opencl: use q8_0 mm for larger matrix

lhez · lhez · commit 057e67c85d11 · 2025-10-14T13:01:54.000-07:00
diff --git a/ggml/src/ggml-opencl/ggml-opencl.cpp b/ggml/src/ggml-opencl/ggml-opencl.cpp
@@ -6980,6 +6980,9 @@ static void ggml_cl_mul_mat(ggml_backend_t backend, const ggml_tensor * src0, co
                 return;
             }
             case GGML_TYPE_Q8_0: {
+                if (ne11 < 32) {
+                    break;
+                }
                 kernel = backend_ctx->kernel_mul_mm_q8_0_f32_l4_lm;
                 nth0 = 128; // calculated as (BM*BN)/(TM*TN)
 

Original file line number	Diff line number	Diff line change
`@@ -6980,6 +6980,9 @@ static void ggml_cl_mul_mat(ggml_backend_t backend, const ggml_tensor * src0, co`
`6980`	`6980`	`return;`
`6981`	`6981`	`}`
`6982`	`6982`	`case GGML_TYPE_Q8_0: {`
	`6983`	`+ if (ne11 < 32) {`
	`6984`	`+ break;`
	`6985`	`+ }`
`6983`	`6986`	`kernel = backend_ctx->kernel_mul_mm_q8_0_f32_l4_lm;`
`6984`	`6987`	`nth0 = 128; // calculated as (BMBN)/(TMTN)`
`6985`	`6988`