Fix q_xxs using mul_mat_q (#7459)

AidanBeltonS · web-flow · commit 95f84d5ce8b4 · 2024-05-27T22:04:51.000+05:30
diff --git a/ggml-sycl.cpp b/ggml-sycl.cpp
@@ -15263,6 +15263,7 @@ static void ggml_sycl_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1
             }
         } else {
             bool use_mul_mat_q = min_compute_capability >= VER_4VEC && ggml_is_quantized(src0->type);
+            use_mul_mat_q = use_mul_mat_q && (src0->type != GGML_TYPE_IQ2_XXS);
 
             if (use_xmx && min_compute_capability >= VER_GEN9 && src1->ne[1] > XMX_MAX_BATCH_SIZE) {
                 use_mul_mat_q = false;

Original file line number	Diff line number	Diff line change
`@@ -15263,6 +15263,7 @@ static void ggml_sycl_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1`
`15263`	`15263`	`}`
`15264`	`15264`	`} else {`
`15265`	`15265`	`bool use_mul_mat_q = min_compute_capability >= VER_4VEC && ggml_is_quantized(src0->type);`
	`15266`	`+ use_mul_mat_q = use_mul_mat_q && (src0->type != GGML_TYPE_IQ2_XXS);`
`15266`	`15267`
`15267`	`15268`	`if (use_xmx && min_compute_capability >= VER_GEN9 && src1->ne[1] > XMX_MAX_BATCH_SIZE) {`
`15268`	`15269`	`use_mul_mat_q = false;`