Merge remote-tracking branch 'origin/develop' into stream-k-initial-impl

.pre-commit-config.yaml

+repos:
+-   repo: local
+    hooks:
+    -   id: clang-format
+        name: clang-format
+        entry: clang-format-10 -i --style=file
+        language: system
+        types_or: [c++, inc]
+    -   id: copyright-year-checker
+        name: copyright-year-checker
+        entry: script/check_copyright_year.sh
+        verbose: false
+        language: script
+        types: [c++]

Dockerfile

@@ -12,24 +12,27 @@ RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
 RUN chmod 1777 /tmp
 RUN apt-get update
 RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
-RUN if [ "$ROCMVERSION" != "5.6" ]; then \
+
+ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
+RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
+
+RUN wget https://repo.radeon.com/amdgpu-install/5.6/ubuntu/focal/amdgpu-install_5.6.50600-1_all.deb  --no-check-certificate
+RUN apt-get update && \
+DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
+    ./amdgpu-install_5.6.50600-1_all.deb
+
+RUN if [ "$ROCMVERSION" != "5.7" ]; then \
         wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
-        sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"; \
-    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "" ]; then \
-         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amd-nonfree-radeon_20.04-1_all.deb" && \
-         apt update && apt-get install -y ./amd-nonfree-radeon_20.04-1_all.deb && \
-         amdgpu-repo --amdgpu-build=1567752 --rocm-build=compute-rocm-dkms-no-npi-hipclang/11914 && \
-         amdgpu-install -y --usecase=rocm --no-dkms; \
-    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "rc3" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
-         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.6-20.04-1_all.deb" && \
-         apt update && apt-get install -y ./amdgpu-install-internal_5.6-20.04-1_all.deb && \
-         sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.6 rel-45  > /etc/apt/sources.list.d/rocm-build.list' && \
-         amdgpu-repo --amdgpu-build=1602498 && amdgpu-install -y --usecase=rocm --no-dkms; \
+        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 [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
+         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
+         apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
+         amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354; \
     fi
 
-RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
-RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
+RUN amdgpu-install -y --usecase=rocm --no-dkms
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \

Jenkinsfile

@@ -11,6 +11,20 @@ def show_node_info() {
     """
 }
 
+def nthreads() {
+    def nproc = sh(returnStdout: true, script: 'nproc')
+    echo "Number of cores: ${nproc}"
+    def n = nproc.toInteger()
+    if (n > 32){
+        n /= 2
+    }
+    if (n > 64){
+        n = 64
+    }
+    echo "Number of threads used for building: ${n}"
+    return n
+}
+
 def runShell(String command){
     def responseCode = sh returnStatus: true, script: "${command} > tmp.txt"
     def output = readFile(file: "tmp.txt")
@@ -19,7 +33,7 @@ def runShell(String command){
 
 def getDockerImageName(){
     def img
-    if (params.ROCMVERSION != "5.6"){
+    if (params.ROCMVERSION != "5.7"){
        if (params.COMPILER_VERSION == "") {
            img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
        }
@@ -219,7 +233,8 @@ def cmake_build(Map conf=[:]){
         """
     def setup_cmd = conf.get("setup_cmd", "${cmake_envs} cmake ${setup_args}   .. ")
     // reduce parallelism when compiling, clang uses too much memory
-    def build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make  -j\$(( \$(nproc) / 2 )) ${config_targets}")
+    def nt = nthreads()
+    def build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make  -j${nt} ${config_targets}")
     def execute_cmd = conf.get("execute_cmd", "")
 
     def cmd = conf.get("cmd", """
@@ -461,7 +476,7 @@ def Build_CK(Map conf=[:]){
                         else{
                             echo "GPU is OK"
                         }
-                        if ( runShell('grep -n "gfx1030" clinfo.log') ){
+                        if ( runShell('grep -n "gfx1030" clinfo.log') || runShell('grep -n "gfx1101" clinfo.log') ){
                             navi_node = 1
                         }
                     }
@@ -482,7 +497,7 @@ def Build_CK(Map conf=[:]){
                         else{
                             echo "GPU is OK"
                         }
-                        if ( runShell('grep -n "gfx1030" clinfo.log') ){
+                        if ( runShell('grep -n "gfx1030" clinfo.log') || runShell('grep -n "gfx1101" clinfo.log') ){
                             navi_node = 1
                         }
                     }
@@ -494,7 +509,8 @@ def Build_CK(Map conf=[:]){
                     cmake_build(conf)
                     dir("build"){
                         //run tests and examples
-                        sh 'make -j\$(( \$(nproc) / 2 )) check'
+                        def nt = nthreads()	
+                        sh 'make -j${nt} check'
                         if (navi_node == 0 ){
                             //we only need the ckProfiler to run the performance tests, so we pack and stash it
                             //do not stash profiler on Navi nodes
@@ -598,7 +614,7 @@ 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
-                                              0 21 * * * % ROCMVERSION=5.5;COMPILER_VERSION=release;COMPILER_COMMIT=
+                                              0 21 * * * % ROCMVERSION=5.6;COMPILER_VERSION=;COMPILER_COMMIT=
                                               0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
 
 pipeline {
@@ -717,7 +733,7 @@ pipeline {
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
                     }
                 }
-                stage("Build CK and run Tests on Navi")
+                stage("Build CK and run Tests on Navi21")
                 {
                     when {
                         beforeAgent true

README.md

@@ -109,6 +109,24 @@ make install
 
 Instructions for using CK as a pre-built kernel library are under [client_example](/client_example)
 
+## Contributing
+
+When you contribute to Composable Kernel, make sure to run `clang-format` on all the changed files. We highly recommend using git hooks that are managed by the `pre-commit` framework. To install hooks, run:
+
+```bash
+sudo script/install_precommit.sh
+```
+
+This way, `pre-commit` will add the appropriate hooks to your local repository and automatically run `clang-format` (and possibly additional checks) before any commit is created.
+
+If you need to uninstall hooks from the repository, you can do so by running the following command:
+
+```bash
+script/uninstall_precommit.sh
+```
+
+If for any reason, you need to temporarily disable precommit hooks, you can add the `--no-verify` option to the `git commit` command.
+
 ## Caveat
 ### Kernel Timing and Verification
 

example/01_gemm/CMakeLists.txt

@@ -45,3 +45,8 @@ if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS
 endif()
 
 add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
+
+if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+  add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
+endif()

example/01_gemm/gemm_xdl_f8.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::f8_t;
+using BDataType        = ck::f8_t;
+using CDataType        = ck::f8_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::f8_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::Default;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+// ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+// ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,              16>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<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_wavelet_fp16.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp"
 
 using ADataType        = ck::half_t;
 using BDataType        = ck::half_t;

example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp

@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 

example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp

@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 

example/09_convnd_fwd/convnd_fwd_dl_int8.cpp

@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 

example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp

@@ -17,7 +17,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"

example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp

@@ -17,7 +17,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"

example/34_batchnorm/CMakeLists.txt

 add_example_executable(example_batchnorm_forward_training batchnorm_forward_training_nhwc.cpp)
+add_example_executable(example_batchnorm_forward_training_obsolete batchnorm_forward_training_nhwc_obsolete.cpp)
 add_example_executable(example_batchnorm_forward_inferring batchnorm_forward_inferring_nhwc.cpp)
 add_example_executable(example_batchnorm_backward batchnorm_backward_nhwc.cpp)

example/34_batchnorm/batchnorm_forward_training_nhwc.cpp

@@ -414,7 +414,7 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
         (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
 
         y_dev.FromDevice(y.mData.data());
-        pass = pass && ck::utils::check_err(y, y_ref);
+        pass = pass && ck::utils::check_err(y, y_ref, "Incorrect normalized output values");
 
         if(updateMovingAverage)
         {
@@ -424,8 +424,12 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
             resultRunningMean_dev.FromDevice(resultRunningMean.mData.data());
             resultRunningVariance_dev.FromDevice(resultRunningVariance.mData.data());
 
-            pass = pass && ck::utils::check_err(resultRunningMean, resultRunningMean_ref);
-            pass = pass && ck::utils::check_err(resultRunningVariance, resultRunningVariance_ref);
+            pass = pass && ck::utils::check_err(resultRunningMean,
+                                                resultRunningMean_ref,
+                                                "Incorrect running mean values");
+            pass = pass && ck::utils::check_err(resultRunningVariance,
+                                                resultRunningVariance_ref,
+                                                "Incorrect running variance values");
         };
 
         if(saveMeanAndInvVariance)
@@ -438,8 +442,11 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
             resultSaveMean_dev.FromDevice(resultSaveMean.mData.data());
             resultSaveInvVariance_dev.FromDevice(resultSaveInvVariance.mData.data());
 
-            pass = pass && ck::utils::check_err(resultSaveMean, resultSaveMean_ref);
-            pass = pass && ck::utils::check_err(resultSaveInvVariance, resultSaveInvVariance_ref);
+            pass = pass && ck::utils::check_err(
+                               resultSaveMean, resultSaveMean_ref, "Incorrect saved mean values");
+            pass = pass && ck::utils::check_err(resultSaveInvVariance,
+                                                resultSaveInvVariance_ref,
+                                                "Incorrect saved invvariance values");
         };
     };
 

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType   = int8_t;
 using WeiDataType  = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType   = int8_t;
 using WeiDataType  = int8_t;

example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp

@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;