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

apply more optimization

Browse source 
- original IM2COL kernel + _ext with MIN()

Signed-off-by: Junhee Yoo <junhee.yoo@navercorp.com>
This commit is contained in:
Junhee Yoo 2024-10-23 11:19:23 +09:00
parent 0084847991
commit bd86c4c4df
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
ggml/src/ggml-metal.m
View file

@ -2625,7 +2625,7 @@ static void ggml_metal_encode_node(
[encoder setBytes:&KW length:sizeof(int32_t) atIndex:15];
const int64_t D = N / M + (N % M > 0 ? 1 : 0);
[encoder dispatchThreadgroups:MTLSizeMake(D * CHW, OH, OW) threadsPerThreadgroup:MTLSizeMake(M, 1, 1)];
[encoder dispatchThreadgroups:MTLSizeMake(D * CHW, OH, OW) threadsPerThreadgroup:MTLSizeMake(MIN((uint64_t)N, M), 1, 1)];
} else {
[encoder dispatchThreadgroups:MTLSizeMake(IC, OH, OW) threadsPerThreadgroup:MTLSizeMake(N, KH, KW)];
}