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

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0ee590e..0deabb1 100644 (file)
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -707,28 +707,42 @@ if (WITH_CUDA)
         message(STATUS "CUDA detected: " ${CUDA_VERSION})\r
 \r
                set(CUDA_COMPUTE_CAPABILITIES " 1.1 1.2 1.3 2.0 " CACHE STRING "Add or remove compute capability")\r
-    set(CUDA_NVCC_FLAGS_ARCH ${CUDA_COMPUTE_CAPABILITIES})\r
-\r
-    set(CUDA_NVCC_FLAGS_NUM "")\r
-\r
-    while(NOT ${CUDA_NVCC_FLAGS_ARCH} STREQUAL "")\r
-        string(REGEX MATCH "[0-9]+.[0-9]+" RESULT_NUM ${CUDA_NVCC_FLAGS_ARCH})\r
-        string(REGEX MATCHALL "[0-9]" RESULT_STR ${RESULT_NUM})\r
-        string(REGEX REPLACE ";" "\ " RESULT ${RESULT_STR})\r
-        list(APPEND CUDA_NVCC_FLAGS_NUM ${RESULT})\r
-        string(REGEX REPLACE "${RESULT_NUM}" "\ " CUDA_NVCC_FLAGS_ARCH_STR ${CUDA_NVCC_FLAGS_ARCH})\r
-        string(STRIP ${CUDA_NVCC_FLAGS_ARCH_STR} CUDA_NVCC_FLAGS_ARCH)\r
-    endwhile()\r
-\r
-    set (OpenCV_CUDA_CC "")\r
-    set (loop_var "")\r
-    foreach( loop_var IN LISTS CUDA_NVCC_FLAGS_NUM)\r
-        set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -gencode arch=compute_${loop_var},code=sm_${loop_var})\r
-        set (OpenCV_CUDA_CC  ${OpenCV_CUDA_CC}  -gencode arch=compute_${loop_var},code=sm_${loop_var})\r
-    endforeach()\r
-\r
-    ### set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${OpenCV_COMPUTE_CAPABILITIES})\r
-  endif()\r
+        set(CUDA_NVCC_FLAGS_ARCH ${CUDA_COMPUTE_CAPABILITIES})\r
+        \r
+        string(REGEX MATCH "1\\.0" NVIDIA_CC_10 ${CUDA_COMPUTE_CAPABILITIES})\r
+        string(REGEX MATCH "1\\.1" NVIDIA_CC_11 ${CUDA_COMPUTE_CAPABILITIES})\r
+        string(REGEX MATCH "1\\.2" NVIDIA_CC_12 ${CUDA_COMPUTE_CAPABILITIES})\r
+        string(REGEX MATCH "1\\.3" NVIDIA_CC_13 ${CUDA_COMPUTE_CAPABILITIES})\r
+        string(REGEX MATCH "2\\.0" NVIDIA_CC_20 ${CUDA_COMPUTE_CAPABILITIES})\r
+        string(REGEX MATCH "2\\.1" NVIDIA_CC_21 ${CUDA_COMPUTE_CAPABILITIES})\r
+        \r
+        string(COMPARE EQUAL "1.0" "${NVIDIA_CC_10}" HAVE_PTX_FOR_NVIDIA_CC_10)\r
+        string(COMPARE EQUAL "1.1" "${NVIDIA_CC_11}" HAVE_PTX_FOR_NVIDIA_CC_11)\r
+        string(COMPARE EQUAL "1.2" "${NVIDIA_CC_12}" HAVE_PTX_FOR_NVIDIA_CC_12)\r
+        string(COMPARE EQUAL "1.3" "${NVIDIA_CC_13}" HAVE_PTX_FOR_NVIDIA_CC_13)\r
+        string(COMPARE EQUAL "2.0" "${NVIDIA_CC_20}" HAVE_PTX_FOR_NVIDIA_CC_20)\r
+        string(COMPARE EQUAL "2.1" "${NVIDIA_CC_21}" HAVE_PTX_FOR_NVIDIA_CC_21)\r
+\r
+        set(CUDA_NVCC_FLAGS_NUM "")\r
+\r
+        while(NOT ${CUDA_NVCC_FLAGS_ARCH} STREQUAL "")\r
+            string(REGEX MATCH "[0-9]+.[0-9]+" RESULT_NUM ${CUDA_NVCC_FLAGS_ARCH})\r
+            string(REGEX MATCHALL "[0-9]" RESULT_STR ${RESULT_NUM})\r
+            string(REGEX REPLACE ";" "\ " RESULT ${RESULT_STR})\r
+            list(APPEND CUDA_NVCC_FLAGS_NUM ${RESULT})\r
+            string(REGEX REPLACE "${RESULT_NUM}" "\ " CUDA_NVCC_FLAGS_ARCH_STR ${CUDA_NVCC_FLAGS_ARCH})\r
+            string(STRIP ${CUDA_NVCC_FLAGS_ARCH_STR} CUDA_NVCC_FLAGS_ARCH)\r
+        endwhile()\r
+\r
+        set (OpenCV_CUDA_CC "")\r
+        set (loop_var "")\r
+        foreach( loop_var IN LISTS CUDA_NVCC_FLAGS_NUM)\r
+            set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} -gencode arch=compute_${loop_var},code=sm_${loop_var})\r
+            set (OpenCV_CUDA_CC  ${OpenCV_CUDA_CC}  -gencode arch=compute_${loop_var},code=sm_${loop_var})\r
+        endforeach()\r
+\r
+        ### set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${OpenCV_COMPUTE_CAPABILITIES})\r
+    endif()\r
 endif()\r
 \r
 \r

diff --git a/cvconfig.h.cmake b/cvconfig.h.cmake
index d21d8c0..9a9a6a3 100644 (file)
--- a/cvconfig.h.cmake
+++ b/cvconfig.h.cmake
@@ -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
 

diff --git a/doc/gpu_initialization.tex b/doc/gpu_initialization.tex
index 92ff89b..7c341ab 100644 (file)
--- a/doc/gpu_initialization.tex
+++ b/doc/gpu_initialization.tex
@@ -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". 
+

diff --git a/modules/gpu/include/opencv2/gpu/gpu.hpp b/modules/gpu/include/opencv2/gpu/gpu.hpp
index 329fa43..5293241 100644 (file)
--- a/modules/gpu/include/opencv2/gpu/gpu.hpp
+++ b/modules/gpu/include/opencv2/gpu/gpu.hpp
@@ -72,6 +72,12 @@ namespace cv
         CV_EXPORTS bool hasNativeDoubleSupport(int device);\r
         CV_EXPORTS bool hasAtomicsSupport(int device);\r
 \r
+        //! Checks if the GPU module was built with PTX support (-arch) of the given CC\r
+        CV_EXPORTS bool hasPtxFor(int major, int minor);\r
+\r
+        //! Checks if the GPU module is PTX compatible with the given NVIDIA device\r
+        CV_EXPORTS bool isCompatibleWith(int device);\r
+\r
         //////////////////////////////// Error handling ////////////////////////\r
 \r
         CV_EXPORTS void error(const char *error_string, const char *file, const int line, const char *func);\r

diff --git a/modules/gpu/src/initialization.cpp b/modules/gpu/src/initialization.cpp
index fae118e..0ce9f6a 100644 (file)
--- a/modules/gpu/src/initialization.cpp
+++ b/modules/gpu/src/initialization.cpp
@@ -122,5 +122,57 @@ CV_EXPORTS bool cv::gpu::hasAtomicsSupport(int device)
     return major > 1 || (major == 1 && minor >= 1);\r
 }\r
 \r
+CV_EXPORTS bool cv::gpu::hasPtxFor(int major, int minor) \r
+{\r
+#ifdef HAVE_PTX_FOR_NVIDIA_CC_10\r
+    if (major == 1 && minor == 0) return true;\r
+#endif\r
+\r
+#ifdef HAVE_PTX_FOR_NVIDIA_CC_11\r
+    if (major == 1 && minor == 1) return true;\r
+#endif\r
+\r
+#ifdef HAVE_PTX_FOR_NVIDIA_CC_12\r
+    if (major == 1 && minor == 2) return true;\r
+#endif\r
+\r
+#ifdef HAVE_PTX_FOR_NVIDIA_CC_13\r
+    if (major == 1 && minor == 3) return true;\r
+#endif\r
+\r
+#ifdef HAVE_PTX_FOR_NVIDIA_CC_20\r
+    if (major == 2 && minor == 0) return true;\r
+#endif\r
+\r
+#ifdef HAVE_PTX_FOR_NVIDIA_CC_21\r
+    if (major == 2 && minor == 1) return true;\r
+#endif\r
+\r
+    return false;\r
+}\r
+\r
+\r
+CV_EXPORTS bool isCompatibleWith(int device)\r
+{\r
+    // According to the CUDA C Programming Guide Version 3.2: "PTX code \r
+    // produced for some specific compute capability can always be compiled to\r
+    // binary code of greater or equal compute capability". \r
+\r
+    int major, minor;\r
+    getComputeCapability(device, major, minor);\r
+\r
+    for (; major >= 1; --major)\r
+    {\r
+        for (; minor >= 0; --minor)\r
+        {\r
+            if (hasPtxFor(major, minor))\r
+                return true;\r
+        }\r
+        minor = 9;\r
+    }\r
+\r
+    return false;\r
+}\r
+\r
 #endif\r
 \r