[aarch64] acl build script updates for multi isa build with armv8-a base architecture #1370
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Hi @malfet , please review and let me know. I have added one patch for ACL scons build script to this repo, please let me know if there is a better place to maintain it than here. |
This was referenced Mar 29, 2023
Closed
|
Here is the ACL github issue to check if this can be made default in the ACL build system for the future ACL releases. |
snadampal
force-pushed
the
pt2.0_acl_armv8.0
branch
from
583cb7f
to
213d56e
Compare
|
Hi @malfet , updated the PR to modify the build files with "sed" directly, instead of the explicit patch. Please let me know if there is any feedback. |
malfet
approved these changes
Apr 14, 2023
|
Thank you very much for the fix. Merged |
|
great! thanks for merging it @malfet . |
jithunnair-amd
added a commit
to ROCm/builder
that referenced
this pull request
Aug 7, 2023
* Add 12.1 workflow for docker image build (pytorch#1367) * add 12.1 workflow for docker image build * add github workflow * update cuDNN to 8.8.1 and location for archive * Do not use ftp (pytorch#1369) * Do not use ftp `s#ftp://#https://#` * Remove no-longer relevant comment * add magma build for CUDA12.1 (pytorch#1368) * add magma build for CUDA12.1 * copy and fix CMake.patch; drop sm_37 for CUDA 12.1 * remove CUDA 11.6 builds (pytorch#1366) * remove CUDA 11.6 builds * remove more 11.6 builds * build libtorch and manywheel for 12.1 (pytorch#1373) * enable nightly CUDA 12.1 builds (pytorch#1374) * enable nightly CUDA 12.1 builds * fix version typo * Fix typo (bracket) in DEPS_LIST setting (pytorch#1377) * Remove special cases for Python 3.11 (pytorch#1381) * Remove special case for Python 3.11 * Remove install torch script * Windows CUDA 12.1 changes (pytorch#1376) * Windows CUDA 12.1 changes * add CUDA version checks for Windows MAGMA builds * use magma branch without fermi arch * fix check for 12.1 (pytorch#1383) * Update CUDA_UPGRADE_GUIDE.MD (pytorch#1384) * add pytorch-cuda constraints for CUDA 12.1 (pytorch#1385) * Fix `cuda-pytorch/meta.yaml` And add `12.1` to the matrix Test plan: `conda build . -c nvidia` and observe https://anaconda.org/malfet/pytorch-cuda/files * [aarch64] acl build script updates for multi isa build with armv8-a base architecture (pytorch#1370) * Compile with C++17 standard in check_binary.sh To unblock pytorch/pytorch#98209 * Use robocopy fix 256char limit (pytorch#1386) * add constraints for pytorch-cuda (pytorch#1391) * [Conda] Update MacOS target to 11.0 As 10.9 was release a decade ago and for that reason yet not supported C++17 standard. Similar to pytorch/pytorch#99857 * Update MACOSX_DEPLOYMENT_TARGET to 10.13 To fix builds, though we should really target 11.0 at the very least * add nvjitlink for Windows builds for CUDA 12.1 (pytorch#1393) * pytorch-cuda: Added `nvjitlink` as dependency * Add `fsspec` to list of packages Added by pytorch/pytorch#99768 * Add ffmpeg build to audio aarch64 (pytorch#1396) * [S3] Add all `cu117.with.pypi` deps to nightly CUDA index * Fix nvjitlink inclusion in 12.1 wheels (pytorch#1397) * Fix nvjitlink inclusion in 12.1 wheels * Fix typo * update winserver driver (pytorch#1388) * Add pyyaml as PyTorch runtime dep (pytorch#1394) Companion PR to pytorch/pytorch#100166 * Fix typo * Temp: Comment out VS2019 installation As it should be part of the AMI * Use VS2022 for libtorch windows tests * Revert "Use VS2022 for libtorch windows tests" This reverts commit 2922d7d. * Revert "Temp: Comment out VS2019 installation" This reverts commit e919e17. * Do not cleanup MSVC and CUDA on Windows non-ephemeral runners (pytorch#1398) * Update CUDA_UPGRADE_GUIDE.MD (pytorch#1399) Update Related PR's * Attempt to fix infinite copy into existing folder Looks like robocopy is confused what to do about symlinks * Handle symlink when using robocopy (pytorch#1400) * Use /xjd instead of /sl when dealing with symlink on Windows https://superuser.com/questions/1642106/how-does-robocopy-handle-file-system-links-symbolic-links-hard-links-and-junct * Skip uninstalling other MSVC versions if they are found (pytorch#1402) * Update driver for cuda 12.1 (pytorch#1403) * Don't prepend system-wide PATH when installing Python for binary smoke test (pytorch#1404) The PATH has already been set and restored manually in the script * Pin MSVC version to 2019 (pytorch#1405) * Pin numpy for windows builds (pytorch#1406) * Pin numpy for windows builds * bump mkl version, remove conda-forge * Change python 3.9 mkl version * Use pinned mkl * Upgrade nightly wheels to rocm5.5 (pytorch#1407) * Add MIOpen db files to wheel * Update magma commits for various branches to include header path updates * Add ROCm5.5 support with Navi31-tuned MIOpen branch * Upgrade nightly wheels to rocm5.5 * Update build_docker.sh for gfx1100 * Update build_docker.sh for gfx1100 --------- Co-authored-by: Jithun Nair <jithun.nair@amd.com> Co-authored-by: Jithun Nair <37884920+jithunnair-amd@users.noreply.github.com> * Update CUDA_UPGRADE_GUIDE.MD with CI update instructions * Pin numpy to 1.21.3 for python 3.10 (pytorch#1409) * Update cuda matrix CUDA_UPGRADE_GUIDE.MD * Pin delocate (pytorch#1411) * Fix pytorch#1410. (pytorch#1412) The libdrm replacement for ROCm images will first search the typical location for the amdgpu.ids file. If failed, it will search as it did before in the python install location. * One-step ROCm manywheel/libtorch docker build (pytorch#1418) * Use rocm/dev-centos-7:<rocm_version>-complete image with full ROCm install * Remove ROCm install step and reinstate magma and MIOpen build steps * Install full package for MIOpen, including headers and db files Retained some of the disk-cleanup-related code from pytorch@cb0912c * Use rocm/dev-ubuntu-20.04:<rocm_version>-complete image with full ROCm install * Remove ROCm install and reinstate magma build from source * Use --offload-arch instead of --amdgpu-target to silence warnings * Use beefier runner instance for ROCm docker builds * Typo * Simplify ROCm targets * Update wheel build scripts for ROCm5.5 (pytorch#1413) * Fix lib search logic and lib list * Add libhipsolver.so dependency for upstream Pytorch (needed since PyTorch PR 97370) * Add MIOpen db files to share path only for ROCm5.5 and above * Add cu12 packages to allow list for download.pytorch.org (pytorch#1420) * Fix ROCm5.4.2 builds breakage (pytorch#1421) * Use MIOpen db logic for ROCm5.5 or later * Remove moved lines * Fix bash logic to add elements of array to existing array * Add nvidia-nvjitlink-cu12 to s3 manage (pytorch#1422) * Add nvidia-nvjitlink-cu12 to s3 manage (pytorch#1423) * Pin openssl for python 3.9 conda package (pytorch#1424) Fixes pytorch/pytorch#103436 Test PR: pytorch/pytorch#103437 * Update CUDA_UPGRADE_GUIDE.MD * Pin openssl for py3.8 (pytorch#1425) * Add pytorch-triton to small wheel (pytorch#1426) * Add poetry and pypi tests (pytorch#1419) * smoke test poetry Add a little more tests test Test poetry test Test poetry on python 3.10 Add more poetry tests Test en us test test Try verboose testing testing try quiet install Code refactooring test move linux pipy validation to workflow test test Fix path try test pipy More torch installations test testing test test test new fix install 2 try poetry nightly test nightly test test Test poetry validation test test_new test * Put back executing this on pull * [Manywheel] Add Python-3.12.0b2 (pytorch#1427) To enable initial experiments with PyTorch builds * smoke test poetry (pytorch#1428) Add a little more tests test Test poetry test Test poetry on python 3.10 Add more poetry tests Test en us test test Try verboose testing testing try quiet install Code refactooring test move linux pipy validation to workflow test test Fix path try test pipy More torch installations test testing test test test new fix install 2 try poetry nightly test nightly test test Test poetry validation test test_new test Put back executing this on pull Print matrix variable test Fix conditional for pypi poetry tests add quptes Add nightly as supplemental requirement Make sure we clone module only for first time Fix python test validate binaries Add repo existance checks test Disable runtime error before final validation fix typo fix cwd * smoke test poetry (pytorch#1429) * s/master/main/ * Update aarch64 Scripts for CI worflow (pytorch#1431) * Run release validation testing once a day, run nightly a little later (pytorch#1434) * Remove CUDA 11.7 builds (pytorch#1408) * remove CUDA 11.7 builds * remove CUDA 11.7 from MAGMA builds * add pytorch-cuda back for 11.7 * add 11.7 back to pytorch-cuda * Add safe directory to all dockerfiles (pytorch#1435) * update cuDNN to 8.9.2.26 for CUDA 12.1 (pytorch#1436) * update NCCL to 2.18.1 (pytorch#1437) Co-authored-by: Nikita Shulga <nshulga@meta.com> * Update CI for aarch64 (pytorch#1438) * Fix wheel macos arm64 validations (pytorch#1441) * Fix wheel validations * Try using upgrade flag instead * try uninstall * test * Try using python3 * use python3 vs python for validation * Fix windows vs other os python execution * Uninstall fix * Remove previous installations on macos-arm64 before smoke testing (pytorch#1444) More arm64 changes test run under environment sleep 15min allow investigate add sleep test test Test test test Arm64 use python fix test testing test tests testing test test * Update build_docker.sh (pytorch#1446) Use [`nvidia/cuda:11.4.3-devel-centos7`](https://hub.docker.com/layers/nvidia/cuda/11.4.3-devel-centos7/images/sha256-e2201a4954dfd65958a6f5272cd80b968902789ff73f26151306907680356db8?context=explore) because `nvidia/cuda:10.2-devel-centos7` was deleted in accordance with [Nvidia's Container Support Policy](https://gitlab.com/nvidia/container-images/cuda/blob/master/doc/support-policy.md): > After a period of Six Months time, the EOL tags WILL BE DELETED from Docker Hub and Nvidia GPU Cloud (NGC). This deletion ensures unsupported tags (and image layers) are not left lying around for customers to continue using after they have long been abandoned. Also delete redundant DEVTOOLSET=7 clause * Fix aarch64 nightly (pytorch#1449) * Update Docker base images for conda and libtorch (pytorch#1448) Followup after pytorch#1446 CUDA-10.2 and moreover CUDA-9.2 docker images are gone per [Nvidia's Container Support Policy](https://gitlab.com/nvidia/container-images/cuda/blob/master/doc/support-policy.md): > After a period of Six Months time, the EOL tags WILL BE DELETED from Docker Hub and Nvidia GPU Cloud (NGC). This deletion ensures unsupported tags (and image layers) are not left lying around for customers to continue using after they have long been abandoned. Also, as all our Docker script install CUDA toolkit anyway, what's the point of using `nvidia/cuda` images at all instead of `centos:7`/`ubuntu:18.04` that former are based on, according to https://gitlab.com/nvidia/container-images/cuda/-/blob/master/dist/11.4.3/centos7/base/Dockerfile Explicitly install `g++` to `libtorch/Docker` base image, as it's needed by `patchelf` Please note, that `libtorch/Docker` can not be completed without buildkit, as `rocm` step depends on `python3` which is not available in `cpu` image * Fix magma installation inside docker container (pytorch#1447) Not sure, what weird version of `wget` is getting installed, but attempt to download https://anaconda.org/pytorch/magma-cuda121/2.6.1/download/linux-64/magma-cuda121-2.6.1-1.tar.bz2 fails with: ``` --2023-07-06 03:18:38-- https://anaconda.org/pytorch/magma-cuda121/2.6.1/download/linux-64/magma-cuda121-2.6.1-1.tar.bz2 Resolving anaconda.org (anaconda.org)... 104.17.93.24, 104.17.92.24, 2606:4700::6811:5d18, ... Connecting to anaconda.org (anaconda.org)|104.17.93.24|:443... connected. ERROR: cannot verify anaconda.org's certificate, issued by ‘/C=US/O=Let's Encrypt/CN=E1’: Issued certificate has expired. To connect to anaconda.org insecurely, use `--no-check-certificate'. ``` Also, switch from NVIDIA container to a stock `centos:7` one, to make containers slimmer and fit on standard GitHub Actions runners. * [Manywheel] Add `/usr/local/cuda` symlink And add `nvcc` to path Regression introduced by pytorch#1447 when NVIDIA image was dropped in favor of base `centos` image * Do not build PyTorch with LLVM (pytorch#1445) As NNC is dead, and llvm dependency has not been updated in last 4 years First step towards fixing pytorch/pytorch#103756 * Remove `DESIRED_CUDA` logic from `check_binary.sh` As [`pytorch/manylinux-builder`](https://hub.docker.com/r/pytorch/manylinux-builder) containers has only one version of CUDA, there is no need to select any Nor setup `LD_LIBRARY_PATH` as it does not match the setup users might have on their system (but keep it for libtorch tests) Should fix crash due to different minor version of cudnn installed in docker container and specified as dependency to a small wheel package, seen here https://github.com/pytorch/pytorch/actions/runs/5478547018/jobs/9980463690 * Revert "Remove `DESIRED_CUDA` logic from `check_binary.sh`" This reverts commit ed9a2ae. * Update CUDA_UPGRADE_GUIDE.MD to add small wheel update Cudnn step * Set CUDA_VERSION in conda Docker environment * Rebuild docker images on release builds (pytorch#1451) * Rebuild docker images on release * Include with-push * Create `/usr/local/cuda` in libtorch builds I.e. applying the same changes as in pytorch@4a7ed14 to libtorch docker builds * Revert "Rebuild docker images on release builds (pytorch#1451)" (pytorch#1452) This reverts commit 2ba03df as it essentially broke all the builds on trunk (fix is coming) * Reland "Rebuild docker images on release builds" This is a reland of pytorch#1451 with an important to branches filter: entries in multi-line array definition should start with `-` otherwise it were attempting to match branch name `main\nrelease/*` I.e. just copy-n-paste example from https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#using-filters Test plan: actionlint .github/workflows/build-manywheel-images.yml Original PR description: Rebuild docker images on release builds. It should also tag images for release here: https://github.com/pytorch/builder/blob/3fc310ac21c9ede8d0ce13ec71096820a41eb9f4/conda/build_docker.sh#L58-L60 This is first step in pinning docker images for release. * Let's try to force the path this way * Remove `DESIRED_CUDA` logic from `check_binary.sh` As [`pytorch/manylinux-builder`](https://hub.docker.com/r/pytorch/manylinux-builder) containers has only one version of CUDA, there is no need to select any Nor setup `LD_LIBRARY_PATH` as it does not match the setup users might have on their system (but keep it for libtorch tests for now) Should fix crash due to different minor version of cudnn installed in docker container and specified as dependency to a small wheel package, seen here https://github.com/pytorch/pytorch/actions/runs/5478547018/jobs/9980463690 * Advance libgfortran version (pytorch#1453) * Update builder images to ROCm5.6 (pytorch#1443) * Update manywheel and libtorch images to rocm5.6 * Add MIOpen branch for ROCm5.6 * Pin miniconda install to py310_23.5.2 for macos and windows (pytorch#1460) * Cleanup unused builder files (pytorch#1459) * Remove unused builder files (pytorch#1461) * Add support for ROCm5.6 for nightly wheels (pytorch#1442) * Add msccl-algorithms directory to PyTorch wheel * Bundle msccl-algorithms into wheel * Use correct src path for msccl-algorithms (cherry picked from commit 95b5af3) * Add hipblaslt dependency for ROCm5.6 onwards * Update build_all_docker.sh to ROCm5.6 * [aarch64][build] Aarch64 lapack fix and ARMCL version update (pytorch#1462) * Fix lapack missing and armcl update * update ARMCL version * Remove unused parameter to limit-win-builds from validation workflows (pytorch#1464) * Run git update-index --chmod=+x on aarch64_ci_build.sh (pytorch#1466) * Fix erroneous logic that was skipping msccl files even for ROCm5.6; update msccl path for ROCm5.7 (cherry picked from commit 36c10cc) --------- Co-authored-by: ptrblck <ptrblck@users.noreply.github.com> Co-authored-by: Nikita Shulga <nshulga@fb.com> Co-authored-by: Andrey Talman <atalman@fb.com> Co-authored-by: Nikita Shulga <nshulga@meta.com> Co-authored-by: snadampal <87143774+snadampal@users.noreply.github.com> Co-authored-by: Richard Zou <zou3519@users.noreply.github.com> Co-authored-by: Huy Do <huydhn@gmail.com> Co-authored-by: Bo Li <110066325+BLOrange-AMD@users.noreply.github.com> Co-authored-by: Jeff Daily <jeff.daily@amd.com> Co-authored-by: Mike Schneider <104035434+xncqr@users.noreply.github.com> Co-authored-by: Eli Uriegas <1700823+seemethere@users.noreply.github.com>
malfet
pushed a commit
that referenced
this pull request
Sep 28, 2023
…#1550) Looking at this PR: #1370 this line: https://github.com/pytorch/builder/pull/1370/files#diff-54480d0a69ca27f54fb0736a9762caa8b03bd4736dcd77190d99ec3033c9bd2fR229 That fixed the issue: pytorch/pytorch#97226 One of the changes is to set ``` arch=armv8a ``` We are experiencing the same issue now: pytorch/pytorch#109312 Hence this fix.
atalman
added a commit
to atalman/builder
that referenced
this pull request
Oct 2, 2023
…pytorch#1550) Looking at this PR: pytorch#1370 this line: https://github.com/pytorch/builder/pull/1370/files#diff-54480d0a69ca27f54fb0736a9762caa8b03bd4736dcd77190d99ec3033c9bd2fR229 That fixed the issue: pytorch/pytorch#97226 One of the changes is to set ``` arch=armv8a ``` We are experiencing the same issue now: pytorch/pytorch#109312 Hence this fix.
atalman
added a commit
that referenced
this pull request
Oct 2, 2023
…Pie (#1562) * [aarch64] set acl_build_flags arch=armv8a, remove editing build flags (#1550) Looking at this PR: #1370 this line: https://github.com/pytorch/builder/pull/1370/files#diff-54480d0a69ca27f54fb0736a9762caa8b03bd4736dcd77190d99ec3033c9bd2fR229 That fixed the issue: pytorch/pytorch#97226 One of the changes is to set ``` arch=armv8a ``` We are experiencing the same issue now: pytorch/pytorch#109312 Hence this fix. * [aarch64] patch mkl-dnn to use 'march=armv8-a' as the default build (#1554) * [aarch64] patch pytorch 2.1 for mkl-dnn fix (#1555) * patch ci script with mkldnn fix (#1556) * Fix path issue when building aarch64 wheels (#1560) --------- Co-authored-by: snadampal <87143774+snadampal@users.noreply.github.com>
jithunnair-amd
added a commit
to ROCm/builder
that referenced
this pull request
Feb 22, 2024
* Set FORCE_RPATH for ROCm (pytorch#1468) * Decouple aarch64 ci setup and build (pytorch#1470) * Run git update-index --chmod=+x aarch64_ci_setup.sh (pytorch#1471) * [aarch64][CICD]Add aarch64 docker image build. (pytorch#1472) * Add aarch64 docker image build * removing ulimit for PT workflow * set aarch64 worker for docker build * Fix `install_conda.sh` By pinning conda version to 23.5.2 as latest(23.7.2 at this time) does not have a compatible version of `git` packages Fixes pytorch#1473 * Remove explicit `conda install cmake` As it's already done as part of `common/install_conda.sh` script * update to CUDA 12.1U1 (pytorch#1476) Should fix pytorch/pytorch#94772 in wheel builds * Use conda version 23.5.2 for conda pytorch build (pytorch#1477) * Use py311 miniconda install (pytorch#1479) * Windows conda build fix (pytorch#1480) * Revert "Use py311 miniconda install (pytorch#1479)" (pytorch#1481) This reverts commit 5585c05. * Remove c/cb folder on windows (pytorch#1482) * Add numpy install - fix windows smoke tests (pytorch#1483) * Add numpy install * Add numpy install * Add hostedtoolcache purge step (pytorch#1484) * Add hostedtoolcache purge step * Change step name * Update CUDA_UPGRADE_GUIDE.MD * update CUDA to 12.1U1 for Windows (pytorch#1485) * Small improvements in build pytorch script (pytorch#1486) * Undo using conda activate (pytorch#1487) * Update meta.yaml (pytorch#1389) * Add pytorch-triton-rocm as an install dependency for ROCm (pytorch#1463) * Add pytorch-triton-rocm as an install dependency for ROCm * Update build_rocm.sh * Add aarch64 to validation framework (pytorch#1474) * Add aarch64 to validation framework (pytorch#1489) * Add aarch64 to validation framework (pytorch#1490) * Add aarch64 to validation framework * Add aarch64 to validation framework * Add aarch64 to validation framework (pytorch#1491) * Add aarch64 to validation framework * Add aarch64 to validation framework * Add aarch64 to validation framework * Temporary disable poetry test (pytorch#1492) * Add torchonly option to validation workflows (pytorch#1494) * Add torchonly option to validation workflows * fix typo * Remove pipy validation temporarily (pytorch#1495) * Remove pipy validation temporarily (pytorch#1496) * Add no-sudo to linux-aarch64 tests (pytorch#1499) * Pass container image to aarch64 test jobs (pytorch#1500) * Add setup aarch64 builds for aarch64 testing (pytorch#1501) * Fix DESIRED_PYTHON setting for aarch64 validations (pytorch#1502) * Use extra-index-url for aarch64 builds (pytorch#1503) * Pypi validation enable (pytorch#1504) * Validation pypi torchonly (pytorch#1505) * Pipy validation workflow (pytorch#1506) * Pipy validation workflow (pytorch#1507) * Pipy validation workflow (pytorch#1508) * Pipy validation workflow (pytorch#1509) * Validate poetry workflow (pytorch#1511) * Validate poetry workflow (pytorch#1512) * Remove linux-aarch64 installation workaround (pytorch#1513) * Temporary change test aarch64 builds (pytorch#1514) * Remove torchonly restictions from aarch64 builds (pytorch#1517) * Fix aarch64 nightly/release version override (pytorch#1518) * Aarch64 fix overrdie passing from CI to build * Aarch64 fix overrdie passing from CI to build * Aarch64 fix overrdie passing from CI to build * Revert "Temporary change test aarch64 builds (pytorch#1514)" (pytorch#1521) This reverts commit 1e281be. * Changes related to OVERRIDE_PACKAGE_VERSION in aarch64 builds (pytorch#1520) (pytorch#1523) * Torchmetrics in S3 Index (pytorch#1522) We will need the stable torchmetrics wheel in the S3 index, since torchrec depends on it. This is similar to how pytorch depends on numpy, etc. and these binaries need to be hosted in our index when uses try to pip install from download.pytorch.org. * [aarch64] update ACL version to v23.05.1 and OpenBLAS to v0.3.20 (pytorch#1488) * Changed runner for linux arm64 (pytorch#1525) * Add torch-tensorrt to S3 PyPI Index (pytorch#1529) As pytorch/tensorrt moves off of CCI onto Nova, we must to host their nightlies on our S3 index. This change allows the indexing to occur correctly for this package. * Enable torch compile for python 3.11 smoke tests (pytorch#1534) * Enable torch compile for python 3.11 smoke tests * Make sure release is covered * Fix typo * add jinja2 (pytorch#1536) * Remove restriction on 3.11 (pytorch#1537) * Revert "add jinja2 (pytorch#1536)" (pytorch#1538) This reverts commit 224a4c5. * S3 Management Job Outside Docker (pytorch#1531) * S3 Management Job Outside Docker * job name * remove failfast * no matrix * inherit secrets * spacing? * random nits * add back secrets * add back matrix * export env vars correctlty * Update update-s3-html.yml * Add fbgemm-gpu to S3 Index (pytorch#1539) * Update builder images to ROCm5.7 (pytorch#1541) * Update docker build images for rocm5.7 * Fix erroneous logic that was skipping msccl files even for ROCm5.6; update msccl path for ROCm5.7 (cherry picked from commit 36c10cc) * missing bzip2 package install for miopen * Revert "missing bzip2 package install for miopen" This reverts commit 8ef5fc9. * ROCm 5.7 MIOpen does not need any patches, do not build from source --------- Co-authored-by: Jeff Daily <jeff.daily@amd.com> * Update docker build convenience scripts to ROCm5.7 (pytorch#1543) * Do not uninstall MIOpen if skipping build-from-source (pytorch#1544) * Install nvtx3 on Windows (pytorch#1547) * Provide file hashes in the URLs to avoid unnecessary file downloads (bandwidth saver) (pytorch#1433) Supply sha256 query parameters using boto3 to avoid hundreds of extra Gigabytes of downloads each day during pipenv and poetry resolution lock cycles. Fixes point 1 in pytorch/pytorch#76557 Fixes pytorch#1347 * Workaround for older files * Bugfixes introduced by pytorch#1433 Replace `obj` with `obj.key` in few places Dismantle pyramid of doom while iterating over objects Test plan: Run `python manage.py whl/test --generate-pep503` * [S3_management] Update boto3 to 1.28.53 * [manage_s3] Download objects metadata concurrently Using `concurrent.futures.ThreadPoolExecutor` This speeds up rebuilding `whl/test` index from 300 sec to 90 sec on my laptop * Make smoke-test runnable without envvars * [aarch64] set acl_build_flags arch=armv8a, remove editing build flags (pytorch#1550) Looking at this PR: pytorch#1370 this line: https://github.com/pytorch/builder/pull/1370/files#diff-54480d0a69ca27f54fb0736a9762caa8b03bd4736dcd77190d99ec3033c9bd2fR229 That fixed the issue: pytorch/pytorch#97226 One of the changes is to set ``` arch=armv8a ``` We are experiencing the same issue now: pytorch/pytorch#109312 Hence this fix. * [BE] Fix all flake8 violations in `smoke_test.py` (pytorch#1553) Namely: - `if(x):` -> `if x:` - `"dev\d+"` -> `"dev\\d+"` - Keep 2 newlines between functions - Add `assert foo is not None` to suppress "variable assigned but not used" warning * [aarch64] patch mkl-dnn to use 'march=armv8-a' as the default build (pytorch#1554) * [aarch64] patch pytorch 2.1 for mkl-dnn fix (pytorch#1555) * patch ci script with mkldnn fix (pytorch#1556) * [BE] Add lint workflow (pytorch#1557) And format `smoke_test.py` with `ruff` Invoke/confgure `ruff` using `lintrunner` Copy lint runner adapters from https://github.com/pytorch/pytorch/tree/main/tools/linter/adapters * [BE] Add `s3_management` to the linted folders (pytorch#1558) Add `PERF401` to list of ignored suggestions, fix the rest. * Fix path issue when building aarch64 wheels (pytorch#1560) * Fix linalg smoke tests (pytorch#1563) * Towards enabling M1 wheel builds Do not try to install MKL on Apple Silicon * And only install llvm-9 on x86 systems * Do not build tests when building natively on M1 * And fix Python-3.8 native compilation on M1 There are no numpy=3.17 for M1 * Release 2.1 update promotion scripts (pytorch#1564) * [BE] Small code cleanup Fold multiple inidices and single index generation into one loop As loop body is the same anyway... * S3_management: Add option to compute sha256 That will be used later to generate sha256 indexes in PEP503 * Remove debug print * [S3_management] Minor improvements - Refactor `fetch_obj_names` into class method - Make sure that object remains public when ACL is computed - Add `has_public_read` and `grant_public_read` class methods * s3_management: compute checksum in cloud I.e. file never gets downloaded on the client, which is a nice thing * [S3Management] Add `undelete_prefix` method That can be used to recover object in a versioned bucket * Validate poetry for release (pytorch#1567) * Validate poetry for release * test * test * fixtypo * Use released version of 3.12 (pytorch#1568) As it was released on Oct 6 2023: https://www.python.org/downloads/release/python-3120/ * Move manywheel builds to `linux.12xlarge.ephemeral` (pytorch#1569) Should be faster(<20 min vs 40+ min) and as secure as using GH ones * Add cuSparseLt-0.5.0 to manywheel images * Use `linux.12xlarge.ephemeral` for conda docker builds (pytorch#1570) As `ubuntu.20.04` often OOM/failed to fetch data from RHEL repo * Revert "Add cuSparseLt-0.5.0 to manywheel images" This reverts commit 00841b6 as cuSparseLT is not compatible with CentOS 7 * Move libtorch docker builder to `linux.12xlarge.ephemeral` (pytorch#1571) As running it on `ubutu22.04` often results in flay infra failures/running out of disk space, for example, from https://github.com/pytorch/builder/actions/runs/6484948230/job/17609933012 ``` cat: write error: No space left on device ``` * Add cuSparseLt-0.4.0 to manywheel images But set USE_CUSPARSELT to 0 by default * Add xformers to the list of indexable packages * Build wheels with cuSparseLt Build libtorch without cuSparseLt so far Factor out `DEPS_LIST` to top level and add cuSparseLt of `USE_CUSPARSELT` is set to 1 Tested in pytorch/pytorch#111245 * Do not build conda with CuSparseLT * Add ROCM_PATH env var to Dockerfile for ROCm5.7 issue with finding HIP (pytorch#1572) * [aarch64_wheel] Minor typing improvements * [aarch64_wheel] Flake8 fix * [aarch64_wheel] Cosmetic changes * [aarch64_wheel] Fix readdir crash Probably fixes pytorch/pytorch#111695 * [S3_management] generate libtorch index.html * [CI] Update ruff to 0.1.1 To keep it in sync with pytorch * Get rid of http://repo.okay.com.mx (pytorch#1575) * [S3_management] Print time it takes to fetch index * [S3_manage] Handle invalid versions * [S3_management] Fix Version on error And fix flake8 lint violation * [S3_Management] Refactor `from_S3` Move `fetch_metadata` into its own method, which could be called later on Make S3Object non-frozen and introduce implicit __hash__ method * [S3_Management] Filter nighly before `fetch_metadata` This reduces time to call `from_S3Index` from 600 to 80 sec * Add option to build -arm64- libtorch binaries * [Docker] Remove trailing whitespace And cause docker rebuild, to overwrite docker build from release/2.1 branch artifacts * [MacOS] Small changes to libtorch naming Intel x86 libtorch builds will have `x86_64` suffix and Apple Silicon ones will have `arm64` ones, but latest will point to Intel ones for now. * Update libtorch/Dockerfile to use Ubuntu-20.04 (pytorch#1578) As 18.04 EOLed * Conda builds should respect `MAX_JOBS` May be this help with OOMs * [S3_management] Fix subpackage urls Make them `lower()` * Advance versions for release 2.1.1 (pytorch#1583) * [aarch64] Release pypi prep script change for aarch64 builds (pytorch#1585) * Changes needed for core enablement of 3.12 binary wheels (pytorch#1586) * Fix aarch64 build on 3.8 (pytorch#1593) * Add some more validation checks for torch.linalg.eigh and torch.compile (pytorch#1580) * Add some more validation checks for torch.linalg.eigh and torch.compile * Update test * Also update smoke_test.py * Fix lint * Revert "Add some more validation checks for torch.linalg.eigh and torch.compile (pytorch#1580)" (pytorch#1594) This reverts commit 4c7fa06. * Release validations using release version matrix (pytorch#1611) * Release pypi prep change (pytorch#1587) * [aarch64] Release pypi prep script change for aarch64 builds * Release versions for testing Testing calling version (pytorch#1588) Upstream/release validations (pytorch#1589) * Testing calling version * add release matrix Upstream/release validations (pytorch#1590) * Testing calling version * add release matrix * test test (pytorch#1591) test (pytorch#1592) Release v1 (pytorch#1595) * test * test Release v1 (pytorch#1596) * test * test * test test (pytorch#1597) Test versions validations (pytorch#1598) * test * basedir Test versions validations (pytorch#1599) * test * basedir * test test (pytorch#1600) * test * test Add release versions everywhere (pytorch#1601) * test * test * test * test test (pytorch#1602) Test version validations (pytorch#1603) * test * test Test version validations (pytorch#1604) * test * test * test tests (pytorch#1605) More tests nov16 (pytorch#1606) * tests * test More tests nov16 (pytorch#1607) * tests * test * test More tests nov16 (pytorch#1608) * tests * test * test * test More tests nov16 (pytorch#1609) * tests * test * test * test * test * fix_lint * fix: typo (pytorch#1581) * desired_cuda -> DESIRED_CUDA (pytorch#1612) * desired_cuda -> DESIRED_CUDA Found with shellcheck * Update manywheel/build_cuda.sh Co-authored-by: Nikita Shulga <2453524+malfet@users.noreply.github.com> --------- Co-authored-by: Nikita Shulga <2453524+malfet@users.noreply.github.com> * [BE] Cleanup build unused code (pytorch#1613) 1. Upload Scripts are not used anymore. We use Github Action upload workflows 2. M1 Builds are now automated 3. build_all.bat run git grep in pytorch and builder - No result * Changes to pypi release promotion scripts introduced for 2.1.0 and 2.1.1 (pytorch#1614) * Changes topypi release promotion scripts introduced during 2.1.1 * typo * Pin miniconda version for Windows To Miniconda3-py311_23.9.0-0-Windows-x86_64.exe * Fix poetry and pypi validations when version is specified (pytorch#1622) * test (pytorch#1617) Fix validations (pytorch#1618) * test * poetry_fix * test Fix validations (pytorch#1619) * test * poetry_fix * test * test * restrict * Validate pypi build only for release (pytorch#1623) * Validate pypi build only for release (pytorch#1624) * [Manywheel] Do not hardcode triton version * [Manywheel][BE] Dedup Triton requirement spec * [Manywheel] Restrict `pytorch-triton` to x86-64 Linux Partially addresses pytorch/pytorch#114042 * Tweak py312 conda requirements * Build PyTorch without TLS for 3.12 Because GLOO still expect OpenSSL-1, but 3.12 is build with OpenSSL-3 * [conda] Skip sympy for 3.12 As at the moment it is only available for Windows %) * [conda] Do not depend on triton for 3.12 yet * Tweak mkl requirements for win+py312 * Add aarch64 conda env lib to LD_LIBRARY_PATH (pytorch#1628) After the change on pytorch#1586, nightly aarch64 wheel fails to find `libopenblas.so` which is now installed under `/opt/conda/envs/aarch64_env/lib/` instead of the base conda `/opt/conda/lib`. Using CPU nightly wheels on aarch64 from Nov 16 then ends up with the error as described in pytorch/pytorch#114862: `Calling torch.geqrf on a CPU tensor requires compiling PyTorch with LAPACK. Please use PyTorch built with LAPACK support`. The error can be found on night build log https://github.com/pytorch/pytorch/actions/runs/6887666324/job/18735230109#step:15:4933 Fixes pytorch/pytorch#114862 I double check `2.1.[0-1]` and the current RC for 2.1.2, the issue is not there because pytorch#1586 only change builder main, thus impacting nightly. ### Testing Build nightly wheel manually on aarch64 runner and confirm that openblas is detected correctly: ``` -- Found a library with BLAS API (open). Full path: (/opt/conda/envs/aarch64_env/lib/libopenblas.so) ... -- USE_BLAS : 1 -- BLAS : open -- BLAS_HAS_SBGEMM : -- USE_LAPACK : 1 -- LAPACK : open ... ``` * Revert "[conda] Skip sympy for 3.12" This reverts commit 88457a1. As sympy has been updated to 1.12 and it now supports Python-3.12 * [aarch64] ACL, OpenBLAS and mkldnn updates for PyTorch 2.2 (pytorch#1627) Note# ~~This PR has a dependency on updating the oneDNN version to v3.3 (via ideep submodule to v3.3)~~ ideep submodule update is done, so, this PR can be merged anytime now. This PR is for: ACL - build with fixed format kernels OpenBLAS - upgrade the version to 0.3.25 numpy - upgrade version to 1.26.2 and mkldnn - cleanup the patches that are already upstreamed. * Validation scripts, install using version (pytorch#1633) * Test Windows static lib (pytorch#1465) Add support for testing Windows Cuda static lib * Pin windows intel-openmp to 2023.2.0 (pytorch#1635) (pytorch#1636) * Torch compile test for python 3.8-3.11 linux only (pytorch#1629) This should fix failure on with Python 3.12 validations: https://github.com/pytorch/builder/actions/runs/7064433251/job/19232483984#step:11:4859 * [aarch64] cleanup mkldnn patching (pytorch#1630) pytorch is moved to oneDNN v3.3.2 and some of the old patches are not applicable any more. * Add `aarch64_linux` to the list of linted files * Actually fix lint this type * Extend test_linalg from smoke_test.py To take device as an argument and run tests on both cpu and cuda * Run smoke_test_linalg during check_binary This is a regression test for pytorch/pytorch#114862 * Fix linalg testing * [BE] Add CI for check_binary.sh changes (pytorch#1637) Make sure latest nightly passes the testing for: - Linux Wheel CPU - Linux Wheel CUDA Tweak script a bit to work correctly with relative path to executable * Keep nightly 20231010 for ExecuTorch alpha 0.1 for now (pytorch#1642) * [Validations] do conda update before starting validations (pytorch#1643) * [Validations] Validate aarch64 if all is slected (pytorch#1644) * Fix validation workflow on aarch64 with conda 23.11.0 and GLIBC_2.25 (pytorch#1645) * Debug aarch64 clone * Debug * Fix validation workflow with conda 23.11.0 and GLIBC_2.25 * Gate the change on linux-aarch64 and keep the old LD_LIBRARY_PATH * Try to unset LD_LIBRARY_PATH in the workflow instead * Fix copy/paste typo * Do not hardcode triton version in builder code (pytorch#1646) * Do not hardcode triton version in builder code * Minor tweak to use pytorch_rootdir * [Lint] Prohibit tabs in shell scripts Fix current violations * Link conda packages with cusparselt Fixes pytorch/pytorch#115085 * aarch64: patch mkl-dnn for xbyak crashes due to /sys not accessible (pytorch#1648) There are platforms with /sys not mounted. skip handling HW caps for such platforms. cherry-pick of: oneapi-src/oneDNN#1773 This fixes the issue# pytorch/pytorch#115482 * Update builder images to ROCm6.0 (pytorch#1647) * Update ROCm versions for docker images * Don't build MIOpen from source for ROCm6.0 * Temporarily use magma fork with ROCm6.0 patch * Update ROCm versions for docker images * Add gfx942 * Update MIOpen repo * Magma PR 42 is merged, so use upstream repo master branch now * gfx942 target only fully supported for ROCm6.0 and above * Avoid finding out std::basic_string_view (pytorch#1528) As pytorch moving to C++17, the binary can contain both "std::basic_string_view" and "std::__cxx11::basic_string<", change the pattern to avoid finding out std::basic_string_view, causing false positives. * Add test ops validation for validation workflows (pytorch#1650) * Add test ops validation * include workflows * Add test ops validation for validation workflows (pytorch#1651) * Add test ops validation for validation workflows (pytorch#1652) * Add test ops validation for validation workflows (pytorch#1653) * Add test ops validation for validation workflows (pytorch#1654) * Add test ops validation for validation workflows (pytorch#1655) * [validations] Add missing required packages (pytorch#1656) * [validations] Perform test_ops only on CUDA binaries (pytorch#1657) * [validations] Adjust timeout for linux jobs (pytorch#1658) * [validations] Restrict testing for python 3.8-3.11 (pytorch#1659) * [validations] Fix use case if INCLUDE_TEST_OPS is not set (pytorch#1660) * Add unit tests and one line reproducers to detect bad pytorch cuda wheels (pytorch#1663) * Add one line reproducers and unit tests that would fail when bad wheels were generated by the compiler(s). nextafter reproducer thanks to @malfet! * cosmetic fixes * fix comments * Fix quotation issues when migrating from python file to one line format (pytorch#1664) Sorry, looks like the last line had an issue while porting it from multi-line python file to one-line. Side question: when does this file get used? Is it only used during release binary generation/testing? * Add nccl version print for cuda related smoke test (pytorch#1667) * Apply nccl test to linux only (pytorch#1669) * Build nccl after installing cuda (pytorch#1670) Fix: pytorch/pytorch#116977 Nccl 2.19.3 don't exist for cuda 11.8 and cuda 12.1. Refer to https://docs.nvidia.com/deeplearning/nccl/release-notes/rel_2-19-3.html#rel_2-19-3 CUDA 12.0, 12.2, 12.3 are supported. Hence we do manual build. Follow this build process: https://github.com/NVIDIA/nccl/tree/v2.19.3-1?tab=readme-ov-file#build We want nccl version be exactly the same as installed here: https://github.com/pytorch/pytorch/blob/main/.github/scripts/generate_binary_build_matrix.py#L45 * Update cusparselt to v0.5.2 (pytorch#1672) This PR adds in support for cuSPARSELt v0.5.2 and updates the cuda 12.1 build step to use it instead of 0.4.0 Also fixes a typo when deleting the cusparselt folder after installing. * Run test ops tests from outside of pytorch root folder (pytorch#1676) * Remove s3 update html job and scripts (pytorch#1677) * [BE] Remove unused nightly_defaults.bat (pytorch#1678) * [Conda] Mark `blas * mkl` as x86 only dependency * [Conda] Download arch appropriate Miniconda By using `$(uname -m)` as suffix, which is arm64 on Apple Silicon and x86 on Intel Macs * [Conda] Do not depend on llvmdev-9 on ARM As earliest available for the platform is llvmdev-11 * [Conda] Set correct developer dir for MacOS runners * [Conda] Add llvm-openmp dependency for ARM64 PyTorch for M1 is finally built with OpenMP, so it needs to depend on it * Use dynamic MKL on Windows (pytorch#1467) Use dynamic MKL on Windows and updated MKL to 2021.4.0 On conda python 3.12 use mkl 2023.1 * Add torchrec to promote s3 script (pytorch#1680) * Add torchrec to promote s3 script * Add torchrec version to release_version.sh * Revert "Dynamic MKL windows" (pytorch#1682) * Revert "Revert "Dynamic MKL windows"" (pytorch#1683) * Add numpy install to windows conda tests (pytorch#1684) * Windows conda test. Install numpy in conda testenv (pytorch#1685) * Add fbgemm to promote s3 script (pytorch#1681) * Release 2.2.0 pypi prep script modifications (pytorch#1686) * [Analytics] add pypi staging validations, remove circleci script (pytorch#1688) * [Analytics] Pypi validations. Add call to check-wheel-contents (pytorch#1689) * Modify Validate Nightly PyPI Wheel Binary Size to pick correct binary (pytorch#1690) * Fix test_ops scripts on release validation testing (pytorch#1691) * Add option to validate only from download.pytorch.org (pytorch#1692) * Exclude pipy and poetry tests when USE_ONLY_DL_PYTORCH_ORG is set (pytorch#1693) * [ROCm] add hipblaslt library files (pytorch#1695) With pytorch/pytorch#114329 merged, we need to include hipblaslt library files within the ROCm wheel. * Minor tweak to fbgemmgpu version to ignore RC suffix (pytorch#1694) * Remove custom PyTorch build dependency logic on 3.11 (pytorch#1697) * Remove custom PyTorch build dependency logic on 3.11 * Add a smoke test for openmp * Pin conda-build to 3.28.4 (pytorch#1698) * ci: aarch64 linux: fix torch performance issue with conda openblas package (pytorch#1696) changing the conda openblas package from pthread version to openmp version to match torch openmp runtime. The pthread version was conflicting with the openmp runtime and causing thread over-subscription and performance degradation. * Add triton version for nightly and release (pytorch#1703) * Bundle PTXAS into 11.8 wheel * Add tensorrt promo script, bump release version for 2.2.1 (pytorch#1706) * Pin Conda to 23.11.0 --------- Co-authored-by: Andrey Talman <atalman@fb.com> Co-authored-by: Mike Schneider <104035434+xncqr@users.noreply.github.com> Co-authored-by: Nikita Shulga <nshulga@meta.com> Co-authored-by: ptrblck <ptrblck@users.noreply.github.com> Co-authored-by: JYX <jyx21@mails.tsinghua.edu.cn> Co-authored-by: Omkar Salpekar <osalpekar@gmail.com> Co-authored-by: snadampal <87143774+snadampal@users.noreply.github.com> Co-authored-by: Danylo Baibak <baibak@meta.com> Co-authored-by: Supadchaya <138070207+spcyppt@users.noreply.github.com> Co-authored-by: Jeff Daily <jeff.daily@amd.com> Co-authored-by: cyy <cyyever@outlook.com> Co-authored-by: Matt Davis <matteius@gmail.com> Co-authored-by: Nikita Shulga <2453524+malfet@users.noreply.github.com> Co-authored-by: Huy Do <huydhn@gmail.com> Co-authored-by: albanD <desmaison.alban@gmail.com> Co-authored-by: Luo Bo <84075753+0x804d8000@users.noreply.github.com> Co-authored-by: Sergii Dymchenko <kit1980@gmail.com> Co-authored-by: Ionuț Manța <ionut@janeasystems.com> Co-authored-by: Wei Wang <143543872+nWEIdia@users.noreply.github.com> Co-authored-by: Jesse Cai <jessecai@fb.com> Co-authored-by: henrylhtsang <91030427+henrylhtsang@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR fixes the issue pytorch/pytorch#97226