metal : extend l2_norm support for non-cont src0 (#19502)

This commit is contained in:
Georgi Gerganov
2026-02-11 14:53:19 +02:00
committed by GitHub
parent ada90bf2ba
commit 9ab072ebbe
5 changed files with 75 additions and 30 deletions
+7 -4
View File
@@ -1480,13 +1480,15 @@ ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_bin_one(ggml_met
ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_l2_norm(ggml_metal_library_t lib, const ggml_tensor * op) {
assert(op->op == GGML_OP_L2_NORM);
GGML_ASSERT(op->src[0]->ne[0] % 4 == 0);
GGML_ASSERT(ggml_is_contiguous_1(op->src[0]));
char base[256];
char name[256];
snprintf(base, 256, "kernel_l2_norm_f32");
const bool is_c4 = op->src[0]->ne[0] % 4 == 0;
const char * t0_str = ggml_type_name(op->src[0]->type);
const char * t_str = ggml_type_name(op->type);
snprintf(base, 256, "kernel_l2_norm_%s_%s%s", t0_str, t_str, is_c4 ? "_4" : "");
snprintf(name, 256, "%s", base);
ggml_metal_pipeline_with_params res = ggml_metal_library_get_pipeline(lib, name);
@@ -1494,6 +1496,7 @@ ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_l2_norm(ggml_met
res = ggml_metal_library_compile_pipeline(lib, base, name, nullptr);
}
res.c4 = is_c4;
res.smem = 32*sizeof(float);
return res;