build_tools/buildkite/cmake/linux/cuda/benchmark.yml

# Copyright 2022 The IREE Authors
#
# Licensed under the Apache License v2.0 with LLVM Exceptions.
# See https://llvm.org/LICENSE.txt for license information.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception

steps:
  - label: ":hammer_and_wrench: Build linux-cuda benchmark tools"
    key: "build-cuda-benchmark-tools"
    commands: |
      docker run --user=$(id -u):$(id -g) \
        --volume="$${HOME?}:$${HOME?}" \
        --volume="/etc/passwd:/etc/passwd:ro" \
        --volume="/etc/group:/etc/group:ro" \
        --volume="$$PWD:$$IREE_DOCKER_WORKDIR" \
        --workdir="$$IREE_DOCKER_WORKDIR" \
        --rm \
        $${DOCKER_IMAGE} \
        build_tools/cmake/build_linux_benchmark_tools.sh linux-cuda
      tar -czvf iree-linux-cuda-tools-${BUILDKITE_BUILD_NUMBER}.tgz \
        build-targets/linux-cuda/tools/iree-benchmark-module \
        build-targets/linux-cuda/tools/build_config.txt
    agents:
      - "queue=build"
    artifact_paths:
      - "iree-linux-cuda-tools-${BUILDKITE_BUILD_NUMBER}.tgz"

  - label: ":stopwatch: Benchmark on Nvidia A100 (GCP-a2-highgpu-1g)"
    key: "run-cuda-benchmark-gcp-gpu-a100"
    depends_on: "build-cuda-benchmark-tools"
    commands: |
      git clean -fdx
      buildkite-agent artifact download \
        "benchmark-suites-linux-cuda-$${BUILDKITE_BUILD_NUMBER}.tgz" ./
      buildkite-agent artifact download \
        "iree-linux-cuda-tools-$${BUILDKITE_BUILD_NUMBER}.tgz" ./
      tar -xzvf "benchmark-suites-linux-cuda-$${BUILDKITE_BUILD_NUMBER}.tgz"
      tar -xzvf "iree-linux-cuda-tools-$${BUILDKITE_BUILD_NUMBER}.tgz"
      python3 build_tools/benchmarks/run_benchmarks_on_linux.py \
        --device_model=GCP-a2-highgpu-1g \
        --normal_benchmark_tool_dir=build-targets/linux-cuda/tools/ \
        --driver_filter_regex=cuda \
        -o "benchmark-results-gcp-gpu-a100-$${BUILDKITE_BUILD_NUMBER}.json" \
        --verbose \
        build-targets/linux-cuda
    agents:
      - "gcp:machine-type=a2-highgpu-1g"
      - "queue=benchmark-cuda"
    artifact_paths:
      - "benchmark-results-gcp-gpu-a100-${BUILDKITE_BUILD_NUMBER}.json"
    timeout_in_minutes: "10"