ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-05 11:30:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix IQK_FA_ALL_QUANTS on AVX2 (#360)

Browse Source 
* Fix IQK_FA_ALL_QUANTS on AVX2

* Make it also work, not just compile

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-04-30 10:45:43 +02:00
committed by GitHub
parent 9ba362706c
commit 4c2bee0bed
1 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -16076,8 +16076,13 @@ struct HelperIQ4nl final : public BaseHelper<step> {
constexpr static int block_size_q = QK8_0;
#else
HelperIQ4nl(const char * data, int stride) : Base(data, stride) {}
#ifdef HAVE_FANCY_SIMD
using block_q8 = block_q8_2;
constexpr static int block_size_q = QK8_2;
#else
using block_q8 = block_q8_0;
constexpr static int block_size_q = QK8_0;
#endif
#endif
// Needed for v * softmax(k * q)
@@ -16974,7 +16979,11 @@ struct FlashQKfp32 {
#ifdef __aarch64__
MAKE_FUNCS(mul_mat_qX_0_q8_0<DequantizerIQ4NL, nq);
#else
#ifdef HAVE_FANCY_SIMD
MAKE_FUNCS(mul_mat_qX_1_q8_2_T<IQ4_NL_Unpacker, nq);
#else
MAKE_FUNCS(mul_mat_qX_0_q8_0_T<IQ4_NL_Unpacker, nq);
#endif
#endif
}
#endif