Skip to content

GitLab

Commit 32806d5f authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'amd-develop' into amd-master

parents e70a4d19 d0f355a3
Hide whitespace changes
Inline Side-by-side
Showing with 493 additions and 36 deletions
.clang-tidy
View file @ 32806d5f
CheckOptions:
- key: bugprone-reserved-identifier.AllowedIdentifiers
value: '__HIP_PLATFORM_HCC__;__HIP_ROCclr__'
value: '__HIP_PLATFORM_HCC__;__HIP_PLATFORM_AMD__;__HIP_ROCclr__'
.github/ISSUE_TEMPLATE/config.yml 0 → 100644
View file @ 32806d5f
blank_issues_enabled: true
.github/ISSUE_TEMPLATE/issue_report.yml 0 → 100644
View file @ 32806d5f
name: Issue Report
description: File a report for ROCm related issues on Linux and Windows. For issues pertaining to documentation or non-bug related, please open a blank issue located below.
title: "[Issue]: "
body:
- type: markdown
attributes:
value: |
Thank you for taking the time to fill out this report!
You can acquire your OS, CPU, GPU (for filling out this report) with the following commands:
Linux:
echo "OS:" && cat /etc/os-release | grep -E "^(NAME=|VERSION=)";
echo "CPU: " && cat /proc/cpuinfo | grep "model name" | sort --unique;
echo "GPU:" && /opt/rocm/bin/rocminfo | grep -E "^\s*(Name|Marketing Name)";
Windows:
(Get-WmiObject Win32_OperatingSystem).Version
(Get-WmiObject win32_Processor).Name
(Get-WmiObject win32_VideoController).Name
- type: textarea
attributes:
label: Problem Description
description: Describe the issue you encountered.
validations:
required: true
- type: input
attributes:
label: Operating System
description: What is the name and version number of the OS?
placeholder: "e.g. Ubuntu 22.04.3 LTS (Jammy Jellyfish)"
validations:
required: true
- type: input
attributes:
label: CPU
description: What CPU did you encounter the issue on?
placeholder: "e.g. AMD Ryzen 9 5900HX with Radeon Graphics"
validations:
required: true
- type: dropdown
attributes:
label: GPU
description: What GPU(s) did you encounter the issue on (you can select multiple GPUs from the list)
multiple: true
options:
- AMD Instinct MI300X
- AMD Instinct MI300A
- AMD Instinct MI300
- AMD Instinct MI250X
- AMD Instinct MI250
- AMD Instinct MI210
- AMD Instinct MI100
- AMD Instinct MI50
- AMD Instinct MI25
- AMD Radeon Pro V620
- AMD Radeon Pro VII
- AMD Radeon RX 7900 XTX
- AMD Radeon VII
- AMD Radeon Pro W7900
- AMD Radeon Pro W7800
- AMD Radeon Pro W6800
- AMD Radeon Pro W6600
- AMD Radeon Pro W5500
- AMD Radeon RX 7900 XT
- AMD Radeon RX 7600
- AMD Radeon RX 6950 XT
- AMD Radeon RX 6900 XT
- AMD Radeon RX 6800 XT
- AMD Radeon RX 6800
- AMD Radeon RX 6750
- AMD Radeon RX 6700 XT
- AMD Radeon RX 6700
- AMD Radeon RX 6650 XT
- AMD Radeon RX 6600 XT
- AMD Radeon RX 6600
- Other
validations:
required: true
- type: input
attributes:
label: Other
description: If you selected Other, please specify
- type: dropdown
attributes:
label: ROCm Version
description: What version(s) of ROCm did you encounter the issue on?
multiple: true
options:
- ROCm 6.0.0
- ROCm 5.7.1
- ROCm 5.7.0
- ROCm 5.6.1
- ROCm 5.6.0
- ROCm 5.5.1
- ROCm 5.5.0
validations:
required: true
- type: dropdown
attributes:
label: ROCm Component
description: (Optional) If this issue relates to a specific ROCm component, it can be mentioned here.
multiple: true
options:
- Other
- AMD Common Language Runtime
- AMD MIGraphX
- AMD System Management Interface
- amdgpu KCL/autoconf
- amdgpu Kernel-mode GPU Driver
- amdgpu-install
- AOMP
- AOMP Extras
- AqlProfile
- build-infra
- chelsio
- clang-ocl
- Composable Kernel
- dkms
- docker / ROCm-docker
- flang
- gpuburn
- half
- HIP
- HIP Examples
- hipBLAS
- hipBLASLt
- HIPCC
- hipCUB
- hip-examples-private
- hipFFT
- hipfort
- HIPIFY
- hipRAND
- hipSOLVER
- hipSPARSE
- hipSPARSELt
- hipTensor
- hip-tests
- HSA Runtime
- infrastructure
- jenkins-utils
- libdrm
- Linux BPI packaging framework
- llvm-project
- Mesa
- meta
- MIOpen
- MIVisionX
- ml-framework-ci
- MLSEQA_TestRepo
- OpenCL API C++ Bindings
- OpenCL API Headers
- OpenCL Conformance Test Suite
- OpenCL ICD Loader
- perftest-p2p
- prototype
- RCCL
- rccl-rdma-sharp-plugins
- rocALUTION
- rocBLAS
- ROCdbgapi
- ROCdebug-agent
- rocFFT
- ROCgdb
- ROCK
- ROCm Documentation/Website
- ROCm Data Center Tool
- ROCm Examples
- ROCm for Windows
- ROCm Performance Primitives
- ROCm System Management Interface Library
- ROCm Thrust
- ROCm Validation Suite
- rocm_bandwidth_test
- rocm-cmake
- rocm-core
- rocm-docs-core
- rocminfo
- rocMLIR
- rocmtools
- rocPRIM
- rocprofiler
- rocRAND
- ROCR-Runtime
- rocSOLVER
- rocSPARSE
- roctracer
- ROCT-Thunk-Interface
- rocWMMA
- Tensile
- umr
- ibv_rc_pingpong-amd
- mellanox
- mpitest
- Pytorch
- Tensorflow
- APEX
- torchvision
- Magma
- type: textarea
attributes:
label: Steps to Reproduce
description: (Optional) Detailed steps to reproduce the issue.
validations:
required: false
- type: textarea
attributes:
label: (Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support
description: The output of rocminfo --support could help to better address the problem.
validations:
required: false
- type: textarea
attributes:
label: Additional Information
description: (Optional) Any additional information that is relevant, e.g. relevant environment variables, dockerfiles, log files, dmesg output (on Linux), etc.
validations:
required: false
.github/dependabot.yml
View file @ 32806d5f
......@@ -10,3 +10,9 @@ updates:
open-pull-requests-limit: 10
schedule:
interval: "daily"
labels:
- "documentation"
- "dependencies"
- "ci:docs-only"
reviewers:
- "samjwu"
.gitignore
View file @ 32806d5f
......@@ -54,5 +54,4 @@ _images/
_static/
_templates/
_toc.yml
docBin/
_doxygen/
.readthedocs.yaml
View file @ 32806d5f
......@@ -3,11 +3,6 @@
version: 2
build:
os: ubuntu-22.04
tools:
python: "3.8"
sphinx:
configuration: docs/conf.py
......@@ -16,3 +11,8 @@ formats: [htmlzip, pdf, epub]
python:
install:
- requirements: docs/sphinx/requirements.txt
build:
os: ubuntu-22.04
tools:
python: "3.8"
CHANGELOG.md
View file @ 32806d5f
......@@ -19,6 +19,7 @@ None
- Support for NHWGC (2D and 3D) grouped convolution backward weight (#769 #804)
- Support for bf16/f32/f16 and NHWGC (2D and 3D) grouped convolution backward data (#757 #799)
- Support for Batched Gemm DL (#732)
- Introduce wrapper sublibrary (limited functionality). (#1071, #1098)
### Changes
- Changed the grouped convolution API to maintain consistency with other convolution kernels (#817)
......
CITATION.cff
View file @ 32806d5f
......@@ -59,9 +59,9 @@ authors:
family-names: Zhou
- given-names: Jianfeng
family-names: Yan
repository-code: 'https://github.com/ROCmSoftwarePlatform/composable_kernel'
repository-code: 'https://github.com/ROCm/composable_kernel'
abstract: 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 progarmming languages, like HIP C++.
keywords:
- 'CK, Composable Kernel, Tensor Coordinate Transformation'
license: MIT
license-url: https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/7fc3ed761aa35709d87c8fbbe41dd368648b3541/LICENSE
license-url: https://github.com/ROCm/composable_kernel/blob/7fc3ed761aa35709d87c8fbbe41dd368648b3541/LICENSE
CMakeLists.txt
View file @ 32806d5f
......@@ -61,6 +61,7 @@ endif()
#for f8/bf8_t type
add_compile_options(-Wno-bit-int-extension)
add_compile_options(-Wno-pass-failed)
if(DL_KERNELS)
add_definitions(-DDL_KERNELS)
......@@ -146,6 +147,33 @@ if(${hip_VERSION_FLAT} GREATER 500723302)
add_compile_options(-fno-offload-uniform-block)
endif()
#
# Seperate linking jobs from compiling
# Too many concurrent linking jobs can break the build
# Copied from LLVM
set(CK_PARALLEL_LINK_JOBS "" CACHE STRING
"Define the maximum number of concurrent link jobs (Ninja only).")
if(CMAKE_GENERATOR MATCHES "Ninja")
if(CK_PARALLEL_LINK_JOBS)
set_property(GLOBAL APPEND PROPERTY JOB_POOLS link_job_pool=${CK_PARALLEL_LINK_JOBS})
set(CMAKE_JOB_POOL_LINK link_job_pool)
endif()
elseif(CK_PARALLEL_LINK_JOBS)
message(WARNING "Job pooling is only available with Ninja generators.")
endif()
# Similar for compiling
set(CK_PARALLEL_COMPILE_JOBS "" CACHE STRING
"Define the maximum number of concurrent compile jobs (Ninja only).")
if(CMAKE_GENERATOR MATCHES "Ninja")
if(CK_PARALLEL_COMPILE_JOBS)
set_property(GLOBAL APPEND PROPERTY JOB_POOLS compile_job_pool=${CK_PARALLEL_COMPILE_JOBS})
set(CMAKE_JOB_POOL_COMPILE compile_job_pool)
endif()
elseif(CK_PARALLEL_COMPILE_JOBS)
message(WARNING "Job pooling is only available with Ninja generators.")
endif()
option(USE_BITINT_EXTENSION_INT4, "Whether to enable clang's BitInt extension to provide int4 data type." OFF)
option(USE_OPT_NAVI3X, "Whether to enable LDS cumode and Wavefront32 mode for NAVI3X silicons." OFF)
......@@ -216,7 +244,11 @@ if( DEFINED CK_OVERRIDE_HIP_VERSION_PATCH )
endif()
message(STATUS "Build with HIP ${HIP_VERSION}")
link_libraries(hip::device)
add_compile_definitions(__HIP_PLATFORM_HCC__=1)
if(CK_hip_VERSION VERSION_GREATER_EQUAL 6.0.23494)
add_compile_definitions(__HIP_PLATFORM_AMD__=1)
else()
add_compile_definitions(__HIP_PLATFORM_HCC__=1)
endif()
## tidy
include(EnableCompilerWarnings)
......
Dockerfile
View file @ 32806d5f
FROM ubuntu:20.04
ARG DEBIAN_FRONTEND=noninteractive
ARG ROCMVERSION=5.7
ARG ROCMVERSION=6.0
ARG compiler_version=""
ARG compiler_commit=""
......@@ -16,8 +16,8 @@ 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 wget https://repo.radeon.com/amdgpu-install/5.7/ubuntu/focal/amdgpu-install_5.7.50700-1_all.deb --no-check-certificate
RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated ./amdgpu-install_5.7.50700-1_all.deb
RUN wget https://repo.radeon.com/amdgpu-install/6.0/ubuntu/focal/amdgpu-install_6.0.60000-1_all.deb --no-check-certificate
RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated ./amdgpu-install_6.0.60000-1_all.deb
RUN 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] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
......
Jenkinsfile
View file @ 32806d5f
......@@ -33,7 +33,7 @@ def runShell(String command){
def getDockerImageName(){
def img
if (params.ROCMVERSION != "6.0"){
if (params.ROCMVERSION != "6.1"){
if (params.COMPILER_VERSION == "") {
img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
}
......@@ -302,7 +302,7 @@ def buildHipClangJob(Map conf=[:]){
def retimage
(retimage, image) = getDockerImage(conf)
gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCmSoftwarePlatform', repo: 'composable_kernel') {
gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel') {
withDockerContainer(image: image, args: dockerOpts + ' -v=/var/jenkins/:/var/jenkins') {
timeout(time: 5, unit: 'HOURS')
{
......@@ -355,7 +355,7 @@ def runCKProfiler(Map conf=[:]){
def variant = env.STAGE_NAME
def retimage
gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCmSoftwarePlatform', repo: 'composable_kernel') {
gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel') {
try {
(retimage, image) = getDockerImage(conf)
withDockerContainer(image: image, args: dockerOpts) {
......@@ -487,7 +487,7 @@ def Build_CK(Map conf=[:]){
def retimage
def navi_node = 0
gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCmSoftwarePlatform', repo: 'composable_kernel') {
gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel') {
try {
(retimage, image) = getDockerImage(conf)
withDockerContainer(image: image, args: dockerOpts) {
......@@ -553,7 +553,7 @@ def Build_CK(Map conf=[:]){
sh """#!/bin/bash
rm -rf "${params.hipTensor_branch}".zip
rm -rf hipTensor-"${params.hipTensor_branch}"
wget https://github.com/ROCmSoftwarePlatform/hipTensor/archive/refs/heads/"${params.hipTensor_branch}".zip
wget https://github.com/ROCm/hipTensor/archive/refs/heads/"${params.hipTensor_branch}".zip
unzip -o "${params.hipTensor_branch}".zip
"""
dir("hipTensor-${params.hipTensor_branch}"){
......@@ -605,7 +605,7 @@ def process_results(Map conf=[:]){
def variant = env.STAGE_NAME
def retimage
gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCmSoftwarePlatform', repo: 'composable_kernel') {
gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCm', repo: 'composable_kernel') {
try {
(retimage, image) = getDockerImage(conf)
}
......@@ -655,8 +655,8 @@ def process_results(Map conf=[:]){
}
//launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCMVERSION=5.7;COMPILER_VERSION=
0 21 * * * % ROCMVERSION=5.7;COMPILER_VERSION=;COMPILER_COMMIT=
CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCMVERSION=6.0;COMPILER_VERSION=
0 21 * * * % ROCMVERSION=6.0;COMPILER_VERSION=;COMPILER_COMMIT=
0 19 * * * % BUILD_DOCKER=true;DL_KERNELS=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=;USE_SCCACHE=false
0 17 * * * % BUILD_DOCKER=true;DL_KERNELS=true;COMPILER_VERSION=amd-mainline-open;COMPILER_COMMIT=;USE_SCCACHE=false''' : ""
......@@ -675,8 +675,8 @@ pipeline {
description: "Force building docker image (default: false), set to true if docker image needs to be updated.")
string(
name: 'ROCMVERSION',
defaultValue: '5.7',
description: 'Specify which ROCM version to use: 5.7 (default).')
defaultValue: '6.0',
description: 'Specify which ROCM version to use: 6.0 (default).')
string(
name: 'COMPILER_VERSION',
defaultValue: '',
......@@ -703,8 +703,8 @@ pipeline {
description: "Use the CK build to verify hipTensor build and tests (default: ON)")
string(
name: 'hipTensor_branch',
defaultValue: 'develop',
description: 'Specify which branch of hipTensor to use (default: develop)')
defaultValue: 'mainline',
description: 'Specify which branch of hipTensor to use (default: mainline)')
booleanParam(
name: "USE_SCCACHE",
defaultValue: true,
......@@ -768,8 +768,15 @@ pipeline {
}
agent{ label rocmnode("gfx908 || gfx90a") }
environment{
setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" -DCMAKE_EXE_LINKER_FLAGS=" -L ${env.WORKSPACE}/script -T hip_fatbin_insert " """
execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
setup_args = """ -DCMAKE_INSTALL_PREFIX=../install \
-DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" \
-DCMAKE_EXE_LINKER_FLAGS=" -L ${env.WORKSPACE}/script -T hip_fatbin_insert " \
-DCMAKE_CXX_FLAGS=" -O3 " """
execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && \
cmake -DCMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" \
-DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" \
-DCMAKE_CXX_COMPILER="${build_compiler()}" \
-DCMAKE_CXX_FLAGS=" -O3 " .. && make -j """
}
steps{
Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
......@@ -784,8 +791,12 @@ pipeline {
}
agent{ label rocmnode("gfx908 || gfx90a") }
environment{
setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a" """
execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a" -DCMAKE_CXX_FLAGS=" -O3 " """
execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && \
cmake -DCMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" \
-DGPU_TARGETS="gfx908;gfx90a" \
-DCMAKE_CXX_COMPILER="${build_compiler()}" \
-DCMAKE_CXX_FLAGS=" -O3 " .. && make -j """
}
steps{
Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
......@@ -800,8 +811,12 @@ pipeline {
}
agent{ label rocmnode("navi21") }
environment{
setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1030" -DDL_KERNELS=ON """
execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1030" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1030" -DDL_KERNELS=ON -DCMAKE_CXX_FLAGS=" -O3 " """
execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && \
cmake -DCMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" \
-DGPU_TARGETS="gfx1030" \
-DCMAKE_CXX_COMPILER="${build_compiler()}" \
-DCMAKE_CXX_FLAGS=" -O3 " .. && make -j """
}
steps{
Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
......@@ -816,8 +831,12 @@ pipeline {
}
agent{ label rocmnode("navi32") }
environment{
setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1101" -DDL_KERNELS=ON """
execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1101" -DDL_KERNELS=ON -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1101" -DDL_KERNELS=ON -DCMAKE_CXX_FLAGS=" -O3 " """
execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && \
cmake -DCMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" \
-DGPU_TARGETS="gfx1101" \
-DCMAKE_CXX_COMPILER="${build_compiler()}" \
-DCMAKE_CXX_FLAGS=" -O3 " .. && make -j """
}
steps{
Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
......
README.md
View file @ 32806d5f
......@@ -32,7 +32,6 @@ 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.
page.
```note
If you use CK, cite us as follows:
......@@ -71,7 +70,7 @@ Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composa
3. Clone CK source code from the GitHub repository and start the build:
```bash
git clone https://github.com/ROCmSoftwarePlatform/composable_kernel.git && \
git clone https://github.com/ROCm/composable_kernel.git && \
cd composable_kernel && \
mkdir build && \
cd build
......
client_example/01_gemm/gemm.cpp
View file @ 32806d5f
......@@ -185,6 +185,7 @@ int main(int argc, char* argv[])
<< best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
// run the best intance
if(found)
{
auto& op_ptr = op_ptrs[best_op_id];
......
client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp
View file @ 32806d5f
......@@ -204,6 +204,7 @@ int main(int argc, char* argv[])
<< best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
// run the best intance
if(found)
{
auto& op_ptr = op_ptrs[best_op_id];
......
client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
View file @ 32806d5f
......@@ -197,6 +197,7 @@ int main(int argc, char* argv[])
<< best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
// run the best intance
if(found)
{
auto& op_ptr = op_ptrs[best_op_id];
......
client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
View file @ 32806d5f
......@@ -190,6 +190,7 @@ int main(int argc, char* argv[])
<< best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
// run the best intance
if(found)
{
auto& op_ptr = op_ptrs[best_op_id];
......
client_example/03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp
View file @ 32806d5f
......@@ -200,6 +200,7 @@ int main(int argc, char* argv[])
<< best_op_name << std::endl;
// run the best intance
if(found)
{
auto& op_ptr = op_ptrs[best_op_id];
......
client_example/05_layernorm/CMakeLists.txt
View file @ 32806d5f
add_executable(client_layernorm2d_bwd_data layernorm2d_bwd_data.cpp)
target_link_libraries(client_layernorm2d_bwd_data PRIVATE composable_kernel::device_other_operations)
add_executable(client_layernorm2d_fwd layernorm2d_fwd.cpp)
target_link_libraries(client_layernorm2d_fwd PRIVATE composable_kernel::device_other_operations)
......
client_example/05_layernorm/layernorm2d_bwd_data.cpp 0 → 100644
View file @ 32806d5f
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iomanip>
#include <vector>
#include <iostream>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
#include "ck/tensor_operation/gpu/device/device_normalization_bwd_data.hpp"
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
#include "ck/library/tensor_operation_instance/gpu/layernorm_bwd_data.hpp"
using DYDataType = float;
using XDataType = float;
using GammaDataType = float;
using MeanInvStdDataType = float;
using DXDataType = float;
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
struct SimpleDeviceMem
{
SimpleDeviceMem() = delete;
SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
{
(void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
}
void* GetDeviceBuffer() { return p_mem_; }
~SimpleDeviceMem() { (void)hipFree(p_mem_); }
void* p_mem_;
};
int main(int argc, char* argv[])
{
ck::index_t M = 1024;
ck::index_t N = 1024;
SimpleDeviceMem dy_dev(sizeof(DYDataType) * M * N);
SimpleDeviceMem x_dev(sizeof(XDataType) * M * N);
SimpleDeviceMem gamma_dev(sizeof(GammaDataType) * N);
SimpleDeviceMem mean_dev(sizeof(MeanInvStdDataType) * M);
SimpleDeviceMem inv_std_dev(sizeof(MeanInvStdDataType) * M);
SimpleDeviceMem dx_dev(sizeof(DXDataType) * M * N);
using DeviceOp = ck::tensor_operation::device::DeviceNormalizationBwdData<DYDataType,
XDataType,
GammaDataType,
MeanInvStdDataType,
DXDataType,
Rank,
NumReduceDim>;
// get device op instances
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
std::string best_op_name;
bool found = false;
int best_op_id = -1;
float best_ave_time = std::numeric_limits<float>::max();
float best_gb_per_sec = 0;
// profile device operation instances
std::cout << "Run all instances and do timing" << std::endl;
for(int i = 0; i < op_ptrs.size(); ++i)
{
auto& op_ptr = op_ptrs[i];
auto argument_ptr = op_ptr->MakeArgumentPointer({M, N}, // lengths
{N, 1}, // dyStrides
{N, 1}, // xStrides
{0, 1}, // gammaStrides
{1, 0}, // meanStrides
{1, 0}, // invStdStrides
{N, 1}, // dxStrides
{1}, // reduceDims
dy_dev.GetDeviceBuffer(),
x_dev.GetDeviceBuffer(),
gamma_dev.GetDeviceBuffer(),
mean_dev.GetDeviceBuffer(),
inv_std_dev.GetDeviceBuffer(),
dx_dev.GetDeviceBuffer());
auto invoker_ptr = op_ptr->MakeInvokerPointer();
std::string op_name = op_ptr->GetTypeString();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
SimpleDeviceMem workspace(workspace_sz);
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
std::size_t num_byte = sizeof(DYDataType) * M * N + sizeof(XDataType) * M * N +
sizeof(GammaDataType) * N + sizeof(MeanInvStdDataType) * M * 2 +
sizeof(DXDataType) * M * N;
float gb_per_sec = num_byte / 1.E6 / ave_time;
std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
<< op_name << std::endl;
if(ave_time < best_ave_time)
{
found = true;
best_op_id = i;
best_op_name = op_name;
best_ave_time = ave_time;
best_gb_per_sec = gb_per_sec;
}
}
else
{
std::cout << op_name << " does not support this problem" << std::endl;
}
}
std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
<< best_op_name << std::endl;
// run the best intance
if(found)
{
auto& op_ptr = op_ptrs[best_op_id];
std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
<< std::endl;
auto argument_ptr = op_ptr->MakeArgumentPointer({M, N}, // lengths
{N, 1}, // dyStrides
{N, 1}, // xStrides
{0, 1}, // gammaStrides
{1, 0}, // meanStrides
{1, 0}, // invStdStrides
{N, 1}, // dxStrides
{1}, // reduceDims
dy_dev.GetDeviceBuffer(),
x_dev.GetDeviceBuffer(),
gamma_dev.GetDeviceBuffer(),
mean_dev.GetDeviceBuffer(),
inv_std_dev.GetDeviceBuffer(),
dx_dev.GetDeviceBuffer());
auto invoker_ptr = op_ptr->MakeInvokerPointer();
if(op_ptr->IsSupportedArgument(argument_ptr.get()))
{
size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
SimpleDeviceMem workspace(workspace_sz);
op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
}
std::cout << "Done" << std::endl;
}
return 0;
}
client_example/05_layernorm/layernorm2d_fwd.cpp
View file @ 32806d5f
......@@ -16,7 +16,7 @@ using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using SaveMeanInvStdDataType = float;
using SaveMeanInvStdDataType = ck::half_t;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
#define SAVE_MEAN_INV_STD
......@@ -150,6 +150,7 @@ int main(int argc, char* argv[])
<< best_op_name << std::endl;
// run the best intance
if(found)
{
auto& op_ptr = op_ptrs[best_op_id];
std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
......
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