@@ -3002,7 +3002,7 @@ static void ggml_vk_load_shaders(vk_device& device) {
30023002 if ((device->architecture == AMD_GCN) && (device->driver_id != vk::DriverId::eAmdProprietary)) {
30033003 m_warptile_mmq = m_warptile_mmq_int = { 256, 64, 64, 32, 16, 16, 2, 2, 2, 1, 16 };
30043004 m_warptile_mmqid = m_warptile_mmqid_int = { 256, 64, 64, 32, 16, 16, 2, 2, 2, 1, 16 };
3005- } else if (device->vendor_id == VK_VENDOR_ID_AMD && device->coopmat_support) {
3005+ } else if (device->vendor_id == VK_VENDOR_ID_AMD && device->coopmat_support && device->driver_id != vk::DriverId::eAmdProprietary ) {
30063006 // This is intentionally using tx_m values, slight performance increase
30073007 l_warptile = { 256, 128, 128, 16, subgroup_size_8, 64, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
30083008 l_warptile_mmq = l_warptile_mmq_int = { 256, 128, 128, 32, subgroup_size_8, 64, 2, tm_m, tn_m, tk_m, subgroup_size_8 };
@@ -5083,7 +5083,7 @@ static vk_device ggml_vk_get_device(size_t idx) {
50835083 switch (device->vendor_id) {
50845084#ifndef GGML_VULKAN_RUN_TESTS
50855085 case VK_VENDOR_ID_AMD:
5086- device->mul_mat_l[i] = device->coopmat_support;
5086+ device->mul_mat_l[i] = device->coopmat_support && device->driver_id != vk::DriverId::eAmdProprietary ;
50875087 device->mul_mat_m[i] = true;
50885088 device->mul_mat_s[i] = true;
50895089 device->mul_mat_id_l[i] = false;
0 commit comments