@@ -2941,9 +2941,9 @@ static void ggml_vk_load_shaders(vk_device& device) {
29412941 ggml_vk_create_pipeline(device, device->pipeline_rope_vision_f16, "rope_vision_f16", rope_vision_f16_len, rope_vision_f16_data, "main", 4, sizeof(vk_op_rope_push_constants), {1, 512, 1}, {}, 1);
29422942 }
29432943
2944- ggml_vk_create_pipeline(device, device->pipeline_out_prod_f32, "out_prod_f32", out_prod_f32_len, out_prod_f32_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, { 0 }, 1);
2945- ggml_vk_create_pipeline(device, device->pipeline_out_prod_q4_0, "out_prod_q4_0", out_prod_q4_0_len, out_prod_q4_0_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, { 0 }, 1, true);
2946- ggml_vk_create_pipeline(device, device->pipeline_out_prod_q8_0, "out_prod_q8_0", out_prod_q8_0_len, out_prod_q8_0_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, { 0 }, 1, true);
2944+ ggml_vk_create_pipeline(device, device->pipeline_out_prod_f32, "out_prod_f32", out_prod_f32_len, out_prod_f32_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, { device->subgroup_size }, 1);
2945+ ggml_vk_create_pipeline(device, device->pipeline_out_prod_q4_0, "out_prod_q4_0", out_prod_q4_0_len, out_prod_q4_0_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
2946+ ggml_vk_create_pipeline(device, device->pipeline_out_prod_q8_0, "out_prod_q8_0", out_prod_q8_0_len, out_prod_q8_0_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, { device->subgroup_size }, 1, true);
29472947
29482948 ggml_vk_create_pipeline(device, device->pipeline_out_prod_f16_f32, "out_prod_f16_f32", out_prod_f16_f32_len, out_prod_f16_f32_data, "main", 3, sizeof(vk_op_binary_push_constants), {512, 1, 1}, { 0 }, 1);
29492949
0 commit comments