metal : various optimizations + refactoring (#16446)

* metal : ssm_scan minor opts

* metal : get_rows optimize

* metal : cpy optimize

* metal : ssm_conv opt

* metal : ssm_scan simplify

* metal : ssm_Scan opt
This commit is contained in:
Georgi Gerganov
2025-10-07 08:21:40 +03:00
committed by GitHub
parent 3df2244df4
commit 8ae32dc9ec
5 changed files with 258 additions and 452 deletions
+14 -8
View File
@@ -338,7 +338,13 @@ ggml_metal_pipeline_t ggml_metal_library_get_pipeline_ssm_conv(ggml_metal_librar
char base[256];
char name[256];
snprintf(base, 256, "kernel_ssm_conv_%s_%s", ggml_type_name(op->src[0]->type), ggml_type_name(op->src[1]->type));
const char * suffix = "";
if (op->src[1]->ne[0] % 4 == 0) {
suffix = "_4";
}
snprintf(base, 256, "kernel_ssm_conv_%s_%s%s", ggml_type_name(op->src[0]->type), ggml_type_name(op->src[1]->type), suffix);
snprintf(name, 256, "%s", base);
ggml_metal_pipeline_t res = ggml_metal_library_get_pipeline(lib, name);
@@ -352,15 +358,15 @@ ggml_metal_pipeline_t ggml_metal_library_get_pipeline_ssm_conv(ggml_metal_librar
}
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_ssm_scan(ggml_metal_library_t lib, const ggml_tensor * op) {
GGML_TENSOR_LOCALS( int32_t, ne0, op->src[0], ne);
char base[256];
char name[256];
if (op->src[3]->ne[0] == 1) {
snprintf(base, 256, "kernel_ssm_scan_group_%s", ggml_type_name(op->src[0]->type));
} else {
snprintf(base, 256, "kernel_ssm_scan_%s", ggml_type_name(op->src[0]->type));
}
snprintf(name, 256, "%s", base);
const int nsg = (ne00 + 31)/32;
snprintf(base, 256, "kernel_ssm_scan_%s", ggml_type_name(op->src[0]->type));
snprintf(name, 256, "%s_nsg=%d", base, nsg);
ggml_metal_pipeline_t res = ggml_metal_library_get_pipeline(lib, name);
if (res) {
@@ -369,7 +375,7 @@ ggml_metal_pipeline_t ggml_metal_library_get_pipeline_ssm_scan(ggml_metal_librar
res = ggml_metal_library_compile_pipeline(lib, base, name, nullptr);
ggml_metal_pipeline_set_smem(res, 32*sizeof(float));
ggml_metal_pipeline_set_smem(res, 32*sizeof(float)*nsg);
return res;
}