zluda device id #4

installer.py

@@ -474,19 +474,22 @@ def install_rocm_zluda():
     #    os.environ.setdefault('TENSORFLOW_PACKAGE', 'tensorflow-rocm')
 
     amd_gpus = []
-    hip_default_device = None
+    device = None
     try:
         amd_gpus = rocm.get_agents()
         if len(amd_gpus) == 0:
-            (log.info if sys.platform == "win32" else log.warning)('ROCm: no agent was found')
+            if sys.platform == "win32":
+                log.warning('You do not have perl or any AMDGPUs. The installer may select a wrong device as compute device.')
+                log.info('ROCm: no agent was found')
+            else:
+                log.warning('ROCm: no agent was found')
         else:
             log.info(f'ROCm: agents={[gpu.name for gpu in amd_gpus]}')
             if args.device_id is None:
-                hip_default_device = amd_gpus[0]
+                device = amd_gpus[0]
                 for idx, gpu in enumerate(amd_gpus):
                     if gpu.arch == rocm.MicroArchitecture.RDNA:
-                        hip_default_device = gpu
-                        log.debug(f'ROCm default agent: idx={idx} gpu={gpu.name}')
+                        device = gpu
                         os.environ.setdefault('HIP_VISIBLE_DEVICES', str(idx))
                         # if os.environ.get('TENSORFLOW_PACKAGE') == 'tensorflow-rocm': # do not use tensorflow-rocm for navi 3x
                         #    os.environ['TENSORFLOW_PACKAGE'] = 'tensorflow==2.13.0'
@@ -495,12 +498,14 @@ def install_rocm_zluda():
             else:
                 device_id = int(args.device_id)
                 if device_id < len(amd_gpus):
-                    hip_default_device = amd_gpus[device_id]
-                    log.debug(f'ROCm agent: id={device_id} gpu={hip_default_device.name}')
+                    device = amd_gpus[device_id]
     except Exception as e:
         log.warning(f'ROCm agent enumerator failed: {e}')
 
-    log.info(f'ROCm: version={rocm.version}')
+    msg = f'ROCm: version={rocm.version}'
+    if device is not None:
+        msg += f', using agent {device.name}'
+    log.info(msg)
     torch_command = ''
     if sys.platform == "win32":
         # TODO after ROCm for Windows is released
@@ -526,7 +531,7 @@ def install_rocm_zluda():
         if error is None:
             try:
                 zluda_installer.load(zluda_path)
-                torch_command = os.environ.get('TORCH_COMMAND', f'torch=={zluda_installer.get_default_torch_version(hip_default_device)} torchvision --index-url https://download.pytorch.org/whl/cu118')
+                torch_command = os.environ.get('TORCH_COMMAND', f'torch=={zluda_installer.get_default_torch_version(device)} torchvision --index-url https://download.pytorch.org/whl/cu118')
                 log.info(f'Using ZLUDA in {zluda_path}')
             except Exception as e:
                 error = e
@@ -553,16 +558,16 @@ def install_rocm_zluda():
             install(ort_package, 'onnxruntime-training')
 
         if 'Flash attention' in opts.get('sdp_options'):
-            install(rocm.get_flash_attention_command(hip_default_device))
+            install(rocm.get_flash_attention_command(device))
         elif not args.experimental:
             uninstall('flash-attn')
 
-        if hip_default_device is not None and rocm.version != "6.2" and rocm.version == rocm.version_torch and rocm.get_blaslt_enabled():
-            log.debug(f'ROCm hipBLASLt: arch={hip_default_device.name} available={hip_default_device.blaslt_supported}')
-            rocm.set_blaslt_enabled(hip_default_device.blaslt_supported)
+        if device is not None and rocm.version != "6.2" and rocm.version == rocm.version_torch and rocm.get_blaslt_enabled():
+            log.debug(f'ROCm hipBLASLt: arch={device.name} available={device.blaslt_supported}')
+            rocm.set_blaslt_enabled(device.blaslt_supported)
 
-    if hip_default_device is not None:
-        os.environ.setdefault('HSA_OVERRIDE_GFX_VERSION', hip_default_device.get_gfx_version())
+    if device is not None:
+        os.environ.setdefault('HSA_OVERRIDE_GFX_VERSION', device.get_gfx_version())
 
     return torch_command