From: Andreas Franek <andreas.franek@igd.fraunhofer.de>
Date: Fri, 30 Oct 2015 17:28:19 +0000 (+0100)
Subject: Simplifies cross compilation with CUDA for ARM boards. CUDA arch is set for all ARM... 
X-Git-Tag: accepted/tizen/6.0/unified/20201030.111113~2163^2
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=01c5a0ed123faa46ffa7b7dba532a91896093fd1;p=platform%2Fupstream%2Fopencv.git

Simplifies cross compilation with CUDA for ARM boards. CUDA arch is set for all ARM targets, --unresolved-symbols=ignore-in-shared-libs is set when cross-compiling for ARM and FindCUDA.cmake was adjusted to find shared library stubs within the CUDA toolkit.
---

diff --git a/cmake/FindCUDA.cmake b/cmake/FindCUDA.cmake
index 5efd36c..51979a3 100644
--- a/cmake/FindCUDA.cmake
+++ b/cmake/FindCUDA.cmake
@@ -685,6 +685,9 @@ macro(cuda_find_library_local_first_with_path_ext _var _names _doc _path_ext )
     # and old paths.
     set(_cuda_64bit_lib_dir "${_path_ext}lib/x64" "${_path_ext}lib64" "${_path_ext}libx64" )
   endif()
+  if(CMAKE_CROSSCOMPILING AND (ARM OR AARCH64))
+    set(_cuda_cross_arm_lib_dir "${_path_ext}lib/stubs")
+  endif()
   if(CUDA_VERSION VERSION_GREATER "6.0")
     set(_cuda_static_lib_names "")
     foreach(name ${_names})
@@ -698,7 +701,7 @@ macro(cuda_find_library_local_first_with_path_ext _var _names _doc _path_ext )
     PATHS "${CUDA_TOOLKIT_TARGET_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}"
     ENV CUDA_PATH
     ENV CUDA_LIB_PATH
-    PATH_SUFFIXES ${_cuda_64bit_lib_dir} "${_path_ext}lib/Win32" "${_path_ext}lib" "${_path_ext}libWin32"
+    PATH_SUFFIXES ${_cuda_64bit_lib_dir} ${_cuda_cross_arm_lib_dir} "${_path_ext}lib/Win32" "${_path_ext}lib" "${_path_ext}libWin32"
     DOC ${_doc}
     NO_DEFAULT_PATH
     )
diff --git a/cmake/OpenCVDetectCUDA.cmake b/cmake/OpenCVDetectCUDA.cmake
index 5789421..c1cb958 100644
--- a/cmake/OpenCVDetectCUDA.cmake
+++ b/cmake/OpenCVDetectCUDA.cmake
@@ -79,14 +79,12 @@ if(CUDA_FOUND)
   endif()
 
   if(NOT DEFINED __cuda_arch_bin)
-    if(ANDROID)
-      if(ARM)
-        set(__cuda_arch_bin "3.2")
-        set(__cuda_arch_ptx "")
-      elseif(AARCH64)
-        set(__cuda_arch_bin "5.3")
-        set(__cuda_arch_ptx "")
-      endif()
+    if(ARM)
+      set(__cuda_arch_bin "3.2")
+      set(__cuda_arch_ptx "")
+    elseif(AARCH64)
+      set(__cuda_arch_bin "5.3")
+      set(__cuda_arch_ptx "")
     else()
       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")
@@ -200,6 +198,10 @@ if(CUDA_FOUND)
       set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xcompiler -fno-finite-math-only)
     endif()
 
+    if(CMAKE_CROSSCOMPILING AND (ARM OR AARCH64))
+      set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -Xlinker --unresolved-symbols=ignore-in-shared-libs)
+    endif()
+
     # disabled because of multiple warnings during building nvcc auto generated files
     if(CMAKE_COMPILER_IS_GNUCXX AND CMAKE_GCC_REGEX_VERSION VERSION_GREATER "4.6.0")
       ocv_warnings_disable(CMAKE_CXX_FLAGS -Wunused-but-set-variable)