* Enable Attention op for ROCM EP.
As a note, potential hipify improvements: (1) handle math
contants (attention_softmax.h), (2) correctly generate transpose
options for the GEMM helpers, consider counterpart/dummy API for
CublasMathModeSetter (attention_impl.cu, attention_impl.cu). After
these improvements, we don't need to manually keep copies of the
above mentioned files any more.
* Clean up debugging code.
* [ROCm] update hipify-perl location
Depending on the ROCm version installed, hipify-perl might not always
live in the hard-coded path of /opt/rocm/bin. Use python 3.3's
shutil.which to locate the script.
* provide alternative locations for hipify-perl if not in PATH
* implement hipify-perl search as a function
This avoids running the logic during module import since all builds
import the amd_hipify module.
* fix flake8 errors
Support for device function pointers is not yet available for ROCm.
Instead, the device function pointers were converted to device functors.
Case statements, lambdas, and macros are used for dispatch; as a result,
all combinations of kernels are compiled with inlined functors. The
basis of this approach can be found in PyTorch.
Lastly, hipify and register Resize and Upsample for ROCm EP.
* 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
* make work for both rocm 4.2 and rocm 4.3.1
* fix rocm 4.3.1 docker image reference
* fix CUDA_VERSION to ROCM_VERSION
* fix ReduceConsts conflict def
* add ifdef to miopen_common.h as well
* trailing ws
* correct batchnorm replacement output order;
remove bn replacement in grad graph builder
* update op defs and kernel class
* implement batch norm internal and grad.
* change saved_var into saved_inv_std
* cuda test case: bn internal
* remove redundant include
* fix comment; add support and UT for 1d input.
* exclude batch_norm_internal in amd_hipify
* run BNInternal UT for CUDA only
* fix CI error
* fix comment errors
* fix error
* add comment for inconsistency with cudnnBN doc
* additional comments for cudnnBN inconsistency
* Pass cuda stream to thrust function to not use default stream.
In the commit 299ace0, ORT has been changed to not use cuda default stream.
* update amd_hipify.py
* remove un-necessary stream sync
Co-authored-by: Weixing Zhang <wezhan@microsoft.com>
* ConvGrad CUDA impl
* Set up the test case for Deberta Conv1D
* Add fp16 test
Co-authored-by: Sherlock Huang <bahuang@OrtTrainingDev3.af05slrtruoetgaxwwjv5nsq5e.px.internal.cloudapp.net>
* Permit dense seq optimization on BERT-L pytorch export by enabling ReduceSumTraining, Equal, and NonZero on AMD
* enable Equal tests
* enable fast_matrix_reduction test case
* optimize cuda conv by fused activation
* remove needless print out
* exclude test from cpu
* handle status error from cudnn 8.x
* add reference to base class
* add hipify
Move CudaKernel from cuda_common.h to a new separate header, cuda_kernel.h. Update include sites to use cuda_kernel.h instead if they need CudaKernel. Inclusions of cuda_common.h are now more lightweight.
Make corresponding changes for ROCM execution provider code.
Other minor cleanup.
Some part of code for reduction kernels has been changed in 858040fa,
which cause failures in rocm build since ROCm EP shares some code with
CUDA EP. This PR is to quick fix this failure by not sharing two files
for now to unblock CI enabling on ROCm EP. Another PR for leveraging
858040fa for ROCm EP will be done later.
* Add kernels for AMD GPU.
This PR is mostly about GPU kernels for ROCm EP. Due to similar GPU programming language (CUDA and HIP and similar math library calls, one principle in ROCM EP design is to share CUDA kernels as much as possible for ROCm. Thus, the script amd_hipify.py has been created for converting CUDA kernels to ROCm HIP kernels automatically during compilation phase. But, for some reasons such as perf issue, syntax difference..., some converted kernels need some manual intervention. These kernels will be checked in the repo physically for now. In order to avoid manual intervention, the plan is to refactor CUDA kernels to make them portable between CUDA EP and ROCm EP as much as possible.
Please refer to "HIP Porting Guide" for details.
* like lamb, multi-tensor-apply needs to be disabled for IsAllFiniteOp and ReduceAllL2, current AMD GPU compiler has perf issue for kernel parameter which is a structure with "pass by value".
* Use hipMemsetAsync and add checks on HIP calls.
* move the generated files to build folder.
Co-authored-by: Jesse Benson <jesseb@microsoft.com>
