adapted dependency settings to new CUDA variables

.github/workflows/continuous-integration.yml

@@ -27,6 +27,10 @@ jobs:
         # excludes debug on this one as it has a segmentation fault during the compilation (!)
         - container: "alicevision/popsift-deps:cuda12.1.0-ubuntu22.04"
           build_tpe: "Debug"
+        - container: "alicevision/popsift-deps:cuda12.1.0-ubuntu22.04"
+          build_tpe: "Release"
+        - container: "alicevision/popsift-deps:cuda11.8.0-ubuntu20.04"
+          build_tpe: "Debug"
 
     container:
       image: ${{ matrix.container }}

CMakeLists.txt

@@ -7,7 +7,7 @@ project(PopSift VERSION 1.0.0 LANGUAGES CXX CUDA)
 # Since CMake 3.18. See https://cmake.org/cmake/help/latest/policy/CMP0104.html
 cmake_policy(SET CMP0104 NEW)
 
-set(CMAKE_CUDA_ARCHITECTURES "all-major"
+set(CMAKE_CUDA_ARCHITECTURES "native"
     CACHE
     STRING "Which CUDA CCs to support: native, all, all-major or an explicit list delimited by semicolons"
     FORCE)

src/CMakeLists.txt

@@ -1,6 +1,6 @@
 
 add_library(popsift SHARED
-        popsift/popsift.cpp popsift/popsift.h
+        popsift/popsift.cu popsift/popsift.h
         popsift/features.cu popsift/features.h
         popsift/sift_constants.cu popsift/sift_constants.h
         popsift/sift_conf.cu popsift/sift_conf.h
@@ -41,8 +41,11 @@ add_library(popsift SHARED
 
 target_link_libraries(popsift
                       PUBLIC
-		      CUDA::nvtx3
+		      CUDA::cudart
+		      CUDA::nvToolsExt
 		      Threads::Threads)
+# expected library to link was: "CUDA::nvtx3" but it appears to be simply nvToolsExt
+# or maybe ${NVTX_LIBRARY} ???
 
 # It is necessary to choose between shared and static, otherwise the CUDA_RUNTIME_LIBRARY target property
 # will not be set.
@@ -58,13 +61,15 @@ set(popsift_generated_dir "${CMAKE_CURRENT_BINARY_DIR}/generated")
 
 # BUILD_INTERFACE allows to include the directory with source only when target is
 # built in the building tree (ie, not from an install location)
+# The CUDA install dir variable has changed from the old CUDA_INCLUDE_DIRS to the new CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES
 target_include_directories(popsift
             PUBLIC
             $<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}/src>
             $<BUILD_INTERFACE:${popsift_generated_dir}>
             $<BUILD_INTERFACE:${popsift_generated_dir}/popsift>
             $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}/>
-            ${CUDA_INCLUDE_DIRS})
+	    ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES}
+	    )
 
 # EXPORTING THE LIBRARY
 #

src/application/CMakeLists.txt

@@ -73,7 +73,7 @@ add_executable(popsift-demo  main.cpp pgmread.cpp pgmread.h)
 set_property(TARGET popsift-demo PROPERTY CXX_STANDARD 11)
 
 target_compile_options(popsift-demo PRIVATE ${PD_COMPILE_OPTIONS} )
-target_include_directories(popsift-demo PUBLIC ${PD_INCLUDE_DIRS})
+target_include_directories(popsift-demo PUBLIC PopSift::popsift ${PD_INCLUDE_DIRS})
 target_compile_definitions(popsift-demo PRIVATE ${Boost_DEFINITIONS})
 target_link_libraries(popsift-demo PUBLIC PopSift::popsift ${PD_LINK_LIBS})
 
@@ -87,7 +87,7 @@ add_executable(popsift-match match.cpp pgmread.cpp pgmread.h)
 set_property(TARGET popsift-match PROPERTY CXX_STANDARD 11)
 
 target_compile_options(popsift-match PRIVATE ${PD_COMPILE_OPTIONS} )
-target_include_directories(popsift-match PUBLIC ${PD_INCLUDE_DIRS})
+target_include_directories(popsift-match PUBLIC PopSift::popsift ${PD_INCLUDE_DIRS})
 target_compile_definitions(popsift-match PRIVATE ${Boost_DEFINITIONS})
 target_link_libraries(popsift-match PUBLIC PopSift::popsift ${PD_LINK_LIBS})
 

src/application/main.cpp

@@ -30,13 +30,6 @@
 #endif
 #include "pgmread.h"
 
-#if POPSIFT_IS_DEFINED(POPSIFT_USE_NVTX)
-#include <nvToolsExtCuda.h>
-#else
-#define nvtxRangePushA(a)
-#define nvtxRangePop()
-#endif
-
 using namespace std;
 
 static bool print_dev_info  = false;
@@ -183,8 +176,6 @@ SiftJob* process_image( const string& inputFile, PopSift& PopSift )
             exit( -1 );
         }
 
-        nvtxRangePushA( "load and convert image - devil" );
-
         ilImage img;
         if( img.Load( inputFile.c_str() ) == false ) {
             cerr << "Could not load image " << inputFile << endl;
@@ -200,25 +191,20 @@ SiftJob* process_image( const string& inputFile, PopSift& PopSift )
 
         image_data = img.GetData();
 
-        nvtxRangePop( ); // "load and convert image - devil"
-
         job = PopSift.enqueue( w, h, image_data );
 
         img.Clear();
     }
     else
 #endif
     {
-        nvtxRangePushA( "load and convert image - pgmread" );
         int w{};
         int h{};
         image_data = readPGMfile( inputFile, w, h );
         if( image_data == nullptr ) {
             exit( EXIT_FAILURE );
         }
 
-        nvtxRangePop( ); // "load and convert image - pgmread"
-
         if( ! float_mode )
         {
             // PopSift.init( w, h );
@@ -251,16 +237,10 @@ void read_job( SiftJob* job, bool really_write )
          << endl;
 
     if( really_write ) {
-        nvtxRangePushA( "Writing features to disk" );
-
         std::ofstream of( "output-features.txt" );
         feature_list->print( of, write_as_uchar );
     }
     delete feature_list;
-
-    if( really_write ) {
-        nvtxRangePop( ); // Writing features to disk
-    }
 }
 
 int main(int argc, char **argv)

src/application/match.cpp

@@ -30,13 +30,6 @@
 #endif
 #include "pgmread.h"
 
-#if POPSIFT_IS_DEFINED(POPSIFT_USE_NVTX)
-#include <nvToolsExtCuda.h>
-#else
-#define nvtxRangePushA(a)
-#define nvtxRangePop()
-#endif
-
 using namespace std;
 
 static bool print_dev_info  {false};
@@ -171,7 +164,6 @@ SiftJob* process_image( const string& inputFile, PopSift& PopSift )
     unsigned char* image_data;
     SiftJob* job;
 
-    nvtxRangePushA( "load and convert image" );
 #ifdef USE_DEVIL
     if( ! pgmread_loading )
     {
@@ -189,8 +181,6 @@ SiftJob* process_image( const string& inputFile, PopSift& PopSift )
         cout << "Loading " << w << " x " << h << " image " << inputFile << endl;
         image_data = img.GetData();
 
-        nvtxRangePop( );
-
         // PopSift.init( w, h );
         job = PopSift.enqueue( w, h, image_data );
 
@@ -206,8 +196,6 @@ SiftJob* process_image( const string& inputFile, PopSift& PopSift )
             exit( EXIT_FAILURE );
         }
 
-        nvtxRangePop( );
-
         // PopSift.init( w, h );
         job = PopSift.enqueue( w, h, image_data );