metal: Q1_0 backend (#21528)

* initial Q1_0 Metal backend

* tuning q1_0 metal kernels

* add Q1_0 to test-backend-ops

* add Q1_0<->F32 copy test

* Apply suggestions from code review

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Pasha Khosravi
2026-04-08 16:07:47 +03:00
committed by GitHub
co-authored by Georgi Gerganov
parent 5764d7c6a6
commit dcdcbad42a
6 changed files with 205 additions and 0 deletions
+10
View File
@@ -736,6 +736,11 @@ ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_mul_mv(ggml_meta
suffix = ne00 % 4 == 0 ? "_4" : "";
}
} break;
case GGML_TYPE_Q1_0:
{
nsg = N_SG_Q1_0;
nr0 = N_R0_Q1_0;
} break;
case GGML_TYPE_Q4_0:
{
nsg = N_SG_Q4_0;
@@ -948,6 +953,11 @@ ggml_metal_pipeline_with_params ggml_metal_library_get_pipeline_mul_mv_id(ggml_m
smem = 32*sizeof(float)*nr0;
suffix = ne00 % 4 == 0 ? "_4" : "";
} break;
case GGML_TYPE_Q1_0:
{
nsg = N_SG_Q1_0;
nr0 = N_R0_Q1_0;
} break;
case GGML_TYPE_Q4_0:
{
nsg = N_SG_Q4_0;