-
Notifications
You must be signed in to change notification settings - Fork 512
171 lines (136 loc) · 5.27 KB
/
fbgemm_gpu_cuda_release.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
# 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.
name: FBGEMM_GPU-CUDA Release Build
on:
# PR Trigger (enabled only for debugging)
#
pull_request:
branches:
- main
# Push Trigger (enable to catch errors coming out of multiple merges)
#
push:
branches:
- main
# Manual Trigger
#
workflow_dispatch:
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: amazonlinux:2023
options: --user root
defaults:
run:
shell: bash
env:
PRELUDE: .github/scripts/setup_env.bash
BUILD_ENV: build_binary
continue-on-error: true
strategy:
# Don't fast-fail all the other builds if one of the them fails
fail-fast: false
matrix:
host-machine: [
{ arch: x86, instance: "linux.24xlarge" },
]
python-version: [ "3.8", "3.9", "3.10", "3.11" ]
cuda-version: [ "11.8.0", "12.1.1" ]
steps:
- name: Setup Build Container
run: yum update -y; yum install -y binutils findutils git pciutils sudo tar wget which
- name: Checkout the Repository
uses: actions/checkout@v3
with:
submodules: true
- name: Display System Info
run: . $PRELUDE; print_system_info
- name: Display GPU Info
run: . $PRELUDE; print_gpu_info
- 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
- name: Install Build Tools
run: . $PRELUDE; install_build_tools $BUILD_ENV
- name: Install CUDA
run: . $PRELUDE; install_cuda $BUILD_ENV ${{ matrix.cuda-version }}
- name: Install PyTorch Test
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV test cuda ${{ matrix.cuda-version }}
- name: Install cuDNN
run: . $PRELUDE; install_cudnn $BUILD_ENV "$(pwd)/build_only/cudnn" ${{ matrix.cuda-version }}
- name: Prepare FBGEMM_GPU Build
run: . $PRELUDE; cd fbgemm_gpu; prepare_fbgemm_gpu_build $BUILD_ENV
- name: Build FBGEMM_GPU
run: . $PRELUDE; cd fbgemm_gpu; build_fbgemm_gpu_package $BUILD_ENV release cuda
- name: Upload Built Wheel as GHA Artifact
uses: actions/upload-artifact@v3
with:
name: fbgemm_gpu_release_cuda_${{ matrix.host-machine.arch }}_${{ matrix.python-version }}_cu${{ matrix.cuda-version }}.whl
path: fbgemm_gpu/dist/fbgemm_gpu-*.whl
# Download the built artifact from GHA, test on GPU, and push to PyPI
test_and_publish_artifact:
runs-on: ${{ matrix.host-machine.instance }}
defaults:
run:
shell: bash
env:
PRELUDE: .github/scripts/setup_env.bash
BUILD_ENV: build_binary
ENFORCE_NVIDIA_GPU: 1
strategy:
fail-fast: false
matrix:
host-machine: [
{ arch: x86, instance: "linux.g5.4xlarge.nvidia.gpu" },
]
python-version: [ "3.8", "3.9", "3.10", "3.11" ]
cuda-version: [ "11.8.0", "12.1.1" ]
# Specify exactly ONE CUDA version for artifact publish
cuda-version-publish: [ "11.8.0" ]
needs: build_artifact
steps:
- name: Checkout the Repository
uses: actions/checkout@v3
with:
submodules: true
- name: Download Wheel Artifact from GHA
uses: actions/download-artifact@v3
with:
name: fbgemm_gpu_release_cuda_${{ matrix.host-machine.arch }}_${{ matrix.python-version }}_cu${{ matrix.cuda-version }}.whl
- name: Install NVIDIA Drivers and NVIDIA-Docker Runtime
uses: pytorch/test-infra/.github/actions/setup-nvidia@main
- name: Display System Info
run: . $PRELUDE; print_system_info; print_ec2_info
- name: Display GPU Info
run: . $PRELUDE; print_gpu_info
- 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 CUDA
run: . $PRELUDE; install_cuda $BUILD_ENV ${{ matrix.cuda-version }}
- name: Install PyTorch Test
run: . $PRELUDE; install_pytorch_pip $BUILD_ENV test cuda ${{ matrix.cuda-version }}
- name: Prepare FBGEMM_GPU Build
run: . $PRELUDE; cd fbgemm_gpu; prepare_fbgemm_gpu_build $BUILD_ENV
- name: Install FBGEMM_GPU
run: . $PRELUDE; install_fbgemm_gpu_package $BUILD_ENV *.whl
- name: Test with PyTest
timeout-minutes: 10
run: . $PRELUDE; cd fbgemm_gpu/test; run_fbgemm_gpu_tests $BUILD_ENV
- name: Push FBGEMM_GPU Binary to PYPI
if: ${{ github.event_name != 'pull_request' && github.event_name != 'push' && matrix.cuda-version == matrix.cuda-version-publish }}
env:
PYPI_TOKEN: ${{ secrets.PYPI_TOKEN }}
run: . $PRELUDE; publish_to_pypi $BUILD_ENV fbgemm_gpu-*.whl "$PYPI_TOKEN"