Skip to content

O
opencv
Commit f3a26568 authored by Alexey Spizhevoy's avatar Alexey Spizhevoy
Browse files
Options

added hasPtxFor and isCompatibleWith functions into gpu module, added docs for them

parent 566befe9
Hide whitespace changes
Inline Side-by-side
Showing with 137 additions and 26 deletions
CMakeLists.txt
View file @ f3a26568
......@@ -707,28 +707,42 @@ if (WITH_CUDA)
message(STATUS "CUDA detected: " ${CUDA_VERSION})
set(CUDA_COMPUTE_CAPABILITIES " 1.1 1.2 1.3 2.0 " CACHE STRING "Add or remove compute capability")
set(CUDA_NVCC_FLAGS_ARCH ${CUDA_COMPUTE_CAPABILITIES})
set(CUDA_NVCC_FLAGS_NUM "")
while(NOT ${CUDA_NVCC_FLAGS_ARCH} STREQUAL "")
string(REGEX MATCH "[0-9]+.[0-9]+" RESULT_NUM ${CUDA_NVCC_FLAGS_ARCH})
string(REGEX MATCHALL "[0-9]" RESULT_STR ${RESULT_NUM})
string(REGEX REPLACE ";" "\ " RESULT ${RESULT_STR})
list(APPEND CUDA_NVCC_FLAGS_NUM ${RESULT})
string(REGEX REPLACE "${RESULT_NUM}" "\ " CUDA_NVCC_FLAGS_ARCH_STR ${CUDA_NVCC_FLAGS_ARCH})
string(STRIP ${CUDA_NVCC_FLAGS_ARCH_STR} CUDA_NVCC_FLAGS_ARCH)
endwhile()
set (OpenCV_CUDA_CC "")
set (loop_var "")
foreach( loop_var IN LISTS CUDA_NVCC_FLAGS_NUM)
set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -gencode arch=compute_${loop_var},code=sm_${loop_var})
set (OpenCV_CUDA_CC ${OpenCV_CUDA_CC} -gencode arch=compute_${loop_var},code=sm_${loop_var})
endforeach()
### set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${OpenCV_COMPUTE_CAPABILITIES})
endif()
set(CUDA_NVCC_FLAGS_ARCH ${CUDA_COMPUTE_CAPABILITIES})
string(REGEX MATCH "1\\.0" NVIDIA_CC_10 ${CUDA_COMPUTE_CAPABILITIES})
string(REGEX MATCH "1\\.1" NVIDIA_CC_11 ${CUDA_COMPUTE_CAPABILITIES})
string(REGEX MATCH "1\\.2" NVIDIA_CC_12 ${CUDA_COMPUTE_CAPABILITIES})
string(REGEX MATCH "1\\.3" NVIDIA_CC_13 ${CUDA_COMPUTE_CAPABILITIES})
string(REGEX MATCH "2\\.0" NVIDIA_CC_20 ${CUDA_COMPUTE_CAPABILITIES})
string(REGEX MATCH "2\\.1" NVIDIA_CC_21 ${CUDA_COMPUTE_CAPABILITIES})
string(COMPARE EQUAL "1.0" "${NVIDIA_CC_10}" HAVE_PTX_FOR_NVIDIA_CC_10)
string(COMPARE EQUAL "1.1" "${NVIDIA_CC_11}" HAVE_PTX_FOR_NVIDIA_CC_11)
string(COMPARE EQUAL "1.2" "${NVIDIA_CC_12}" HAVE_PTX_FOR_NVIDIA_CC_12)
string(COMPARE EQUAL "1.3" "${NVIDIA_CC_13}" HAVE_PTX_FOR_NVIDIA_CC_13)
string(COMPARE EQUAL "2.0" "${NVIDIA_CC_20}" HAVE_PTX_FOR_NVIDIA_CC_20)
string(COMPARE EQUAL "2.1" "${NVIDIA_CC_21}" HAVE_PTX_FOR_NVIDIA_CC_21)
set(CUDA_NVCC_FLAGS_NUM "")
while(NOT ${CUDA_NVCC_FLAGS_ARCH} STREQUAL "")
string(REGEX MATCH "[0-9]+.[0-9]+" RESULT_NUM ${CUDA_NVCC_FLAGS_ARCH})
string(REGEX MATCHALL "[0-9]" RESULT_STR ${RESULT_NUM})
string(REGEX REPLACE ";" "\ " RESULT ${RESULT_STR})
list(APPEND CUDA_NVCC_FLAGS_NUM ${RESULT})
string(REGEX REPLACE "${RESULT_NUM}" "\ " CUDA_NVCC_FLAGS_ARCH_STR ${CUDA_NVCC_FLAGS_ARCH})
string(STRIP ${CUDA_NVCC_FLAGS_ARCH_STR} CUDA_NVCC_FLAGS_ARCH)
endwhile()
set (OpenCV_CUDA_CC "")
set (loop_var "")
foreach( loop_var IN LISTS CUDA_NVCC_FLAGS_NUM)
set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -gencode arch=compute_${loop_var},code=sm_${loop_var})
set (OpenCV_CUDA_CC ${OpenCV_CUDA_CC} -gencode arch=compute_${loop_var},code=sm_${loop_var})
endforeach()
### set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${OpenCV_COMPUTE_CAPABILITIES})
endif()
endif()
......
cvconfig.h.cmake
View file @ f3a26568
......@@ -163,6 +163,24 @@
/* NVidia Cuda Runtime API*/
#cmakedefine HAVE_CUDA
/* The project was generated with 1.0 NVIDIA device arch support */
#cmakedefine HAVE_PTX_FOR_NVIDIA_CC_10
/* The project was generated with 1.1 NVIDIA device arch support */
#cmakedefine HAVE_PTX_FOR_NVIDIA_CC_11
/* The project was generated with 1.2 NVIDIA device arch support */
#cmakedefine HAVE_PTX_FOR_NVIDIA_CC_12
/* The project was generated with 1.3 NVIDIA device arch support */
#cmakedefine HAVE_PTX_FOR_NVIDIA_CC_13
/* The project was generated with 2.0 NVIDIA device arch support */
#cmakedefine HAVE_PTX_FOR_NVIDIA_CC_20
/* The project was generated with 2.1 NVIDIA device arch support */
#cmakedefine HAVE_PTX_FOR_NVIDIA_CC_21
/* VideoInput library */
#cmakedefine HAVE_VIDEOINPUT
doc/gpu_initialization.tex
View file @ f3a26568
......@@ -52,7 +52,7 @@ Returns free and total memory for the current device.
\cvCppFunc{gpu::hasNativeDoubleSupport}
Returns true if the specified GPU has native double support, false otherwise.
Returns true, if the specified GPU has native double support, otherwise false.
\cvdefCpp{bool hasNativeDoubleSupport(int device);}
\begin{description}
......@@ -61,9 +61,31 @@ Returns true if the specified GPU has native double support, false otherwise.
\cvCppFunc{gpu::hasAtomicsSupport}
Returns true if the specified GPU has atomics support, false otherwise.
Returns true, if the specified GPU has atomics support, otherwise false.
\cvdefCpp{bool hasAtomicsSupport(int device);}
\begin{description}
\cvarg{device}{GPU index. Can be obtained via \cvCppCross{gpu::getDevice}.}
\end{description}
\ No newline at end of file
\end{description}
\cvCppFunc{gpu::hasPtxFor}
Returns true, if the GPU module was built with PTX support of the given compute capability, otherwise false.
\cvdefCpp{bool hasPtxFor(int major, int minor);}
\begin{description}
\cvarg{major}{Major CC version.}
\cvarg{minor}{Minor CC version.}
\end{description}
\cvCppFunc{gpu::isCompatibleWith}
Returns true, if the GPU module is PTX compatible with the given NVIDIA GPU device, otherwise false.
\cvdefCpp{bool isCompatibleWith(int device);}
\begin{description}
\cvarg{device}{GPU index. Can be obtained via \cvCppCross{gpu::getDevice}.}
\end{description}
According to the CUDA C Programming Guide Version 3.2: "PTX code produced for some specific compute capability can always be compiled to binary code of greater or equal compute capability".
modules/gpu/include/opencv2/gpu/gpu.hpp
View file @ f3a26568
......@@ -72,6 +72,12 @@ namespace cv
CV_EXPORTS bool hasNativeDoubleSupport(int device);
CV_EXPORTS bool hasAtomicsSupport(int device);
//! Checks if the GPU module was built with PTX support (-arch) of the given CC
CV_EXPORTS bool hasPtxFor(int major, int minor);
//! Checks if the GPU module is PTX compatible with the given NVIDIA device
CV_EXPORTS bool isCompatibleWith(int device);
//////////////////////////////// Error handling ////////////////////////
CV_EXPORTS void error(const char *error_string, const char *file, const int line, const char *func);
......
modules/gpu/src/initialization.cpp
View file @ f3a26568
......@@ -122,5 +122,57 @@ CV_EXPORTS bool cv::gpu::hasAtomicsSupport(int device)
return major > 1 || (major == 1 && minor >= 1);
}
CV_EXPORTS bool cv::gpu::hasPtxFor(int major, int minor)
{
#ifdef HAVE_PTX_FOR_NVIDIA_CC_10
if (major == 1 && minor == 0) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_11
if (major == 1 && minor == 1) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_12
if (major == 1 && minor == 2) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_13
if (major == 1 && minor == 3) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_20
if (major == 2 && minor == 0) return true;
#endif
#ifdef HAVE_PTX_FOR_NVIDIA_CC_21
if (major == 2 && minor == 1) return true;
#endif
return false;
}
CV_EXPORTS bool isCompatibleWith(int device)
{
// According to the CUDA C Programming Guide Version 3.2: "PTX code
// produced for some specific compute capability can always be compiled to
// binary code of greater or equal compute capability".
int major, minor;
getComputeCapability(device, major, minor);
for (; major >= 1; --major)
{
for (; minor >= 0; --minor)
{
if (hasPtxFor(major, minor))
return true;
}
minor = 9;
}
return false;
}
#endif
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment