Skip to content

GitLab

Commit 2298a1a4 authored by illsilin's avatar illsilin
Browse files

sync from public

parents 965b7ba4 2f088b87
Expand all Show whitespace changes
Inline Side-by-side
Showing with 428 additions and 334 deletions
.github/CODEOWNERS
View file @ 2298a1a4
* @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
* @junliume @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca
# Documentation files
docs/ @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
*.md @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
*.rst @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
.readthedocs.yaml @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
docs/ @ROCm/rocm-documentation @junliume @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca
*.md @ROCm/rocm-documentation @junliume @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca
*.rst @ROCm/rocm-documentation @junliume @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca
.readthedocs.yaml @ROCm/rocm-documentation @junliume @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca
# Header directory for Doxygen documentation
library/include/ @ROCm/rocm-documentation @junliume @illsilin @carlushuang @aosewski @poyenc @geyyer @bartekxk
library/include/ @ROCm/rocm-documentation @junliume @illsilin @carlushuang @qianfengz @aosewski @poyenc @geyyer @bartekxk @andriy-ca
CONTRIBUTORS.md
View file @ 2298a1a4
[Back to the main page](./README.md)
# Composable Kernel Developers and Contributors
This is the list of developers and contributors to Composable Kernel library
......
Dockerfile
View file @ 2298a1a4
FROM ubuntu:20.04
ARG DEBIAN_FRONTEND=noninteractive
ARG ROCMVERSION=6.2
ARG ROCMVERSION=6.3
ARG compiler_version=""
ARG compiler_commit=""
ARG CK_SCCACHE=""
RUN set -xe
ARG DEB_ROCM_REPO=http://repo.radeon.com/rocm/apt/.apt_$ROCMVERSION/
RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
# Add rocm repository
RUN chmod 1777 /tmp
RUN apt-get update
RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
RUN if [ "$ROCMVERSION" != "6.3" ]; then \
sh -c "wget https://repo.radeon.com/amdgpu-install/$ROCMVERSION/ubuntu/focal/amdgpu-install_6.2.60200-1_all.deb --no-check-certificate" && \
apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated ./amdgpu-install_6.2.60200-1_all.deb && \
# Add rocm repository
RUN set -xe && \
useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins && \
apt-get update && apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl && \
curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
RUN if [ "$ROCMVERSION" != "6.4" ]; then \
sh -c "wget https://repo.radeon.com/amdgpu-install/$ROCMVERSION/ubuntu/focal/amdgpu-install_6.3.60300-1_all.deb --no-check-certificate" && \
apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated ./amdgpu-install_6.3.60300-1_all.deb && \
wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
elif [ "$ROCMVERSION" = "6.3" ] && [ "$compiler_version" = "rc1" ]; then \
sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_6.3-20.04-1_all.deb --no-check-certificate" && \
apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install dialog libpopt0 rsync && DEBIAN_FRONTEND=noninteractive apt-get install ./amdgpu-install-internal_6.3-20.04-1_all.deb && \
sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 6.3 rel-20 > /etc/apt/sources.list.d/rocm-build.list' && \
amdgpu-repo --amdgpu-build=2074281; \
fi
RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
RUN amdgpu-install -y --usecase=rocm --no-dkms
RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list" && \
amdgpu-install -y --usecase=rocm --no-dkms
## Sccache binary built from source for ROCm, only install if CK_SCCACHE is defined
ARG SCCACHE_REPO_URL=http://compute-artifactory.amd.com/artifactory/rocm-generic-experimental/rocm-sccache
......@@ -76,68 +67,47 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
clang-format-12 \
kmod && \
apt-get clean && \
rm -rf /var/lib/apt/lists/*
rm -rf /var/lib/apt/lists/* && \
rm -rf amdgpu-install* && \
# Remove unnecessary rocm components that take a lot of space
apt-get remove -y rocblas rocfft rocsparse composablekernel-dev
# hipTensor requires rocm-llvm-dev for rocm versions > 6.0.1
RUN if [ "$ROCMVERSION" = "6.1" ]; then \
sh -c "apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated rocm-llvm-dev"; \
fi
# Update the cmake to version 3.27.5
RUN pip install --upgrade cmake==3.27.5
RUN pip install --upgrade cmake==3.27.5 && \
#Install latest ccache
RUN git clone https://github.com/ccache/ccache.git && \
cd ccache && mkdir build && cd build && cmake .. && make install
git clone https://github.com/ccache/ccache.git && \
cd ccache && mkdir build && cd build && cmake .. && make install && \
#Install ninja build tracing tools
RUN wget -qO /usr/local/bin/ninja.gz https://github.com/ninja-build/ninja/releases/latest/download/ninja-linux.zip
RUN gunzip /usr/local/bin/ninja.gz
RUN chmod a+x /usr/local/bin/ninja
RUN git clone https://github.com/nico/ninjatracing.git
cd / && \
wget -qO /usr/local/bin/ninja.gz https://github.com/ninja-build/ninja/releases/latest/download/ninja-linux.zip && \
gunzip /usr/local/bin/ninja.gz && \
chmod a+x /usr/local/bin/ninja && \
git clone https://github.com/nico/ninjatracing.git && \
#Install latest cppcheck
RUN git clone https://github.com/danmar/cppcheck.git && \
cd cppcheck && mkdir build && cd build && cmake .. && cmake --build .
WORKDIR /
# Setup ubsan environment to printstacktrace
RUN ln -s /usr/bin/llvm-symbolizer-3.8 /usr/local/bin/llvm-symbolizer
ENV UBSAN_OPTIONS=print_stacktrace=1
git clone https://github.com/danmar/cppcheck.git && \
cd cppcheck && mkdir build && cd build && cmake .. && cmake --build . && \
cd / && \
# Install an init system
RUN wget https://github.com/Yelp/dumb-init/releases/download/v1.2.0/dumb-init_1.2.0_amd64.deb
RUN dpkg -i dumb-init_*.deb && rm dumb-init_*.deb
ARG PREFIX=/opt/rocm
wget https://github.com/Yelp/dumb-init/releases/download/v1.2.0/dumb-init_1.2.0_amd64.deb && \
dpkg -i dumb-init_*.deb && rm dumb-init_*.deb && \
# Install packages for processing the performance results
RUN pip3 install --upgrade pip
RUN pip3 install sqlalchemy==1.4.46
RUN pip3 install pymysql
RUN pip3 install pandas==2.0.3
RUN pip3 install setuptools-rust
RUN pip3 install sshtunnel==0.4.0
# Setup ubsan environment to printstacktrace
ENV UBSAN_OPTIONS=print_stacktrace=1
ENV LC_ALL=C.UTF-8
ENV LANG=C.UTF-8
RUN groupadd -f render
pip3 install --upgrade pip && \
pip3 install sqlalchemy==1.4.46 pymysql pandas==2.0.3 setuptools-rust sshtunnel==0.4.0 && \
# Add render group
groupadd -f render && \
# Install the new rocm-cmake version
RUN git clone -b master https://github.com/ROCm/rocm-cmake.git && \
git clone -b master https://github.com/ROCm/rocm-cmake.git && \
cd rocm-cmake && mkdir build && cd build && \
cmake .. && cmake --build . && cmake --build . --target install
WORKDIR /
# Add alternative compilers, if necessary
ENV compiler_version=$compiler_version
ENV compiler_commit=$compiler_commit
RUN sh -c "echo compiler version = '$compiler_version'"
RUN sh -c "echo compiler commit = '$compiler_commit'"
ARG DISABLE_CACHE=0
RUN sh -c "echo compiler version = '$compiler_version'" && \
sh -c "echo compiler commit = '$compiler_commit'"
RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline-open" ] ) && [ "$compiler_commit" = "" ]; then \
RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline" ] ) && [ "$compiler_commit" = "" ]; then \
git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
cd llvm-project && mkdir build && cd build && \
cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
......@@ -145,16 +115,10 @@ RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd
else echo "using the release compiler"; \
fi
RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline-open" ] ) && [ "$compiler_commit" != "" ]; then \
RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline" ] ) && [ "$compiler_commit" != "" ]; then \
git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
make -j 8 ; \
else echo "using the release compiler"; \
fi
#clean-up the deb package
RUN sh -c "rm -rf amdgpu-install*"
#ENV HIP_CLANG_PATH='/llvm-project/build/bin'
#RUN sh -c "echo HIP_CLANG_PATH = '$HIP_CLANG_PATH'"
Dockerfile.compiler 0 → 100644
View file @ 2298a1a4
ARG BASE_DOCKER="rocm/composable_kernel:ck_ub20.04_rocm6.3"
FROM $BASE_DOCKER
ARG compiler_version=""
ARG compiler_commit=""
# Add alternative compilers, if necessary
ENV compiler_version=$compiler_version
ENV compiler_commit=$compiler_commit
RUN sh -c "echo compiler version = '$compiler_version'" && \
sh -c "echo compiler commit = '$compiler_commit'"
RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline" ] ) && [ "$compiler_commit" = "" ]; then \
git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
cd llvm-project && mkdir build && cd build && \
cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
make -j 16 ; \
else echo "using the release compiler"; \
fi
RUN if ( [ "$compiler_version" = "amd-staging" ] || [ "$compiler_version" = "amd-mainline" ] ) && [ "$compiler_commit" != "" ]; then \
git clone -b "$compiler_version" https://github.com/ROCm/llvm-project.git && \
cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
make -j 16 ; \
else echo "using the release compiler"; \
fi
Jenkinsfile
View file @ 2298a1a4
This diff is collapsed.
README.md
View file @ 2298a1a4
......@@ -26,23 +26,15 @@ The current CK library is structured into four layers:
## General information
To build our documentation locally, use the following code:
``` bash
cd docs
pip3 install -r sphinx/requirements.txt
python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
```
You can find a list of our developers and contributors on our [Contributors](/CONTRIBUTORS.md) page.
```note
If you use CK, cite us as follows:
* [Realizing Tensor Operators Using Coordinate Transformations and Tile Based Programming](???):
This paper will be available on arXiv soon.
* [CITATION.cff](/CITATION.cff)
```
* [CK supported operations](include/ck/README.md)
* [CK Tile supported operations](include/ck_tile/README.md)
* [CK wrapper](client_example/25_wrapper/README.md)
* [CK codegen](codegen/README.md)
* [CK profiler](profiler/README.md)
* [Examples (Custom use of CK supported operations)](example/README.md)
* [Client examples (Use of CK supported operations with instance factory)](client_example/README.md)
* [Terminology](/TERMINOLOGY.md)
* [Contributors](/CONTRIBUTORS.md)
CK is released under the **[MIT license](/LICENSE)**.
......@@ -137,6 +129,14 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa
You can find instructions for running ckProfiler in [profiler](/profiler).
* Build our documentation locally:
``` bash
cd docs
pip3 install -r sphinx/requirements.txt
python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
```
Note the `-j` option for building with multiple threads in parallel, which speeds up the build significantly.
However, `-j` launches unlimited number of threads, which can cause the build to run out of memory and
crash. On average, you should expect each thread to use ~2Gb of RAM.
......@@ -154,8 +154,7 @@ Additional cmake flags can be used to significantly speed-up the build:
other platforms have faster instances, such as `xdl` or `wmma`, available.
* `CK_USE_FP8_ON_UNSUPPORTED_ARCH` (default is OFF) must be set to ON in order to build instances,
such as `gemm_universal` and `gemm_multiply_multiply` for fp8 data type for GPU targets which do not
have native support for fp8 data type, such as gfx908 or gfx90a. These instances are useful on
such as `gemm_universal`, `gemm_universal_streamk` and `gemm_multiply_multiply` for fp8 data type for GPU targets which do not have native support for fp8 data type, such as gfx908 or gfx90a. These instances are useful on
architectures like the MI100/MI200 for the functional support only.
## Using sccache for building
......
TERMINOLOGY.md 0 → 100644
View file @ 2298a1a4
[Back to the main page](./README.md)
# Composable Kernel terminology
\ No newline at end of file
client_example/25_wrapper/README.md
View file @ 2298a1a4
[Back to the main page](../../README.md)
# Composable Kernel wrapper GEMM tutorial
This tutorial demonstrates how to implement matrix multiplication using Composable Kernel (CK)
wrapper. We present the base version of GEMM without most of the available optimizations; however,
it's worth noting that CK has kernels with different optimizations.
This tutorial demonstrates how to implement matrix multiplication using Composable Kernel (CK) wrapper. We present the base version of GEMM without most of the available optimizations; however, it's worth noting that CK has kernels with different optimizations.
To implement these optimizations, you can use the CK wrapper or directly use available instances in
CK. You can also refer to the
[optimized GEMM example](https://github.com/ROCm/composable_kernel/blob/develop/client_example/25_wrapper/wrapper_optimized_gemm.cpp),
that uses CK wrapper based on the
[`gridwise_gemm_xdlops_v2r3`](https://github.com/ROCm/composable_kernel/blob/develop/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp) implementation.
To implement these optimizations, you can use the CK wrapper or directly use available instances in CK. You can also refer to the [optimized GEMM example](https://github.com/ROCm/composable_kernel/blob/develop/client_example/25_wrapper/wrapper_optimized_gemm.cpp), that uses CK wrapper based on the [`gridwise_gemm_xdlops_v2r3`](https://github.com/ROCm/composable_kernel/blob/develop/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp) implementation.
The kernel definition should look similar to:
......
client_example/31_grouped_gemm_bf16Aint8B/grouped_gemm_multiply_bias_fastgelu_xdl_bf16_i8.cpp
View file @ 2298a1a4
......@@ -121,7 +121,7 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
constexpr ck::index_t NumDTensor = 2;
using GroupedGemmKernelArgument =
ck::tensor_operation::device::GroupedGemmTileLoopKernelArguments<NumDTensor>;
ck::tensor_operation::device::GroupedGemmKernelArgument<NumDTensor>;
std::vector<GroupedGemmKernelArgument> grouped_gemm_kernel_args_;
grouped_gemm_kernel_args_.reserve(group_count);
......
client_example/31_grouped_gemm_bf16Aint8B/grouped_gemm_multiply_xdl_bf16_i8.cpp
View file @ 2298a1a4
......@@ -120,7 +120,7 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
constexpr ck::index_t NumDTensor = 1;
using GroupedGemmKernelArgument =
ck::tensor_operation::device::GroupedGemmTileLoopKernelArguments<NumDTensor>;
ck::tensor_operation::device::GroupedGemmKernelArgument<NumDTensor>;
std::vector<GroupedGemmKernelArgument> grouped_gemm_kernel_args_;
grouped_gemm_kernel_args_.reserve(group_count);
......
client_example/README.md
View file @ 2298a1a4
[Back to the main page](../README.md)
# Composable Kernel client examples
##
Client application links to CK library, and therefore CK library needs to be installed before building client applications.
......
codegen/CMakeLists.txt
View file @ 2298a1a4
......@@ -7,6 +7,7 @@ set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
set(CK_ROOT ${CMAKE_CURRENT_SOURCE_DIR}/..)
configure_file(${CK_ROOT}/include/ck/config.h.in ${CK_ROOT}/include/ck/config.h)
find_package(ROCM)
include(ROCMInstallTargets)
......
codegen/README.md 0 → 100644
View file @ 2298a1a4
[Back to the main page](../README.md)
# Composable Kernel codegen
\ No newline at end of file
docs/sphinx/requirements.in
View file @ 2298a1a4
rocm-docs-core==1.8.4
rocm-docs-core==1.11.0
sphinxcontrib-bibtex==2.6.3
docs/sphinx/requirements.txt
View file @ 2298a1a4
......@@ -103,7 +103,7 @@ requests==2.32.3
# via
# pygithub
# sphinx
rocm-docs-core==1.8.4
rocm-docs-core==1.11.0
# via -r requirements.in
six==1.16.0
# via pybtex
......
example/01_gemm/CMakeLists.txt
View file @ 2298a1a4
......@@ -77,9 +77,16 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
add_example_executable(example_gemm_xdl_fp8_streamk_v3 gemm_xdl_fp8_streamk_v3.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_streamk_v3)
add_example_executable(example_gemm_xdl_fp16_fp8 gemm_xdl_fp16_fp8.cpp)
add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
add_custom_target(example_gemm_wmma)
add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
add_example_executable(example_gemm_wmma_bf16 gemm_wmma_bf16.cpp)
add_example_dependencies(example_gemm_wmma example_gemm_wmma_bf16)
add_example_executable(example_gemm_wmma_int8 gemm_wmma_int8.cpp)
add_example_dependencies(example_gemm_wmma example_gemm_wmma_int8)
example/01_gemm/common.hpp
View file @ 2298a1a4
......@@ -44,7 +44,7 @@ struct ProblemSizeStreamK final
ck::index_t StrideB = -1;
ck::index_t StrideC = -1;
ck::index_t NumSKBlocks = -1;
ck::index_t NumSKBlocks = -1; // number of stream-k blocks
};
struct ProblemSizeStreamK_universal final
{
......
example/01_gemm/gemm_wmma_bf16.cpp 0 → 100644
View file @ 2298a1a4
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp"
using ADataType = ck::bhalf_t;
using BDataType = ck::bhalf_t;
using AccDataType = float;
using CShuffleDataType = float;
using CDataType = ck::bhalf_t;
using ALayout = Row;
using BLayout = Col;
using CLayout = Row;
using AElementOp = PassThrough;
using BElementOp = PassThrough;
using CElementOp = PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// clang-format off
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmWmma_CShuffle
< ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
CShuffleDataType,
AElementOp,
BElementOp,
CElementOp,
GemmDefault,
1, // Prefetch stage
128, // BlockSize
64, // MPerBlock
128, // NPerBlock
64, // KPerBlock
2, // K1
16, // MPerWmma
16, // NPerWmma
2, // M-Repeat // M-PerWmma / M-Repeat = M-Wave
4, // N-Repeat // N-PerWmma / N-Repeat = N-Wave
S<4, 32, 1>,
S<1, 0, 2>,
S<1, 0, 2>,
2,
2,
2,
true,
S<4, 32, 1>,
S<1, 0, 2>,
S<1, 0, 2>,
2,
2,
2,
true,
1, // C shuffle (M Repeat) Per store
1, // C shuffle (N Repeat) Per store
S<1, 32, 1, 4>,
8>;
// clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_wmma_int8.cpp 0 → 100644
View file @ 2298a1a4
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp"
using ADataType = int8_t;
using BDataType = int8_t;
using AccDataType = int32_t;
using CShuffleDataType = int32_t;
using CDataType = int8_t;
using ALayout = Row;
using BLayout = Col;
using CLayout = Row;
using AElementOp = PassThrough;
using BElementOp = PassThrough;
using CElementOp = PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
// clang-format off
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmWmma_CShuffle
< ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
CShuffleDataType,
AElementOp,
BElementOp,
CElementOp,
GemmDefault,
1, // Prefetch stage
128, // BlockSize
64, // MPerBlock
128, // NPerBlock
64, // KPerBlock
2, // K1
16, // MPerWmma
16, // NPerWmma
2, // M-Repeat // M-PerWmma / M-Repeat = M-Wave
4, // N-Repeat // N-PerWmma / N-Repeat = N-Wave
S<4, 32, 1>,
S<1, 0, 2>,
S<1, 0, 2>,
2,
2,
2,
true,
S<4, 32, 1>,
S<1, 0, 2>,
S<1, 0, 2>,
2,
2,
2,
true,
1, // C shuffle (M Repeat) Per store
1, // C shuffle (N Repeat) Per store
S<1, 32, 1, 4>,
8>;
// clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example.inc"
int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
example/01_gemm/gemm_xdl_fp16_streamk_v3.cpp
View file @ 2298a1a4
......@@ -8,7 +8,7 @@
using ADataType = ck::half_t;
using BDataType = ck::half_t;
using AccDataType = float;
using CShuffleDataType = ck::half_t;
using CShuffleDataType = float;
using CDataType = ck::half_t;
using ALayout = Row;
......@@ -43,6 +43,17 @@ using DeviceGemmV2_Streamk_Instance =
using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
using ReferenceGemmInstanceGPU = ck::tensor_operation::device::ReferenceGemm<ALayout,
BLayout,
CLayout,
ADataType,
BDataType,
CDataType,
AccDataType,
AElementOp,
BElementOp,
CElementOp>;
#include "run_gemm_example_streamk_v2.inc"
int main(int argc, char* argv[]) { return !run_gemm_universal_streamk_example(argc, argv); }
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment