CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

metal : new q4_0 matrix-vector kernel (#2188)

Browse Source 
Prefetch data to improve GPU utilization. ~48% faster for 33B model.
This commit is contained in:
Shouzheng Liu
2023-07-12 16:10:55 -04:00
committed by GitHub
parent 975221e954
commit 1cbf561466
2 changed files with 61 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
ggml-metal.m
View File

@@ -739,7 +739,10 @@ void ggml_metal_graph_compute(
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:13];
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:14];
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1) {
if (src0t == GGML_TYPE_Q4_0) {
[encoder dispatchThreadgroups:MTLSizeMake(ne01 / 8+((ne01 % 8) & 0x01), ne11, 1) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
else if (src0t == GGML_TYPE_Q4_1) {
[encoder setThreadgroupMemoryLength:nth0*nth1*sizeof(float) atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne11, 1) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}