More changes for HPX support

UO-OACISS · Jul 24, 2020 · f61a4df · f61a4df
1 parent 7ac9c4f
commit f61a4df
Show file tree

Hide file tree

Showing 5 changed files with 27 additions and 21 deletions.
diff --git a/cmake/Modules/APEX_SetupCUPTI.cmake b/cmake/Modules/APEX_SetupCUPTI.cmake
@@ -9,15 +9,20 @@ if(APEX_WITH_CUDA)
 
   # Add an imported target
   add_library(cuda INTERFACE IMPORTED)
-  add_library(${CUDA_cupti_LIBRARY} INTERFACE IMPORTED)
+  add_library(cupti INTERFACE IMPORTED)
+  hpx_info("apex" "Building APEX with CUDA/CUPTI support.")
   set_property(TARGET cupti PROPERTY
     INTERFACE_INCLUDE_DIRECTORIES ${CUDAToolkit_INCLUDE_DIR})
-  #set_property(TARGET cuda PROPERTY
-    #INTERFACE_LINK_LIBRARIES ${CUDA_LIBRARY})
-  #set_property(TARGET cupti PROPERTY
-    #INTERFACE_LINK_LIBRARIES ${CUDA_cupti_LIBRARY})
+  set_property(TARGET cuda PROPERTY
+    INTERFACE_LINK_LIBRARIES ${CUDA_LIBRARY})
+  set_property(TARGET cupti PROPERTY
+    INTERFACE_LINK_LIBRARIES ${CUDA_cupti_LIBRARY})
+
+  # Add the right definitions to the apex_flags target
+  target_compile_definitions(apex_flags INTERFACE APEX_WITH_CUDA)
 
-  list(APPEND _apex_imported_targets cuda cupti)
+  list(APPEND _apex_imported_targets cuda)
+  list(APPEND _apex_imported_targets cupti)
 
 else()
   add_custom_target(project_cuda)

diff --git a/src/apex/CMakeLists.hpx b/src/apex/CMakeLists.hpx
@@ -59,13 +59,6 @@ else()
   set(APEX_ARCH_X86 FALSE)
 endif()
 
-if(APEX_USE_CUDA)
-    enable_language(CUDA)
-    include(APEX_SetupCUPTI)
-    set(cupti_sources activity_trace_async.cpp)
-    include_directories (${CUDAToolkit_INCLUDE_DIR})
-endif(APEX_USE_CUDA)
-
 # If not x86, don't use RDTSC
 if(NOT APEX_ARCH_X86)
   target_compile_definitions(apex_flags INTERFACE APEX_USE_CLOCK_TIMESTAMP=1)
@@ -175,6 +168,12 @@ endif()
 # We create one target per imported library (cleaner) #
 set(_apex_imported_targets "")
 
+include(APEX_SetupCUPTI)
+if(APEX_WITH_CUDA)
+    set(cupti_sources activity_trace_async.cpp)
+    #include_directories (${CUDAToolkit_INCLUDE_DIR})
+endif(APEX_WITH_CUDA)
+
 # Setup PAPI
 include(APEX_SetupPAPI)
 
@@ -343,8 +342,10 @@ target_include_directories(apex PUBLIC
 target_link_libraries(apex PRIVATE apex_flags)
 
 # Link to all imported targets
+hpx_info("apex" "Will build APEX with imported targets:")
 foreach(lib IN LISTS _apex_imported_targets)
   if (TARGET "${lib}")
+    hpx_info("apex" "    ${lib}")
     target_link_libraries(apex PRIVATE ${lib})
   endif()
 endforeach()

diff --git a/src/apex/CMakeLists.standalone b/src/apex/CMakeLists.standalone
@@ -25,9 +25,9 @@ if (OTF2_FOUND)
 SET(OTF2_SOURCE otf2_listener.cpp)
 endif(OTF2_FOUND)
 
-if (APEX_USE_CUDA)
+if (APEX_WITH_CUDA)
 SET(CUPTI_SOURCE activity_trace_async.cpp)
-endif(APEX_USE_CUDA)
+endif(APEX_WITH_CUDA)
 
 # Try to keep this in alphabetical order
 SET(all_SOURCE
@@ -80,9 +80,9 @@ if (MPI_CXX_FOUND)
   include_directories (${MPI_CXX_INCLUDE_PATH})
 endif(MPI_CXX_FOUND)
 
-if (APEX_USE_CUDA)
+if (APEX_WITH_CUDA)
   include_directories (${CUDAToolkit_INCLUDE_DIR})
-endif(APEX_USE_CUDA)
+endif(APEX_WITH_CUDA)
 
 # If we are building libapex.so, we want to include all the other libraries,
 # so that we can LD_PRELOAD this library with all requirements met.

diff --git a/src/apex/apex.cpp b/src/apex/apex.cpp
@@ -1355,7 +1355,7 @@ void flushTrace(void);
 
 void finalize()
 {
-#ifdef APEX_USE_CUDA
+#ifdef APEX_WITH_CUDA
     flushTrace();
 #endif
 #ifdef APEX_WITH_JUPYTER_SUPPORT

diff --git a/src/apex/profiler.hpp b/src/apex/profiler.hpp
@@ -42,7 +42,7 @@ class disabled_profiler_exception : public std::exception {
     }
 };
 
-#if !defined(APEX_USE_CLOCK_TIMESTAMP) && !defined(APEX_USE_CUDA)
+#if !defined(APEX_USE_CLOCK_TIMESTAMP) && !defined(APEX_WITH_CUDA)
 template<std::intmax_t clock_freq>
 struct rdtsc_clock {
     typedef unsigned long long rep;
@@ -62,7 +62,7 @@ struct rdtsc_clock {
 };
 #endif
 
-#if defined(APEX_USE_CLOCK_TIMESTAMP) || defined(APEX_USE_CUDA)
+#if defined(APEX_USE_CLOCK_TIMESTAMP) || defined(APEX_WITH_CUDA)
 #define MYCLOCK std::chrono::CLOCK_TYPE
 #else
 typedef rdtsc_clock<1> OneHzClock;
@@ -215,7 +215,7 @@ class profiler {
 
     /* This function returns 1/X, where "X" is the MHz rating of the CPU. */
     static double get_cpu_mhz () {
-#if defined(APEX_USE_CLOCK_TIMESTAMP) || defined (APEX_USE_CUDA)
+#if defined(APEX_USE_CLOCK_TIMESTAMP) || defined (APEX_WITH_CUDA)
         return 1.0;
 #else
         static double ticks_per_period = 0.0;