Merge pull request #88 from fabric-testbed/87-gpu-as-multi-function-p…

…ci-devices support gpus as multi function pci devices
fabric-testbed · Feb 27, 2024 · 0ce0f99 · 0ce0f99
2 parents 8e94af3 + 4b2b8df
commit 0ce0f99
Show file tree

Hide file tree

Showing 3 changed files with 19 additions and 7 deletions.
diff --git a/fimutil/__init__.py b/fimutil/__init__.py
@@ -3,5 +3,5 @@
 This is a package of Information Model utilitied for FABRIC
 for scanning different types of sites
 """
-__VERSION__ = "1.6.3"
+__VERSION__ = "1.6.4"
 __version__ = __VERSION__
diff --git a/fimutil/ralph/gpu.py b/fimutil/ralph/gpu.py
@@ -16,8 +16,8 @@ class GPU:
     """
     Model: str
     Description: str
-    BDF: str
-    NUMA: str
+    BDF: list
+    NUMA: list
 
     @staticmethod
     def find_gpus(node_raw_json) -> List[Any]:
@@ -33,8 +33,18 @@ def find_gpus(node_raw_json) -> List[Any]:
                     logging.debug(f'Detected GPU {gpu_model}')
                     model = gpu_model
                     description = custom_fields[field]
-                    bdf = custom_fields[field + '_pci_id']
+                    # May contain multiple PCI devices depending on GPU type
+                    # GPUs at CIEN rack have both an audio and video PCI device associated
+                    # [root@cien-w1 ~]# lspci  | grep 25:00.
+                    # 25:00.0 VGA compatible controller: NVIDIA Corporation AD102GL [RTX 6000 Ada Generation] (rev a1)
+                    # 25:00.1 Audio device: NVIDIA Corporation AD102 High Definition Audio Controller (rev a1)
+                    # bdf field woul show as 25:00.0 25:00.1
+                    # Extract PCI devices information
+                    bdf_values = custom_fields.get(f'{field}_pci_id', '').split()
+                    bdf = [bdf.strip() for bdf in bdf_values]
                     # -1 means unknown
-                    numa = custom_fields.get(field + '_numa_node', '-1')
+                    # Extract NUMA information
+                    numa_val = custom_fields.get(f'{field}_numa_node', '-1')
+                    numa = [numa_val] * len(bdf) if bdf else [numa_val]
                     ret.append(GPU(model, description, bdf, numa))
         return ret
diff --git a/pyproject.toml b/pyproject.toml
@@ -4,7 +4,9 @@ build-backend = "flit_core.buildapi"
 
 [project]
 name = "fim-utils"
-authors = [{name = "Ilya Baldin", email = "ibaldin@renci.org"}, {name = "Xi Yang", email="xiyang@es.net"}, {name="Hussamuddin Nasir", email="nasir@netlab.uky.edu"}]
+authors = [{name = "Ilya Baldin", email = "ibaldin@renci.org"},
+           {name = "Xi Yang", email="xiyang@es.net"},
+           {name="Hussamuddin Nasir", email="nasir@netlab.uky.edu"}]
 readme = "README.md"
 license = {file = "LICENSE"}
 classifiers = ["License :: OSI Approved :: MIT License",
@@ -13,7 +15,7 @@ classifiers = ["License :: OSI Approved :: MIT License",
 dynamic = ["version", "description"]
 requires-python = '>=3.9'
 dependencies = [
-	"fabric_fim >= 1.6.0",
+	"fabric_fim >= 1.6.2",
     "pyjq == 2.6.0",
 	"jsonpath_ng == 1.5.3",
 	]