### Description
<!-- Describe your changes. -->
### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->
### Description
Added CUDNN Frontend and used it for NHWC convolutions, and optionally
fuse activation.
#### Backward compatible
- For model existed with FusedConv, model can still run.
- If ORT is built with cuDNN 8, cuDNN frontend will not be built into
binary. Old kernels (using cudnn backend APIs) are used.
#### Major Changes
- For cuDNN 9, we will enable cudnn frontend to fuse convolution and
bias when a provider option `fuse_conv_bias=1`.
- Remove the fusion of FusedConv from graph transformer for CUDA
provider, so there will not be FusedConv be added to graph for CUDA EP
in the future.
- Update cmake files regarding to cudnn settings. The search order of
CUDNN installation in build are like the following:
* environment variable `CUDNN_PATH`
* `onnxruntime_CUDNN_HOME` cmake extra defines. If a build starts from
build.py/build.sh, user can pass it through `--cudnn_home` parameter, or
by environment variable `CUDNN_HOME` if `--cudnn_home` not used.
* cudnn python package installation directory like
python3.xx/site-packages/nvidia/cudnn
* CUDA installation path
#### Potential Issues
- If ORT is built with cuDNN 8, FusedConv fusion is no longer done
automatically, so some model might have performance regression. If user
still wants FusedConv operator for performance reason, they can still
have multiple ways to walkaround: like use older version of onnxruntime;
or use older version of ORT to save optimized onnx, then run with latest
version of ORT. We believe that majority users have moved to cudnn 9
when 1.20 release (since the default in ORT and PyTorch is cudnn 9 for 3
months when 1.20 release), so the impact is small.
- cuDNN graph uses TF32 by default, and user cannot disable TF32 through
the use_tf32 cuda provider option. If user encounters accuracy issue
(like in testing), user has to set environment variable
`NVIDIA_TF32_OVERRIDE=0` to disable TF32. Need update the document of
use_tf32 later.
#### Follow ups
This is one of PRs that target to enable NHWC convolution in CUDA EP by
default if device supports it. There are other changes will follow up to
make it possible.
(1) Enable `prefer_nhwc` by default for device with sm >= 70.
(2) Change `fuse_conv_bias=1` by default after more testing.
(3) Add other NHWC operators (like Resize or UpSample).
### Motivation and Context
The new CUDNN Frontend library provides the functionality to fuse
operations and provides new heuristics for kernel selection. Here it
fuses the convolution with the pointwise bias operation. On the [NVIDIA
ResNet50](https://pytorch.org/hub/nvidia_deeplearningexamples_resnet50/)
we get a performance boost from 49.1144 ms to 42.4643 ms per inference
on a 2560x1440 input (`onnxruntime_perf_test -e cuda -I -q -r 100-d 1 -i
'prefer_nhwc|1' resnet50.onnx`).
---------
Co-authored-by: Tianlei Wu <tlwu@microsoft.com>
Co-authored-by: Maximilian Mueller <maximilianm@nvidia.com>
### Description
Remove the "--enable_language_interop_ops" build flag, because the code
is incompatible with the latest numpy, and the build flag is not used
anywhere except a macOS CI pipeline. It does not seem to have a ship
plan.
### Motivation and Context
The build error was:
```
onnxruntime/core/language_interop_ops/pyop/pyop.cc:122:85: error: no member named 'elsize' in '_PyArray_Descr'
static_cast<int64_t>(PyArray_DescrFromType(type)->elsize),
~~~~~~~~~~~~~~~~~~~~~~~~~~~ ^
```
### Description
Enable creating dedicated build for on device training. With this PR we
can build a lean binary for on device training using flag
--enable_training_apis. This binary includes only the essentials like
training ops, optimizers etc and NOT features like Aten fallback,
strided tensors, gradient builders etc . This binary also removes all
the deprecated components like training::TrainingSession and OrtTrainer
etc
### Motivation and Context
This enables our partners to create a lean binary for on device
training.
### Description
1. Renames all references of on device training to training apis. This
is to keep the naming general. Nothing really prevents us from using the
same apis on servers\non-edge devices.
2. Update ENABLE_TRAINING option: With this PR when this option is
enabled, training apis and torch interop is also enabled.
3. Refactoring for onnxruntime_ENABLE_TRAINING_TORCH_INTEROP option:
- Removed user facing option
- Setting onnxruntime_ENABLE_TRAINING_TORCH_INTEROP to ON when
onnxruntime_ENABLE_TRAINING is ON as we always build with torch interop.
Once this PR is merged when --enable_training is selected we will do a
"FULL Build" for training (with all the training entry points and
features).
Training entry points include:
1. ORTModule
2. Training APIs
Features include:
1. ATen Fallback
2. All Training OPs includes communication and collectives
3. Strided Tensor Support
4. Python Op (torch interop)
5. ONNXBlock (Front end tools for training artifacts prep when using
trianing apis)
### Motivation and Context
Intention is to simply the options for building training enabled builds.
This is part of the larger work item to create dedicated build for
learning on the edge scenarios with just training apis enabled.
### Description
Use target name for flatbuffers.
Add version range for flatbuffers. It is similar to #13870
### Motivation and Context
To fix a build error:
```
CMake Error at onnxruntime_graph.cmake:88 (add_dependencies):
The dependency target "flatbuffers" of target "onnxruntime_graph" does not
exist.
Call Stack (most recent call first):
CMakeLists.txt:1490 (include)
```
It happens when flatbuffers library is already installed. For example,
on Ubuntu people may get it from apt-get. But, the one provided by
Ubuntu 20.04 is not compatible with our code. The one in Ubuntu 22.04
works fine.
## Description
1. Convert some git submodules to cmake external projects
2. Update nsync from
[1.23.0](https://github.com/google/nsync/releases/tag/1.23.0) to
[1.25.0](https://github.com/google/nsync/releases/tag/1.25.0)
3. Update re2 from 2021-06-01 to 2022-06-01
4. Update wil from an old commit to 1.0.220914.1 tag
5. Update gtest to a newer commit so that it can optionally leverage
absl/re2 for parsing command line flags.
The following git submodules are deleted:
1. FP16
2. safeint
3. XNNPACK
4. cxxopts
5. dlpack
7. flatbuffers
8. googlebenchmark
9. json
10. mimalloc
11. mp11
12. pthreadpool
More will come.
## Motivation and Context
There are 3 ways of integrating 3rd party C/C++ libraries into ONNX
Runtime:
1. Install them to a system location, then use cmake's find_package
module to locate them.
2. Use git submodules
6. Use cmake's external projects(externalproject_add).
At first when this project was just started, we considered both option 2
and option 3. We preferred option 2 because:
1. It's easier to handle authentication. At first this project was not
open source, and it had some other non-public dependencies. If we use
git submodule, ADO will handle authentication smoothly. Otherwise we
need to manually pass tokens around and be very careful on not exposing
them in build logs.
2. At that time, cmake fetched dependencies after "cmake" finished
generating vcprojects/makefiles. So it was very difficult to make cflags
consistent. Since cmake 3.11, it has a new command: FetchContent, which
fetches dependencies when it generates vcprojects/makefiles just before
add_subdirectories, so the parent project's variables/settings can be
easily passed to the child projects.
And when the project went on, we had some new concerns:
1. As we started to have more and more EPs and build configs, the number
of submodules grew quickly. For more developers, most ORT submodules are
not relevant to them. They shouldn't need to download all of them.
2. It is impossible to let two different build configs use two different
versions of the same dependency. For example, right now we have protobuf
3.18.3 in the submodules. Then every EP must use the same version.
Whenever we have a need to upgrade protobuf, we need to coordinate
across the whole team and many external developers. I can't manage it
anymore.
3. Some projects want to manage the dependencies in a different way,
either because of their preference or because of compliance
requirements. For example, some Microsoft teams want to use vcpkg, but
we don't want to force every user of onnxruntime using vcpkg.
7. Someone wants to dynamically link to protobuf, but our build script
only does static link.
8. Hard to handle security vulnerabilities. For example, whenever
protobuf has a security patch, we have a lot of things to do. But if we
allowed people to build ORT with a different version of protobuf without
changing ORT"s source code, the customer who build ORT from source will
be able to act on such things in a quicker way. They will not need to
wait ORT having a patch release.
9. Every time we do a release, github will also publish a source file
zip file and a source file tarball for us. But they are not usable,
because they miss submodules.
### New features
After this change, users will be able to:
1. Build the dependencies in the way they want, then install them to
somewhere(for example, /usr or a temp folder).
2. Or download the dependencies by using cmake commands from these
dependencies official website
3. Similar to the above, but use your private mirrors to migrate supply
chain risks.
4. Use different versions of the dependencies, as long as our source
code is compatible with them. For example, you may use you can't use
protobuf 3.20.x as they need code changes in ONNX Runtime.
6. Only download the things the current build needs.
10. Avoid building external dependencies again and again in every build.
### Breaking change
The onnxruntime_PREFER_SYSTEM_LIB build option is removed you could think from now
it is default ON. If you don't like the new behavior, you can set FETCHCONTENT_TRY_FIND_PACKAGE_MODE to NEVER.
Besides, for who relied on the onnxruntime_PREFER_SYSTEM_LIB build
option, please be aware that this PR will change find_package calls from
Module mode to Config mode. For example, in the past if you have
installed protobuf from apt-get from ubuntu 20.04's official repo,
find_package can find it and use it. But after this PR, it won't. This
is because that protobuf version provided by Ubuntu 20.04 is too old to
support the "config mode". It can be resolved by getting a newer version
of protobuf from somewhere.
1. Update CK to its latest develop branch
2. `-mllvm -amdgpu-early-inline-all=true` is critical to CK's
performance, ensure it is properly configured.
- The flags are propagated from target `hip-lang::device`'s
`INTERFACE_COMPILE_OPTIONS`, we must not manually add the flags.
- Instead, we must ensure this target is properly configured by checking
_CMAKE_HIP_DEVICE_RUNTIME_TARGET is set.
TL,DR
`hip-lang::device` sometime will be not be properly configured if our
`CMAKE_PREFIX_PATH` is not configured carefully. In the CI docker, the
configuration is in good state, but on dev machine it is not, which then
silently result poor performance for kernels. We fixed it in this PR and
add a guard to avoid unsuccessful future editing and to prevent
convoluted debugging process.
`_CMAKE_HIP_DEVICE_RUNTIME_TARGET ` is shared in
`/opt/rocm/lib/cmake/hip-lang/hip-lang-config.cmake` and it is internal
to
[CMake](https://gitlab.kitware.com/cmake/cmake/-/merge_requests/6121/diffs),
the variable name will not be changed in the foreseeable future.
* C API version 0.001
* fix linker issues
* fixes for save checkpoint api
* plus fixes based on tests
* plus test_runner and other changes
* Plus cosmetic updates
* remove unnecessary headers
* plus some updates
* plus more changes
Co-authored-by: Ashwini Khade <askhade@microsoft.com@orttrainingdev10.d32nl1ml4oruzj4qz3bqlggovf.px.internal.cloudapp.net>
* lr_scheduler implementation
(cherry picked from commit d9c2552b3a3b2ff38ee0a14770257aa1169f6fa9)
* refactor Module/Optimizer constructor.
* add intermidiate API layer bridging public interfaces with internal ones.
* synthetic data loader
* make end to end run pass
* avoid many session input copy (CPU to GPU)
some clean up
* NVTX for runner
* minor fix after sync
* revert to let Module/Optimizer handle session creation.
* fix tests & test file folder consolidation
* refine based on comments & fix cpplint
* typos
* Checkpoint API Implementation
* fix build issues
* fix undefined reference for ParseData of type string.
* refinements
* resolve some comments
* expose python api
* make save and load test pass
* some clean up
* make optimizer save/load test pass
* make custom property save/load test pass
* formatting
* fix comments - fix wave - code placement, remove legacy ckpt logic dependency, remove external data support
* fix comment - wave 2 - Remove ParseData/ParseStringData, Use UnpackTensor, Simplify CheckpointProperty usage
* fix comment - wave 3 - rename all api_test namespace to api
* fix comment - wave 4 - load/save trainable/nontrainable param seperately.
* Rename Load/SaveORTCheckpoint
* renaming API && remove CheckpointUntils. api::LoadCheckpoint/SaveCheckpoint is the exposed interfaces.
* revert unnecessary format change for onnxruntime/core/framework/tensorprotoutils.h/cc
* formatting
* re-org the class folders for better dependency managerment
* save_checkpoint accpeting TensorProto as inputs
* More clean up
* clean up the naming
* refactor a bit type constraints on custom property
* fix comment - file read/write && report error when file read/write failed
* extract LoopDir to FilterFilesFromDirectory
* fix build
* add api test runner
* add build flag for training_api
* address review comments
* some fixes
* address more comments
* make the build pass by filling in empty implementation
* fix more
* re-hipify all rocm EP sources
* fix all other files affected by re-hipify
* add cuda_provider_factory.h to amd_hipify.py
* do not use cudnn_conv_algo_search in ROCm EP, missing reduce min registration
* Fix ReduceConsts template specialization introduced in #9101.
Fixes the error when building for ROCm 4.3.1:
error: too many template headers for onnxruntime::rocm::ReduceConsts<__half>::One (should be 0)
* fix flake8 error in amd_hipify.py
* speed up hipify with concurrent.futures
* flake8 fix in amd_hipify.py
* fix build - python.h not found
* disable --build_shared_lib for ortmodule tests
* fix
* fix the build flag
* disable --build_shared_lib for training path (not only for ortmodule)
* fix missing test model files
* disable test CApiTest.test_custom_op_library when ENABLE_TRAINING_TORCH_INTEROP is ON
* enable custom_op_library build
* fix build
* fix
* merge master and fix build failure
* build onnx_test_runner when onnxruntime_ENABLE_TRAINING_TORCH_INTEROP is ON
* resolve comments
* use --enable_training_torch_interop to replace "onnxruntime_ENABLE_TRAINING_TORCH_INTEROP=ON"
* clean up builds for interop_torch
* add python dependency for executables
* disable onnxruntime_ENABLE_TRAINING_TORCH_INTEROP by default; enable it in ortmodule GPU training pipeline only
* disable training unrelated tests when torch interop is enabled
* simplify the python dependency.
* clean up and fix
1. Update manylinux build scripts. This will add [PEP600](https://www.python.org/dev/peps/pep-0600/)(manylinux2 tags) support. numpy has adopted this new feature, we should do the same. The old build script files were copied from https://github.com/pypa/manylinux, but they has been deleted and replaced in the upstream repo. The manylinux repo doesn't have a manylinux2014 branch anymore. So I'm removing the obsolete code, sync the files with the latest master.
2. Update GPU CUDA version from 11.0 to 11.1(after a discussion with PMs).
3. Delete tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_cuda10_2. (Merged the content to tools/ci_build/github/linux/docker/Dockerfile.manylinux2014_cuda11)
4. Modernize the cmake code of how to locate python devel files. It was suggested in https://github.com/onnx/onnx/pull/1631 .
5. Remove `onnxruntime_MSVC_STATIC_RUNTIME` and `onnxruntime_GCC_STATIC_CPP_RUNTIME` build options. Now cmake has builtin support for it. Starting from cmake 3.15, we can use `CMAKE_MSVC_RUNTIME_LIBRARY` cmake variable to choose which MSVC runtime library we want to use.
6. Update Ubuntu docker images that used in our CI build from Ubuntu 18.04 to Ubuntu 20.04.
7. Update GCC version in CUDA 11.1 pipelines from 8.x to 9.3.1
8. Split Linux GPU CI pipeline to two jobs: build the code on a CPU machine then run the tests on another GPU machines. In the past we didn't test our python packages. We only tested the pre-packed files. So we didn't catch the rpath issue in CI build.
9. Add a CentOS machine pool and test our Linux GPU build on real CentOS machines.
10. Rework ARM64 Linux GPU python packaging pipeline. Previously it uses cross-compiling therefore we must static link to C Runtime. But now have pluggable EP API and it doesn't support static link. So I changed to use qemu emulation instead. Now the build is 10x slower than before. But it is more extensible.
* First iteration of making cuda a shared provider.
Separated out shared OpKernel change, so doing this to merge with that change.
* More cuda shared library refactoring
* More cuda shared library refactoring
* More build options tested, converted the training ops over.
* Fix merge breaks
* Fix submodules
* Fix submodules
* Fix submodules
* Fix python
* Fix compile errors
* Duplicate symbol fix
* Test fix for ROCM provider
* Another ROCM test workaround
* ROCM Build Test
* ROCM build fix
* ROCM
* ROCM
* ROCM
* ROCM
* ROCM
* ROCM test
* Reduce header dependencies
* Remove redundant namespace
* Test fix for linux
* Fix linux build
* Fix Eigen build error
* Fix unused parameter warning
* Test link error
* Another linker test
* Linker test
* Linker test
* Another test
* Another build test
* Fix linux link error
* Build test
* Fix control flow ops to use common base class with core code
* Remove extra qualifiers
* Fix template syntax for linux
* Fix cuda memory leak
* Fix pybind
* Test disabling cast
* Cleanup
* Restore cuda in test
* Remove more header dependencies
* Test not adding cuda provider to session
* Make GetProviderInfo_CUDA throw
* No-op cuda provider creation
* Fix some setup issues
* Fix memory cleanup on unload
* Diagnostics
* Don't unload library
* Add diagnostics
* Fix deleting registry at right time.
* Test disabling profiler
* Fix merge break
* Revert profiler change
* Move unloading of shared providers into Environment
* Free more global allocations before library unloads
* Add more diagnostics
* Move unloading back to the OrtEnv as there are multiple Environments created during a session.
Remove some library dependencies for tests.
* Fix more cmake files
* ERROR -> WARNING
* Fix python shutdown
* Test not using dml in pipeline
* Change python version and disable dml
* Update python version
* Test adding unload method for shared providers
* Disable DLL test
* Python test
* Revert "Python test"
This reverts commit c7ec2cfe98.
* Revert "Disable DLL test"
This reverts commit e901cb93aa.
* Revert "Test adding unload method for shared providers"
This reverts commit c427b78799.
* Point to RyanWinGPU
* Revert python version
* Fix id_to_allocator_map
* Another python exit test
* Remove extra debug messages
Try a more clean python shutdown through DllMain
* Revert DllMain idea, it didn't work
* Merge conflicts
* Fix merge with master issues.
* Comments
* Undo edit to file
* Cleanup + new training ops
* Revert yml changes
* Fix another merge error
* ROCM fix
* ROCM fix v2
* Put back Linux hack, it is necessary
* Stupid fixes
* Fix submodule out of sync
* ROCM fix 3
* ROCM 4
* Test java fix
* Fix typos
* Java test on my VM
* Fix build error
* Spotless fix
* Leave temp file around to load properly
* Fix cleanup on exit
* Fix break
* Java comments
* Remove LongformerAttentionBase workaround
* Spotless fix
* Switch yml back to regular build pool
* Revert "Switch yml back to regular build pool"
This reverts commit be35fc2a5a.
* Code review feedback
* Fix errors due to merge
* Spotless fix
* Fix minimal build
* Java fix for non cuda case
* Java fix for CPU build
* Fix Nuphar?
* Fix nuphar 2
* Fix formatting
* Revert "Remove LongformerAttentionBase workaround"
This reverts commit 648679b370.
* Training fix
* Another java fix
* Formatting
* Formatting
* For orttraining
* Last orttraining build fix...
* training fixes
* Fix test provider error
* Missing pass command
* Removed in wrong spot
* Python typo
* Python typos
* Python crash on exit, possibly due to unloading of libraries.
* Remove test_execution_provider from training build
Only enable python atexit on windows
Remove assert on provider library exit
* Still can't unload providers in python, alas.
* Disable Nvtx temporarily
* MPI Kernels for Training
* MPI Kernels part 2
* Patch through INcclService
* Oops, wrong CMakeLists
* Missing namespace
* Fix missing ()
* Move INcclService::GetInstance around to link nicer
* Missing }
* Missing MPI libraries for Cuda
* Add extra GetType functions used by MPI
* Missing Nccl library
* Remove LOGS statements as a test
* Add in a couple more missing GetType methods
* Update comments
* Missed a logging reference in mpi_context.h
* Convert aten_op to shared (due to marge with master)
* Test moving DistributedRunContext instance into shared provider layer
(with purpose error to verify it's being built properly)
* Test passed, now with fix
* Missing static
* Oops, scope DistributedRunContext to just NCCL
* Merge related issues and code review feedback.
* Merge error
* Bump to rel-1.9.1 (#7684)
* Formatting
* Code review feedback for Java build on non Windows
* Remove cupti library dependency from core library
* Test Java pipeline fix
* Linux build fix
* Revert "Linux build fix"
This reverts commit a73a811516.
* Revert "Remove cupti library dependency from core library"
This reverts commit 6a889ee8bf.
* Packaging pipeline fixes to copy cuda shared provider for tensorrt & standard packages
* Add cuda to Tensorrt nuget package
* onnxruntime_common still has a cuda header dependency
Co-authored-by: ashbhandare <ash.bhandare@gmail.com>
1. Merge Nuget CPU pipeline, Java CPU pipeline, C-API pipeline into a single one.
2. Enable compile warnings for cuda files(*.cu) on Windows.
3. Enable static code analyze for the Windows builds in these jobs. For example, this is our first time scanning the JNI code.
4. Fix some warnings in the training code.
5. Enable code sign for Java. Previously we forgot it.
6. Update TPN.txt to remove Jemalloc.
The ROCm EP is designed and implemented based on AMD GPU software stack named ROCm. Here is the link for the details about ROCm: https://rocmdocs.amd.com/en/latest/
ROCm EP was created based on the following things:
1. AMD GPU programming language: HIP
2. AMD GPU HIP language runtime: amdhip64
3. BLAS: rocBLAS, hipBLAS
4. DNN: miOpen
5. Collective Communication library: RCCL
6. cub: hipCub
7. …
Current status:
BERT-L and GPT2 training can be ran on AMD GPU with data parallel.
Next:
1. Make more GPU code be sharable between ROCm EP and CUDA EP since HIP language and HIP runtime API are very close to CUDA.
2. Continue improving the implementation.
3. Continue GPU kernel optimization.
4. Support model parallelism on ROCm EP.
……
The rocm kernels have been removed from this commit and will be in a separate PR. Since the original PR was too big(~180 files), it was suggested to split the PR into two parts, one is rocm-kernels, the other is non rocm kernels.
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
Co-authored-by: sabreshao <sabre.shao@amd.com>
Co-authored-by: anghostcici <11013544+anghostcici@users.noreply.github.com>
Co-authored-by: Suffian Khan <sukha@microsoft.com>
Co-authored-by: Edward Chen <18449977+edgchen1@users.noreply.github.com>
* Move fbs include from header to cc
* add initial cmake for flatbuffers
* Move most flatbuffers util to ort_flatbuffers
* move code around
* fix
* move test/perf runner to use flatbuffer directly instead of model
* minor update
* Fix build break
* Clean up includes and foward decl
* Fix traning CI build breaks
* Addressed PR comment, replaced some include with forward decls
* Remove ORT_MUST_USE_RESULT temporarily
* ORT on CUDA 11
1. Seperate HOROVOD and MPI
2. Seperate NCCL from HOROVOD in CMakeLists.txt
2. Remove dependency on external cub
3. cudnnSetRNNDescriptor is changed in cuDNN 8.0
* polish the code about MPI/NCCL in CMakeLists.txt and build.py
* check CUDA version
* ${MPI_INCLUDE_DIRS} should be PUBLIC
* sm30, sm50 are deprecated in CUDA 11 Toolkit
* update change based on code review feedback.
* add sm_52
* improve MPI/NCCL build path
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
- Add support for ENABLE_LANGUAGE_INTEROP_OPS in training build which is enabled for nightly builds
- Fix passing of environment variables to `sudo docker run` in build definitions
- Fix setup.py package naming logic
* dashboard integration - first phase
* change a field
* perf scripts
* addressing PR comments
* address comments and fix build
* minor
* make GetConfigFromData() const
* more update for comments
* addressing comments
* more on addressing comments
* minor
* fix build
* add condition check
* more on comments
* retrun status
* remove batch size
* on comments
* rename pkg path
* rename pkg path
* additional commentss
Co-authored-by: Ethan Tao <ettao@microsoft.com>
