refactor: enhance detector

Signed-off-by: thxCode <thxcode0824@gmail.com>

Author: thxCode

gpustack_runtime/deployer/__utils__.py

@@ -140,14 +140,12 @@ def _get_backend_version_and_variant(backend: str) -> (str, str):
     version = ""
     variant = ""
 
-    manufacturer = backend_to_manufacturer(backend)
-    if not manufacturer:
-        return version, variant
-
     devices = detect_devices(fast=False)
     if not devices:
         return version, variant
 
+    manufacturer = backend_to_manufacturer(backend)
+
     for device in devices:
         if device.manufacturer != manufacturer:
             continue

gpustack_runtime/detector/__types__.py

@@ -70,7 +70,7 @@ def __str__(self):
 """
 
 
-def manufacturer_to_backend(manufacturer: ManufacturerEnum) -> str | None:
+def manufacturer_to_backend(manufacturer: ManufacturerEnum) -> str:
     """
     Convert manufacturer to runtime backend,
     e.g., NVIDIA -> cuda, AMD -> rocm.
@@ -82,13 +82,17 @@ def manufacturer_to_backend(manufacturer: ManufacturerEnum) -> str | None:
         manufacturer: The manufacturer of the device.
 
     Returns:
-        The corresponding runtime backend. None if the manufacturer is unknown.
+        The corresponding runtime backend.
+        Return "unknown" if the manufacturer is unknown.
 
     """
-    return _MANUFACTURER_BACKEND_MAPPING.get(manufacturer)
+    backend = _MANUFACTURER_BACKEND_MAPPING.get(manufacturer)
+    if backend:
+        return backend
+    return ManufacturerEnum.UNKNOWN.value
 
 
-def backend_to_manufacturer(backend: str) -> ManufacturerEnum | None:
+def backend_to_manufacturer(backend: str) -> ManufacturerEnum:
     """
     Convert runtime backend to manufacturer,
     e.g., cuda -> NVIDIA, rocm -> AMD.
@@ -100,13 +104,14 @@ def backend_to_manufacturer(backend: str) -> ManufacturerEnum | None:
         backend: The runtime backend.
 
     Returns:
-        The corresponding manufacturer. None if the backend is unknown.
+        The corresponding manufacturer.
+        Return ManufacturerEnum.Unknown if the backend is unknown.
 
     """
     for manufacturer, mapped_backend in _MANUFACTURER_BACKEND_MAPPING.items():
         if mapped_backend == backend:
             return manufacturer
-    return None
+    return ManufacturerEnum.UNKNOWN
 
 
 def supported_manufacturers() -> list[ManufacturerEnum]:
@@ -241,7 +246,7 @@ def __init__(self, manufacturer: ManufacturerEnum):
         self.manufacturer = manufacturer
 
     @property
-    def backend(self) -> str | None:
+    def backend(self) -> str:
         """
         The backend name of the detector, e.g., 'cuda', 'rocm'.
         """

gpustack_runtime/detector/amd.py

@@ -91,20 +91,20 @@ def detect(self) -> Devices | None:
             for dev_idx, dev in enumerate(devs):
                 dev_index = dev_idx
 
-                dev_card = None
+                dev_card_id = None
                 if hasattr(pyamdsmi, "amdsmi_get_gpu_kfd_info"):
                     dev_kfd_info = pyamdsmi.amdsmi_get_gpu_kfd_info(dev)
-                    dev_card = dev_kfd_info.get("node_id")
+                    dev_card_id = dev_kfd_info.get("node_id")
                 else:
                     with contextlib.suppress(pyrocmsmi.ROCMSMIError):
                         pyrocmsmi.rsmi_init()
-                        dev_card = pyrocmsmi.rsmi_dev_node_id_get(dev_idx)
+                        dev_card_id = pyrocmsmi.rsmi_dev_node_id_get(dev_idx)
 
                 dev_gpudev_info = None
-                if dev_card is not None:
+                if dev_card_id is not None:
                     with contextlib.suppress(pyamdgpu.AMDGPUError):
                         _, _, dev_gpudev = pyamdgpu.amdgpu_device_initialize(
-                            dev_card,
+                            dev_card_id,
                         )
                         dev_gpudev_info = pyamdgpu.amdgpu_query_gpu_info(dev_gpudev)
                         pyamdgpu.amdgpu_device_deinitialize(dev_gpudev)
@@ -122,8 +122,6 @@ def detect(self) -> Devices | None:
                     with contextlib.suppress(pyrocmsmi.ROCMSMIError):
                         pyrocmsmi.rsmi_init()
                         dev_cc = pyrocmsmi.rsmi_dev_target_graphics_version_get(dev_idx)
-                if dev_cc:
-                    dev_cc = dev_cc[3:]  # Strip "gfx" prefix
 
                 dev_gpu_metrics_info = pyamdsmi.amdsmi_get_gpu_metrics_info(dev)
                 dev_cores = (
@@ -152,6 +150,7 @@ def detect(self) -> Devices | None:
                 dev_appendix = {
                     "arch_family": _get_arch_family(dev_gpudev_info),
                     "vgpu": dev_compute_partition is not None,
+                    "card_id": dev_card_id,
                 }
 
                 ret.append(

gpustack_runtime/detector/hygon.py

@@ -86,8 +86,6 @@ def detect(self) -> Devices | None:
                 dev_uuid = pyrocmsmi.rsmi_dev_unique_id_get(dev_idx)
                 dev_name = pyrocmsmi.rsmi_dev_name_get(dev_idx)
                 dev_cc = pyrocmsmi.rsmi_dev_target_graphics_version_get(dev_idx)
-                if dev_cc:
-                    dev_cc = dev_cc[3:]  # Strip "gfx" prefix
 
                 dev_cores = None
                 dev_cores_util = pyrocmsmi.rsmi_dev_busy_percent_get(dev_idx)

gpustack_runtime/detector/pyrocmsmi/__init__.py

@@ -197,7 +197,12 @@ def rsmi_dev_target_graphics_version_get(device=0):
     _rocmsmiCheckReturn(ret)
     version = str(c_version.value)
     if len(version) == 4:
-        version = hex(c_version.value)[2:]
+        dev_name = rsmi_dev_name_get(device)
+        if "Instinct MI2" in dev_name:
+            hex_part = str(hex(int(version[2:]))).replace("0x", "")
+            version = version[:2] + hex_part
+    else:
+        version = str(c_version.value // 10 + c_version.value % 10)
     return "gfx" + version