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

metal : move mm_id indices to shared mem (#5982)

Browse source
This commit is contained in:
Georgi Gerganov 2024-03-10 23:12:48 +02:00 • committed by GitHub
parent 7ab7b733bb
commit bb6d00bbf9
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 6 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml-metal.m
View file

@ -1642,8 +1642,8 @@ static enum ggml_status ggml_metal_graph_compute(
// TODO: make this more general
GGML_ASSERT(n_as <= 8);
// max size of the src1ids array in the kernel stack
GGML_ASSERT(ne11 <= 512);
// max size of the src1ids array in the kernel shared buffer
GGML_ASSERT(ne11 <= 4096);
const int64_t ne20 = src2 ? src2->ne[0] : 0;
const int64_t ne21 = src2 ? src2->ne[1] : 0;
@ -1741,7 +1741,7 @@ static enum ggml_status ggml_metal_graph_compute(
[encoder setBuffer:id_src_cur offset:offs_src_cur atIndex:19 + j];
}
[encoder setThreadgroupMemoryLength:8192 atIndex:0];
[encoder setThreadgroupMemoryLength:GGML_PAD(8192 + 2*ne11, 16) atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake((ne11 + 31)/32, (ne21 + 63)/64, n_as*ne12*ne13) threadsPerThreadgroup:MTLSizeMake(128, 1, 1)];
} else {