]> asedeno.scripts.mit.edu Git - linux.git/blobdiff - drivers/gpu/drm/amd/amdkfd/kfd_topology.c
drm/amdkfd: Update number of compute unit from KGD
[linux.git] / drivers / gpu / drm / amd / amdkfd / kfd_topology.c
index cb0303a323d25041b04a386bf99042f03974a1a0..ca2e51a2fe75c712ecfcf4b44c23a51a90f93832 100644 (file)
@@ -133,8 +133,7 @@ static void kfd_populated_cu_info_gpu(struct kfd_topology_device *dev,
        dev->node_props.max_slots_scratch_cu = cu->max_slots_scatch_cu;
        if (cu->hsa_capability & CRAT_CU_FLAGS_HOT_PLUGGABLE)
                dev->node_props.capability |= HSA_CAP_HOT_PLUGGABLE;
-       pr_info("CU GPU: simds=%d id_base=%d\n", cu->num_simd_cores,
-                               cu->processor_id_low);
+       pr_info("CU GPU: id_base=%d\n", cu->processor_id_low);
 }
 
 /* kfd_parse_subtype_cu is called when the topology mutex is already acquired */
@@ -1124,6 +1123,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 {
        uint32_t gpu_id;
        struct kfd_topology_device *dev;
+       struct kfd_cu_info cu_info;
        int res;
 
        gpu_id = kfd_generate_gpu_id(gpu);
@@ -1161,6 +1161,9 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
 
        dev->gpu_id = gpu_id;
        gpu->id = gpu_id;
+       dev->gpu->kfd2kgd->get_cu_info(dev->gpu->kgd, &cu_info);
+       dev->node_props.simd_count = dev->node_props.simd_per_cu *
+                       cu_info.cu_active_number;
        dev->node_props.vendor_id = gpu->pdev->vendor;
        dev->node_props.device_id = gpu->pdev->device;
        dev->node_props.location_id = (gpu->pdev->bus->number << 24) +