Merge pull request #1887 from jet47:android-cuda-6.0

CMakeLists.txt

@@ -127,9 +127,9 @@ endif()
 OCV_OPTION(WITH_1394           "Include IEEE1394 support"                    ON   IF (NOT ANDROID AND NOT IOS) )
 OCV_OPTION(WITH_AVFOUNDATION   "Use AVFoundation for Video I/O"              ON   IF IOS)
 OCV_OPTION(WITH_CARBON         "Use Carbon for UI instead of Cocoa"          OFF  IF APPLE )
-OCV_OPTION(WITH_CUDA           "Include NVidia Cuda Runtime support"         ON   IF (CMAKE_VERSION VERSION_GREATER "2.8" AND NOT ANDROID AND NOT IOS) )
+OCV_OPTION(WITH_CUDA           "Include NVidia Cuda Runtime support"         ON   IF (CMAKE_VERSION VERSION_GREATER "2.8" AND NOT IOS) )
-OCV_OPTION(WITH_CUFFT          "Include NVidia Cuda Fast Fourier Transform (FFT) library support"            ON  IF (CMAKE_VERSION VERSION_GREATER "2.8" AND NOT ANDROID AND NOT IOS) )
+OCV_OPTION(WITH_CUFFT          "Include NVidia Cuda Fast Fourier Transform (FFT) library support"            ON  IF (CMAKE_VERSION VERSION_GREATER "2.8" AND NOT IOS) )
-OCV_OPTION(WITH_CUBLAS         "Include NVidia Cuda Basic Linear Algebra Subprograms (BLAS) library support" OFF IF (CMAKE_VERSION VERSION_GREATER "2.8" AND NOT ANDROID AND NOT IOS) )
+OCV_OPTION(WITH_CUBLAS         "Include NVidia Cuda Basic Linear Algebra Subprograms (BLAS) library support" OFF IF (CMAKE_VERSION VERSION_GREATER "2.8" AND NOT IOS) )
 OCV_OPTION(WITH_NVCUVID        "Include NVidia Video Decoding library support"                               OFF IF (CMAKE_VERSION VERSION_GREATER "2.8" AND NOT ANDROID AND NOT IOS AND NOT APPLE) )
 OCV_OPTION(WITH_EIGEN          "Include Eigen2/Eigen3 support"               ON)
 OCV_OPTION(WITH_VFW            "Include Video for Windows support"           ON   IF WIN32 )

cmake/FindCUDA.cmake

@@ -759,11 +759,15 @@ endif()
 set(CUDA_VERSION_STRING "${CUDA_VERSION}")
 
 # Support for arm cross compilation with CUDA 5.5
-if(CUDA_VERSION VERSION_GREATER "5.0" AND CMAKE_CROSSCOMPILING AND ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm" AND EXISTS "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabihf")
+set(__cuda_toolkit_target_dir_initial "${CUDA_TOOLKIT_ROOT_DIR}")
-  set(CUDA_TOOLKIT_TARGET_DIR "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabihf" CACHE PATH "Toolkit target location.")
+if(CUDA_VERSION VERSION_GREATER "5.0" AND CMAKE_CROSSCOMPILING AND ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm")
-else()
+  if(ANDROID AND EXISTS "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-androideabi")
-  set(CUDA_TOOLKIT_TARGET_DIR "${CUDA_TOOLKIT_ROOT_DIR}" CACHE PATH "Toolkit target location.")
+    set(__cuda_toolkit_target_dir_initial "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-androideabi")
+  elseif(EXISTS "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabihf")
+    set(__cuda_toolkit_target_dir_initial "${CUDA_TOOLKIT_ROOT_DIR}/targets/armv7-linux-gnueabihf")
+  endif()
 endif()
+set(CUDA_TOOLKIT_TARGET_DIR "${__cuda_toolkit_target_dir_initial}" CACHE PATH "Toolkit target location.")
 mark_as_advanced(CUDA_TOOLKIT_TARGET_DIR)
 
 # Target CPU architecture

cmake/OpenCVDetectCUDA.cmake

@@ -90,12 +90,17 @@ if(CUDA_FOUND)
   endif()
 
   if(NOT DEFINED __cuda_arch_bin)
-    if(${CUDA_VERSION} VERSION_LESS "5.0")
+    if(ANDROID)
-      set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")
+      set(__cuda_arch_bin "3.2")
+      set(__cuda_arch_ptx "")
     else()
-      set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")
+      if(${CUDA_VERSION} VERSION_LESS "5.0")
+        set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0")
+      else()
+        set(__cuda_arch_bin "1.1 1.2 1.3 2.0 2.1(2.0) 3.0 3.5")
+      endif()
+      set(__cuda_arch_ptx "3.0")
     endif()
-    set(__cuda_arch_ptx "3.0")
   endif()
 
   set(CUDA_ARCH_BIN ${__cuda_arch_bin} CACHE STRING "Specify 'real' GPU architectures to build binaries for, BIN(PTX) format is supported")
@@ -147,6 +152,11 @@ if(CUDA_FOUND)
   set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})
   set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
 
+  if(ANDROID)
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xptxas;-dlcm=ca")
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-target-os-variant=Android")
+  endif()
+
   message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
 
   OCV_OPTION(CUDA_FAST_MATH "Enable --use_fast_math for CUDA compiler " OFF)
@@ -169,6 +179,14 @@ if(CUDA_FOUND)
 
       # we remove -Wsign-promo as it generates warnings under linux
       string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
+
+      # we remove -Wno-delete-non-virtual-dtor because it's used for C++ compiler
+      # but NVCC uses C compiler by default
+      string(REPLACE "-Wno-delete-non-virtual-dtor" "" ${var} "${${var}}")
+
+      # we remove -frtti because it's used for C++ compiler
+      # but NVCC uses C compiler by default
+      string(REPLACE "-frtti" "" ${var} "${${var}}")
     endforeach()
 
     if(BUILD_SHARED_LIBS)

modules/gpu/CMakeLists.txt

@@ -1,4 +1,4 @@
-if(ANDROID OR IOS)
+if(IOS)
   ocv_module_disable(gpu)
 endif()
 

samples/cpp/tutorial_code/gpu/gpu-basics-similarity/gpu-basics-similarity.cpp

@@ -69,7 +69,7 @@ int main(int, char *argv[])
     BufferPSNR bufferPSNR;
     BufferMSSIM bufferMSSIM;
 
-    int TIMES;
+    int TIMES = 10;
     stringstream sstr(argv[3]);
     sstr >> TIMES;
     double time, result = 0;