Merge pull request #10391 from rakhmets/topic/nvml-fabric-info-v1

UCT/CUDA/CUDA_IPC: Switched to nvmlGpuFabricInfo v1 - v1.18.x
openucx · Dec 19, 2024 · ed3cebd · ed3cebd
2 parents 0dc692c + a1a097a
commit ed3cebd
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 13 deletions.
diff --git a/config/m4/cuda.m4 b/config/m4/cuda.m4
@@ -66,10 +66,10 @@ AS_IF([test "x$cuda_checked" != "xyes"],
                                     [AC_MSG_ERROR([libnvidia-ml not found. Install appropriate nvidia-driver package])])
                               cuda_happy="no"])])
 
-         # Check for nvmlDeviceGetGpuFabricInfoV
-         AC_CHECK_DECLS([nvmlDeviceGetGpuFabricInfoV],
+         # Check for nvmlDeviceGetGpuFabricInfo
+         AC_CHECK_DECLS([nvmlDeviceGetGpuFabricInfo],
                         [AC_DEFINE([HAVE_NVML_FABRIC_INFO], 1, [Enable NVML GPU fabric info support])],
-                        [AC_MSG_NOTICE([nvmlDeviceGetGpuFabricInfoV function not found in libnvidia-ml. MNNVL support will be disabled.])],
+                        [AC_MSG_NOTICE([nvmlDeviceGetGpuFabricInfo function not found in libnvidia-ml. MNNVL support will be disabled.])],
                         [[#include <nvml.h>]])
 
 

diff --git a/src/uct/cuda/cuda_ipc/cuda_ipc_md.c b/src/uct/cuda/cuda_ipc/cuda_ipc_md.c
@@ -416,7 +416,7 @@ uct_cuda_ipc_md_check_fabric_info(uct_cuda_ipc_md_t *md,
     static int mnnvl_supported = 0;
 #else
     static int mnnvl_supported = -1;
-    nvmlGpuFabricInfoV_t fabric_info;
+    nvmlGpuFabricInfo_t fabric_info;
     nvmlDevice_t device;
     ucs_status_t status;
     char buf[64];
@@ -436,19 +436,17 @@ uct_cuda_ipc_md_check_fabric_info(uct_cuda_ipc_md_t *md,
         goto out_not_supported;
     }
 
-    fabric_info.version = nvmlGpuFabricInfo_v2;
-    status              = UCT_NVML_FUNC_LOG_ERR(
-                             nvmlDeviceGetGpuFabricInfoV(device, &fabric_info));
+    status = UCT_NVML_FUNC_LOG_ERR(
+            nvmlDeviceGetGpuFabricInfo(device, &fabric_info));
     if (status != UCS_OK) {
         goto out_not_supported;
     }
 
-    ucs_debug("fabric_info: healthmask=%u state=%u status=%u clique=%u uuid=%s",
-              fabric_info.healthMask, fabric_info.state, fabric_info.status,
-              fabric_info.cliqueId,
-              ucs_str_dump_hex(
-                  fabric_info.clusterUuid, NVML_GPU_FABRIC_UUID_LEN, buf,
-                  sizeof(buf), SIZE_MAX));
+    ucs_debug("fabric_info: state=%u status=%u uuid=%s", fabric_info.state,
+              fabric_info.status,
+              ucs_str_dump_hex(fabric_info.clusterUuid,
+                               NVML_GPU_FABRIC_UUID_LEN, buf, sizeof(buf),
+                               SIZE_MAX));
 
     if ((fabric_info.state == NVML_GPU_FABRIC_STATE_COMPLETED) &&
         (fabric_info.status == NVML_SUCCESS)) {