* Use dim 0 as faster dim for writing mean/var/count workspace in batchnorm multiblock method [performance] * Add CountDataType as template parameter in blockwise_welford * Add utility/get_shift.hpp * Add BatchNorm multiblock single-kernel implementation * Add smem inline assembly based implementation of gms_init/gms_barrier/gms_reset for gfx90a * Renaming in device_batchnorm_forward_impl.hpp * Tiny fix in the batchnorm_fwd profiler * Revert "Add smem inline assembly based implementation of gms_init/gms_barrier/gms_reset for gfx90a" This reverts commitd16d00919c. * Use the old two-kernel batchnorm multiblock method for gfx1030 * Use the old two-kernel batchnorm multiblock method for gfx908 * use the single-kernel batchnorm multiblock method only for gfx90a * Remove get_wave_id() from utility/get_id.hpp since it is not used * Set true for testing running mean/variance and saving mean/invvariance in the examples * Fix to copy-right words * Remove un-needed including in utility/get_id.hpp * Add comments to workgroup_synchronization.hpp * Remove un-used codes in gridwise_multiblock_batchnorm_forward.hpp * Renaming in the kernels * Remove un-used kernel file [ROCm/composable_kernel commit:8f5cafaf04]
Composable Kernel
Methodology
Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
CK utilizes two concepts to achieve performance portability and code maintainability:
- A tile-based programming model
- Algorithm complexity reduction for complex ML operators, using innovative technique we call "Tensor Coordinate Transformation".
Code Structure
Current CK library are structured into 4 layers:
- "Templated Tile Operators" layer
- "Templated Kernel and Invoker" layer
- "Instantiated Kernel and Invoker" layer
- "Client API" layer
Documentation
Run the steps below to build documentation locally.
cd docs
pip3 install -r sphinx/requirements.txt
python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
Contributors
The list of developers and contributors is here: Contributors
Citation
If you use CK, please use following citations:
- CK paper will be freely available on arXiv soon: Realizing Tensor Operators Using Coordinate Transformations and Tile Based Programming
- CITATION.cff
License
CK is released under the MIT license. License File
Build CK
Build docker image
DOCKER_BUILDKIT=1 docker build -t ck:latest -f Dockerfile .
Launch docker
docker run \
-it \
--privileged \
--group-add sudo \
-w /root/workspace \
-v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace \
ck:latest \
/bin/bash
Build CK
mkdir build && cd build
# Need to specify target ID, example below is for gfx908 and gfx90a
cmake \
-D CMAKE_PREFIX_PATH=/opt/rocm \
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
-D CMAKE_CXX_FLAGS="-O3" \
-D CMAKE_BUILD_TYPE=Release \
-D GPU_TARGETS="gfx908;gfx90a" \
..
Build examples and tests
make -j examples tests
make test
Instructions for running each individual examples are under example
Build ckProfiler
make -j ckProfiler
Instructions for running ckProfiler are under profiler
Install CK
make install
Using CK as pre-built kernel library
Instructions for using CK as a pre-built kernel library are under client_example
Caveat
Kernel Timing and Verification
CK's own kernel timer will warn up kernel once, and then run it multiple times to get average kernel time. For some kernels that use atomic add, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time. CK's own timer and verification in each example and ckProfiler can be enabled or disabled from command line.

