Merge branch 'development' of github.com:CNugteren/CLBlast into development

Conflicts:
	README.md

5 files changed, 28 insertions, 44 deletions

diff --git a/.appveyor.yml b/.appveyor.yml
index 8597e43e..eb7f1c97 100644
--- a/.appveyor.yml
+++ b/.appveyor.yml
@@ -1,8 +1,8 @@
 environment:
   global:
-    CLBLAST_ROOT: "%APPVEYOR_BUILD_FOLDER%\\bin\\clblast"
+    CLBLAST_ROOT: "%APPVEYOR_BUILD_FOLDER%\\..\\bin\\clblast"
     OPENCL_REGISTRY: "https://www.khronos.org/registry/cl"
-    OPENCL_ROOT: "%APPVEYOR_BUILD_FOLDER%\\bin\\opencl"
+    OPENCL_ROOT: "%APPVEYOR_BUILD_FOLDER%\\..\\bin\\opencl"
 
 platform:
   - x64
diff --git a/.travis.yml b/.travis.yml
index 8e1a80db..0465afa4 100644
--- a/.travis.yml
+++ b/.travis.yml
@@ -17,49 +17,21 @@ addons:
       - kubuntu-backports
     packages:
       - cmake
+      - ocl-icd-opencl-dev
 
 env:
   global:
     - CLBLAST_ROOT=${TRAVIS_BUILD_DIR}/bin/clblast
-    - OPENCL_REGISTRY=https://www.khronos.org/registry/cl
-    - OPENCL_ROOT=${TRAVIS_BUILD_DIR}/bin/opencl
 
 before_install:
   - cmake --version;
   - ${CC} --version;
   - ${CXX} --version;
 
-install:
-  # The following linux logic is necessary because of Travis's move to the GCE platform, which does not
-  # currently contain packages for fglrx: https://github.com/travis-ci/travis-ci/issues/5221
-  # We build our own linkable .so file
-  - if [ ${TRAVIS_OS_NAME} == "linux" ]; then
-      mkdir -p ${OPENCL_ROOT};
-      pushd ${OPENCL_ROOT};
-      travis_retry git clone --depth 1 https://github.com/KhronosGroup/OpenCL-ICD-Loader.git;
-      mv ./OpenCL-ICD-Loader/* .;
-      travis_retry git clone --depth 1 https://github.com/KhronosGroup/OpenCL-Headers.git inc/CL;
-      pushd inc/CL;
-      travis_retry wget -w 1 -np -nd -nv -A h,hpp ${OPENCL_REGISTRY}/api/2.1/cl.hpp;
-      popd;
-      mkdir -p lib;
-      pushd lib;
-      cmake -G "Unix Makefiles" ..;
-      make;
-      cp ./bin/libOpenCL.so .;
-      popd;
-      pushd inc/CL;
-      travis_retry git fetch origin opencl12:opencl12;
-      git checkout opencl12;
-      popd;
-      mv inc/ include/;
-      popd;
-    fi
-
 before_script:
   - mkdir -p ${CLBLAST_ROOT}
   - pushd ${CLBLAST_ROOT}
-  - cmake -DOPENCL_ROOT=${OPENCL_ROOT} -DTESTS=ON -DCLIENTS=ON ${TRAVIS_BUILD_DIR}
+  - cmake -DTESTS=ON -DCLIENTS=ON ${TRAVIS_BUILD_DIR}
 
 script:
   - make
diff --git a/CHANGELOG b/CHANGELOG
index 7973655b..28e671f8 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -9,6 +9,7 @@ Development version (next release)
 - Fixed a bug in the OpenCL kernels: now placing __kernel before __attribute__
 - Added an option (-warm_up) to do a warm-up run before timing in the performance clients
 - Improved performance significantly of rotated GEMV computations
+- Various minor fixes and enhancements
 - Added tuned parameters for various devices (see README)
 
 Version 0.8.0
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 95d1d500..7393c6e7 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -127,11 +127,6 @@ endif()
 
 # ==================================================================================================
 
-# Includes directories: CLBlast and OpenCL
-include_directories(${clblast_SOURCE_DIR}/include ${clblast_SOURCE_DIR}/src ${OPENCL_INCLUDE_DIRS})
-
-# ==================================================================================================
-
 # Sets the supported routines and the used kernels. New routines and kernels should be added here.
 set(KERNELS copy_fast copy_pad transpose_fast transpose_pad xaxpy xdot xger xgemm xgemv)
 set(SAMPLE_PROGRAMS_CPP sgemm)
@@ -173,17 +168,27 @@ endforeach()
 add_library(clblast SHARED ${SOURCES})
 target_link_libraries(clblast ${OPENCL_LIBRARIES})
 
+# Includes directories: CLBlast and OpenCL
+target_include_directories(clblast PUBLIC
+                           $<BUILD_INTERFACE:${clblast_SOURCE_DIR}/include>
+                           $<BUILD_INTERFACE:${clblast_SOURCE_DIR}/src>
+                           $<INSTALL_INTERFACE:include>
+                           ${OPENCL_INCLUDE_DIRS})
+
 # Sets the proper __declspec(dllexport) keyword for Visual Studio when the library is built
 if(MSVC)
   target_compile_definitions(clblast PRIVATE COMPILING_DLL=1) # requires at least CMake 2.8.11
 endif()
 
 # Installs the library
-install(TARGETS clblast DESTINATION lib)
+install(TARGETS clblast EXPORT CLBlast DESTINATION lib)
 install(FILES include/clblast.h DESTINATION include)
 install(FILES include/clblast_c.h DESTINATION include)
 install(FILES include/clblast_half.h DESTINATION include)
 
+# Installs the config for find_package in dependent projects
+install(EXPORT CLBlast DESTINATION lib/cmake/CLBLast FILE CLBlastConfig.cmake)
+
 # ==================================================================================================
 
 # Sets a default platform ($DEVICEPLATFORM) and device ($DEFAULT_DEVICE) to run tuners and tests on
@@ -225,9 +230,6 @@ endif()
 # the CLTune library (not included as part of the source).
 if(TUNERS)
 
-  # Includes CLTune
-  include_directories(${CLTUNE_INCLUDE_DIRS})
-
   # Visual Studio requires the sources of non-exported objects/libraries
   set(TUNERS_COMMON )
   if(MSVC)
@@ -238,6 +240,7 @@ if(TUNERS)
   foreach(KERNEL ${KERNELS})
     add_executable(clblast_tuner_${KERNEL} ${TUNERS_COMMON} src/tuning/kernels/${KERNEL}.cpp)
     target_link_libraries(clblast_tuner_${KERNEL} clblast ${CLTUNE_LIBRARIES} ${OPENCL_LIBRARIES})
+    target_include_directories(clblast_tuner_${KERNEL} PUBLIC ${CLTUNE_INCLUDE_DIRS})
     install(TARGETS clblast_tuner_${KERNEL} DESTINATION bin)
   endforeach()
 
@@ -281,9 +284,6 @@ if(CLIENTS OR TESTS)
     endif()
   endif()
 
-  # Sets the include directories
-  include_directories(${clblast_SOURCE_DIR} ${REF_INCLUDES})
-
 endif()
 
 # ==================================================================================================
@@ -299,6 +299,11 @@ if(CLIENTS)
   else()
     # Creates the common performance-tests objects (requires CMake 2.8.8)
     add_library(test_performance_common OBJECT test/performance/client.cpp)
+
+    # Adds CLBlast's interface include paths because we can't link to CLBlast here
+    target_include_directories(test_performance_common PRIVATE
+                               $<TARGET_PROPERTY:clblast,INTERFACE_INCLUDE_DIRECTORIES>
+                               ${clblast_SOURCE_DIR})
     set(CLIENTS_COMMON ${CLIENTS_COMMON} $<TARGET_OBJECTS:test_performance_common>)
   endif()
 
@@ -321,6 +326,7 @@ if(CLIENTS)
   endforeach()
   foreach(ROUTINE ${ROUTINES})
     target_link_libraries(clblast_client_${ROUTINE} clblast ${REF_LIBRARIES} ${OPENCL_LIBRARIES})
+    target_include_directories(clblast_client_${ROUTINE} PUBLIC ${clblast_SOURCE_DIR} ${REF_INCLUDES})
     install(TARGETS clblast_client_${ROUTINE} DESTINATION bin)
   endforeach()
 
@@ -342,6 +348,9 @@ if(TESTS)
     # Creates the common correctness-tests objects (requires CMake 2.8.8)
     add_library(test_correctness_common OBJECT
                 test/correctness/tester.cpp test/correctness/testblas.cpp)
+    target_include_directories(test_correctness_common PUBLIC
+                               $<TARGET_PROPERTY:clblast,INTERFACE_INCLUDE_DIRECTORIES>
+                               ${clblast_SOURCE_DIR})
     set(TESTS_COMMON ${TESTS_COMMON} $<TARGET_OBJECTS:test_correctness_common>)
   endif()
 
@@ -365,6 +374,7 @@ if(TESTS)
   foreach(ROUTINE ${ROUTINES})
     target_link_libraries(clblast_test_${ROUTINE} clblast ${REF_LIBRARIES} ${OPENCL_LIBRARIES})
     install(TARGETS clblast_test_${ROUTINE} DESTINATION bin)
+    target_include_directories(clblast_test_${ROUTINE} PUBLIC ${clblast_SOURCE_DIR} ${REF_INCLUDES})
     add_test(clblast_test_${ROUTINE} clblast_test_${ROUTINE} ${DEVICEPLATFORM})
   endforeach()
 
diff --git a/README.md b/README.md
index 04a8e470..feff1d40 100644
--- a/README.md
+++ b/README.md
@@ -286,6 +286,7 @@ The contributing authors (code, pull requests, testing) so far are:
 * [Marco Hutter](https://github.com/gpus)
 * [Hugh Perkins](https://github.com/hughperkins)
 * [Gian-Carlo Pascutto](https://github.com/gcp)
+* [Ivan Shapovalov](https://github.com/intelfx)
 * [Dimitri VA](https://github.com/dvasschemacq)
 
 Tuning and testing on a variety of OpenCL devices was made possible by: