Merge remote-tracking branch 'origin/cderb/gtest_package' into sl/con…

…v_igemm_dynamic_dlops_env_var_update

CMakeLists.txt

@@ -91,8 +91,7 @@ option(BUILD_SHARED_LIBS "Create shared libraries" ON)
 
 if(MIOPEN_ENABLE_SQLITE)
     # MIOpen now depends on SQLite as well
-    find_package(PkgConfig)
-    pkg_check_modules(SQLITE3 REQUIRED sqlite3)
+    find_package(SQLite3 REQUIRED)
 endif()
 find_package(BZip2)
 find_package(nlohmann_json 3.9.1 REQUIRED)

Dockerfile

@@ -111,13 +111,4 @@ RUN pip3 install -r /doc-requirements.txt
 # Composable Kernel requires this version cmake
 RUN pip3 install --upgrade cmake==3.27.5
 
-# Use parallel job to accelerate tensile build
-# Workaround for Tensile with TargetID feature
-ARG USE_TARGETID="OFF"
-RUN if [ "$USE_TARGETID" = "ON" ] ; then export HIPCC_LINK_FLAGS_APPEND='-O3 -parallel-jobs=4' && export HIPCC_COMPILE_FLAGS_APPEND='-O3 -Wno-format-nonliteral -parallel-jobs=4' && rm -f /usr/bin/hipcc; fi
-
-# install last released miopentensile in default (master), install latest commits when MIOTENSILE_VER="latest" (develop)
-ARG MIOTENSILE_VER="default"
-RUN if [ "$USE_TARGETID" = "OFF" ] ; then echo "MIOpenTensile is not installed."; elif [ "$MIOTENSILE_VER" = "latest" ] ; then cget -p $PREFIX install ROCmSoftwarePlatform/MIOpenTensile@94a9047741d16a8eccd290131b78fb1aa69cdcdf; else cget -p $PREFIX install ROCmSoftwarePlatform/MIOpenTensile@94a9047741d16a8eccd290131b78fb1aa69cdcdf; fi
-
 RUN groupadd -f render

Jenkinsfile

@@ -192,10 +192,8 @@ def getDockerImage(Map conf=[:])
     env.DOCKER_BUILDKIT=1
     def prefixpath = conf.get("prefixpath", "/opt/rocm") // one image for each prefix 1: /usr/local 2:/opt/rocm
     def gpu_arch = "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102" // prebuilt dockers should have all the architectures enabled so one image can be used for all stages
-    def miotensile_version = conf.get("miotensile_version", "default") // deprecated
-    def target_id = conf.get("target_id", "OFF") // deprecated
     def mlir_build = conf.get("mlir_build", "ON") // always ON
-    def dockerArgs = "--build-arg BUILDKIT_INLINE_CACHE=1 --build-arg PREFIX=${prefixpath} --build-arg GPU_ARCH='${gpu_arch}' --build-arg MIOTENSILE_VER='${miotensile_version}' --build-arg USE_TARGETID='${target_id}' --build-arg USE_MLIR='${mlir_build}' "
+    def dockerArgs = "--build-arg BUILDKIT_INLINE_CACHE=1 --build-arg PREFIX=${prefixpath} --build-arg GPU_ARCH='${gpu_arch}' --build-arg USE_MLIR='${mlir_build}' "
     if(env.CCACHE_HOST)
     {
         def check_host = sh(script:"""(printf "PING\r\n";) | nc -N ${env.CCACHE_HOST} 6379 """, returnStdout: true).trim()
@@ -436,15 +434,15 @@ pipeline {
             description: "")
         booleanParam(
             name: "BUILD_SMOKE_FP32",
-            defaultValue: env.BRANCH_NAME == env.NIGHTLY_BRANCH ? true : false,
+            defaultValue: true, //env.BRANCH_NAME == env.NIGHTLY_BRANCH ? true : false,
             description: "")
         booleanParam(
             name: "BUILD_SMOKE_AUX1",
-            defaultValue: env.BRANCH_NAME == env.NIGHTLY_BRANCH ? true : false,
+            defaultValue: true, //env.BRANCH_NAME == env.NIGHTLY_BRANCH ? true : false,
             description: "")
         booleanParam(
             name: "BUILD_SMOKE_FP16_BF16_INT8",
-            defaultValue: env.BRANCH_NAME == env.NIGHTLY_BRANCH ? true : false,
+            defaultValue: true, //env.BRANCH_NAME == env.NIGHTLY_BRANCH ? true : false,
             description: "")
         booleanParam(
             name: "BUILD_FULL_TESTS",
@@ -535,10 +533,10 @@ pipeline {
         Smoke_targets = "check MIOpenDriver"
         NOCOMGR_flags   = " -DMIOPEN_USE_COMGR=Off"
     }
-    triggers{
-
-        cron(env.BRANCH_NAME == env.NIGHTLY_BRANCH ? env.NIGHTLY_SCHEDULE : '')
-    }
+    //triggers{
+    //    
+    //    cron(env.BRANCH_NAME == env.NIGHTLY_BRANCH ? env.NIGHTLY_SCHEDULE : '')
+    //}
     stages{
         stage('Build Docker'){
             when {
@@ -924,7 +922,7 @@ pipeline {
                     environment{
                         setup_flags="-DDISCRETE_GTEST=1 -DMIOPEN_TEST_DBSYNC=1"
                         config_targets='test_db_sync'
-                        execute_cmd='./bin/test_db_sync'
+                        execute_cmd='MIOPEN_TEST_DBSYNC=1 ./bin/test_db_sync'
                     }
                     steps{
                         buildHipClangJobAndReboot(lfs_pull: true, setup_flags: setup_flags, config_targets: config_targets, execute_cmd: execute_cmd,
@@ -943,7 +941,7 @@ pipeline {
                     environment{
                         setup_flags="-DDISCRETE_GTEST=1 -DMIOPEN_TEST_DBSYNC=1"
                         config_targets='test_db_sync'
-                        execute_cmd='./bin/test_db_sync'
+                        execute_cmd='MIOPEN_TEST_DBSYNC=1 ./bin/test_db_sync'
                     }
                     steps{
                         buildHipClangJobAndReboot(lfs_pull: true, setup_flags: setup_flags, config_targets: config_targets, execute_cmd: execute_cmd,

README.md

@@ -33,17 +33,13 @@ python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
   * HIP -
     * HIP and HCC libraries and header files.
   * OpenCL - OpenCL libraries and header files.
-* [MIOpenGEMM](https://github.com/ROCmSoftwarePlatform/MIOpenGEMM) - enable various functionalities including transposed and dilated convolutions.
-  * This is optional on the HIP backend, and required on the OpenCL backend.
-  * Users can enable this library using the cmake configuration flag `-DMIOPEN_USE_MIOPENGEMM=On`, which is enabled by default when OpenCL backend is chosen.
 * [ROCm cmake](https://github.com/RadeonOpenCompute/rocm-cmake) - provide cmake modules for common build tasks needed for the ROCM software stack.
 * [Half](http://half.sourceforge.net/) - IEEE 754-based half-precision floating point library
 * [Boost](http://www.boost.org/)
   * MIOpen uses `boost-system` and `boost-filesystem` packages to enable persistent [kernel cache](https://rocm.docs.amd.com/projects/MIOpen/en/latest/cache.html)
   * Version 1.79 is recommended, older version may need patches to work on newer systems, e.g. boost1{69,70,72} w/glibc-2.34
 * [SQLite3](https://sqlite.org/index.html) - reading and writing performance database
 * lbzip2 - multi-threaded compress or decompress utility
-* [MIOpenTENSILE](https://github.com/ROCmSoftwarePlatform/MIOpenTensile) - users can enable this library using the cmake configuration flag`-DMIOPEN_USE_MIOPENTENSILE=On`. (deprecated after ROCm 5.1.1)
 * [rocBLAS](https://github.com/ROCm/rocBLAS) - AMD library for Basic Linear Algebra Subprograms (BLAS) on the ROCm platform.
   * Minimum version branch for pre-ROCm 3.5 [master-rocm-2.10](https://github.com/ROCm/rocBLAS/tree/master-rocm-2.10)
   * Minimum version branch for post-ROCm 3.5 [master-rocm-3.5](https://github.com/ROCm/rocBLAS/releases/tag/rocm-3.5.0)
@@ -106,8 +102,6 @@ This prefix can used to specify the dependency path during the configuration pha
 
 * MIOpen's HIP backend uses [rocBLAS](https://github.com/ROCm/rocBLAS) by default. Users can install rocBLAS minimum release by using `apt-get install rocblas`. To disable using rocBLAS set the configuration flag `-DMIOPEN_USE_ROCBLAS=Off`. rocBLAS is *not* available for the OpenCL backend.
 
-* MIOpen's OpenCL backend uses [MIOpenGEMM](https://github.com/ROCmSoftwarePlatform/MIOpenGEMM) by default. Users can install MIOpenGEMM minimum release by using `apt-get install miopengemm`.
-
 ## Building MIOpen from source
 
 ### Configuring with cmake

docs/DebugAndLogging.md

@@ -168,8 +168,7 @@ The `ROCBLAS_LAYER` environmental variable can be set to output GEMM information
 * `ROCBLAS_LAYER=2` - is set to 2, then there is bench logging
 * `ROCBLAS_LAYER=3` - is set to 3, then there is both trace and bench logging
 
-Additionally, using environment variable "MIOPEN_GEMM_ENFORCE_BACKEND", can override the default behavior. The default behavior which is to use
-both MIOpenGEMM and rocBlas depending on the input configuration:
+Additionally, the environment variable "MIOPEN_GEMM_ENFORCE_BACKEND" can be set to override default GEMM backend (Default GEMM backend is rocBLAS):
 
 * `MIOPEN_GEMM_ENFORCE_BACKEND=1`, use rocBLAS if enabled
 * `MIOPEN_GEMM_ENFORCE_BACKEND=2`, reserved

docs/install.md

@@ -6,9 +6,6 @@
   * HIP - 
     * HIP and HCC libraries and header files.
   * OpenCL - OpenCL libraries and header files.
-* [MIOpenGEMM](https://github.com/ROCmSoftwarePlatform/MIOpenGEMM) - enable various functionalities including transposed and dilated convolutions. 
-  * This is optional on the HIP backend, and required on the OpenCL backend.
-  * Users can enable this library using the cmake configuration flag `-DMIOPEN_USE_MIOPENGEMM=On`, which is enabled by default when OpenCL backend is chosen.
 * [ROCm cmake](https://github.com/RadeonOpenCompute/rocm-cmake) - provide cmake modules for common build tasks needed for the ROCM software stack.
 * [Half](http://half.sourceforge.net/) - IEEE 754-based half-precision floating point library
 * [Boost](http://www.boost.org/) 
@@ -72,5 +69,3 @@ cmake -P install_deps.cmake --minimum --prefix /root/MIOpen/install_dir
 This prefix can used to specify the dependency path during the configuration phase using the `CMAKE_PREFIX_PATH`.
 
 * MIOpen's HIP backend uses [rocBLAS](https://github.com/ROCm/rocBLAS) by default. Users can install rocBLAS minimum release by using `apt-get install rocblas`. To disable using rocBLAS set the configuration flag `-DMIOPEN_USE_ROCBLAS=Off`. rocBLAS is *not* available for the OpenCL backend.
-
-* MIOpen's OpenCL backend uses [MIOpenGEMM](https://github.com/ROCm/MIOpenGEMM) by default. Users can install MIOpenGEMM minimum release by using `apt-get install miopengemm`.

docs/sphinx/requirements.in

@@ -1 +1 @@
-rocm-docs-core==0.30.1
+rocm-docs-core==0.30.2

docs/sphinx/requirements.txt

@@ -100,7 +100,7 @@ requests==2.31.0
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core==0.30.1
+rocm-docs-core==0.30.2
     # via -r requirements.in
 smmap==5.0.0
     # via gitdb

include/miopen/config.h.in

@@ -29,8 +29,6 @@
 #cmakedefine01 MIOPEN_BACKEND_OPENCL
 #cmakedefine01 MIOPEN_BACKEND_HIP
 #cmakedefine01 MIOPEN_MODE_NOGPU
-#cmakedefine01 MIOPEN_USE_MIOPENTENSILE
-#cmakedefine01 MIOPEN_USE_MIOPENGEMM
 #cmakedefine01 MIOPEN_USE_ROCBLAS
 #cmakedefine01 MIOPEN_BUILD_DEV
 #cmakedefine01 MIOPEN_GPU_SYNC
@@ -86,7 +84,7 @@
 #cmakedefine MIOPEN_OFFLOADBUNDLER_BIN "@MIOPEN_OFFLOADBUNDLER_BIN@"
 #cmakedefine MIOPEN_CACHE_DIR "@MIOPEN_CACHE_DIR@"
 
-#define MIOPEN_USE_GEMM (MIOPEN_USE_MIOPENTENSILE || MIOPEN_USE_MIOPENGEMM || MIOPEN_USE_ROCBLAS)
+#define MIOPEN_USE_GEMM (MIOPEN_USE_ROCBLAS)
 
 // Usage of "defined" operator within macro expansion is undefined behavior,
 // so "defined(NDEBUG)" cannot be used there... unlike the following macro:

requirements.txt

@@ -1,5 +1,6 @@
 sqlite3@3.43.2 -DCMAKE_POSITION_INDEPENDENT_CODE=On
 boost@1.83 -DCMAKE_POSITION_INDEPENDENT_CODE=On --build -DCMAKE_CXX_FLAGS=" -std=c++14 -Wno-enum-constexpr-conversion -Wno-deprecated-builtins -Wno-deprecated-declarations "
+facebook/[email protected] -X subdir -DCMAKE_DIR=build/cmake
 ROCmSoftwarePlatform/half@10abd99e7815f0ca5d892f58dd7d15a23b7cf92c --build
 ROCmSoftwarePlatform/[email protected] -H sha256:a5f62769d28a73e60bc8d61022820f050e97c977c8f6f6275488db31512e1f42 -DBUILD_FAT_LIBROCKCOMPILER=1 -DCMAKE_IGNORE_PATH=/opt/conda/envs/py_3.9 -DCMAKE_IGNORE_PREFIX_PATH=/opt/conda
 nlohmann/[email protected] -DJSON_MultipleHeaders=ON -DJSON_BuildTests=Off

src/CMakeLists.txt

@@ -237,7 +237,6 @@ set( MIOpen_Source
     solver/fft.cpp
     solver/gemm.cpp
     solver/gemm_bwd.cpp
-    solver/gemm_common.cpp
     solver/gemm_wrw.cpp
     solver/norm/forward_layernorm.cpp
     solver/norm/forward_layernorm2d_ck.cpp
@@ -829,12 +828,7 @@ endif()
 #######################################
 if(MIOPEN_ENABLE_SQLITE)
     # MIOpen depends on SQLite
-    target_include_directories(MIOpen SYSTEM PRIVATE ${SQLITE3_STATIC_INCLUDE_DIRS})
-    target_include_directories(MIOpen SYSTEM INTERFACE $<BUILD_INTERFACE:${SQLITE3_STATIC_INCLUDE_DIRS}>)
-    target_compile_options(MIOpen PRIVATE ${SQLITE3_STATIC_CFLAGS})
-    target_compile_options(MIOpen INTERFACE $<BUILD_INTERFACE:${SQLITE3_STATIC_CFLAGS}>)
-    target_link_libraries(MIOpen PRIVATE ${SQLITE3_STATIC_LDFLAGS})
-    target_link_libraries(MIOpen INTERFACE $<BUILD_INTERFACE:${SQLITE3_STATIC_LDFLAGS}>)
+    target_link_libraries(MIOpen PRIVATE SQLite::SQLite3)
 endif()
 ############################################################
 # MIOpen depends on librt for Boost.Interprocess

src/anyramdb.cpp

@@ -28,7 +28,6 @@
 
 #include <miopen/errors.hpp>
 #include <miopen/logger.hpp>
-#include <miopen/md5.hpp>
 
 #include <boost/filesystem/operations.hpp>
 #include <boost/filesystem.hpp>