mirrors/llama.cpp
0
0 Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-10-28 09:45:55 +08:00
Code Activity

Guard __nv_cvt_fp8x2_to_bf162raw on CTK 13.4

Browse Source 
Only cvt.rn.bf16x2.e4m3x2 is exposed from 13.2 onwards, so
93e98b9203dfacb85f9ac665f5f0482a8c6b4192 was incorrect fix
This commit is contained in:
Oliver Simons committed 2026-10-27 18:47:47 +01:00
1 parent 3ba7956378
commit 645b60eeb1
1 file changed
+1 -1
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-cuda/mmvf.cu
+1 -1
View File
@@ -10,7 +10,7 @@ using mmvf_y_t = std::conditional_t<std::is_same_v<T, ggml_fp8_e4m3_t>, nv_bfloa
#if defined(FP8_AVAILABLE) && !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)
static __device__ __forceinline__ nv_bfloat162 mmvf_f8x2_e4m3_to_bf162(__nv_fp8x2_storage_t x) {
#if CUDART_VERSION >= 13020
#if CUDART_VERSION >= 13040
return static_cast<nv_bfloat162>(__nv_cvt_fp8x2_to_bf162raw(x, __NV_E4M3));
#else
return make_bfloat162(