Added CUDA fast math flag

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 83bea07..25bf7c8 100644 (file)
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -768,6 +768,7 @@ if(HAVE_CUDA)
   status("    Use CUBLAS:"           HAVE_CUBLAS THEN YES ELSE NO)
   status("    NVIDIA GPU arch:"      ${OPENCV_CUDA_ARCH_BIN})
   status("    NVIDIA PTX archs:"     ${OPENCV_CUDA_ARCH_PTX})
+  status("    Use fast math:"        CUDA_FAST_MATH THEN YES ELSE NO)   
 endif()
 
 # ========================== python ==========================

diff --git a/cmake/OpenCVDetectCUDA.cmake b/cmake/OpenCVDetectCUDA.cmake
index 9b3fcd1..a777a46 100644 (file)
--- a/cmake/OpenCVDetectCUDA.cmake
+++ b/cmake/OpenCVDetectCUDA.cmake
@@ -78,6 +78,14 @@ if(CUDA_FOUND)
   set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
 
   message(STATUS "CUDA NVCC target flags: ${CUDA_NVCC_FLAGS}")
+  
+  OCV_OPTION(CUDA_FAST_MATH  "Enable --use_fast_math for CUDA compiler " OFF)
+  
+  if(ENABLE_CUDA_MATH)
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --use_fast_math)    
+  endif()
+  
+  mark_as_advanced(CUDA_BUILD_CUBIN CUDA_BUILD_EMULATION CUDA_VERBOSE_BUILD CUDA_SDK_ROOT_DIR)
 
   unset(CUDA_npp_LIBRARY CACHE)
   find_cuda_helper_libs(npp)