ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 09:09:50 +00:00
Code Issues Packages Projects Releases Wiki Activity

Attempt to fix AVX2 FA (#807)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-09-30 08:06:53 +02:00
committed by GitHub
parent 9932e6b102
commit 475223079c
2 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/iqk/fa/iqk_fa_templates.h
View File

@@ -226,7 +226,8 @@ struct HelperQ8KV final : public BaseHelper {
struct HelperQ80 final : public BaseHelper {
using Base = BaseHelper;
constexpr static ggml_type type = GGML_TYPE_Q8_0;
#ifdef HAVE_FANCY_SIMD
//#ifdef HAVE_FANCY_SIMD
#ifdef __AVX2__
using block_q8 = block_q8_2;
constexpr static int block_size_q = QK8_2;
#else

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

@@ -3299,9 +3299,12 @@ inline std::pair<mul_mat_t, int> mul_mat_kernel(int int_typeA, int nq) {
if (nq == 4) return std::make_pair(mul_mat_qX_0_q8_2_Tx<Q8_0_1_Unpacker, 4, k_step>, 4);
MAKE_FUNCS(mul_mat_qX_1_q8_2_T<Q8_0_1_Unpacker, nq);
#else
if (nq == 1) return std::make_pair(mul_mat_qX_0_q8_0_Tx<Q8_0_Unpacker, 1, k_step>, 1);
if (nq == 2) return std::make_pair(mul_mat_qX_0_q8_0_Tx<Q8_0_Unpacker, 2, k_step>, 2);
if (nq == 4) return std::make_pair(mul_mat_qX_0_q8_0_Tx<Q8_0_Unpacker, 4, k_step>, 4);
//if (nq == 1) return std::make_pair(mul_mat_qX_0_q8_0_Tx<Q8_0_Unpacker, 1, k_step>, 1);
//if (nq == 2) return std::make_pair(mul_mat_qX_0_q8_0_Tx<Q8_0_Unpacker, 2, k_step>, 2);
//if (nq == 4) return std::make_pair(mul_mat_qX_0_q8_0_Tx<Q8_0_Unpacker, 4, k_step>, 4);
if (nq == 1) return std::make_pair(mul_mat_qX_0_q8_0_T<Q8_0_Unpacker, 1, block_q8_2>, 1);
if (nq == 2) return std::make_pair(mul_mat_qX_0_q8_0_T<Q8_0_Unpacker, 2, block_q8_2>, 2);
if (nq == 4) return std::make_pair(mul_mat_qX_0_q8_0_T<Q8_0_Unpacker, 4, block_q8_2>, 4);
if (nq == 3) return std::make_pair(mul_mat_qX_0_q8_0_T<Q8_0_Unpacker, 3, block_q8_2>, 3);
if (nq == 5) return std::make_pair(mul_mat_qX_0_q8_0_T<Q8_0_Unpacker, 5, block_q8_2>, 5);
if (nq == 6) return std::make_pair(mul_mat_qX_0_q8_0_T<Q8_0_Unpacker, 6, block_q8_2>, 6);