refactor: device field

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

Author: thxCode

gpustack_runtime/detector/__types__.py

@@ -159,29 +159,18 @@ class Device:
     """
     Driver version of the device.
     """
-    driver_version_tuple: list[int | str] | None = None
-    """
-    Driver version tuple of the device.
-    None if `driver_version` is missed.
-    """
     runtime_version: str | None = None
     """
-    Runtime version of the device.
+    Runtime version in major[.minor] of the device.
     """
-    runtime_version_tuple: list[int | str] | None = None
+    runtime_version_original: str | None = None
     """
-    Runtime version tuple of the device.
-    None if `runtime_version` is missed.
+    Original runtime version string of the device.
     """
     compute_capability: str | None = None
     """
     Compute capability of the device.
     """
-    compute_capability_tuple: list[int | str] | None = None
-    """
-    Compute capability tuple of the device.
-    None if `compute_capability` is missed.
-    """
     cores: int | None = None
     """
     Total cores of the device.
@@ -190,15 +179,15 @@ class Device:
     """
     Core utilization of the device in percentage.
     """
-    memory: int | float = 0
+    memory: int = 0
     """
     Total memory of the device in MiB.
     """
-    memory_used: int | float = 0
+    memory_used: int = 0
     """
     Used memory of the device in MiB.
     """
-    memory_utilization: int | float = 0
+    memory_utilization: float = 0
     """
     Memory utilization of the device in percentage.
     """

gpustack_runtime/detector/__utils__.py

@@ -368,3 +368,92 @@ def safe_str(value: Any, default: str = "") -> str:
         return str(value)
     except (ValueError, TypeError):
         return default
+
+
+def kibibyte_to_mebibyte(value: int) -> int:
+    """
+    Convert KiB to MiB.
+
+    Args:
+        value:
+            The value in kilobytes.
+
+    Returns:
+        The value in MiB, or 0 if the input is None or negative.
+
+    """
+    if value is None or value < 0:
+        return 0
+
+    try:
+        return value >> 10
+    except (ValueError, TypeError, OverflowError):
+        return 0
+
+
+def byte_to_mebibyte(value: int) -> int:
+    """
+    Convert bytes to MiB.
+
+    Args:
+        value:
+            The value in bytes.
+
+    Returns:
+        The value in MiB, or 0 if the input is None or negative.
+
+    """
+    if value is None or value < 0:
+        return 0
+
+    try:
+        return value >> 20
+    except (ValueError, TypeError, OverflowError):
+        return 0
+
+
+def get_brief_version(version: str | None) -> str | None:
+    """
+    Get a brief version string,
+    e.g., "11.2.152" -> "11.2".
+
+    Args:
+        version:
+            The full version string.
+
+    Returns:
+        The brief version string, or None if the input is None or empty.
+
+    """
+    if not version:
+        return None
+
+    splits = version.split(".", 3)
+    if len(splits) >= 2:
+        return ".".join(splits[:2])
+    if len(splits) == 1:
+        return splits[0]
+    return None
+
+
+def get_utilization(used: int | None, total: int | None) -> float:
+    """
+    Calculate utilization percentage.
+
+    Args:
+        used:
+            The used value.
+        total:
+            The total value.
+
+    Returns:
+        The utilization percentage, rounded to two decimal places.
+
+    """
+    if used is None or total is None or used < 0 or total <= 0:
+        return 0.0
+    try:
+        result = (used / total) * 100
+    except (OverflowError, ZeroDivisionError):
+        return 0.0
+    return round(result, 2)

gpustack_runtime/detector/amd.py

@@ -7,7 +7,13 @@
 from .. import envs
 from . import pyamdgpu, pyamdsmi, pyrocmsmi
 from .__types__ import Detector, Device, Devices, ManufacturerEnum
-from .__utils__ import PCIDevice, get_device_files, get_pci_devices
+from .__utils__ import (
+    PCIDevice,
+    get_brief_version,
+    get_device_files,
+    get_pci_devices,
+    get_utilization,
+)
 
 logger = logging.getLogger(__name__)
 
@@ -79,12 +85,8 @@ def detect(self) -> Devices | None:
         try:
             pyamdsmi.amdsmi_init()
 
-            sys_runtime_ver = pyamdsmi.amdsmi_get_rocm_version_major_minor()
-            sys_runtime_ver_t = (
-                [int(v) if v.isdigit() else v for v in sys_runtime_ver.split(".")]
-                if sys_runtime_ver
-                else None
-            )
+            sys_runtime_ver_original = pyamdsmi.amdsmi_get_rocm_original_version()
+            sys_runtime_ver = get_brief_version(sys_runtime_ver_original)
 
             devs = pyamdsmi.amdsmi_get_processor_handles()
             dev_files = get_device_files(
@@ -108,17 +110,11 @@ def detect(self) -> Devices | None:
 
                 dev_gpu_driver_info = pyamdsmi.amdsmi_get_gpu_driver_info(dev)
                 dev_driver_ver = dev_gpu_driver_info.get("driver_version")
-                dev_driver_ver_t = (
-                    [int(v) if v.isdigit() else v for v in dev_driver_ver.split(".")]
-                    if dev_driver_ver
-                    else None
-                )
 
                 dev_gpu_asic_info = pyamdsmi.amdsmi_get_gpu_asic_info(dev)
                 dev_uuid = dev_gpu_asic_info.get("asic_serial")
                 dev_name = dev_gpu_asic_info.get("market_name")
                 dev_cc = None
-                dev_cc_t = None
                 if hasattr(dev_gpu_asic_info, "target_graphics_version"):
                     dev_cc = dev_gpu_asic_info.target_graphics_version
                 else:
@@ -127,7 +123,6 @@ def detect(self) -> Devices | None:
                         dev_cc = pyrocmsmi.rsmi_dev_target_graphics_version_get(dev_idx)
                 if dev_cc:
                     dev_cc = dev_cc[3:]  # Strip "gfx" prefix
-                    dev_cc_t = [int(v) if v.isdigit() else v for v in dev_cc.split(".")]
 
                 dev_gpu_metrics_info = pyamdsmi.amdsmi_get_gpu_metrics_info(dev)
                 dev_cores = (
@@ -165,18 +160,14 @@ def detect(self) -> Devices | None:
                         name=dev_name,
                         uuid=dev_uuid,
                         driver_version=dev_driver_ver,
-                        driver_version_tuple=dev_driver_ver_t,
                         runtime_version=sys_runtime_ver,
-                        runtime_version_tuple=sys_runtime_ver_t,
+                        runtime_version_original=sys_runtime_ver_original,
                         compute_capability=dev_cc,
-                        compute_capability_tuple=dev_cc_t,
                         cores=dev_cores,
                         cores_utilization=dev_cores_util,
                         memory=dev_mem,
                         memory_used=dev_mem_used,
-                        memory_utilization=(
-                            (dev_mem_used * 100 // dev_mem) if dev_mem > 0 else 0
-                        ),
+                        memory_utilization=get_utilization(dev_mem_used, dev_mem),
                         temperature=dev_temp,
                         power=dev_power,
                         power_used=dev_power_used,

gpustack_runtime/detector/ascend.py

@@ -7,7 +7,7 @@
 from .. import envs
 from . import pyacl, pydcmi
 from .__types__ import Detector, Device, Devices, ManufacturerEnum
-from .__utils__ import PCIDevice, get_pci_devices
+from .__utils__ import PCIDevice, get_brief_version, get_pci_devices, get_utilization
 
 logger = logging.getLogger(__name__)
 
@@ -79,16 +79,9 @@ def detect(self) -> Devices | None:
             pydcmi.dcmi_init()
 
             sys_driver_ver = pydcmi.dcmi_get_driver_version()
-            sys_driver_ver_t = [
-                int(v) if v.isdigit() else v for v in sys_driver_ver.split(".")
-            ]
-
-            sys_runtime_ver = pyacl.aclsysGetCANNVersion()
-            sys_runtime_ver_t = (
-                [int(v) if v.isdigit() else v for v in sys_runtime_ver.split(".")]
-                if sys_runtime_ver
-                else None
-            )
+
+            sys_runtime_ver_original = pyacl.aclsysGetCANNVersion()
+            sys_runtime_ver = get_brief_version(sys_runtime_ver_original)
 
             _, card_list = pydcmi.dcmi_get_card_list()
             for dev_card_id in card_list:
@@ -179,16 +172,13 @@ def detect(self) -> Devices | None:
                             name=dev_name,
                             uuid=dev_uuid.upper(),
                             driver_version=sys_driver_ver,
-                            driver_version_tuple=sys_driver_ver_t,
                             runtime_version=sys_runtime_ver,
-                            runtime_version_tuple=sys_runtime_ver_t,
+                            runtime_version_original=sys_runtime_ver_original,
                             cores=dev_cores_aicore,
                             cores_utilization=dev_util_aicore,
                             memory=dev_mem,
                             memory_used=dev_mem_used,
-                            memory_utilization=(
-                                (dev_mem_used / dev_mem) * 100 if dev_mem > 0 else 0
-                            ),
+                            memory_utilization=get_utilization(dev_mem_used, dev_mem),
                             temperature=dev_temp,
                             power_used=(
                                 dev_power_used / 10  # Convert from 0.1W to W

gpustack_runtime/detector/cambricon.py

@@ -10,6 +10,7 @@
     PCIDevice,
     execute_shell_command,
     get_pci_devices,
+    get_utilization,
     safe_float,
     safe_int,
     support_command,
@@ -96,8 +97,8 @@ def detect(self) -> Devices | None:
                 dev_cores_util = safe_float(dev_util_info.get("MLUAverage", 0))
 
                 dev_mem_usage_info = dev_info.get("PhysicalMemUsage", {})
-                dev_mem = safe_int(dev_mem_usage_info.get("Total", 0)) << 20
-                dev_mem_used = safe_int(dev_mem_usage_info.get("Used", 0)) << 20
+                dev_mem = safe_int(dev_mem_usage_info.get("Total", 0))  # MiB
+                dev_mem_used = safe_int(dev_mem_usage_info.get("Used", 0))  # MiB
 
                 dev_temp_info = dev_info.get("Temperature", {})
                 dev_temp = safe_float(dev_temp_info.get("Chip", 0))
@@ -115,9 +116,7 @@ def detect(self) -> Devices | None:
                         cores_utilization=dev_cores_util,
                         memory=dev_mem,
                         memory_used=dev_mem_used,
-                        memory_utilization=(
-                            (dev_mem_used * 100 // dev_mem) if dev_mem > 0 else 0
-                        ),
+                        memory_utilization=get_utilization(dev_mem_used, dev_mem),
                         temperature=dev_temp,
                         appendix=dev_appendix,
                     ),

gpustack_runtime/detector/hygon.py

@@ -6,7 +6,7 @@
 from .. import envs
 from . import pyrocmsmi
 from .__types__ import Detector, Device, Devices, ManufacturerEnum
-from .__utils__ import PCIDevice, get_pci_devices
+from .__utils__ import PCIDevice, byte_to_mebibyte, get_pci_devices, get_utilization
 
 logger = logging.getLogger(__name__)
 
@@ -78,11 +78,6 @@ def detect(self) -> Devices | None:
             pyrocmsmi.rsmi_init()
 
             sys_driver_ver = pyrocmsmi.rsmi_driver_version_get()
-            sys_driver_ver_t = (
-                [int(v) if v.isdigit() else v for v in sys_driver_ver.split(".")]
-                if sys_driver_ver
-                else None
-            )
 
             devs_count = pyrocmsmi.rsmi_num_monitor_devices()
             for dev_idx in range(devs_count):
@@ -91,15 +86,15 @@ 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)
-                dev_cc_t = None
                 if dev_cc:
                     dev_cc = dev_cc[3:]  # Strip "gfx" prefix
-                    dev_cc_t = [int(v) if v.isdigit() else v for v in dev_cc.split(".")]
 
                 dev_cores = None
                 dev_cores_util = pyrocmsmi.rsmi_dev_busy_percent_get(dev_idx)
-                dev_mem = pyrocmsmi.rsmi_dev_memory_total_get(dev_idx)
-                dev_mem_used = pyrocmsmi.rsmi_dev_memory_usage_get(dev_idx)
+                dev_mem = byte_to_mebibyte(pyrocmsmi.rsmi_dev_memory_total_get(dev_idx))
+                dev_mem_used = byte_to_mebibyte(
+                    pyrocmsmi.rsmi_dev_memory_usage_get(dev_idx),
+                )
                 dev_temp = pyrocmsmi.rsmi_dev_temp_metric_get(dev_idx)
 
                 dev_power = pyrocmsmi.rsmi_dev_power_cap_get(dev_idx)
@@ -116,16 +111,12 @@ def detect(self) -> Devices | None:
                         name=dev_name,
                         uuid=dev_uuid,
                         driver_version=sys_driver_ver,
-                        driver_version_tuple=sys_driver_ver_t,
                         compute_capability=dev_cc,
-                        compute_capability_tuple=dev_cc_t,
                         cores=dev_cores,
                         cores_utilization=dev_cores_util,
-                        memory=(dev_mem >> 20 if dev_mem > 0 else 0),
-                        memory_used=(dev_mem_used >> 20 if dev_mem_used > 0 else 0),
-                        memory_utilization=(
-                            (dev_mem_used * 100 // dev_mem) if dev_mem > 0 else 0
-                        ),
+                        memory=dev_mem,
+                        memory_used=dev_mem_used,
+                        memory_utilization=get_utilization(dev_mem_used, dev_mem),
                         temperature=dev_temp,
                         power=dev_power,
                         power_used=dev_power_used,

gpustack_runtime/detector/iluvatar.py

@@ -10,6 +10,7 @@
     PCIDevice,
     execute_command,
     get_pci_devices,
+    get_utilization,
     safe_float,
     safe_int,
     support_command,
@@ -99,8 +100,8 @@ def detect(self) -> Devices | None:
 
                 dev_cores_util = safe_float(row[2].split()[0])
 
-                dev_mem = safe_int(row[3].split()[0]) << 20
-                dev_mem_used = safe_int(row[4].split()[0]) << 20
+                dev_mem = safe_int(row[3].split()[0])
+                dev_mem_used = safe_int(row[4].split()[0])
 
                 dev_temp = safe_float(row[5].split()[0])
 
@@ -119,9 +120,7 @@ def detect(self) -> Devices | None:
                         cores_utilization=dev_cores_util,
                         memory=dev_mem,
                         memory_used=dev_mem_used,
-                        memory_utilization=(
-                            (dev_mem_used / dev_mem) * 100 if dev_mem > 0 else 0
-                        ),
+                        memory_utilization=get_utilization(dev_mem_used, dev_mem),
                         temperature=dev_temp,
                         power=dev_power,
                         power_used=dev_power_used,