Crosscompile-enabled CMake compilation for Caffe.

.gitignore

@@ -86,8 +86,10 @@ cmake_build
 build
 build_opencl
 build_cuda
+build_arm
 .build_opencl
 .build_cuda
+.build_arm
 .build
 .build_debug/*
 .build_release/*

CMakeLists.txt

@@ -87,12 +87,14 @@ caffe_option(BUILD_python_layer "Build the Caffe Python layer" ON)
 caffe_option(USE_OPENCV "Build with OpenCV support" ON)
 caffe_option(USE_LEVELDB "Build with levelDB" ON)
 caffe_option(USE_LMDB "Build with lmdb" ON)
+caffe_option(USE_HDF5 "Build with hdf5" ON)
 caffe_option(ALLOW_LMDB_NOLOCK "Allow MDB_NOLOCK when reading LMDB files (only if necessary)" OFF)
 caffe_option(USE_OPENMP "Link with OpenMP (when your BLAS wants OpenMP and you get linker errors)" OFF)
 caffe_option(USE_FFT "Build with fftw3 or/and clFFT" OFF)
 caffe_option(USE_SQLITE "Build with SQLITE kernel cache" ON)
 caffe_option(USE_GEMMLOWP "Build with Gemmlowp support" OFF)
 caffe_option(USE_NATIVE_MARCH "Build with -march=native flag, if supported" ON)
+caffe_option(USE_ARM_CROSS_COMPILE "Cross compile for Asus Tinkerboard or Raspberry Pi" OFF)
 
 # ---[ Option to force a persistent storage path
 set(CAFFE_STORAGE_PATH_OVERRIDE "" CACHE STRING "Caffe persistent storage path override (for sqlite cache)")
@@ -105,8 +107,32 @@ if(CPU_ONLY)
   set(USE_LIBDNN OFF)
   set(USE_CLBLAS OFF)
   set(USE_CLBLAST OFF)
+  set(USE_HSA OFF)
+  set(USE_HIP OFF)
 endif()
 
+# ---[ Cross compile for ARM (Asus Tinkerboard, Rasperry Pi)
+if(USE_ARM_CROSS_COMPILE)
+  set(CMAKE_CROSSCOMPILING TRUE)
+  set(CMAKE_PREFIX_PATH "${CMAKE_FIND_ROOT_PATH}")
+  set(CMAKE_SYSROOT "${CMAKE_FIND_ROOT_PATH}")
+  set(CMAKE_SYSROOT_COMPILE "${CMAKE_FIND_ROOT_PATH}")
+  set(CMAKE_CXX_IMPLICIT_INCLUDE_DIRECTORIES "${CMAKE_FIND_ROOT_PATH}/usr/include")
+  set(CMAKE_C_IMPLICIT_INCLUDE_DIRECTORIES "${CMAKE_FIND_ROOT_PATH}/usr/include")
+  set(CMAKE_CUDA_IMPLICIT_INCLUDE_DIRECTORIES "${CMAKE_FIND_ROOT_PATH}/usr/include")
+  set(CMAKE_INSTALL_OLDINCLUDEDIR "${CMAKE_FIND_ROOT_PATH}/usr/include" CACHE STRING "Header files" FORCE)
+  set(USE_HDF5 OFF CACHE BOOL "Build with hdf5" FORCE)
+  set(BOOST_INCLUDEDIR "${CMAKE_FIND_ROOT_PATH}/usr/include" CACHE STRING "Boost inculde path")
+  set(BOOST_LIBRARYDIR "${CMAKE_FIND_ROOT_PATH}/usr/lib/arm-linux-gnueabihf" CACHE STRING "Caffe logical version")
+  set_property(GLOBAL PROPERTY TARGET_SUPPORTS_SHARED_LIBS TRUE)
+  unset(COMPILER_SUPPORTS_ARMV7ANEON CACHE)
+  CHECK_CXX_COMPILER_FLAG("-march=armv7-a" COMPILER_SUPPORTS_ARMV7ANEON)
+  if(COMPILER_SUPPORTS_ARMV7ANEON)
+    message("Using armv7a-neon target")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=armv7-a")
+  endif()
+endif()  
+
 if(USE_ISAAC)
   set(USE_CLBLAS ON)
 endif()
@@ -132,7 +158,6 @@ endif()
 include(cmake/WindowsDownloadPrebuiltDependencies.cmake)
 
 # This code is taken from https://github.com/sh1r0/caffe-android-lib
-caffe_option(USE_HDF5 "Build with hdf5" ON)
 if(ANDROID)
   set(CMAKE_FIND_LIBRARY_SUFFIXES ".a;.so")
   caffe_enable_cpp11_support()
@@ -147,21 +172,19 @@ if(DISABLE_DEVICE_HOST_UNIFIED_MEMORY)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DDISABLE_DEVICE_HOST_UNIFIED_MEMORY")
 endif()
 
-if(UNIX OR APPLE)
+if(UNIX OR APPLE OR USE_ARM_CROSS_COMPILE)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC -Wall -std=c++11 -DCMAKE_BUILD")
 endif()
 
 if(USE_NATIVE_MARCH)
+  unset(COMPILER_SUPPORTS_MARCH_NATIVE CACHE)
   CHECK_CXX_COMPILER_FLAG("-march=native" COMPILER_SUPPORTS_MARCH_NATIVE)
   if(COMPILER_SUPPORTS_MARCH_NATIVE)
+    message("Using native target")
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
   endif()
 endif()
 
-if(DISABLE_DOUBLE_SUPPORT)
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DDISABLE_DOUBLE_SUPPORT")
-endif()
-
 if(MSVC)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP")
 endif()

cmake/Dependencies.cmake

@@ -65,10 +65,6 @@ else()
   endif()
 endif()
 
-include_directories(SYSTEM ${HDF5_INCLUDE_DIRS} ${HDF5_HL_INCLUDE_DIR})
-list(APPEND Caffe_LINKER_LIBS PUBLIC ${HDF5_LIBRARIES} ${HDF5_HL_LIBRARIES})
-list(APPEND Caffe_INCLUDE_DIRS PUBLIC ${HDF5_INCLUDE_DIRS})
-
 # This code is taken from https://github.com/sh1r0/caffe-android-lib
 if(USE_HDF5)
   find_package(HDF5 COMPONENTS HL REQUIRED)
@@ -242,7 +238,7 @@ endif()
 
 # ---[ clBLAS
 if (USE_CLBLAS AND NOT USE_ISAAC)
-  find_package(clBLAS)
+  find_package(clBLAS REQUIRED)
   if (NOT CLBLAS_FOUND)
     message(FATAL_ERROR "clBLAS required but not found.")
   endif()
@@ -253,7 +249,7 @@ endif()
 
 # ---[ ISAAC
 if (USE_ISAAC)
-  find_package(ISAAC)
+  find_package(ISAAC REQUIRED)
   if (NOT ISAAC_FOUND)
     message(FATAL_ERROR "ISAAC required but not found.")
   endif()
@@ -264,8 +260,10 @@ endif()
 # ---[ CLBlast
 if (USE_CLBLAST)
   find_package(CLBlast REQUIRED)
-  message(STATUS "CLBlast found")
-  list(APPEND Caffe_LINKER_LIBS PUBLIC clblast)
+  if (NOT CLBLAST_FOUND)
+    message(FATAL_ERROR "CLBlast required but not found.")
+  endif()
+  list(APPEND Caffe_LINKER_LIBS PUBLIC ${CLBLAST_LIBRARY})
   list(APPEND Caffe_DEFINITIONS PUBLIC -DUSE_CLBLAST)
 endif()
 

cmake/Modules/FindCLBlast.cmake

@@ -0,0 +1,60 @@
+set(CLBLAST_INCLUDE_SEARCH_PATHS
+  /usr/include
+  /usr/local/include
+  /usr/local/include/clblast
+  /usr/local/include  
+  clblast
+  CLBlast
+  ..
+  ../clblast
+  ../CLBlast
+)
+
+set(CLBLAST_LIBRARY_SEARCH_PATHS
+  /lib/
+  /lib64/
+  /usr/lib
+  /usr/lib/clblast
+  /usr/lib64
+  /usr/local/lib
+  /usr/local/lib64
+  clblast/build
+  CLBlast/build
+  ..
+  ../clblast/build
+  ../CLBlast/build
+)
+
+find_path(CLBLAST_INCLUDE_DIR NAMES clblast.h PATHS ${CLBLAST_INCLUDE_SEARCH_PATHS})
+find_library(CLBLAST_LIBRARY NAMES clblast PATHS ${CLBLAST_LIBRARY_SEARCH_PATHS})
+
+set(CLBLAST_FOUND ON)
+
+#    Check include files
+if(NOT CLBLAST_INCLUDE_DIR)
+  set(CLBLAST_FOUND OFF)
+  message(STATUS "Could not find CLBLAST include. Turning CLBLAST_FOUND off")
+endif()
+
+#    Check libraries
+if(NOT CLBLAST_LIBRARY)
+  set(CLBLAST_FOUND OFF)
+  message(STATUS "Could not find CLBLAST lib. Turning CLBLAST_FOUND off")
+endif()
+
+if (CLBLAST_FOUND)
+  if (NOT CLBLAST_FIND_QUIETLY)
+    message(STATUS "Found CLBLAST libraries: ${CLBLAST_LIBRARY}")
+    message(STATUS "Found CLBLAST include: ${CLBLAST_INCLUDE_DIR}")
+  endif (NOT CLBLAST_FIND_QUIETLY)
+else (CLBLAST_FOUND)
+  if (CLBLAST_FIND_REQUIRED)
+    message(FATAL_ERROR "Could not find CLBLAST")
+  endif (CLBLAST_FIND_REQUIRED)
+endif (CLBLAST_FOUND)
+
+mark_as_advanced(
+  CLBLAST_INCLUDE_DIR
+  CLBLAST_LIBRARY
+  CLBLAST
+)

cmake/Modules/FindGFlags.cmake

@@ -15,7 +15,7 @@ set(GFLAGS_ROOT_DIR "" CACHE PATH "Folder contains Gflags")
 
 # We are testing only a couple of files in the include directories
 find_path(GFLAGS_INCLUDE_DIR gflags/gflags.h
-    PATHS ${GFLAGS_ROOT_DIR})
+    PATHS ${GFLAGS_ROOT_DIR} /usr/include)
 
 if(MSVC)
     # rely on gflags-config.cmake

cmake/Modules/FindGlog.cmake

@@ -14,7 +14,7 @@ include(FindPackageHandleStandardArgs)
 set(GLOG_ROOT_DIR "" CACHE PATH "Folder contains Google glog")
 
 find_path(GLOG_INCLUDE_DIR glog/logging.h
-    PATHS ${GLOG_ROOT_DIR})
+    PATHS ${GLOG_ROOT_DIR} /usr/include)
 
 if(MSVC)
     # rely on glog-config.cmake

cmake/Modules/FindLMDB.cmake

@@ -15,8 +15,8 @@
 if(MSVC)
   find_package(LMDB NO_MODULE)
 else()
-  find_path(LMDB_INCLUDE_DIR NAMES  lmdb.h PATHS "$ENV{LMDB_DIR}/include")
-  find_library(LMDB_LIBRARIES NAMES lmdb   PATHS "$ENV{LMDB_DIR}/lib" )
+  find_path(LMDB_INCLUDE_DIR NAMES  lmdb.h PATHS "$ENV{LMDB_DIR}/include" "/usr/include")
+  find_library(LMDB_LIBRARIES NAMES lmdb   PATHS "$ENV{LMDB_DIR}/lib" "/usr/lib" "/usr/lib64")
 endif()
 
 include(FindPackageHandleStandardArgs)

cmake/Modules/FindLevelDB.cmake

@@ -11,13 +11,13 @@ if(MSVC)
   set(LevelDB_LIBRARY ${LevelDB_LIBRARIES})
 else()
   find_path(LevelDB_INCLUDE NAMES leveldb/db.h
-                            PATHS $ENV{LEVELDB_ROOT}/include /opt/local/include /usr/local/include /usr/include
+                            PATHS "$ENV{LEVELDB_ROOT}/include" "/opt/local/include" "/usr/local/include" "/usr/include"
                             DOC "Path in which the file leveldb/db.h is located." )
 
   # Look for the library.
   find_library(LevelDB_LIBRARY NAMES leveldb
-                              PATHS /usr/lib $ENV{LEVELDB_ROOT}/lib
-                              DOC "Path to leveldb library." )
+                               PATHS "$ENV{LEVELDB_ROOT}/lib" "/usr/lib" "/usr/lib64"
+                               DOC "Path to leveldb library." )
 endif()
 
 include(FindPackageHandleStandardArgs)

cmake/Modules/FindProtobuf.cmake

@@ -208,7 +208,7 @@ endif()
 # Find the include directory
 find_path(PROTOBUF_INCLUDE_DIR
     google/protobuf/service.h
-    PATHS ${PROTOBUF_SRC_ROOT_FOLDER}/src
+    PATHS ${PROTOBUF_SRC_ROOT_FOLDER}/src /usr/include
 )
 mark_as_advanced(PROTOBUF_INCLUDE_DIR)
 

cmake/Modules/FindSQLite.cmake

@@ -16,10 +16,10 @@
 # SQLITE3_FOUND	- True if sqlite found.
 
 # Look for the header file.
-FIND_PATH(SQLITE3_INCLUDE_DIR NAMES sqlite3.h sqlite.h)
+FIND_PATH(SQLITE3_INCLUDE_DIR NAMES sqlite3.h sqlite.h PATHS "/usr/include")
 
 # Look for the library.
-FIND_LIBRARY(SQLITE3_LIBRARY NAMES sqlite3 sqlite)
+FIND_LIBRARY(SQLITE3_LIBRARY NAMES sqlite3 sqlite PATHS "/usr/lib" "/usr/lib64")
 
 # Handle the QUIETLY and REQUIRED arguments and set SQLITE3_FOUND to TRUE if all listed variables are TRUE.
 INCLUDE(FindPackageHandleStandardArgs)

cmake/Modules/FindSnappy.cmake

@@ -12,10 +12,10 @@ if(MSVC)
   find_package(Snappy NO_MODULE)
 else()
   find_path(Snappy_INCLUDE_DIR NAMES snappy.h
-                              PATHS ${SNAPPY_ROOT_DIR} ${SNAPPY_ROOT_DIR}/include)
+                              PATHS "${SNAPPY_ROOT_DIR}" "${SNAPPY_ROOT_DIR}/include" "/usr/include")
 
   find_library(Snappy_LIBRARIES NAMES snappy
-                                PATHS ${SNAPPY_ROOT_DIR} ${SNAPPY_ROOT_DIR}/lib)
+                                PATHS "${SNAPPY_ROOT_DIR}" "${SNAPPY_ROOT_DIR}/lib" "/usr/lib" "/usr/lib64")
 endif()
 
 include(FindPackageHandleStandardArgs)

cmake/Summary.cmake

@@ -178,4 +178,9 @@ function(caffe_print_configuration_summary)
   caffe_status("Install:")
   caffe_status("  Install path      :   ${CMAKE_INSTALL_PREFIX}")
   caffe_status("")
+  #get_cmake_property(_variableNames VARIABLES)
+  #list (SORT _variableNames)
+  #  foreach (_variableName ${_variableNames})
+  #	message(STATUS "${_variableName}=${${_variableName}}")
+  #endforeach()
 endfunction()

include/caffe/backend/hip/hip_device.hpp

@@ -8,7 +8,7 @@ namespace caffe {
 
 #ifdef USE_HIP
 
-class hip_device : public device {
+class HipDevice : public Device {
  public:
  protected:
  private:

src/caffe/CMakeLists.txt

@@ -133,7 +133,7 @@ endif()
 configure_file(${caffe_export_hdr_in} ${caffe_export_hdr})
 
 # ---[ Tests
- add_subdirectory(test)
+add_subdirectory(test)
 
 # ---[ Install
 install(DIRECTORY ${Caffe_INCLUDE_DIR}/caffe DESTINATION ${CMAKE_INSTALL_INCLUDEDIR})

src/caffe/libdnn/libdnn.cpp

@@ -53,6 +53,14 @@ string LibDNN<MItype, MOtype>::generate_gemm_core(
        << " = Bsub[k][col + " << (i * rtsn)
        << "];" << std::endl;
   }
+  if (is_integer_type<MItype>()) {
+    ss << "if (tidm == 0) {" << std::endl;
+    for (int i = 0; i < vwn; ++i) {
+      ss << "Bsubcols[col + " << (i * rtsn) << "] += VEC_" << vwn << "_" << i
+         << "(Breg[wn]);" << std::endl;
+    }
+    ss << "}" << std::endl;
+  }
   ss << "}" << std::endl;
 
   // Perform the computation
@@ -63,6 +71,14 @@ string LibDNN<MItype, MOtype>::generate_gemm_core(
     ss << "VEC_" << vwm << "_" << i << "(Areg)" << " = Asub[row + " << (i*rtsm)
        << "][k];" << std::endl;
   }
+  if (is_integer_type<MItype>()) {
+    ss << "if (tidn == 0) {" << std::endl;
+    for (int i = 0; i < vwm; ++i) {
+      ss << "Asubrows[row + " << (i*rtsm) << "] += VEC_" << vwm << "_" << i
+         << "(Areg);" << std::endl;
+    }
+    ss << "}" << std::endl;
+  }
   if (dterm) {
     if (unroll) {
       for (int i = 0; i < vwm; ++i) {

src/caffe/libdnn/libdnn_blas_3.cpp

@@ -339,7 +339,7 @@ string LibDNNBlas<MItype, MOtype>::generate_gemm_source(
     ss << this->generate_gemm_core(tuner, false, alpha_term,
                                    alpha_exactly_one);
 
-    if (is_integer_type<MItype>()) {
+    /*if (is_integer_type<MItype>()) {
       // Add up columns of A
       ss << "for (int_tp k = 0; k < TSK; ++k) {" << std::endl;
       ss << "if (tidn == 0) {" << std::endl;
@@ -356,7 +356,7 @@ string LibDNNBlas<MItype, MOtype>::generate_gemm_source(
          << std::endl;
       ss << "}}" << std::endl;
       ss << "}" << std::endl;
-    }
+    }*/
 
     // Synchronize before loading the next tile
     ss << program->local_barrier() << std::endl;

src/caffe/libdnn/libdnn_conv.cpp

@@ -61,7 +61,7 @@ LibDNNConv<MItype, MOtype>::LibDNNConv(LibDNNConvConfig config)
   for (int id = 0; id < 2; ++id) {
     vector<int_tp> workgroup_sizes;
     for (int_tp i = 0; i < this->dev_ptr_->workgroup_size(id);
-            i += 4) {
+            i += 1) {
       workgroup_sizes.push_back(i);
     }
     fw_tuner_->add_set_param <int_tp>("workgroup_size_" + std::to_string(id),
@@ -956,7 +956,7 @@ string LibDNNConv<MItype, MOtype>::generate_fw_kernels(string name) {
 
   ss << this->generate_gemm_core(fw_tuner_, false, false, false) << std::endl;
 
-  if (is_integer_type<MItype>()) {
+  /*if (is_integer_type<MItype>()) {
     // Add up columns of A
     ss << "for (int_tp k = 0; k < TSK; ++k) {" << std::endl;
     ss << "if (tidn == 0) {" << std::endl;
@@ -973,7 +973,7 @@ string LibDNNConv<MItype, MOtype>::generate_fw_kernels(string name) {
        << std::endl;
     ss << "}}" << std::endl;
     ss << "}" << std::endl;
-  }
+  }*/
 
   // Synchronize before loading the next tile
   ss << this->program_->local_barrier() << std::endl;