diff options
| author | Iwan Kawrakow <iwan.kawrakow@gmail.com> | 2024-06-11 09:12:22 +0200 |
|---|---|---|
| committer | Iwan Kawrakow <iwan.kawrakow@gmail.com> | 2024-06-22 12:02:50 +0300 |
| commit | ad53eabf87816705f49501a54c7555c337bb47ce (patch) | |
| tree | d5500fa26360272673c599eb6bbec4046023d806 /sgemm.cpp | |
| parent | 3593891f390b376f7127efe3bd7eda9baffe6609 (diff) | |
iqk_mul_mat: be independent of llamafile_sgemm (WIP)
* Remove iqk_mul_mat from llamafile_sgemm
* Pass tensor types and strides to iqk_mul_mat
It is marked WIP because only tested on __aarch64__
Diffstat (limited to 'sgemm.cpp')
| -rw-r--r-- | sgemm.cpp | 23 |
1 files changed, 0 insertions, 23 deletions
@@ -51,9 +51,6 @@ #include "sgemm.h" #include "ggml-impl.h" #include "ggml-quants.h" -#if GGML_USE_IQK_MULMAT -#include "iqk_mul_mat.h" -#endif #ifdef _MSC_VER #define NOINLINE __declspec(noinline) @@ -868,26 +865,6 @@ bool llamafile_sgemm(int64_t m, int64_t n, int64_t k, const void *A, int64_t lda if (Ctype != GGML_TYPE_F32) return false; -#if GGML_USE_IQK_MULMAT -#if defined __AVX2__ && defined __FMA__ - bool is_accepted_float_type = k >= 32 && - ((Atype == GGML_TYPE_F16 && Btype == GGML_TYPE_F32) || (Atype == GGML_TYPE_F32 && Btype == GGML_TYPE_F16)); -#elif defined __ARM_FEATURE_FP16_VECTOR_ARITHMETIC && defined __ARM_FEATURE_FMA - bool is_accepted_float_type = k >= 32 && Atype == GGML_TYPE_F16 && Btype == GGML_TYPE_F16; -#else - bool is_accepted_float_type = false; -#endif - if (task == GGML_TASK_TYPE_INIT && is_accepted_float_type) { - return true; - } - - if (task == GGML_TASK_TYPE_COMPUTE && is_accepted_float_type) { - if (iqk_mul_mat(m, n, k, Atype, A, B, (float *)C, ldc, ith, nth)) { - return true; - } - } -#endif - switch (Atype) { case GGML_TYPE_F32: { |