CMake updates/fixes + parallel-overhead updates (#16)

- OMNITRACE_INSTALL_EXAMPLES option - Fix lulesh standalone HIP compilation - Fix transpose standalone HIP compilation - Tweaks to parallel-overhead
ROCm · May 31, 2022 · a9ff15f · a9ff15f
1 parent 8b97c70
commit a9ff15f
Show file tree

Hide file tree

Showing 12 changed files with 129 additions and 37 deletions.
diff --git a/examples/.gitignore b/examples/.gitignore
@@ -0,0 +1,2 @@
+/build*
+/install*
diff --git a/examples/CMakeLists.txt b/examples/CMakeLists.txt
@@ -14,6 +14,13 @@ if(CMAKE_PROJECT_NAME STREQUAL "omnitrace")
     set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
     set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
     set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
+    omnitrace_add_option(OMNITRACE_INSTALL_EXAMPLES "Install omnitrace examples" OFF)
+else()
+    option(OMNITRACE_INSTALL_EXAMPLES "Install omnitrace examples" ON)
+endif()
+
+if(OMNITRACE_INSTALL_EXAMPLES)
+    include(GNUInstallDirs)
 endif()
 
 add_subdirectory(transpose)

diff --git a/examples/code-coverage/CMakeLists.txt b/examples/code-coverage/CMakeLists.txt
@@ -10,6 +10,9 @@ add_executable(code-coverage code-coverage.cpp)
 target_link_libraries(code-coverage PRIVATE Threads::Threads)
 target_compile_options(code-coverage PRIVATE ${_FLAGS})
 
-if(NOT CMAKE_PROJECT_NAME STREQUAL "omnitrace")
-    install(TARGETS code-coverage DESTINATION bin)
+if(OMNITRACE_INSTALL_EXAMPLES)
+    install(
+        TARGETS code-coverage
+        DESTINATION bin
+        COMPONENT omnitrace-examples)
 endif()
diff --git a/examples/lulesh/CMakeLists.txt b/examples/lulesh/CMakeLists.txt
@@ -7,8 +7,20 @@ list(INSERT CMAKE_MODULE_PATH 0 ${PROJECT_SOURCE_DIR}/cmake/Modules)
 option(LULESH_BUILD_KOKKOS "Build Kokkos from submodule" ON)
 if(LULESH_BUILD_KOKKOS)
     add_subdirectory(external)
-    if(LULESH_USE_CUDA OR LULESH_USE_HIP)
+    if(LULESH_USE_CUDA)
         kokkos_compilation(PROJECT)
+    elseif(LULESH_USE_HIP AND NOT "${CMAKE_CXX_COMPILER}" MATCHES "hipcc")
+        if(NOT HIPCC_EXECUTABLE)
+            find_package(hip QUIET)
+
+            find_program(
+                HIPCC_EXECUTABLE
+                NAMES hipcc
+                HINTS ${ROCM_PATH} ENV ROCM_PATH /opt/rocm
+                PATHS ${ROCM_PATH} ENV ROCM_PATH /opt/rocm)
+            mark_as_advanced(HIPCC_EXECUTABLE)
+        endif()
+        kokkos_compilation(PROJECT COMPILER ${HIPCC_EXECUTABLE})
     endif()
 else()
     find_package(Kokkos REQUIRED COMPONENTS separable_compilation)
@@ -44,6 +56,18 @@ add_executable(${PROJECT_NAME} ${sources} ${headers})
 target_include_directories(${PROJECT_NAME} PRIVATE ${PROJECT_SOURCE_DIR}/includes)
 target_link_libraries(${PROJECT_NAME} PRIVATE Kokkos::kokkos lulesh-mpi)
 
-if(NOT CMAKE_PROJECT_NAME STREQUAL "omnitrace")
-    install(TARGETS ${PROJECT_NAME} DESTINATION bin)
+if(OMNITRACE_INSTALL_EXAMPLES)
+    if(LULESH_BUILD_KOKKOS)
+        install(
+            TARGETS kokkoscore kokkoscontainers
+            DESTINATION ${CMAKE_INSTALL_LIBDIR}
+            COMPONENT omnitrace-examples)
+        set_target_properties(
+            ${PROJECT_NAME} PROPERTIES INSTALL_RPATH
+                                       "\$ORIGIN/../${CMAKE_INSTALL_LIBDIR}")
+    endif()
+    install(
+        TARGETS ${PROJECT_NAME}
+        DESTINATION bin
+        COMPONENT omnitrace-examples)
 endif()
diff --git a/examples/lulesh/external/CMakeLists.txt b/examples/lulesh/external/CMakeLists.txt
@@ -30,4 +30,4 @@ checkout_git_submodule(
 
 set(CMAKE_SKIP_INSTALL_ALL_DEPENDENCY ON)
 
-add_subdirectory(kokkos)
+add_subdirectory(kokkos EXCLUDE_FROM_ALL)
diff --git a/examples/mpi/CMakeLists.txt b/examples/mpi/CMakeLists.txt
@@ -16,6 +16,9 @@ endif()
 
 target_link_libraries(mpi-example PRIVATE MPI::MPI_CXX)
 
-if(NOT CMAKE_PROJECT_NAME STREQUAL "omnitrace")
-    install(TARGETS mpi-example DESTINATION bin)
+if(OMNITRACE_INSTALL_EXAMPLES)
+    install(
+        TARGETS mpi-example
+        DESTINATION bin
+        COMPONENT omnitrace-examples)
 endif()
diff --git a/examples/openmp/CMakeLists.txt b/examples/openmp/CMakeLists.txt
@@ -36,6 +36,9 @@ else()
     endif()
 endif()
 
-if(NOT CMAKE_PROJECT_NAME STREQUAL "omnitrace")
-    install(TARGETS openmp-cg openmp-lu DESTINATION bin)
+if(OMNITRACE_INSTALL_EXAMPLES)
+    install(
+        TARGETS openmp-cg openmp-lu
+        DESTINATION bin
+        COMPONENT omnitrace-examples)
 endif()
diff --git a/examples/parallel-overhead/CMakeLists.txt b/examples/parallel-overhead/CMakeLists.txt
@@ -11,6 +11,9 @@ add_executable(parallel-overhead-locks parallel-overhead.cpp)
 target_link_libraries(parallel-overhead-locks PRIVATE Threads::Threads)
 target_compile_definitions(parallel-overhead-locks PRIVATE USE_LOCKS=1)
 
-if(NOT CMAKE_PROJECT_NAME STREQUAL "omnitrace")
-    install(TARGETS parallel-overhead parallel-overhead-locks DESTINATION bin)
+if(OMNITRACE_INSTALL_EXAMPLES)
+    install(
+        TARGETS parallel-overhead parallel-overhead-locks
+        DESTINATION bin
+        COMPONENT omnitrace-examples)
 endif()
diff --git a/examples/parallel-overhead/parallel-overhead.cpp b/examples/parallel-overhead/parallel-overhead.cpp
@@ -1,18 +1,23 @@
 
+#include <atomic>
 #include <cstdio>
 #include <cstdlib>
 #include <pthread.h>
+#include <random>
 #include <string>
 #include <thread>
 #include <vector>
 
-#if defined(USE_LOCKS)
+#if !defined(USE_LOCKS)
+#    define USE_LOCKS 0
+#endif
+
+#if USE_LOCKS > 0
 #    include <mutex>
 using auto_lock_t = std::unique_lock<std::mutex>;
 long       total  = 0;
 std::mutex mtx{};
 #else
-#    include <atomic>
 std::atomic<long> total{ 0 };
 #endif
 
@@ -31,17 +36,29 @@ fib(long n)
 void
 run(size_t nitr, long n)
 {
-#if defined(USE_LOCKS)
+    static std::atomic<int> _tids{ 0 };
+    auto                    _tid = ++_tids;
+
+    std::default_random_engine          eng(std::random_device{}() * (100 + _tid));
+    std::uniform_int_distribution<long> distr{ n - 2, n + 2 };
+
+    auto _get_n = [&]() { return distr(eng); };
+
+    printf("[%i] number of iterations: %zu\n", _tid, nitr);
+
+#if USE_LOCKS > 0
     for(size_t i = 0; i < nitr; ++i)
     {
-        auto        _v = fib(n);
+        auto        _v = fib(_get_n());
         auto_lock_t _lk{ mtx };
         total += _v;
     }
 #else
     long local = 0;
     for(size_t i = 0; i < nitr; ++i)
-        local += fib(n);
+    {
+        local += fib(_get_n());
+    }
     total += local;
 #endif
 }
@@ -64,8 +81,11 @@ main(int argc, char** argv)
     printf("\n[%s] Threads: %zu\n[%s] Iterations: %zu\n[%s] fibonacci(%li)...\n",
            _name.c_str(), nthread, _name.c_str(), nitr, _name.c_str(), nfib);
 
+    bool run_on_main_thread = (USE_LOCKS == 0);
+    auto nwait              = nthread + ((run_on_main_thread) ? 1 : 0);
+
     pthread_barrier_t _barrier;
-    pthread_barrier_init(&_barrier, nullptr, nthread);
+    pthread_barrier_init(&_barrier, nullptr, nwait);
 
     auto _run = [&_barrier](size_t nitr, long n) {
         pthread_barrier_wait(&_barrier);
@@ -75,15 +95,13 @@ main(int argc, char** argv)
     std::vector<std::thread> threads{};
     for(size_t i = 0; i < nthread; ++i)
     {
-        size_t _nitr = ((i % 2) == 1) ? (nitr - (0.1 * nitr)) : (nitr + (0.1 * nitr));
-        _nitr        = std::max<size_t>(_nitr, 1);
-        threads.emplace_back(_run, _nitr, nfib);
+        threads.emplace_back(_run, nitr, nfib);
     }
 
-#if !defined(USE_LOCKS)
-    auto _nitr = std::max<size_t>(nitr - 0.25 * nitr, 1);
-    run(_nitr, nfib - 0.1 * nfib);
-#endif
+    if(run_on_main_thread)
+    {
+        _run(nitr, nfib);
+    }
 
     for(auto& itr : threads)
         itr.join();

diff --git a/examples/python/CMakeLists.txt b/examples/python/CMakeLists.txt
@@ -4,14 +4,17 @@ project(omnitrace-python)
 
 set(PYTHON_FILES builtin.py external.py source.py)
 
-if(NOT CMAKE_PROJECT_NAME STREQUAL "omnitrace")
+if(OMNITRACE_INSTALL_EXAMPLES)
     find_package(Python3 COMPONENTS Interpreter)
     if(Python3_FOUND)
         set(PYTHON_EXECUTABLE "${Python3_EXECUTABLE}")
         foreach(_FILE ${PYTHON_FILES})
             configure_file(${PROJECT_SOURCE_DIR}/${_FILE} ${PROJECT_BINARY_DIR}/${_FILE}
                            @ONLY)
-            install(PROGRAMS ${PROJECT_BINARY_DIR}/${_FILE} DESTINATION bin)
+            install(
+                PROGRAMS ${PROJECT_BINARY_DIR}/${_FILE}
+                DESTINATION bin
+                COMPONENT omnitrace-examples)
         endforeach()
     endif()
 endif()
diff --git a/examples/transpose/CMakeLists.txt b/examples/transpose/CMakeLists.txt
@@ -2,17 +2,32 @@ cmake_minimum_required(VERSION 3.16 FATAL_ERROR)
 
 project(omnitrace-transpose LANGUAGES CXX)
 
-find_program(HIPCC_EXECUTABLE NAMES hipcc)
+find_package(hip QUIET)
+
+find_program(
+    HIPCC_EXECUTABLE
+    NAMES hipcc
+    HINTS ${ROCM_PATH} ENV ROCM_PATH /opt/rocm
+    PATHS ${ROCM_PATH} ENV ROCM_PATH /opt/rocm)
 mark_as_advanced(HIPCC_EXECUTABLE)
 
 if(NOT HIPCC_EXECUTABLE)
     message(AUTHOR_WARNING "hipcc could not be found. Cannot build transpose target")
     return()
 endif()
 
-if(NOT CMAKE_CXX_COMPILER_IS_HIPCC
-   AND HIPCC_EXECUTABLE
-   AND NOT COMMAND omnitrace_custom_compilation)
+if(NOT CMAKE_CXX_COMPILER_IS_HIPCC AND HIPCC_EXECUTABLE)
+    if(CMAKE_CXX_COMPILER STREQUAL HIPCC_EXECUTABLE OR "${CMAKE_CXX_COMPILER}" MATCHES
+                                                       "hipcc")
+        set(CMAKE_CXX_COMPILER_IS_HIPCC
+            1
+            CACHE BOOL "HIP compiler")
+    endif()
+endif()
+
+if((NOT CMAKE_CXX_COMPILER_IS_HIPCC OR (NOT CMAKE_CXX_COMPILER_ID MATCHES "Clang"
+                                        AND NOT hip_FOUND))
+   AND (NOT COMMAND omnitrace_custom_compilation AND NOT HIPCC_EXECUTABLE))
     message(AUTHOR_WARNING "transpose target could not be built")
     return()
 endif()
@@ -25,8 +40,13 @@ endif()
 
 add_executable(transpose transpose.cpp)
 
-if(CMAKE_CXX_COMPILER_IS_CLANG AND TARGET omnitrace::omnitrace-compile-options)
-    target_link_libraries(transpose PRIVATE omnitrace::omnitrace-compile-options)
+if(CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+    target_link_libraries(
+        transpose
+        PRIVATE
+            $<IF:$<TARGET_EXISTS:omnitrace::omnitrace-compile-options>,omnitrace::omnitrace-compile-options,>
+            $<IF:$<TARGET_EXISTS:hip::host>,hip::host,>
+            $<IF:$<TARGET_EXISTS:hip::device>,hip::device,>)
 else()
     target_compile_options(transpose PRIVATE -W -Wall)
 endif()
@@ -41,6 +61,9 @@ if(NOT CMAKE_CXX_COMPILER_IS_HIPCC AND HIPCC_EXECUTABLE)
     omnitrace_custom_compilation(COMPILER ${HIPCC_EXECUTABLE} TARGET transpose)
 endif()
 
-if(NOT CMAKE_PROJECT_NAME STREQUAL "omnitrace")
-    install(TARGETS transpose DESTINATION bin)
+if(OMNITRACE_INSTALL_EXAMPLES)
+    install(
+        TARGETS transpose
+        DESTINATION bin
+        COMPONENT omnitrace-examples)
 endif()
diff --git a/examples/user-api/CMakeLists.txt b/examples/user-api/CMakeLists.txt
@@ -10,6 +10,9 @@ endif()
 add_executable(user-api user-api.cpp)
 target_link_libraries(user-api PRIVATE Threads::Threads omnitrace::omnitrace-user-library)
 
-if(NOT CMAKE_PROJECT_NAME STREQUAL "omnitrace")
-    install(TARGETS user-api DESTINATION bin)
+if(OMNITRACE_INSTALL_EXAMPLES)
+    install(
+        TARGETS user-api
+        DESTINATION bin
+        COMPONENT omnitrace-examples)
 endif()