Update GPU vendors

superbench/benchmarks/micro_benchmarks/ib_validation_performance.py

@@ -282,7 +282,7 @@ def __prepare_general_ib_command_params(self, msg_size, device='cpu'):
             gpu = GPU()
             if gpu.vendor == 'nvidia':
                 gpu_dev = f'--use_cuda={self._args.gpu_dev}'
-            elif gpu.vendor == 'amd':
+            elif gpu.vendor == 'amd' or gpu.vendor == 'hygon':
                 gpu_dev = f'--use_rocm={self._args.gpu_dev}'
             else:
                 self._result.set_return_code(ReturnCode.INVALID_ARGUMENT)

superbench/common/utils/device_manager.py

@@ -12,7 +12,7 @@
 gpu = GPU()
 if gpu.vendor == 'nvidia' or gpu.vendor == 'nvidia-graphics':
     import py3nvml.py3nvml as nvml
-elif gpu.vendor == 'amd' or gpu.vendor == 'amd-graphics':
+elif gpu.vendor == 'amd' or gpu.vendor == 'amd-graphics' or gpu.vendor == 'hygon':
     import amdsmi as rocml
 
 
@@ -456,5 +456,5 @@ def get_device_ecc_error(self, idx):
 device_manager: Optional[DeviceManager] = DeviceManager()
 if gpu.vendor == 'nvidia' or gpu.vendor == 'nvidia-graphics':
     device_manager = NvidiaDeviceManager()
-elif gpu.vendor == 'amd' or gpu.vendor == 'amd-graphics':
+elif gpu.vendor == 'amd' or gpu.vendor == 'amd-graphics' or gpu.vendor == 'hygon':
     device_manager = AmdDeviceManager()

superbench/tools/system_info.py

@@ -286,6 +286,13 @@ def get_gpu_amd(self):
 
         return gpu_dict
 
+    def get_gpu_hygon(self):
+        """Get hygon gpu info."""
+        gpu_dict = self.get_gpu_amd()
+        if gpu_dict:
+            gpu_dict['accelerator_vendor'] = 'hygon'
+        return gpu_dict
+
     def get_gpu(self):
         """Get gpu info and identify gpu type(nvidia/amd).
 
@@ -296,6 +303,8 @@ def get_gpu(self):
             if Path('/dev/nvidiactl').is_char_device() and Path('/dev/nvidia-uvm').is_char_device():
                 return self.get_gpu_nvidia()
             if Path('/dev/kfd').is_char_device() and Path('/dev/dri').is_dir():
+                if Path('/usr/local/hyhal').exists() or Path('/opt/hyhal').exists():
+                    return self.get_gpu_hygon()
                 return self.get_gpu_amd()
         except Exception:
             logger.exception('Error: get gpu info failed')

tests/benchmarks/micro_benchmarks/test_ib_traffic_performance.py

@@ -224,6 +224,11 @@ def test_ib_traffic_performance(self, mock_gpu):
         expect_command = expect_command.replace('cuda', 'rocm')
         command = benchmark._bin_name + benchmark._commands[0].split(benchmark._bin_name)[1]
         assert (command == expect_command)
+        mock_gpu.return_value = 'hygon'
+        benchmark = benchmark_class(benchmark_name, parameters=parameters)
+        ret = benchmark._preprocess()
+        command = benchmark._bin_name + benchmark._commands[0].split(benchmark._bin_name)[1]
+        assert (command == expect_command)
 
         parameters = '--command ib_read_lat --ib_dev mlx5_0 --iters 2000 --msg_size 33554432 ' + \
             '--pattern one-to-one --hostfile hostfile --gpu_dev 0 --direction gpu-to-gpu'