pytorch/.github/workflows/inductor-perf-test-nightly-rocm.yml

name: inductor-perf-nightly-rocm

on:
  push:
    tags:
      - ciflow/inductor-perf-test-nightly-rocm/*
  schedule:
    - cron: 0 7 * * 0,3
  # NB: GitHub has an upper limit of 10 inputs here, so before we can sort it
  # out, let try to run torchao cudagraphs_low_precision as part of cudagraphs
  workflow_dispatch:
    inputs:
      training:
        description: Run training (on by default)?
        required: false
        type: boolean
        default: true
      inference:
        description: Run inference (on by default)?
        required: false
        type: boolean
        default: true
      default:
        description: Run inductor_default?
        required: false
        type: boolean
        default: false
      dynamic:
        description: Run inductor_dynamic_shapes?
        required: false
        type: boolean
        default: false
      cppwrapper:
        description: Run inductor_cpp_wrapper?
        required: false
        type: boolean
        default: false
      cudagraphs:
        description: Run inductor_cudagraphs?
        required: false
        type: boolean
        default: true
      freezing_cudagraphs:
        description: Run inductor_cudagraphs with freezing for inference?
        required: false
        type: boolean
        default: false
      aotinductor:
        description: Run aot_inductor for inference?
        required: false
        type: boolean
        default: false
      maxautotune:
        description: Run inductor_max_autotune?
        required: false
        type: boolean
        default: false
      benchmark_configs:
        description: The list of configs used the benchmark
        required: false
        type: string
        default: inductor_huggingface_perf_rocm,inductor_timm_perf_rocm,inductor_torchbench_perf_rocm

concurrency:
  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
  cancel-in-progress: true

permissions: read-all

jobs:
  get-label-type:
    name: get-label-type
    uses: pytorch/pytorch/.github/workflows/_runner-determinator.yml@main
    if: ${{ (github.event_name != 'schedule' || github.repository == 'pytorch/pytorch') && github.repository_owner == 'pytorch' }}
    with:
      triggering_actor: ${{ github.triggering_actor }}
      issue_owner: ${{ github.event.pull_request.user.login || github.event.issue.user.login }}
      curr_branch: ${{ github.head_ref || github.ref_name }}
      curr_ref_type: ${{ github.ref_type }}
      opt_out_experiments: lf

  linux-jammy-rocm-py3_10-inductor-benchmark-build:
    if: github.repository_owner == 'pytorch'
    name: rocm-py3_10-inductor-benchmark-build
    uses: ./.github/workflows/_linux-build.yml
    with:
      build-environment: linux-jammy-rocm-py3_10
      docker-image-name: ci-image:pytorch-linux-jammy-rocm-n-py3-benchmarks
      test-matrix: |
        { include: [
          { config: "inductor_huggingface_perf_rocm", shard: 1, num_shards: 4, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_huggingface_perf_rocm", shard: 2, num_shards: 4, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_huggingface_perf_rocm", shard: 3, num_shards: 4, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_huggingface_perf_rocm", shard: 4, num_shards: 4, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_timm_perf_rocm", shard: 1, num_shards: 5, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_timm_perf_rocm", shard: 2, num_shards: 5, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_timm_perf_rocm", shard: 3, num_shards: 5, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_timm_perf_rocm", shard: 4, num_shards: 5, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_timm_perf_rocm", shard: 5, num_shards: 5, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_torchbench_perf_rocm", shard: 1, num_shards: 8, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_torchbench_perf_rocm", shard: 2, num_shards: 8, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_torchbench_perf_rocm", shard: 3, num_shards: 8, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_torchbench_perf_rocm", shard: 4, num_shards: 8, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_torchbench_perf_rocm", shard: 5, num_shards: 8, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_torchbench_perf_rocm", shard: 6, num_shards: 8, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_torchbench_perf_rocm", shard: 7, num_shards: 8, runner: "linux.rocm.gpu.gfx942.1" },
          { config: "inductor_torchbench_perf_rocm", shard: 8, num_shards: 8, runner: "linux.rocm.gpu.gfx942.1" },
        ]}
    secrets: inherit

  linux-jammy-rocm-py3_10-inductor-benchmark-test:
    permissions:
      id-token: write
      contents: read
    name: rocm-py3_10-inductor-benchmark-test
    uses: ./.github/workflows/_rocm-test.yml
    needs: linux-jammy-rocm-py3_10-inductor-benchmark-build
    with:
      build-environment: linux-jammy-rocm-py3_10
      dashboard-tag: training-true-inference-true-default-true-dynamic-true-cudagraphs-true-cppwrapper-true-aotinductor-true-freezing_cudagraphs-true-cudagraphs_low_precision-true
      docker-image: ${{ needs.linux-jammy-rocm-py3_10-inductor-benchmark-build.outputs.docker-image }}
      test-matrix: ${{ needs.linux-jammy-rocm-py3_10-inductor-benchmark-build.outputs.test-matrix }}
      timeout-minutes: 720
      # Disable monitor in perf tests for more investigation
      disable-monitor: true
      monitor-log-interval: 10
      monitor-data-collect-interval: 2
    secrets: inherit