never referenced
  constexpr int32_t kWarpSize = 32;
                    ^

Remark: The warnings can be suppressed with "-diag-suppress <warning-number>"
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/151427
Approved by: https://github.com/Skylion007, https://github.com/malfet
This commit is contained in:
FFFrog 2025-04-17 12:20:49 +08:00 committed by PyTorch MergeBot
parent a05cc9f494
commit 49c91b4be9
1 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
aten/src/ATen/native/cuda/int4mm.cu
View File

@ -127,6 +127,14 @@ inline __host__ __device__ uint32_t getAlignmentRoundUp(const void* p) {
return diff == 0 ? 0 : uint32_t(Align) - diff;
}
#if defined (__gfx90a__) || defined(__gfx942__)
#define CDNA2_OR_LATER 1
#else
#define CDNA2_OR_LATER 0
#endif
#if (defined(USE_ROCM) && ROCM_VERSION >= 50700) || ((defined(CUDA_VERSION) && CUDA_VERSION >= 12000) && (!defined(__CUDA_ARCH__) || (__CUDA_ARCH__ >= 800)))
#if defined(USE_ROCM)
// TODO: Support RDNA
constexpr int32_t kWarpSize = 64;
@ -142,14 +150,6 @@ static bool isCDNA2orLater(int index) {
constexpr int32_t kWarpSize = 32;
#endif
#if defined (__gfx90a__) || defined(__gfx942__)
#define CDNA2_OR_LATER 1
#else
#define CDNA2_OR_LATER 0
#endif
#if (defined(USE_ROCM) && ROCM_VERSION >= 50700) || ((defined(CUDA_VERSION) && CUDA_VERSION >= 12000) && (!defined(__CUDA_ARCH__) || (__CUDA_ARCH__ >= 800)))
// f16 vector types
struct __align__(2) f16x1 {
__half vals[1];