From 13a6d0b92a2ff891acfd0563cb1a8e9003cbe7a0 Mon Sep 17 00:00:00 2001
From: Alexey Spizhevoy <no@email>
Date: Thu, 27 Jan 2011 08:26:10 +0000
Subject: [PATCH] fixed parsing GPU archs in BIN(PTX) format

---
 CMakeLists.txt                     | 18 ++++++++++++++----
 cvconfig.h.cmake                   |  8 ++++----
 modules/gpu/src/initialization.cpp | 10 +++++-----
 modules/gpu/src/precomp.hpp        |  2 +-
 4 files changed, 24 insertions(+), 14 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 586f8eb5ec..df8e9a2165 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -712,24 +712,32 @@ if(WITH_CUDA)
         set(CUDA_ARCH_PTX "1.1 1.3" CACHE STRING "Specify 'virtual' PTX architectures to build PTX intermediate code for")              
         
         # These variables are used in config templates
-        string(REGEX REPLACE "\\." "" ARCH_GPU_NO_POINTS "${CUDA_ARCH_BIN}")
+        string(REGEX REPLACE "\\." "" ARCH_BIN_NO_POINTS "${CUDA_ARCH_BIN}")
         string(REGEX REPLACE "\\." "" ARCH_PTX_NO_POINTS "${CUDA_ARCH_PTX}")
         
         # Ckeck if user specified 1.0 compute capability
         string(REGEX MATCH "1.0" HAS_ARCH_10 "${CUDA_ARCH_BIN} ${CUDA_ARCH_PTX}")
+        set(CUDA_ARCH_BIN_OR_PTX_10 0)
         if(NOT ${HAS_ARCH_10} STREQUAL "")
-            set(OPENCV_ARCH_GPU_OR_PTX_10 1)
+            set(CUDA_ARCH_BIN_OR_PTX_10 1)
         endif()
         
-        set(NVCC_FLAGS_EXTRA "")
+        # Flags to be set
+        set(NVCC_FLAGS_EXTRA "")      
+        
+        # These variables are passed into the template
+        set(OPENCV_CUDA_ARCH_BIN "") 
+        set(OPENCV_CUDA_ARCH_PTX "") 
         
         # Tell nvcc to add binaries for the specified GPUs
-        string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_GPU_NO_POINTS}")
+        string(REGEX MATCHALL "[0-9()]+" ARCH_LIST "${ARCH_BIN_NO_POINTS}")
         foreach(ARCH IN LISTS ARCH_LIST)
             if (ARCH MATCHES "([0-9]+)\\(([0-9]+)\\)")
                 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${CMAKE_MATCH_2},code=sm_${CMAKE_MATCH_1})
+                set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${CMAKE_MATCH_1}")
             else()
                 set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=sm_${ARCH})
+                set(OPENCV_CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN} ${ARCH}")
             endif()
         endforeach()
         
@@ -737,8 +745,10 @@ if(WITH_CUDA)
         string(REGEX MATCHALL "[0-9]+" ARCH_LIST "${ARCH_PTX_NO_POINTS}")
         foreach(ARCH IN LISTS ARCH_LIST)
             set(NVCC_FLAGS_EXTRA ${NVCC_FLAGS_EXTRA} -gencode arch=compute_${ARCH},code=compute_${ARCH})
+            set(OPENCV_CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX} ${ARCH}")
         endforeach()               
         
+        # Wil; be processed in other scripts
         set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ${NVCC_FLAGS_EXTRA})        
         set(OpenCV_CUDA_CC "${NVCC_FLAGS_EXTRA}")
         
diff --git a/cvconfig.h.cmake b/cvconfig.h.cmake
index 5b35d247df..e04cc47da0 100644
--- a/cvconfig.h.cmake
+++ b/cvconfig.h.cmake
@@ -167,13 +167,13 @@
 #cmakedefine HAVE_CUDA
 
 /* Compile for 'real' NVIDIA GPU architectures */
-#define OPENCV_ARCH_GPU "${ARCH_GPU_NO_POINTS}"
+#define CUDA_ARCH_BIN "${OPENCV_CUDA_ARCH_BIN}"
 
 /* Compile for 'virtual' NVIDIA PTX architectures */
-#define OPENCV_ARCH_PTX "${ARCH_PTX_NO_POINTS}"
+#define CUDA_ARCH_PTX "${OPENCV_CUDA_ARCH_PTX}"
 
-/* Create PTX or CUBIN for 1.0 compute capability */
-#cmakedefine OPENCV_ARCH_GPU_OR_PTX_10
+/* Create PTX or BIN for 1.0 compute capability */
+#cmakedefine CUDA_ARCH_BIN_OR_PTX_10
 
 /* VideoInput library */
 #cmakedefine HAVE_VIDEOINPUT
diff --git a/modules/gpu/src/initialization.cpp b/modules/gpu/src/initialization.cpp
index 0104bd02a2..17bd150f0f 100644
--- a/modules/gpu/src/initialization.cpp
+++ b/modules/gpu/src/initialization.cpp
@@ -164,33 +164,33 @@ namespace
 
 CV_EXPORTS bool cv::gpu::hasPtxVersion(int major, int minor)
 {
-    return ::compare(OPENCV_ARCH_PTX, major * 10 + minor, std::equal_to<int>());
+    return ::compare(CUDA_ARCH_PTX, major * 10 + minor, std::equal_to<int>());
 }
 
 
 CV_EXPORTS bool cv::gpu::hasLessOrEqualPtxVersion(int major, int minor)
 {
-    return ::compare(OPENCV_ARCH_PTX, major * 10 + minor, 
+    return ::compare(CUDA_ARCH_PTX, major * 10 + minor, 
                      std::less_equal<int>());
 }
 
 
 CV_EXPORTS bool cv::gpu::hasGreaterOrEqualPtxVersion(int major, int minor)
 {
-    return ::compare(OPENCV_ARCH_PTX, major * 10 + minor, 
+    return ::compare(CUDA_ARCH_PTX, major * 10 + minor, 
                      std::greater_equal<int>());
 }
 
 
 CV_EXPORTS bool cv::gpu::hasCubinVersion(int major, int minor)
 {
-    return ::compare(OPENCV_ARCH_GPU, major * 10 + minor, std::equal_to<int>());
+    return ::compare(CUDA_ARCH_BIN, major * 10 + minor, std::equal_to<int>());
 }
 
 
 CV_EXPORTS bool cv::gpu::hasGreaterOrEqualCubinVersion(int major, int minor)
 {
-    return ::compare(OPENCV_ARCH_GPU, major * 10 + minor, 
+    return ::compare(CUDA_ARCH_BIN, major * 10 + minor, 
                      std::greater_equal<int>());
 }
 
diff --git a/modules/gpu/src/precomp.hpp b/modules/gpu/src/precomp.hpp
index 624e67fce1..d5ad3cc727 100644
--- a/modules/gpu/src/precomp.hpp
+++ b/modules/gpu/src/precomp.hpp
@@ -85,7 +85,7 @@
     #error "Insufficient NPP version, please update it."
 #endif
 
-#if defined(OPENCV_ARCH_GPU_OR_PTX_10)
+#if defined(CUDA_ARCH_BIN_OR_PTX_10)
     #error "OpenCV GPU module doesn't support NVIDIA compute capability 1.0"
 #endif