multi_gpu_cpp_test.yaml

# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

# Reusable workflow that runs cuOpt's multi-GPU (NCCL) C++ gtests on a 2-GPU
# runner. It mirrors the single-GPU cpp-test flow (ci/test_cpp.sh): install the
# libcuopt conda test package, then run the gtests — but only those that require
# multiple GPUs. Called by both pr.yaml and test.yaml so the runner label,
# container image and test script are defined in exactly one place.
#
# NCCL communication can only be validated with multiple GPUs (a single device
# cannot host multiple ranks of one communicator). The 2-GPU runner group
# (nv-gpu-amd64-rtxpro6000-2gpu) is enabled for NVIDIA/cuopt in the RAPIDS
# runner configuration.

name: multi-gpu cpp tests

on:
  workflow_call:
    inputs:
      build_type:
        required: true
        type: string
      branch:
        type: string
      date:
        type: string
      sha:
        type: string
      node_type:
        description: "custom-job node_type suffix; must be a >=2-GPU runner."
        type: string
        default: "gpu-rtxpro6000-latest-2"
      arch:
        type: string
        default: "amd64"
      container_image:
        type: string
        default: "rapidsai/ci-conda:26.08-latest"
      script:
        type: string
        default: "ci/test_cpp_multi_gpu.sh"
      continue-on-error:
        description: "Treat failures as non-blocking (useful while the MG suite is maturing)."
        type: boolean
        default: false

permissions: {}

jobs:
  multi-gpu-cpp-tests:
    permissions:
      actions: read
      contents: read
      id-token: write
      packages: read
      pull-requests: read
    uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@main
    with:
      build_type: ${{ inputs.build_type }}
      branch: ${{ inputs.branch }}
      date: ${{ inputs.date }}
      sha: ${{ inputs.sha }}
      node_type: ${{ inputs.node_type }}
      arch: ${{ inputs.arch }}
      container_image: ${{ inputs.container_image }}
      script: ${{ inputs.script }}
      continue-on-error: ${{ inputs.continue-on-error }}