Skip to content

Commit 314454b

Browse files
committed
test johannes fix of the tile kernel, may revert
1 parent 34d243b commit 314454b

File tree

1 file changed

+2
-2
lines changed

1 file changed

+2
-2
lines changed

ggml/src/ggml-cuda/common.cuh

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -223,7 +223,7 @@ static const char * cu_get_error_str(CUresult err) {
223223
#define FP16_AVAILABLE
224224
#endif // defined(GGML_USE_HIP) || defined(GGML_USE_MUSA) || __CUDA_ARCH__ >= GGML_CUDA_CC_PASCAL
225225

226-
#if defined(FP16_AVAILABLE) && __CUDA_ARCH__ > GGML_CUDA_CC_VOLTA
226+
#if defined(FP16_AVAILABLE) && __CUDA_ARCH__ != 610
227227
#define FAST_FP16_AVAILABLE
228228
#endif // defined(FP16_AVAILABLE) && __CUDA_ARCH__ != 610
229229

@@ -263,7 +263,7 @@ static bool fp16_available(const int cc) {
263263

264264
static bool fast_fp16_available(const int cc) {
265265
return GGML_CUDA_CC_IS_AMD(cc) ||
266-
(GGML_CUDA_CC_IS_NVIDIA(cc) && fp16_available(cc) && cc > GGML_CUDA_CC_VOLTA) ||
266+
(GGML_CUDA_CC_IS_NVIDIA(cc) && fp16_available(cc) && cc > 610) ||
267267
(GGML_CUDA_CC_IS_MTHREADS(cc) && fp16_available(cc));
268268
}
269269

0 commit comments

Comments
 (0)