From 3e06ef49630793c86e4deb736fe999f4e484f6d1 Mon Sep 17 00:00:00 2001 From: Nicolas Castet Date: Tue, 27 Jan 2026 09:51:13 -0800 Subject: [PATCH] Fix incorrect MNNVL fabric check Signed-off-by: Nicolas Castet --- .../comm_gemm_overlap/userbuffers/userbuffers-host.cpp | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/transformer_engine/common/comm_gemm_overlap/userbuffers/userbuffers-host.cpp b/transformer_engine/common/comm_gemm_overlap/userbuffers/userbuffers-host.cpp index 9c597be306..6ff9d63a2d 100644 --- a/transformer_engine/common/comm_gemm_overlap/userbuffers/userbuffers-host.cpp +++ b/transformer_engine/common/comm_gemm_overlap/userbuffers/userbuffers-host.cpp @@ -122,10 +122,11 @@ bool has_mnnvl_fabric(int device_id) { NVTE_CALL_CHECK_CUDA_NVML(nvmlDeviceGetHandleByIndex_v2, device_id, &local_device); nvmlGpuFabricInfoV_t fabricInfo = {}; fabricInfo.version = nvmlGpuFabricInfo_v2; - fabricInfo.clusterUuid[0] = '\0'; NVTE_CALL_CHECK_CUDA_NVML(nvmlDeviceGetGpuFabricInfoV, local_device, &fabricInfo); NVTE_CALL_CHECK_CUDA_NVML(nvmlShutdown); - if (fabricInfo.state >= NVML_GPU_FABRIC_STATE_COMPLETED && fabricInfo.clusterUuid[0] != '\0') { + const unsigned char zero_uuid[NVML_GPU_FABRIC_UUID_LEN] = {0}; + if (fabricInfo.state == NVML_GPU_FABRIC_STATE_COMPLETED && + memcmp(fabricInfo.clusterUuid, zero_uuid, NVML_GPU_FABRIC_UUID_LEN) != 0) { mnnvl_fabric_support = true; } }