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

metal : print extra compute pipeline info

Browse source
This commit is contained in:
Georgi Gerganov 2023-08-23 11:25:26 +03:00
parent b34ab74094
commit a0dc47a501
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 37 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

7
ggml-metal.m
View file

@ -167,7 +167,9 @@ struct ggml_metal_context * ggml_metal_init(int n_cb) {
#define GGML_METAL_ADD_KERNEL(name) \
ctx->function_##name = [ctx->library newFunctionWithName:@"kernel_"#name]; \
ctx->pipeline_##name = [ctx->device newComputePipelineStateWithFunction:ctx->function_##name error:&error]; \
fprintf(stderr, "%s: loaded %-32s %16p\n", __func__, "kernel_"#name, (void *) ctx->pipeline_##name); \
fprintf(stderr, "%s: loaded %-32s %16p | th_max = %4d | th_width = %4d\n", __func__, "kernel_"#name, (void *) ctx->pipeline_##name, \
(int) ctx->pipeline_##name.maxTotalThreadsPerThreadgroup, \
(int) ctx->pipeline_##name.threadExecutionWidth); \
if (error) { \
fprintf(stderr, "%s: load pipeline error: %s\n", __func__, [[error description] UTF8String]); \
return NULL; \
@ -768,8 +770,7 @@ void ggml_metal_graph_compute(
[encoder setBytes:&gqa length:sizeof(gqa) atIndex:10];
[encoder setThreadgroupMemoryLength:8192 atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake( (ne11+31)/32, (ne01+63) / 64, ne12) threadsPerThreadgroup:MTLSizeMake(128, 1, 1)];
}
else {
} else {
int nth0 = 32;
int nth1 = 1;