gpustack-runtime 0.2.0.post3__tar.gz → 0.2.0.post4__tar.gz

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: gpustack-runtime
-Version: 0.2.0.post3
+Version: 0.2.0.post4
 Summary: GPUStack Runtime is library for detecting GPU resources and launching GPU workloads.
 Project-URL: Homepage, https://github.com/gpustack/runtime
 Project-URL: Bug Tracker, https://github.com/gpustack/gpustack/issues

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/_version.py

@@ -27,8 +27,8 @@ version_tuple: VERSION_TUPLE
 __commit_id__: COMMIT_ID
 commit_id: COMMIT_ID
 
-__version__ = version = '0.2.0.post3'
-__version_tuple__ = version_tuple = (0, 2, 0, 'post3')
+__version__ = version = '0.2.0.post4'
+__version_tuple__ = version_tuple = (0, 2, 0, 'post4')
 try:
     from ._version_appendix import git_commit
     __commit_id__ = commit_id = git_commit

gpustack_runtime-0.2.0.post4/gpustack_runtime/_version_appendix.py

@@ -0,0 +1 @@
+git_commit = "992585c"

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/detector/__utils__.py

@@ -13,6 +13,8 @@ from functools import lru_cache
 from pathlib import Path
 from typing import Any
 
+from gpustack_runtime import envs
+
 
 @dataclass
 class PCIDevice:
@@ -69,7 +71,10 @@ def get_pci_devices(
         return pci_devices
 
     if not class_prefix:
-        class_prefix = ["0x02", "0x03", "0x12", "0x0b"]
+        class_prefix = []
+        for prefix in envs.GPUSTACK_RUNTIME_DETECT_PCI_CLASS_PREFIXES:
+            if prefix:
+                class_prefix.append(f"0x{prefix.lower()}")
 
     if address and isinstance(address, str):
         address = [address]

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/detector/amd.py

@@ -7,7 +7,7 @@ from pathlib import Path
 
 from .. import envs
 from ..logging import debug_log_exception, debug_log_warning
-from . import Topology, pyamdgpu, pyamdsmi, pyhsa, pyrocmcore, pyrocmsmi
+from . import Topology, pyamdgpu, pyamdsmi, pyhsa, pyrocmsmi
 from .__types__ import (
     Detector,
     Device,
@@ -94,7 +94,10 @@ class AMDDetector(Detector):
         ret: Devices = []
 
         try:
-            hsa_agents = {hsa_agent.uuid: hsa_agent for hsa_agent in pyhsa.get_agents()}
+            hsa_agents = {
+                hsa_agent.bdf or hsa_agent.uuid: hsa_agent
+                for hsa_agent in pyhsa.get_agents()
+            }
 
             pyamdsmi.amdsmi_init()
             try:
@@ -102,7 +105,7 @@ class AMDDetector(Detector):
             except Exception:
                 debug_log_exception(logger, "Failed to initialize ROCm SMI")
 
-            sys_runtime_ver_original = pyrocmcore.getROCmVersion()
+            sys_runtime_ver_original = pyamdsmi.amdsmi_get_rocm_version()
             sys_runtime_ver = get_brief_version(sys_runtime_ver_original)
 
             devs = pyamdsmi.amdsmi_get_processor_handles()
@@ -115,7 +118,13 @@ class AMDDetector(Detector):
                     dev_uuid = f"GPU-{(asic_serial[2:]).lower()}"
                 else:
                     dev_uuid = f"GPU-{pyrocmsmi.rsmi_dev_unique_id_get(dev_idx)[2:]}"
-                dev_hsa_agent = hsa_agents.get(dev_uuid, pyhsa.Agent())
+
+                dev_bdf = pyamdsmi.amdsmi_get_gpu_device_bdf(dev)
+                dev_card_id, dev_renderd_id = _get_card_and_renderd_id(dev_bdf)
+
+                dev_hsa_agent = (
+                    hsa_agents.get(dev_bdf) or hsa_agents.get(dev_uuid) or pyhsa.Agent()
+                )
 
                 dev_gpu_driver_info = pyamdsmi.amdsmi_get_gpu_driver_info(dev)
                 dev_driver_ver = dev_gpu_driver_info.get("driver_version")
@@ -134,9 +143,6 @@ class AMDDetector(Detector):
                                 dev_idx,
                             )
 
-                dev_bdf = pyamdsmi.amdsmi_get_gpu_device_bdf(dev)
-                dev_card_id, dev_renderd_id = _get_card_and_renderd_id(dev_bdf)
-
                 dev_cores = dev_hsa_agent.compute_units
                 dev_asic_family_id = dev_hsa_agent.asic_family_id
                 if (

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/detector/ascend.py

@@ -2,11 +2,14 @@ from __future__ import annotations as __future_annotations__
 
 import contextlib
 import logging
+import os
+import re
 from functools import lru_cache
+from pathlib import Path
 
 from .. import envs
 from ..logging import debug_log_exception, debug_log_warning
-from . import pyacl, pydcmi
+from . import pydcmi
 from .__types__ import (
     Detector,
     Device,
@@ -111,7 +114,7 @@ class AscendDetector(Detector):
 
             sys_driver_ver = pydcmi.dcmi_get_driver_version()
 
-            sys_runtime_ver_original = pyacl.aclsysGetVersion()
+            sys_runtime_ver_original = _get_toolkit_version()
             sys_runtime_ver = get_brief_version(sys_runtime_ver_original)
 
             _, card_list = pydcmi.dcmi_get_card_list()
@@ -210,10 +213,7 @@ class AscendDetector(Detector):
                             dev_numa = map_cpu_affinity_to_numa_node(dev_cpu_affinity)
 
                     dev_appendix = {
-                        "arch_family": (
-                            pyacl.aclrtGetSocName()
-                            or _guess_soc_name_from_dev_name(dev_name)
-                        ),
+                        "arch_family": _guess_soc_name_from_dev_name(dev_name),
                         "vgpu": dev_is_vgpu,
                         "bdf": dev_bdf,
                         "card_id": dev_card_id,
@@ -478,6 +478,58 @@ def _get_device_virtual_info(
     return None
 
 
+def _get_toolkit_home() -> Path:
+    """
+    Resolve the Ascend toolkit home directory.
+
+    Returns:
+        The path to the Ascend toolkit home.
+
+    """
+    # Example ASCEND_TOOLKIT_HOME
+    # - /usr/local/Ascend/cann
+    # - /usr/local/Ascend/ascend-toolkit/latest/runtime
+    toolkit_home = os.getenv("ASCEND_TOOLKIT_HOME")
+    if toolkit_home:
+        return Path(toolkit_home)
+
+    default_home = Path("/usr/local/Ascend/cann")
+    if default_home.is_dir():
+        return default_home
+
+    return Path("/usr/local/Ascend/ascend-toolkit/latest/runtime")
+
+
+def _get_toolkit_version() -> str | None:
+    """
+    Read the Ascend toolkit (CANN) version from known version.info files.
+
+    Returns:
+        The toolkit version string, or None if not found.
+
+    """
+    prefix = "Version="
+    toolkit_home = _get_toolkit_home()
+
+    for vf in (
+        toolkit_home / "version.info",
+        toolkit_home / "share" / "info" / "runtime" / "version.info",
+    ):
+        if not vf.is_file():
+            continue
+
+        try:
+            content = vf.read_text()
+        except OSError:
+            continue
+
+        for line in content.splitlines():
+            if line.startswith(prefix):
+                return line[len(prefix) :].strip()
+
+    return None
+
+
 # Borrowed from https://gitcode.com/Ascend/pytorch/blob/master/torch_npu/csrc/core/npu/NpuVariables.cpp#L13-L40 and
 # https://gitcode.com/Ascend/pytorch/blob/master/torch_npu/csrc/core/npu/NpuVariables.h#L5-L34.
 # Ascend product category, please refer to:
@@ -518,6 +570,11 @@ _soc_name_version_mapping: dict[str, int] = {
 }
 
 
+_910A_REGEX = re.compile(r"^910")
+_910B_REGEX = re.compile(r"^(910B\d|A2G\d)")
+_310P_REGEX = re.compile(r"^(310P\d?|I2\d?)")
+
+
 def _guess_soc_name_from_dev_name(dev_name: str) -> str | None:
     """
     Guess the SoC name from the device name.
@@ -530,11 +587,23 @@ def _guess_soc_name_from_dev_name(dev_name: str) -> str | None:
         The guessed SoC name, or None if not found.
 
     """
+    dev_name = dev_name.strip()
     if dev_name.startswith("Ascend"):
-        dev_name = dev_name[6:].strip()
+        dev_name = dev_name[len("Ascend") :].strip()
     soc_name = f"Ascend{dev_name}"
     if soc_name in _soc_name_version_mapping:
         return soc_name
+
+    # https://gitcode.com/Ascend/mind-cluster/blob/master/component/ascend-common/devmanager/common/utils.go#L159-L176
+    if _310P_REGEX.match(dev_name):
+        return "Ascend310P1"
+    if "310B" in dev_name:
+        return "Ascend310B1"
+    if _910B_REGEX.match(dev_name):
+        return "Ascend910B1"
+    if _910A_REGEX.match(dev_name):
+        return "Ascend910A"
+
     return None
 
 

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/detector/hygon.py

@@ -7,7 +7,7 @@ from pathlib import Path
 
 from .. import envs
 from ..logging import debug_log_exception, debug_log_warning
-from . import Topology, pyamdgpu, pyhsa, pyrocmcore, pyrocmsmi
+from . import Topology, pyamdgpu, pyhsa, pyrocmsmi
 from .__types__ import (
     Detector,
     Device,
@@ -27,6 +27,7 @@ from .__utils__ import (
     get_utilization,
     map_numa_node_to_cpu_affinity,
 )
+from .amd import _get_arch_family
 
 logger = logging.getLogger(__name__)
 
@@ -94,7 +95,10 @@ class HygonDetector(Detector):
         ret: Devices = []
 
         try:
-            hsa_agents = {hsa_agent.uuid: hsa_agent for hsa_agent in pyhsa.get_agents()}
+            hsa_agents = {
+                hsa_agent.bdf or hsa_agent.uuid: hsa_agent
+                for hsa_agent in pyhsa.get_agents()
+            }
 
             pyrocmsmi.rsmi_init()
 
@@ -108,7 +112,7 @@ class HygonDetector(Detector):
                         sys_driver_ver = path.read_text().strip()
                     break
 
-            sys_runtime_ver_original = pyrocmcore.getROCmVersion()
+            sys_runtime_ver_original = pyrocmsmi.rsmi_get_rocm_version()
             sys_runtime_ver = get_brief_version(sys_runtime_ver_original)
 
             devs_count = pyrocmsmi.rsmi_num_monitor_devices()
@@ -116,7 +120,13 @@ class HygonDetector(Detector):
                 dev_index = dev_idx
 
                 dev_uuid = f"GPU-{pyrocmsmi.rsmi_dev_unique_id_get(dev_idx)[2:]}"
-                dev_hsa_agent = hsa_agents.get(dev_uuid, pyhsa.Agent())
+
+                dev_bdf = pyrocmsmi.rsmi_dev_pci_id_get(dev_idx)
+                dev_card_id, dev_renderd_id = _get_card_and_renderd_id(dev_bdf)
+
+                dev_hsa_agent = (
+                    hsa_agents.get(dev_bdf) or hsa_agents.get(dev_uuid) or pyhsa.Agent()
+                )
 
                 dev_name = dev_hsa_agent.name
                 if not dev_name:
@@ -127,17 +137,20 @@ class HygonDetector(Detector):
                     with contextlib.suppress(pyrocmsmi.ROCMSMIError):
                         dev_cc = pyrocmsmi.rsmi_dev_target_graphics_version_get(dev_idx)
 
-                dev_bdf = pyrocmsmi.rsmi_dev_pci_id_get(dev_idx)
-                dev_card_id, dev_renderd_id = _get_card_and_renderd_id(dev_bdf)
-
                 dev_cores = dev_hsa_agent.compute_units
-                if not dev_cores and dev_card_id is not None:
+                dev_asic_family_id = dev_hsa_agent.asic_family_id
+                if (
+                    not dev_cores or not dev_asic_family_id
+                ) and dev_card_id is not None:
                     with (
                         contextlib.suppress(pyamdgpu.AMDGPUError),
                         pyamdgpu.amdgpu_device(dev_card_id) as dev_gpudev,
                     ):
                         dev_gpudev_info = pyamdgpu.amdgpu_query_gpu_info(dev_gpudev)
-                        dev_cores = dev_gpudev_info.cu_active_number
+                        if not dev_cores:
+                            dev_cores = dev_gpudev_info.cu_active_number
+                        if not dev_asic_family_id:
+                            dev_asic_family_id = dev_gpudev_info.family_id
 
                 dev_cores_util = pyrocmsmi.rsmi_dev_busy_percent_get(dev_idx)
                 dev_temp = pyrocmsmi.rsmi_dev_temp_metric_get(dev_idx)
@@ -177,6 +190,7 @@ class HygonDetector(Detector):
                         )
 
                 dev_appendix = {
+                    "arch_family": _get_arch_family(dev_asic_family_id),
                     "vgpu": dev_is_vgpu,
                     "bdf": dev_bdf,
                 }
@@ -305,8 +319,6 @@ class HygonDetector(Detector):
                                     )
                                 else:
                                     distance = TopologyDistanceEnum.SYS
-                            case pyrocmsmi.ROCMSMI_IOLINK_TYPE_XGMI:
-                                distance = TopologyDistanceEnum.LINK
                             case _:
                                 if link_hops == 0:
                                     distance = TopologyDistanceEnum.SELF

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/detector/nvidia.py

@@ -12,7 +12,7 @@ from pathlib import Path
 
 from .. import envs
 from ..logging import debug_log_exception, debug_log_warning
-from . import DeviceMemoryStatusEnum, Topology, pycuda, pynvml
+from . import DeviceMemoryStatusEnum, Topology, pynvml
 from .__types__ import Detector, Device, Devices, ManufacturerEnum, TopologyDistanceEnum
 from .__utils__ import (
     PCIDevice,
@@ -97,11 +97,6 @@ class NVIDIADetector(Detector):
             pci_devs = NVIDIADetector.detect_pci_devices()
 
             pynvml.nvmlInit()
-            if not envs.GPUSTACK_RUNTIME_DETECT_NO_TOOLKIT_CALL:
-                try:
-                    pycuda.cuInit()
-                except pycuda.CUDAError:
-                    debug_log_exception(logger, "Failed to initialize CUDA")
 
             sys_driver_ver = pynvml.nvmlSystemGetDriverVersion()
 
@@ -173,13 +168,8 @@ class NVIDIADetector(Detector):
                     dev_uuid = pynvml.nvmlDeviceGetUUID(dev)
 
                     dev_cores = None
-                    if not envs.GPUSTACK_RUNTIME_DETECT_NO_TOOLKIT_CALL:
-                        with contextlib.suppress(pycuda.CUDAError):
-                            dev_gpudev = pycuda.cuDeviceGet(dev_idx)
-                            dev_cores = pycuda.cuDeviceGetAttribute(
-                                dev_gpudev,
-                                pycuda.CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
-                            )
+                    with contextlib.suppress(pynvml.NVMLError):
+                        dev_cores = pynvml.nvmlDeviceGetNumGpuCores(dev)
 
                     dev_cores_util = _get_sm_util_from_gpm_metrics(dev)
                     if dev_cores_util is None:

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/detector/pyamdsmi/__init__.py

@@ -74,3 +74,38 @@ except Exception:
 
     def amdsmi_shut_down():
         pass
+
+
+# Example ROCM_HOME/ROCM_PATH
+# - /opt/rocm
+_rocm_home = Path(os.getenv("ROCM_HOME", os.getenv("ROCM_PATH") or "/opt/rocm"))
+
+
+# NB: Defined after the `from amdsmi import *` above on purpose:
+# the amdsmi package may export its own amdsmi_get_rocm_version
+# (returning a tuple), which this file-based implementation must override.
+def amdsmi_get_rocm_version() -> str | None:
+    """
+    Read the ROCm version from known files under the ROCm installation.
+
+    Returns:
+        The ROCm version string, or None if not found.
+
+    """
+    for vf in (
+        _rocm_home / ".info" / "version",
+        _rocm_home / ".info" / "version-rocm",
+        _rocm_home / ".info" / "version-dev",
+        _rocm_home / ".info" / "version-libs",
+    ):
+        if not vf.is_file():
+            continue
+
+        try:
+            version = vf.read_text().strip()
+        except OSError:
+            continue
+        if version:
+            return version
+
+    return None

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/detector/pyhsa/__init__.py

@@ -453,10 +453,24 @@ def has_agent_get_asic_family_id(agent):
     return c_family_id.value
 
 
+def hsa_agent_get_info_bdf(agent):
+    c_bdfid = c_uint64()
+    hsa_agent_get_info(agent, HSA_AMD_AGENT_INFO_BDFID, byref(c_bdfid))
+    # BDFID = ((DOMAIN & 0xFFFFFFFF) << 32) | ((BUS & 0xFF) << 8)
+    #         | ((DEVICE & 0x1F) << 3) | (FUNCTION & 0x7)
+    bdfid = c_bdfid.value
+    domain = (bdfid >> 32) & 0xFFFFFFFF
+    bus = (bdfid >> 8) & 0xFF
+    device_id = (bdfid >> 3) & 0x1F
+    function = bdfid & 0x7
+    return f"{domain:04x}:{bus:02x}:{device_id:02x}.{function:x}"
+
+
 @dataclass
 class Agent:
     device_type: int | None = None
     device_id: str | None = None
+    bdf: str | None = None
     uuid: str | None = None
     name: str | None = None
     compute_capability: str | None = None
@@ -476,6 +490,9 @@ def get_agents() -> list[Agent]:
                 return 0
 
             agent_device_id = hex(hsa_agent_get_info_chip_id(agent))
+            agent_bdf = None
+            with contextlib.suppress(HSAError):
+                agent_bdf = hsa_agent_get_info_bdf(agent)
             agent_uuid = hsa_agent_get_info_uuid(agent)
             agent_name = hsa_agent_get_info_product_name(agent)
             agent_compute_capability = hsa_agent_get_info_name(agent)
@@ -488,6 +505,7 @@ def get_agents() -> list[Agent]:
                 Agent(
                     device_type=agent_device_type,
                     device_id=agent_device_id,
+                    bdf=agent_bdf,
                     uuid=agent_uuid,
                     name=agent_name,
                     compute_capability=agent_compute_capability,

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/detector/pyrocmsmi/__init__.py

@@ -207,6 +207,37 @@ def rsmi_shutdown():
         _libInitializedException = None
 
 
+def rsmi_get_rocm_version() -> str | None:
+    """
+    Read the ROCm version from known files under the ROCm installation.
+
+    Returns:
+        The ROCm version string, or None if not found.
+
+    """
+    for vf in (
+        rocm_path / ".info" / "version",
+        rocm_path / ".info" / "version-rocm",
+        rocm_path / ".info" / "version-dev",
+        rocm_path / ".info" / "version-libs",
+        rocm_path.parent / ".info" / "version",
+        rocm_path.parent / ".info" / "version-rocm",
+        rocm_path.parent / ".info" / "version-dev",
+        rocm_path.parent / ".info" / "version-libs",
+    ):
+        if not vf.is_file():
+            continue
+
+        try:
+            version = vf.read_text().strip()
+        except OSError:
+            continue
+        if version:
+            return version
+
+    return None
+
+
 @convertStrBytes
 def rsmi_driver_version_get():
     component = rsmi_sw_component_t.RSMI_SW_COMP_DRIVER

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/gpustack_runtime/envs.py

@@ -39,14 +39,10 @@ if TYPE_CHECKING:
     Set true to disable PCI check during detection.
     Useful for WSL environments, where PCI information may not be available.
     """
-    GPUSTACK_RUNTIME_DETECT_NO_TOOLKIT_CALL: bool = False
+    GPUSTACK_RUNTIME_DETECT_PCI_CLASS_PREFIXES: set[str] | None = None
     """
-    Set true to disable toolkit calls during detection.
-    Device detection typically involves calling platform-side management libraries and platform-side toolkit to retrieve extra information.
-    For example, during NVIDIA detection, the NVML and CUDA are called, with CUDA used to retrieve GPU cores.
-    However, if certain toolchains are not correctly installed in the environment,
-    such as the Nvidia Fabric Manager being missing, calling the CUDA can cause blocking.
-    Enabling this parameter can prevent blocking events.
+    The PCI class prefixes to check during detection.
+    When `GPUSTACK_RUNTIME_DETECT_NO_PCI_CHECK` is false, the detector will check if the PCI class of the device starts with any of the given prefixes.
     """
     GPUSTACK_RUNTIME_DETECT_NO_HEALTH_CHECK: bool = True
     """
@@ -374,11 +370,15 @@ variables: dict[str, Callable[[], Any]] = {
         lambda: to_bool(getenv("GPUSTACK_RUNTIME_DETECT_NO_PCI_CHECK")),
         lambda: any(x in get_os_release() for x in ("microsoft", "wsl")),
     ),
-    "GPUSTACK_RUNTIME_DETECT_NO_TOOLKIT_CALL": lambda: to_bool(
-        getenv(
-            "GPUSTACK_RUNTIME_DETECT_NO_TOOLKIT_CALL",
-            "0",
+    "GPUSTACK_RUNTIME_DETECT_PCI_CLASS_PREFIXES": lambda: to_set(
+        getenvs(
+            [
+                "GPUSTACK_RUNTIME_DETECT_PCI_CLASS_PREFIXES",
+                "GPUSTACK_PCI_CLASS_PREFIXES",
+            ],
+            "02,03,0b,12",
         ),
+        sep=",",
     ),
     "GPUSTACK_RUNTIME_DETECT_NO_HEALTH_CHECK": lambda: to_bool(
         getenv(

{gpustack_runtime-0.2.0.post3 → gpustack_runtime-0.2.0.post4}/ruff.toml

@@ -121,10 +121,8 @@ parametrize-names-type = "csv"
 
 [lint.per-file-ignores]
 "gpustack_runtime/deployer/__patches__.py" = ["ALL"]
-"gpustack_runtime/detector/pyacl/*.py" = ["ALL"]
 "gpustack_runtime/detector/pyamdsmi/*.py" = ["ALL"]
 "gpustack_runtime/detector/pyamdgpu/*.py" = ["ALL"]
-"gpustack_runtime/detector/pycuda/*.py" = ["ALL"]
 "gpustack_runtime/detector/pydcmi/*.py" = ["ALL"]
 "gpustack_runtime/detector/pyhgml/*.py" = ["ALL"]
 "gpustack_runtime/detector/pyhsa/*.py" = ["ALL"]
@@ -132,7 +130,6 @@ parametrize-names-type = "csv"
 "gpustack_runtime/detector/pymtml/*.py" = ["ALL"]
 "gpustack_runtime/detector/pymxsml/*.py" = ["ALL"]
 "gpustack_runtime/detector/pynvml/*.py" = ["ALL"]
-"gpustack_runtime/detector/pyrocmcore/*.py" = ["ALL"]
 "gpustack_runtime/detector/pyrocmsmi/*.py" = ["ALL"]
 "tests/*.py" = ["D", "S101"]
 

gpustack_runtime-0.2.0.post3/gpustack_runtime/_version_appendix.py

@@ -1 +0,0 @@
-git_commit = "de1ceac"