forked from pytorch/FBGEMM
-
Notifications
You must be signed in to change notification settings - Fork 4
200 lines (162 loc) · 6.11 KB
/
fbgemm_gpu_ci_rocm.yml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
# This workflow is used for FBGEMM_GPU-ROCm CI as well as nightly builds of
# FBGEMM_GPU-ROCm against PyTorch-ROCm Nightly.
name: FBGEMM_GPU-ROCm CI
on:
# PR Trigger (enabled for regression checks and debugging)
#
pull_request:
branches:
- main
# Push Trigger (enable to catch errors coming out of multiple merges)
#
push:
branches:
- main
# Cron Trigger (UTC)
#
# Based on the Conda page for PyTorch-nightly, the GPU nightly releases appear
# around 02:30 PST every day (roughly 2 hours after the CPU releases)
#
schedule:
- cron: '45 12 * * *'
# Manual Trigger
#
workflow_dispatch:
inputs:
publish_to_pypi:
description: Publish Artifact to PyPI
type: boolean
required: false
default: false
concurrency:
# Cancel previous runs in the PR if a new commit is pushed
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
cancel-in-progress: true
jobs:
# Build on CPU hosts and upload to GHA
build_artifact:
runs-on: ${{ matrix.host-machine.instance }}
container:
image: ${{ matrix.container-image }}
options: --user root
defaults:
run:
shell: bash
env:
PRELUDE: .github/scripts/setup_env.bash
BUILD_ENV: build_binary
BUILD_VARIANT: rocm
strategy:
fail-fast: false
matrix:
host-machine: [
{ arch: x86, instance: "gfx90a" },
]
container-image: [ "ubuntu:20.04" ]
python-version: [ "3.12" ]
rocm-version: [ "6.1" ]
compiler: [ "gcc", "clang" ]
steps:
- name: Setup Build Container
run: |
apt update -y
apt install -y binutils git pciutils sudo wget
git config --global --add safe.directory '*'
- name: Checkout the Repository
uses: actions/checkout@v4
- name: Display System Info
run: . $PRELUDE; print_system_info
- name: Display GPU Info
run: . $PRELUDE; print_gpu_info
- name: Free Disk Space
run: . $PRELUDE; free_disk_space
- name: Setup Miniconda
run: . $PRELUDE; setup_miniconda $HOME/miniconda
- name: Create Conda Environment
run: . $PRELUDE; create_conda_environment $BUILD_ENV ${{ matrix.python-version }}
- name: Install C/C++ Compilers
run: . $PRELUDE; install_cxx_compiler $BUILD_ENV ${{ matrix.compiler }}
- name: Install Build Tools
run: . $PRELUDE; install_build_tools $BUILD_ENV
- name: Install ROCm
run: . $PRELUDE; install_rocm_ubuntu $BUILD_ENV ${{ matrix.rocm-version }}
- name: Install PyTorch-ROCm Nightly
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV nightly rocm/${{ matrix.rocm-version }}
- name: Collect PyTorch Environment Info
if: ${{ success() || failure() }}
run: if . $PRELUDE && which conda; then collect_pytorch_env_info $BUILD_ENV; fi
- name: Prepare FBGEMM_GPU Build
run: . $PRELUDE; cd fbgemm_gpu; prepare_fbgemm_gpu_build $BUILD_ENV
- name: Build FBGEMM_GPU Wheel
run: . $PRELUDE; cd fbgemm_gpu; build_fbgemm_gpu_package $BUILD_ENV nightly rocm
- name: Upload Built Wheel as GHA Artifact
uses: actions/upload-artifact@v4
with:
name: fbgemm_gpu_nightly_rocm_${{ matrix.host-machine.arch }}_${{ matrix.compiler }}_py${{ matrix.python-version }}_rocm${{ matrix.rocm-version }}.whl
path: fbgemm_gpu/dist/*.whl
if-no-files-found: error
# Download the built artifact from GHA, test on GPU, and push to PyPI
test_and_publish_artifact:
runs-on: ${{ matrix.host-machine.instance }}
container:
image: "rocm/dev-ubuntu-20.04:${{ matrix.rocm-version }}-complete"
options: --user root --device=/dev/kfd --device=/dev/dri --ipc=host --shm-size 16G --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined
defaults:
run:
shell: bash
env:
PRELUDE: .github/scripts/setup_env.bash
BUILD_ENV: build_binary
BUILD_VARIANT: rocm
ENFORCE_ROCM_DEVICE: 1
strategy:
fail-fast: false
matrix:
host-machine: [
{ arch: x86, instance: "gfx90a" },
]
# ROCm machines are limited, so we only test a subset of Python versions
python-version: [ "3.12" ]
rocm-version: [ "6.1" ]
compiler: [ "gcc", "clang" ]
needs: build_artifact
steps:
- name: Setup Build Container
run: |
apt update -y
apt install -y git wget
git config --global --add safe.directory '*'
- name: Checkout the Repository
uses: actions/checkout@v3
- name: Download Wheel Artifact from GHA
uses: actions/download-artifact@v4
with:
name: fbgemm_gpu_nightly_rocm_${{ matrix.host-machine.arch }}_${{ matrix.compiler }}_py${{ matrix.python-version }}_rocm${{ matrix.rocm-version }}.whl
- name: Display System Info
run: . $PRELUDE; print_system_info
- name: Display GPU Info
run: . $PRELUDE; print_gpu_info
- name: Free Disk Space
run: . $PRELUDE; free_disk_space
- name: Setup Miniconda
run: . $PRELUDE; setup_miniconda $HOME/miniconda
- name: Create Conda Environment
run: . $PRELUDE; create_conda_environment $BUILD_ENV ${{ matrix.python-version }}
- name: Install ROCm AMD-SMI
run: . $PRELUDE; install_rocm_amdsmi_ubuntu $BUILD_ENV
- name: Install PyTorch-ROCm Nightly
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV nightly rocm/${{ matrix.rocm-version }}
- name: Collect PyTorch Environment Info
if: ${{ success() || failure() }}
run: if . $PRELUDE && which conda; then collect_pytorch_env_info $BUILD_ENV; fi
- name: Prepare FBGEMM_GPU Build
run: . $PRELUDE; cd fbgemm_gpu; prepare_fbgemm_gpu_build $BUILD_ENV
- name: Install FBGEMM_GPU Wheel
run: . $PRELUDE; install_fbgemm_gpu_wheel $BUILD_ENV *.whl
- name: Test with PyTest
timeout-minutes: 60
run: . $PRELUDE; test_all_fbgemm_gpu_modules $BUILD_ENV