Per the RFC and numerous conversations on Discord, this rebuilds the
torch-mlir CI and discontinues the infra and coupling to the binary
releases
(https://discourse.llvm.org/t/rfc-discontinuing-pytorch-1-binary-releases/76371).
I iterated on this to get latency back to about what it was with the old
(much larger and non-ephemeral) runners: About 4m - 4.5m for an
incremental change.
Behind the scenes changes:
* Uses a new runner pool operated by AMD. It is currently set to manual
scaling and has two runners (32-core, 64GiB RAM) while we get some
traction. We can either fiddle with some auto-scaling or use a schedule
to give it an increase during certain high traffic hours.
* Builds are now completely isolated and cannot have run-to-run
interference like we were getting before (i.e. lock file/permissions
stuff).
* The GHA runner is installed directly into a manylinux 2.28 container
with upgraded dev tools. This eliminates the need to do sub-invocations
of docker on Linux in order to run on the same OS that is used to build
wheels.
* While not using it now, this setup was cloned from another project
that posts the built artifacts to the job and fans out testing. Might be
useful here later.
* Uses a special git cache that lets us have ephemeral runners and still
check out the repo and deps (incl. llvm) in ~13s.
* Running in an Azure VM Scale Set.
In-repo changes:
* Disables (but does not yet delete):
* Old buildAndTest.yml jobs
* releaseSnapshotPackage.yml
* Adds a new `ci.yml` pipeline and scripts the steps in `build_tools/ci`
(by decomposing the existing `build_linux_packages.sh` for in-tree
builds and modularizing it a bit better).
* Test framework changes:
* Adds a `TORCH_MLIR_TEST_CONCURRENCY` env var that can be used to bound
the multiprocess concurrency. Ended up not using this in the final
version but is useful to have as a knob.
* Changes the default concurrency to `nproc * 0.8 + 1` vs `nproc * 1.1`.
We're running on systems with significantly less virtual memory and I
did a bit of fiddling to find a good tradeoff.
* Changed multiprocess mode to spawn instead of fork. Otherwise, I was
getting instability (as discussed on discord).
* Added MLIR configuration to disable multithreaded contexts globally
for the project. Constantly spawning `nproc * nproc` threads (more than
that actually) was OOM'ing.
* Added a test timeout of 5 minutes. If a multiprocess worker crashes,
the framework can get wedged indefinitely (and then will just be reaped
after multiple hours). We should fix this, but this at least keeps the
CI pool from wedging with stuck jobs.
Functional changes needing followup:
* No matter what I did, I couldn't get the LTC tests to work, and I'm
not 100% sure they were being run in the old setup as the scripts were a
bit twisty. I disabled them and left a comment.
* Dropped out-of-tree build variants. These were not providing much
signal and increase CI needs by 50%.
* Dropped MacOS and Windows builds. Now that we are "just a library" and
not building releases, there is less pressure to test these commit by
commit. Further, since we bump torch-mlir to known good commits on these
platforms, it has been a long time since either of these jobs have
provided much signal (and they take ~an hour+ to run). We can add them
back later post-submit if ever needed.
Prior to this, the concurrency rules for presubmits (which cancel eagerly) were being applied to main. The result was that landing a second patch would cancel the CI on the one prior.
This lifts the core of the jit_ir_importer and ltc out of the pt1
project, making them peers to it. As a side-effect of this layering, now
the "MLIR bits" (dialects, etc) are not commingled with the various
parts of the pt1 project, allowing pt1 and ltc to overlay cleanly onto a
more fundamental "just MLIR" Python core. Prior to this, the Python
namespace was polluted to the point that this could not happen.
That "just MLIR" Python core will be introduced in a followup, which
will create the space to upstream the FX and ONNX pure Python importers.
This primary non-NFC change to the API is:
* `torch_mlir.dialects.torch.importer.jit_ir` ->
`torch_mlir.jit_ir_importer`.
The rest is source code layering so that we can make the pt1 project
optional without losing the other features.
Progress on #2546.
This is a first step towards the structure we discussed here:
https://gist.github.com/stellaraccident/931b068aaf7fa56f34069426740ebf20
There are two primary goals:
1. Separate the core project (C++ dialects and conversions) from the
hard PyTorch dependencies. We move all such things into projects/pt1 as
a starting point since they are presently entangled with PT1-era APIs.
Additional work can be done to disentangle components from that
(specifically LTC is identified as likely ultimately living in a
`projects/ltc`).
2. Create space for native PyTorch2 Dynamo-based infra to be upstreamed
without needing to co-exist with the original TorchScript path.
Very little changes in this path with respect to build layering or
options. These can be updated in a followup without commingling
directory structure changes.
This also takes steps toward a couple of other layering enhancements:
* Removes the llvm-external-projects/torch-mlir-dialects sub-project,
collapsing it into the main tree.
* Audits and fixes up the core C++ build to account for issues found
while moving things. This is just an opportunistic pass through but
roughly ~halves the number of build actions for the project from the
high 4000's to the low 2000's.
It deviates from the discussed plan by having a `projects/` tree instead
of `compat/`. As I was thinking about it, this will better accommodate
the follow-on code movement.
Once things are roughly in place and the CI passing, followups will
focus on more in-situ fixes and cleanups.
The CI jobs that use stable PyTorch are currently not required to pass
in order for a patch to get merged in `main`. This commit makes sure
that if a CI job for stable PyTorch fails, it does not cancel the
other required jobs.
* CI: Skip (redundant) libtorch build when using stable PyTorch version
When we use PyTorch stable builds, there is no need to build libtorch
from source, making the stable-pytorch-with-torch-binary-OFF
configuration redundant with stable-pytorch-with-torch-binary-ON. This
patch drops the redundant configuration from CI.
* CI: Simplify guard conditions for creating and using libtorch cache
Whether libtorch is enabled or not is predicated on a host of conditions
such as the platform, in-tree versus out-of-tree build, and stable
versus nightly PyTorch builds. Instead of repeating these conditions to
guard whether to create or use the libtorch cache artifacts (and getting
them almost incorrect), this patch predicates the relevant pipeline
steps to whether libtorch is enabled, thus making the conditions far
simpler.
* feat: split pytorch requirements into stable and nightly
* fix: add true to tests to see full output
* refactor: add comments to explain true statement
* feat: move some tests to experimental mode
* refactor: refactor pipeline into more fine grained difference
* feat: add version differentiation for some tests
* feat: activate more configs
* refactor: change implementation to use less requirement files
* refactor: remove contraints used for testing
* fix: revert some requirement file names
* refactor: remove unnecessary ninja install
* fix: fix version parsing
* refactor: remove dependency on torchvision in main requirements file
* refactor: remove index url
* style: remove unnecesary line switch
* fix: readd index url
We previously used a fork of the action/cache repository for the PyTorch
cache since the actions/cache repo did not support read-only caches.
Now that actions/cache supports separate read and write steps, this
patch switches back to the actions/cache repo.
This patch, by itself, doesn't fix caching on Windows, but once a new
release of ccache is available, caching for Windows builds should start
working again (validated by building ccache from source and using it
with LLVM builds).
Ccache rejects caching when either the `/Zi` or `/ZI` flags are used
during compilation on Windows, since these flags tell the compiler to
embed debug information in a PDB file (separate from the object file
produced by the compiler). In particular, our CI builds add the `/Zi`
flag, making ccache mark these compiler invocations as uncacheable.
But what caused our CI to add debug flags, especially when we specified
`-DCMAKE_BUILD_TYPE=Release`? On Windows, unless we specify the
`--config Release` flag during the CMake build step, CMake assumes a
debug build. So all this while, we had been producing debug builds of
torch-mlir for every PR! No doubt it took so long to build the Windows
binaries.
The reason for having to specify the configuration during the _build_
step (as opposed to the _configure_ step) of CMake on Windows is that
CMake's Visual Studio generators will produce _both_ Release and Debug
profiles during the CMake configure step (thus requiring a build-time
value that tells CMake whether to build in Release or Debug mode).
Luckily, on Linux and macOS, the `--config` flag seems to be simply
ignored, instead of causing build errors.
Strangely, based on cursory tests, it seems like on Windows we need to
specify the Relase configuration as both `-DCMAKE_BUILD_TYPE=Release` as
well as `--config Release`. Dropping either made my build switch to a
Debug configuration.
Additionally, there is a bug in ccache v4.8 (although this is addressed
in trunk) that causes ccache to reject caching if the compiler
invocation includes any flag that starts with `/Z`, including /`Zc`,
which is added by LLVM's HandleLLVMOptions.cmake and which isn't related
to debug info or PDB files. The next release of ccache should include
the fix, which is to reject caching only for `/Zi` and `/ZI` flags and
not all flags that start with `/Z`.
As a side note, debugging this problem was possible because of ccache's
log file, which is enabled by: `ccache --set-config="log_file=log.txt"`.
Despite using sudo to delete the workspace directory, we still
occasionally run into checkout errors. This patch thus drops the
deletion of the workspace prior to checkout. It also restricts the
number of parallel jobs in the submodule fetch step to just one, to try
and resolve the checkout issue ("index.lock: File exists.").
We have recently started seeing errors like:
```
Synchronizing submodule url for 'externals/llvm-project'
Synchronizing submodule url for 'externals/mlir-hlo'
/usr/bin/git -c protocol.version=2 submodule update --init --force --depth=1
Error: fatal: Unable to create '/home/anush/actions-runner/_work/torch-mlir/torch-mlir/.git/modules/externals/llvm-project/index.lock': File exists.
```
As a workaround, this patch removes the workspace directory before the
checkout step.
This patch replaces all MHLO operations with their StableHLO
counterparts and adds a validation pass to ensure that no MHLO operations
remain before translating all Stablehlo operations to the MHLO dialect
for further lowering to the Linalg dialect.
This patch also updates all lit tests so that they refer to the
`convert-torch-to-stablehlo` pass and so that they check for StableHLO
operations.
There appear to be two problems with the caching layer in our CI runs:
(a) the sizes of some of the caches have grown to multiples of the
300 MB limit and (b) caching on Windows seems to be provide little to no
benefit.
To help understand the reasons for these problems, this patch adds a
line item to the list of steps run in CI to dump the ccache
configuration and statistics just prior to uploading the cache artifact.
This patch makes a few small, but key, changes to enable ccache on
Windows. First, it replaces the hendrikmuhs/ccache-action action with
command line invocations to the ccache binary, since the action has two
bugs, one of which causes CI to refer to different ccache artifacts
before versus after the build on Windows whereas the other bug can
sometimes cause the action to incorrectly infer that the cache is empty.
Second, this patch slightly alters the cache key, so that our old cache
artifacts, which have grown too big, are eventually discarded in favor
of the new, smaller cache artifacts. Along the way, this patch also
keeps the RollPyTorch's cache artifact separate from the regular build's
cache artifact so as to keep these artifacts small, and also because the
RollPyTorch action is off the critical path for most contributors.
Finally, this patch makes small changes to the CMake file so that on
Windows, the ccache binary is added as a prefix, as recommended on the
[ccache Wiki](https://github.com/ccache/ccache/wiki/MS-Visual-Studio).
* ci: update versions of external actions
Node.js 12 actions are deprecated and will eventually go away, so this
patch bumps the old actions to their latest versions that use Node.js
16.
* ci: replace deprecated action with bash commands
The llvm/actions/install-ninja action uses Node.js 12, which is
deprecated. Since that action is not updated to work with Node.js 16,
this patch replaces that action with equivalent bash commands to install
Ninja.
* ci: use smaller ccache artifacts to reduce evictions
Over time, our ccache sizes have grown quite large (some as large as
1.3 GB), which results in us routinely exceeding GitHub's limits, thus
triggering frequent cache evictions. As a result, cache downloads and
uploads take unnecessary long, in addition to fewer cache entries being
available.
Based on experiments on a clean cache state, it appears that we need
less than 300 MB of (compressed) ccache artifacts for each build type.
Anything larger than that will accrue changes from the past that aren't
needed.
To alleviate the cache burden, this patch sets the maximum ccache size
to be 300 MB. This change should not affect the success or failure of
our builds. I will monitor the build times to check whether this change
causes any performance degradation.
* ci: use consistent platform identifiers
Prior to this patch, some of our builds ran on `ubuntu-latest`, while
some others ran on `ubuntu-20.04` and others ran on `ubuntu-22.04`, with
similar situations for macOS and windows. This patch instead sets all
Linux builds to run on `ubuntu-latest`, all macOS builds to run on
`macos-latest`, and all Windows builds to run on `windows-latest`, to
make debugging future CI failures a little easier.
Until recently, we had to either risk feature branches creating PyTorch
build caches (which were unusable by the main branch or other parallel
feature branches because of GitHub's rules around sharing caches among
branches) or we had to limit the PyTorch build caches to only the main
branch, causing CI runs on feature branches to be terribly slow because
they had to rebuild PyTorch each time.
This patch enables the best of both worlds, by using a fork
(github.com/ashay/cache) of the GitHub's cache action, where the fork
adds an option (called `save`) which, when set, uploads a new cache
entry. We thus set this `save` flag only when we're building PyTorch
from source in Torch-MLIR's main branch, whereas all other builds set
this `save` flag to `false`.
The ability to conditionally update the cache has been an oft-requested
feature on the original (github.com/actions/cache) repository and
multiple unmerged PRs exist to allow conditional cache updates, so it is
likely that using the fork is only a temporary solution.
This patch is part of a larger set of improvements to the CI/build
system. In the code, we refer to the version as the string that
contains the release identifier such as 1.14.0.dev20221028, so calling
the file that contains the commit hash as pytorch-version.txt creates
confusion. For the sake of simplicity, this patch renames that file to
be pytorch-hash.txt.
If PyTorch build caches are created on a branch other than the main
branch, then GitHub does not share those caches with the main branch,
making every CI run that runs for each PR slow. This patch resolves the
problem by letting only the main branch create and use PyTorch build
caches.
* ci: cache PyTorch source builds
This patch reduces the time spent in regular CI builds by caching
PyTorch source builds. Specifically, this patch:
1. Makes CI lookup the cache entry for the PyTorch commit hash in
pytorch-version.txt
2. If lookup was successful, CI fetches the previously-generated WHL
file into the build_tools/python/wheelhouse directory
3. CI sets the `TM_PYTORCH_INSTALL_WITHOUT_REBUILD` variable to `true`
4. The build_libtorch.sh script then uses the downloaded WHL file
instead of rebuilding PyTorch
* ci: warm up PyTorch source cache during daily RollPyTorch action
This patch makes the RollPyTorch action write the updated WHL file to
the cache, so that it can be later retrieved by CI that runs for each
PR. We deliberately add the caching step to the end of the action since
the RollPyTorch action never needs to read from the cache, although
executing this step earlier in the process should not cause problems
either.
Instead of letting the auto-update script either fail because of script
errors or letting it commit bad versions, this patch makes the update
process manual, for now. Once the script stabilizes, I will its
re-enable periodic execution.
* Move CIs to use docker builds
Now that #1234 has landed and anyone can run CI / Release builds locally move GHA to use the same flow.
* update names
* Update comments
We use it for more than TorchScript testing now. This is a purely
mechanical change to adjust some file paths to remove "torchscript".
The most perceptible change here is that now e2e tests are run with
```
./tools/e2e_test.sh
instead of:
./tools/torchscript_e2e_test.sh
```
* Disable LTC by default until upstream revert relands
Tracked with the WIP https://github.com/llvm/torch-mlir/pull/1292
* Disable LTC e2e tests temporarily
* Update setup.py
Disable LTC in setup.py temporarily until upstream is fixed.
When we renamed the directory containing submodules from `external` to
`externals`, we accidentally left the original name in the Github
workflow. This patch fixes the problem.
My earlier[ PR](https://github.com/llvm/torch-mlir/pull/1213) had (among other things) decoupled ubuntu and macos builds into separate matrix runs. This is not working well due to limited number of MacOS GHA VMs causing long queue times and backlog. There are two reasons causing this backlog:
1. macos arm64 builds with pytorch source are getting erratically cancelled due to resource / network constraints. This is addressed with this: https://github.com/llvm/torch-mlir/pull/1215
> "macos-arm64 (in-tree, OFF) The hosted runner: GitHub Actions 3 lost communication with the server. Anything in your workflow that terminates the runner process, starves it for CPU/Memory, or blocks its network access can cause this error."
2. macos runs don't fail-fast when ubuntu runs fail due to being in separate matrix setups. This PR couples them again.
* mac m1 cross compile
Add support for M1 cross compile
* Remove redundant ExecutionEngine
It is registered as part of RegisterEverything
* nuke non-universal zstd
disable LTC
* Replace CHECK_EQ with TORCH_CHECK_EQ
* Check value of TORCH_MLIR_USE_INSTALLED_PYTORCH during LTC build
* Update LTC XFAIL with NewZerosModule ops
* Explicitly blacklist _like ops
* Automatically blacklist new_/_like ops
* Prune away unused Python dependencies from LTC
* Add flag to disable LTC
* Autogen dummy _REFERENCE_LAZY_BACKEND library when LTC is disabled
* Implement compute_shape_var
* Removed Var tests from XFAIL Set
* XFAIL tests using _local_scalar_dense or index.Tensor
* Add StdDim tests to XFAIL set
* Autogen aten::cat
Stella Laurenzo
Ashay Rane
Ramiro Leal-Cavazos
maxbartel
powderluv
Sean Silva
Tanyo Kwok
Henry Tu
Sambhav Jain
Antonio Kim
