gpustack-runtime 0.1.39.post2__py3-none-any.whl → 0.1.40__py3-none-any.whl

gpustack_runtime/detector/amd.py

@@ -16,6 +16,7 @@ from .__utils__ import (
     get_brief_version,
     get_numa_node_by_bdf,
     get_pci_devices,
+    get_physical_function_by_bdf,
     get_utilization,
     map_numa_node_to_cpu_affinity,
 )
@@ -107,7 +108,11 @@ class AMDDetector(Detector):
                     asic_serial = dev_gpu_asic_info.get("asic_serial")
                     dev_uuid = f"GPU-{(asic_serial[2:]).lower()}"
                 else:
-                    dev_uuid = f"GPU-{pyrocmsmi.rsmi_dev_unique_id_get(dev_idx)[2:]}"
+                    dev_uuid = ""
+                    with contextlib.suppress(pyrocmsmi.ROCMSMIError):
+                        dev_uuid = (
+                            f"GPU-{pyrocmsmi.rsmi_dev_unique_id_get(dev_idx)[2:]}"
+                        )
                 dev_hsa_agent = hsa_agents.get(dev_uuid, pyhsa.Agent())
 
                 dev_gpu_driver_info = pyamdsmi.amdsmi_get_gpu_driver_info(dev)
@@ -119,8 +124,13 @@ class AMDDetector(Detector):
 
                 dev_cc = dev_hsa_agent.compute_capability
                 if not dev_cc:
-                    with contextlib.suppress(pyrocmsmi.ROCMSMIError):
-                        dev_cc = pyrocmsmi.rsmi_dev_target_graphics_version_get(dev_idx)
+                    if "target_graphics_version" in dev_gpu_asic_info:
+                        dev_cc = dev_gpu_asic_info.get("target_graphics_version")
+                    else:
+                        with contextlib.suppress(pyrocmsmi.ROCMSMIError):
+                            dev_cc = pyrocmsmi.rsmi_dev_target_graphics_version_get(
+                                dev_idx,
+                            )
 
                 dev_bdf = None
                 dev_card_id = None
@@ -195,15 +205,13 @@ class AMDDetector(Detector):
                         dev_power = pyrocmsmi.rsmi_dev_power_cap_get(dev_idx)
                         dev_power_used = pyrocmsmi.rsmi_dev_power_get(dev_idx)
 
-                dev_compute_partition = None
-                with contextlib.suppress(pyamdsmi.AmdSmiException):
-                    dev_compute_partition = pyamdsmi.amdsmi_get_gpu_compute_partition(
-                        dev,
-                    )
+                dev_is_vgpu = False
+                if dev_bdf:
+                    dev_is_vgpu = get_physical_function_by_bdf(dev_bdf) != dev_bdf
 
                 dev_appendix = {
                     "arch_family": _get_arch_family(dev_asic_family_id),
-                    "vgpu": dev_compute_partition is not None,
+                    "vgpu": dev_is_vgpu,
                 }
                 if dev_bdf:
                     dev_appendix["bdf"] = dev_bdf

gpustack_runtime/detector/hygon.py

@@ -16,6 +16,7 @@ from .__utils__ import (
     get_brief_version,
     get_numa_node_by_bdf,
     get_pci_devices,
+    get_physical_function_by_bdf,
     get_utilization,
     map_numa_node_to_cpu_affinity,
 )
@@ -156,8 +157,12 @@ class HygonDetector(Detector):
                 dev_power = pyrocmsmi.rsmi_dev_power_cap_get(dev_idx)
                 dev_power_used = pyrocmsmi.rsmi_dev_power_get(dev_idx)
 
+                dev_is_vgpu = False
+                if dev_bdf:
+                    dev_is_vgpu = get_physical_function_by_bdf(dev_bdf) != dev_bdf
+
                 dev_appendix = {
-                    "vgpu": False,
+                    "vgpu": dev_is_vgpu,
                 }
                 if dev_bdf is not None:
                     dev_appendix["bdf"] = dev_bdf

gpustack_runtime/detector/iluvatar.py

@@ -23,9 +23,9 @@ from .__utils__ import (
     get_numa_node_by_bdf,
     get_numa_nodeset_size,
     get_pci_devices,
+    get_physical_function_by_bdf,
     get_utilization,
     map_numa_node_to_cpu_affinity,
-    support_command,
 )
 
 logger = logging.getLogger(__name__)
@@ -54,8 +54,14 @@ class IluvatarDetector(Detector):
         pci_devs = IluvatarDetector.detect_pci_devices()
         if not pci_devs and not envs.GPUSTACK_RUNTIME_DETECT_NO_PCI_CHECK:
             logger.debug("No Iluvatar PCI devices found")
+            return supported
 
-        supported = support_command("ixsmi")
+        try:
+            pyixml.nvmlInit()
+            pyixml.nvmlShutdown()
+            supported = True
+        except pyixml.NVMLError:
+            debug_log_exception(logger, "Failed to initialize IXML library")
 
         return supported
 
@@ -73,7 +79,7 @@ class IluvatarDetector(Detector):
 
     def detect(self) -> Devices | None:
         """
-        Detect Iluvatar GPUs using ixsmi tool.
+        Detect Iluvatar GPUs using pyixml.
 
         Returns:
             A list of detected Iluvatar GPU devices,
@@ -165,13 +171,20 @@ class IluvatarDetector(Detector):
                     if dev_cc_t:
                         dev_cc = ".".join(map(str, dev_cc_t))
 
+                dev_bdf = None
+                with contextlib.suppress(pyixml.NVMLError):
+                    dev_pci_info = pyixml.nvmlDeviceGetPciInfo(dev)
+                    dev_bdf = str(dev_pci_info.busIdLegacy).lower()
+
                 dev_is_vgpu = False
-                dev_pci_info = pyixml.nvmlDeviceGetPciInfo(dev)
+                if dev_bdf:
+                    dev_is_vgpu = get_physical_function_by_bdf(dev_bdf) != dev_bdf
 
                 dev_appendix = {
                     "vgpu": dev_is_vgpu,
-                    "bdf": str(dev_pci_info.busIdLegacy).lower(),
                 }
+                if dev_bdf:
+                    dev_appendix["bdf"] = dev_bdf
 
                 ret.append(
                     Device(
@@ -229,6 +242,8 @@ class IluvatarDetector(Detector):
         )
 
         try:
+            pyixml.nvmlInit()
+
             for i, dev_i in enumerate(devices):
                 dev_i_handle = pyixml.nvmlDeviceGetHandleByUUID(dev_i.uuid)
 

gpustack_runtime/detector/mthreads.py

@@ -3,9 +3,10 @@ from __future__ import annotations
 import logging
 from functools import lru_cache
 
+import pymtml
+
 from .. import envs
 from ..logging import debug_log_exception, debug_log_warning
-from . import pymtml
 from .__types__ import (
     Detector,
     Device,
@@ -105,9 +106,8 @@ class MThreadsDetector(Detector):
 
         try:
             pymtml.mtmlLibraryInit()
-
-            sys_driver_ver = pymtml.mtmlSystemGetDriverVersion()
-
+            system = pymtml.mtmlLibraryInitSystem()
+            sys_driver_ver = pymtml.mtmlSystemGetDriverVersion(system)
             dev_count = pymtml.mtmlLibraryCountDevice()
             for dev_idx in range(dev_count):
                 dev_index = dev_idx
@@ -139,25 +139,20 @@ class MThreadsDetector(Detector):
 
                 dev_mem = 0
                 dev_mem_used = 0
-                devmem = pymtml.mtmlDeviceInitMemory(dev)
-                try:
+                with pymtml.mtmlMemoryContext(dev) as devmem:
                     dev_mem = byte_to_mebibyte(  # byte to MiB
                         pymtml.mtmlMemoryGetTotal(devmem),
                     )
                     dev_mem_used = byte_to_mebibyte(  # byte to MiB
                         pymtml.mtmlMemoryGetUsed(devmem),
                     )
-                finally:
-                    pymtml.mtmlDeviceFreeMemory(devmem)
 
                 dev_cores_util = None
                 dev_temp = None
-                devgpu = pymtml.mtmlDeviceInitGpu(dev)
-                try:
+                with pymtml.mtmlGpuContext(dev) as devgpu:
                     dev_cores_util = pymtml.mtmlGpuGetUtilization(devgpu)
                     dev_temp = pymtml.mtmlGpuGetTemperature(devgpu)
-                finally:
-                    pymtml.mtmlDeviceFreeGpu(devgpu)
+
                 if dev_cores_util is None:
                     debug_log_warning(
                         logger,
@@ -198,6 +193,7 @@ class MThreadsDetector(Detector):
             debug_log_exception(logger, "Failed to process devices fetching")
             raise
         finally:
+            pymtml.mtmlLibraryFreeSystem(system)
             pymtml.mtmlLibraryShutDown()
 
         return ret