replace ccache with sccache, pin package versions

CMakeLists.txt

@@ -373,9 +373,10 @@ include_directories(BEFORE
 
 SET(BUILD_DEV ON CACHE BOOL "BUILD_DEV")
 if(BUILD_DEV)
-    add_compile_options(-Werror)
-    add_compile_options(-Weverything)
+    add_compile_options(-Werror -Weverything)
 endif()
+#add flags to reduce the size of binaries
+add_compile_options(-Oz -flto=thin)
 message("CMAKE_CXX_FLAGS: ${CMAKE_CXX_FLAGS}")
 
 add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -C ${CMAKE_CFG_INTDIR})
@@ -390,35 +391,27 @@ IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu
     file(READ "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}/CMakeLists.txt" cmake_instance)
     set(add_inst 0)
     if(("${cmake_instance}" MATCHES "fp8" OR "${cmake_instance}" MATCHES "_f8") AND DTYPES MATCHES "fp8")
-        #message("fp8 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "bf8" OR "${cmake_instance}" MATCHES "_b8") AND DTYPES MATCHES "bf8")
-        #message("bf8 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
-        #message("fp16 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "fp32" OR "${cmake_instance}" MATCHES "_f32") AND DTYPES MATCHES "fp32")
-        #message("fp32 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
-        #message("fp64 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "bf16" OR "${cmake_instance}" MATCHES "_b16") AND DTYPES MATCHES "bf16")
-        #message("bf16 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "int8" OR "${cmake_instance}" MATCHES "_i8") AND DTYPES MATCHES "int8")
-        #message("int8 instance found!")
         set(add_inst 1)
     endif()
     if(NOT "${cmake_instance}" MATCHES "DTYPES")
-        #message("instance should be built for all types!")
         set(add_inst 1)
     endif()
     if(add_inst EQUAL 1 OR NOT DEFINED DTYPES)

Dockerfile

@@ -26,10 +26,17 @@ RUN wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
 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
 
+## Sccache binary built from source for ROCm
+ARG SCCACHE_REPO_URL=http://compute-artifactory.amd.com/artifactory/rocm-generic-experimental/rocm-sccache
+ENV SCCACHE_INSTALL_LOCATION=/usr/local/.cargo/bin
+RUN mkdir -p ${SCCACHE_INSTALL_LOCATION} && \
+curl ${SCCACHE_REPO_URL}/portable/0.2.16/sccache-0.2.16-alpha.1-rocm --output ${SCCACHE_INSTALL_LOCATION}/sccache && \
+chmod +x ${SCCACHE_INSTALL_LOCATION}/sccache
+ENV PATH=$PATH:${SCCACHE_INSTALL_LOCATION}
+
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
     build-essential \
-    ccache \
     cmake \
     git \
     hip-rocclr \
@@ -61,7 +68,7 @@ RUN gunzip /usr/local/bin/ninja.gz
 RUN chmod a+x /usr/local/bin/ninja
 RUN git clone https://github.com/nico/ninjatracing.git
 # Update the cmake to the latest version
-RUN pip install --upgrade cmake
+RUN pip install --upgrade cmake=3.27.5
 
 # Setup ubsan environment to printstacktrace
 RUN ln -s /usr/bin/llvm-symbolizer-3.8 /usr/local/bin/llvm-symbolizer
@@ -75,10 +82,10 @@ ARG PREFIX=/opt/rocm
 # 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
+RUN pip3 install pymysql=1.4.6
+RUN pip3 install pandas=2.0.3
 RUN pip3 install setuptools-rust
-RUN pip3 install sshtunnel
+RUN pip3 install sshtunnel=0.4.0
 # Setup ubsan environment to printstacktrace
 ENV UBSAN_OPTIONS=print_stacktrace=1
 
@@ -114,6 +121,8 @@ RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" != "" ];
     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'"

Jenkinsfile

@@ -221,7 +221,7 @@ def cmake_build(Map conf=[:]){
     }
     if(env.CK_CCACHE)
     {
-        setup_args = " -DCMAKE_CXX_COMPILER_LAUNCHER='ccache' -DCMAKE_C_COMPILER_LAUNCHER='ccache' " + setup_args
+        setup_args = " -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache " + setup_args
     }
     echo "ccache server: ${env.CK_CCACHE}"
 

README.md

@@ -86,7 +86,6 @@ cmake
 If GPU_TARGETS is not set on the cmake command line, CK will be built for all targets supported by the 
 current compiler.
 
-
 Additional cmake flags can be used to significantly speed-up the build:
 
 INSTANCES_ONLY (by default is OFF) must be set to ON in order to build only the instances and library
@@ -98,11 +97,31 @@ of select data types only. Currently, building of int8 instances is taking a lot
 DL_KERNELS (by default is OFF) must be set to ON in order to build the gemm_dl and batched_gemm_multi_d_dl 
 instances. Those instances are only needed for the NAVI2x platforms.
 
+### Using sccache for building
+
+The default CK dockers come with pre-installed version of sccache which supports clang being used as hip-compiler
+" -x hip". Using sccache can help reduce the time to re-build the code from hours to 1 - 2 minutes. In order to
+invoke sccache, you need to run
+
+```bash
+ sccache --start-server
+```
+and add the following flags to the cmake command line:
+
+```bash
+ -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache
+```
+
 ### Build examples and tests
 
 ```bash
  make -j examples tests
- make test
+```
+
+### Build and run all examples and tests
+
+```bash
+ make -j check
 ```
 
 Instructions for running each individual examples are under [example](/example)