request prealloc_y sync after quantization

0cc4m · 0cc4m · commit b153aac38508 · 2025-11-01T20:40:30.000Z
diff --git a/ggml/src/ggml-vulkan/ggml-vulkan.cpp b/ggml/src/ggml-vulkan/ggml-vulkan.cpp
@@ -7493,7 +7493,7 @@ 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;
     }
 }
@@ -7793,7 +7793,7 @@ 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
`@@ -7493,7 +7493,7 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context&`
`7493`	`7493`	`if (x_non_contig \|\| qx_needs_dequant) {`
`7494`	`7494`	`ctx->prealloc_x_need_sync = true;`
`7495`	`7495`	`}`
`7496`		`- if (y_non_contig) {`
	`7496`	`+ if (y_non_contig \|\| quantize_y) {`
`7497`	`7497`	`ctx->prealloc_y_need_sync = true;`
`7498`	`7498`	`}`
`7499`	`7499`	`}`
`@@ -7793,7 +7793,7 @@ static void ggml_vk_mul_mat_vec_id_q_f16(ggml_backend_vk_context * ctx, vk_conte`
`7793`	`7793`	`if (x_non_contig) {`
`7794`	`7794`	`ctx->prealloc_x_need_sync = true;`
`7795`	`7795`	`}`
`7796`		`- if (y_non_contig) {`
	`7796`	`+ if (y_non_contig \|\| quantize_y) {`
`7797`	`7797`	`ctx->prealloc_y_need_sync = true;`
`7798`	`7798`	`}`
`7799`	`7799`	`}`