ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-26 08:04:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

iq3_ks: Fix CUDA dot product

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-10-09 18:09:35 +03:00
parent 252c6b2d82
commit 0317ba5a01
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-cuda/iqk_mmvq.cu
View File

@@ -559,9 +559,9 @@ __device__ __forceinline__ float vec_dot_iq3_ks_q8_1(
const int hshift = 4*(1-ib128);
const uint16_t * values1 = iq3k_table + ((bq3->scales[4*ib128+0] << 6) & 0x40);
const uint16_t * values2 = iq3k_table + ((bq3->scales[4*ib128+0] << 6) & 0x40);
const uint16_t * values3 = iq3k_table + ((bq3->scales[4*ib128+0] << 6) & 0x40);
const uint16_t * values4 = iq3k_table + ((bq3->scales[4*ib128+0] << 6) & 0x40);
const uint16_t * values2 = iq3k_table + ((bq3->scales[4*ib128+1] << 6) & 0x40);
const uint16_t * values3 = iq3k_table + ((bq3->scales[4*ib128+2] << 6) & 0x40);
const uint16_t * values4 = iq3k_table + ((bq3->scales[4*ib128+3] << 6) & 0x40);
const int * q8;
int sumi[4] = {0, 0, 0, 0};