Fixes for CUDA version of CLBlast

CNugteren · Jun 3, 2018 · bd1715a · bd1715a
1 parent 4f594e3
commit bd1715a
Show file tree

Hide file tree

Showing 2 changed files with 17 additions and 13 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -424,12 +424,14 @@ if(TUNERS)
     target_include_directories(clblast_tuner_${KERNEL} PUBLIC $<TARGET_PROPERTY:clblast,INTERFACE_INCLUDE_DIRECTORIES> ${API_INCLUDE_DIRS})
     install(TARGETS clblast_tuner_${KERNEL} DESTINATION bin)
   endforeach()
-  foreach(ROUTINE_TUNER ${ROUTINE_TUNERS})
-    add_executable(clblast_tuner_routine_${ROUTINE_TUNER} ${TUNERS_COMMON} src/tuning/routines/${ROUTINE_TUNER}.cpp test/test_utilities.cpp)
-    target_link_libraries(clblast_tuner_routine_${ROUTINE_TUNER} clblast)
-    target_include_directories(clblast_tuner_routine_${ROUTINE_TUNER} PUBLIC $<TARGET_PROPERTY:clblast,INTERFACE_INCLUDE_DIRECTORIES> ${API_INCLUDE_DIRS} ${clblast_SOURCE_DIR})
-    install(TARGETS clblast_tuner_routine_${ROUTINE_TUNER} DESTINATION bin)
-  endforeach()
+  if(OPENCL)
+    foreach(ROUTINE_TUNER ${ROUTINE_TUNERS})
+      add_executable(clblast_tuner_routine_${ROUTINE_TUNER} ${TUNERS_COMMON} src/tuning/routines/${ROUTINE_TUNER}.cpp test/test_utilities.cpp)
+      target_link_libraries(clblast_tuner_routine_${ROUTINE_TUNER} clblast)
+      target_include_directories(clblast_tuner_routine_${ROUTINE_TUNER} PUBLIC $<TARGET_PROPERTY:clblast,INTERFACE_INCLUDE_DIRECTORIES> ${API_INCLUDE_DIRS} ${clblast_SOURCE_DIR})
+      install(TARGETS clblast_tuner_routine_${ROUTINE_TUNER} DESTINATION bin)
+    endforeach()
+  endif()
 
   # Adds 'alltuners' target: runs all tuners for all precisions
   set(ALLTUNERS )
@@ -440,12 +442,14 @@ if(TUNERS)
     endforeach()
     set(ALLTUNERSDEPENDS clblast_tuner_${KERNEL})
   endforeach()
-  foreach(ROUTINE_TUNER ${ROUTINE_TUNERS})
-    foreach(PRECISION ${PRECISIONS})
-      set(ALLTUNERS ${ALLTUNERS} COMMAND clblast_tuner_routine_${ROUTINE_TUNER} -precision ${PRECISION})
+  if(OPENCL)
+    foreach(ROUTINE_TUNER ${ROUTINE_TUNERS})
+      foreach(PRECISION ${PRECISIONS})
+        set(ALLTUNERS ${ALLTUNERS} COMMAND clblast_tuner_routine_${ROUTINE_TUNER} -precision ${PRECISION})
+      endforeach()
+      set(ALLTUNERSDEPENDS clblast_tuner_routine_${ROUTINE_TUNER})
     endforeach()
-    set(ALLTUNERSDEPENDS clblast_tuner_routine_${ROUTINE_TUNER})
-  endforeach()
+  endif()
   add_custom_target(alltuners ${ALLTUNERS} DEPENDS ${ALLTUNERSDEPENDS})
 
 endif()

diff --git a/src/cupp11.hpp b/src/cupp11.hpp
@@ -678,8 +678,8 @@ class Kernel {
   }
 
   // Regular constructor with memory management
-  explicit Kernel(const Program &program, const std::string &name): name_(name) {
-    CheckError(cuModuleGetFunction(&kernel_, program.GetModule(), name.c_str()));
+  explicit Kernel(const std::shared_ptr<Program> program, const std::string &name): name_(name) {
+    CheckError(cuModuleGetFunction(&kernel_, program->GetModule(), name.c_str()));
   }
 
   // Sets a kernel argument at the indicated position. This stores both the value of the argument