request prealloc_y sync after quantization

0cc4m · 0cc4m · commit 1b78909c991c · 2025-11-07T19:58:29.000+01:00
diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -7502,7 +7502,7 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context&
     if (x_non_contig || qx_needs_dequant) {
         ctx->prealloc_x_need_sync = true;
     }
-    if (y_non_contig) {
+    if (y_non_contig || quantize_y) {
         ctx->prealloc_y_need_sync = true;
     }
 }
@@ -7803,7 +7803,7 @@ static void ggml_vk_mul_mat_vec_id_q_f16(ggml_backend_vk_context * ctx, vk_conte
     if (x_non_contig) {
         ctx->prealloc_x_need_sync = true;
     }
-    if (y_non_contig) {
+    if (y_non_contig || quantize_y) {
         ctx->prealloc_y_need_sync = true;
     }
 }

Original file line number	Diff line number	Diff line change
`@@ -7502,7 +7502,7 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context&`
`7502`	`7502`	`if (x_non_contig \|\| qx_needs_dequant) {`
`7503`	`7503`	`ctx->prealloc_x_need_sync = true;`
`7504`	`7504`	`}`
`7505`		`- if (y_non_contig) {`
	`7505`	`+ if (y_non_contig \|\| quantize_y) {`
`7506`	`7506`	`ctx->prealloc_y_need_sync = true;`
`7507`	`7507`	`}`
`7508`	`7508`	`}`
`@@ -7803,7 +7803,7 @@ static void ggml_vk_mul_mat_vec_id_q_f16(ggml_backend_vk_context * ctx, vk_conte`
`7803`	`7803`	`if (x_non_contig) {`
`7804`	`7804`	`ctx->prealloc_x_need_sync = true;`
`7805`	`7805`	`}`
`7806`		`- if (y_non_contig) {`
	`7806`	`+ if (y_non_contig \|\| quantize_y) {`
`7807`	`7807`	`ctx->prealloc_y_need_sync = true;`
`7808`	`7808`	`}`
`7809`	`7809`	`}`