vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Use two memcpy calls for q5_0 buffer transfer

Browse source
This commit is contained in:
0cc4m 2023-04-30 10:44:48 +02:00
parent 369d903eda
commit e69c924ad1
1 changed files with 3 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-opencl.c
View file

@ -3,6 +3,7 @@
#define CL_TARGET_OPENCL_VERSION 110
#include <clblast_c.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
@ -308,7 +309,8 @@ void ggml_cl_sgemm_wrapper(
cl_host_b = (cl_block_q5_0*) malloc(sizeof(cl_block_q5_0) * global / 32);
for (size_t i = 0; i < global / 32; i++) {
cl_host_b[i].d = ggml_fp16_to_fp32(b[i].d);
memcpy(&cl_host_b[i].qh, b[i].qh, sizeof(uint32_t) + QK5_0 / 2);
memcpy(&cl_host_b[i].qh, b[i].qh, sizeof(uint32_t));
memcpy(&cl_host_b[i].qs, b[i].qs, QK5_0 / 2);
}
host_b = (const float*) cl_host_b;
size_qb = global * (sizeof(float) + sizeof(uint32_t) + local) / 32;