Add backend shared lib build to cmake

ericwait · May 7, 2024 · 64f97a8 · 64f97a8
1 parent e87a6a5
commit 64f97a8
Show file tree

Hide file tree

Showing 6 changed files with 173 additions and 63 deletions.
diff --git a/src/c/Cuda/CHelpers.cpp b/src/c/Cuda/CHelpers.cpp
@@ -13,9 +13,9 @@ float* createEllipsoidKernel(Vec<std::size_t> radii, Vec<std::size_t>& kernelDim
 	std::memset(kernel,0,sizeof(float)*kernelDims.product());
 
 	Vec<int> mid((kernelDims-1)/2);
-	Vec<float> dimScale = Vec<float>(1,1,1) / Vec<float>(radii.pwr(2));
+	Vec<float> dimScale = Vec<float>(1.0f) / Vec<float>(radii.pwr(2));
 
-	Vec<int> cur(0,0,0);
+	Vec<int> cur(0);
 	for (cur.z=0; cur.z<kernelDims.z; ++cur.z)
 	{
 		for (cur.y=0; cur.y<kernelDims.y; ++cur.y)

diff --git a/src/c/Cuda/CHelpers.h b/src/c/Cuda/CHelpers.h
@@ -1,19 +1,27 @@
 #pragma once
 
-#include "Vec.h"
-#include <memory.h>
-#include <complex>
+#ifdef _WIN32
+    #define DLL_EXPORT_API __declspec(dllexport)
+    #define DLL_IMPORT_API __declspec(dllimport)
+#else
+    #define DLL_EXPORT_API __attribute__((__visibility("default")))
+    #define DLL_IMPORT_API 
+#endif 
 
 #ifdef IMAGE_PROCESSOR_DLL
-#ifdef IMAGE_PROCESSOR_INTERNAL
-#define IMAGE_PROCESSOR_API __declspec(dllexport)
-#else
-#define IMAGE_PROCESSOR_API __declspec(dllimport)
-#endif // IMAGE_PROCESSOR_INTERNAL
+    #ifdef IMAGE_PROCESSOR_EXPORT
+        #define IMAGE_PROCESSOR_API DLL_EXPORT_API
+    #else
+        #define IMAGE_PROCESSOR_API DLL_IMPORT_API
+    #endif // IMAGE_PROCESSOR_EXPORT
 #else
-#define IMAGE_PROCESSOR_API
+    #define IMAGE_PROCESSOR_API
 #endif // IMAGE_PROCESSOR_DLL
 
+#include "Vec.h"
+#include <memory.h>
+#include <complex>
+
 IMAGE_PROCESSOR_API float* createEllipsoidKernel(Vec<std::size_t> radii, Vec<std::size_t>& kernelDims);
 
 IMAGE_PROCESSOR_API int calcOtsuThreshold(const double* normHistogram, int numBins);
diff --git a/src/c/Cuda/CMakeLists.txt b/src/c/Cuda/CMakeLists.txt
@@ -1,9 +1,8 @@
 # Setup backend Hydra CUDA library (static) for cuda building
-add_library(HydraCudaLib STATIC "")
+add_library(HydraCudaStatic STATIC "")
 
-target_compile_features(HydraCudaLib PRIVATE cxx_std_11)
-target_link_libraries(HydraCudaLib PRIVATE OpenMP::OpenMP_CXX)
-set_property(TARGET HydraCudaLib PROPERTY CUDA_STANDARD 11)
+target_compile_features(HydraCudaStatic PRIVATE cxx_std_11)
+target_link_libraries(HydraCudaStatic PRIVATE OpenMP::OpenMP_CXX)
 
 # Set a variable to turn on/off PROCESS_MUTEX support
 set(USE_PROCESS_MUTEX ON CACHE BOOL "Use process-level mutex to guard GPU calls")
@@ -18,15 +17,16 @@ set(USE_PROCESS_MUTEX ON CACHE BOOL "Use process-level mutex to guard GPU calls"
 # string(APPEND CMAKE_CUDA_FLAGS " -gencode=arch=compute_61,code=sm_61")
 # string(APPEND CMAKE_CUDA_FLAGS " -gencode=arch=compute_70,code=compute_70")
 
-set_target_properties(HydraCudaLib PROPERTIES CUDA_ARCHITECTURES "52;60;61;70")
-
-set_target_properties(HydraCudaLib
+set_target_properties(HydraCudaStatic
     PROPERTIES 
         POSITION_INDEPENDENT_CODE ON
+        CUDA_STANDARD 11
         CUDA_SEPARABLE_COMPILATION ON
-        CUDA_RESOLVE_DEVICE_SYMBOLS ON)
+        CUDA_RESOLVE_DEVICE_SYMBOLS ON
+        CUDA_ARCHITECTURES "52;60;61;70"
+)
 
-target_sources(HydraCudaLib
+target_sources(HydraCudaStatic
     PUBLIC
         BufferConversions.h
         CHelpers.h
@@ -89,7 +89,101 @@ target_sources(HydraCudaLib
         KernelIterator.cu
 )
 
-target_include_directories(HydraCudaLib 
+target_include_directories(HydraCudaStatic 
+    PUBLIC
+        $<INSTALL_INTERFACE:include>
+        $<BUILD_INTERFACE:${CMAKE_CURRENT_LIST_DIR}>
+        $<BUILD_INTERFACE:${CMAKE_CURRENT_LIST_DIR}/..>
+    PRIVATE
+        ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES}
+)
+
+# Backend shared lib for dynamic c++ linking
+add_library(HydraCuda SHARED "")
+
+target_compile_features(HydraCuda PRIVATE cxx_std_11)
+target_link_libraries(HydraCuda 
+    PRIVATE
+        OpenMP::OpenMP_CXX
+)
+
+# Set a variable to turn on/off PROCESS_MUTEX support
+# set(USE_PROCESS_MUTEX ON CACHE BOOL "Use process-level mutex to guard GPU calls")
+set_target_properties(HydraCuda 
+    PROPERTIES
+        CUDA_STANDARD 11
+        CUDA_ARCHITECTURES "52;60;61;70"
+        CUDA_SEPARABLE_COMPILATION ON
+        CUDA_RESOLVE_DEVICE_SYMBOLS ON
+        CXX_VISIBILITY_PRESET hidden
+        POSITION_INDEPENDENT_CODE ON
+        LIBRARY_OUTPUT_DIRECTORY $<BUILD_INTERFACE:${PROJECT_BINARY_DIR}>
+        LIBRARY_OUTPUT_DIRECTORY $<INSTALL_INTERFACE:lib>
+)
+
+target_sources(HydraCuda
+    PUBLIC
+        BufferConversions.h
+        CHelpers.h
+        CommandLineProgress.h
+        CudaDeviceInfo.h
+        CudaDeviceStats.h
+        CudaUtilities.h
+        CWrapperAutogen.h
+        CWrappers.h
+        Defines.h
+        ImageChunk.h
+        ImageView.h
+        KernelGenerators.h
+        Vec.h
+    PRIVATE
+        CudaAddTwoImages.cuh
+        CudaClosure.cuh
+        CudaDeviceCount.cuh
+        CudaDeviceImages.cuh
+        CudaElementWiseDifference.cuh
+        CudaEntropyFilter.cuh
+        CudaGaussian.cuh
+        CudaGetMinMax.cuh
+        CudaHighPassFilter.cuh
+        CudaIdentityFilter.cuh
+        CudaImageContainer.cuh
+        CudaImageContainerClean.cuh
+        CudaLoG.cuh
+        CudaMaxFilter.cuh
+        CudaMeanAndVariance.cuh
+        CudaMeanFilter.cuh
+        CudaMedianFilter.cuh
+        CudaMemoryStats.cuh
+        CudaMinFilter.cuh
+        CudaMinMax.cuh
+        CudaMultiplySum.cuh
+        CudaNLMeans.cuh
+        CudaOpener.cuh
+        CudaStdFilter.cuh
+        CudaSum.cuh
+        CudaVarFilter.cuh
+        CudaWienerFilter.cuh
+        ImageDimensions.cuh
+        Kernel.cuh
+        KernelIterator.cuh
+        SeparableMultiplySum.cuh
+
+        CHelpers.cpp
+        CudaDeviceInfo.cpp
+        CudaDeviceStats.cpp
+        CudaUtilities.cpp
+        GaussianKernel.cpp
+        ImageChunk.cpp
+        LoGKernel.cpp
+
+        CWrapperAutogen.cu
+        CWrappers.cu
+        Kernel.cu
+        KernelIterator.cu
+)
+
+target_include_directories(HydraCuda
     PUBLIC
         $<INSTALL_INTERFACE:include>
         $<BUILD_INTERFACE:${CMAKE_CURRENT_LIST_DIR}>

diff --git a/src/c/Cuda/CWrappers.h b/src/c/Cuda/CWrappers.h
@@ -1,40 +1,48 @@
-#pragma once
-#include "Vec.h"
-#include "CudaDeviceStats.h"
-#include "ImageView.h"
-
-#include <limits>
-#include <algorithm>
-
-
-#ifdef IMAGE_PROCESSOR_DLL
-#ifdef IMAGE_PROCESSOR_INTERNAL
-#define IMAGE_PROCESSOR_API __declspec(dllexport)
-#else
-#define IMAGE_PROCESSOR_API __declspec(dllimport)
-#endif // IMAGE_PROCESSOR_INTERNAL
-#else
-#define IMAGE_PROCESSOR_API
-#endif // IMAGE_PROCESSOR_DLL
-
-
-#include "CWrapperAutogen.h"
-
-
-IMAGE_PROCESSOR_API void clearDevice();
-
-IMAGE_PROCESSOR_API int deviceCount();
-IMAGE_PROCESSOR_API int deviceStats(DevStats** stats);
-IMAGE_PROCESSOR_API int memoryStats(std::size_t** stats);
-
-
-/// Example wrapper header calls 
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<bool> imageIn, ImageView<bool> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<char> imageIn, ImageView<char> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<unsigned char> imageIn, ImageView<unsigned char> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<short> imageIn, ImageView<short> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<unsigned short> imageIn, ImageView<unsigned short> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<int> imageIn, ImageView<int> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<unsigned int> imageIn, ImageView<unsigned int> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<float> imageIn, ImageView<float> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
-//IMAGE_PROCESSOR_API void fooFilter(const ImageView<double> imageIn, ImageView<double> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+#pragma once
+
+#ifdef _WIN32
+    #define DLL_EXPORT_API __declspec(dllexport)
+    #define DLL_IMPORT_API __declspec(dllimport)
+#else
+    #define DLL_EXPORT_API __attribute__((visibility("default")))
+    #define DLL_IMPORT_API 
+#endif 
+
+#ifdef IMAGE_PROCESSOR_DLL
+    #ifdef IMAGE_PROCESSOR_EXPORT
+        #define IMAGE_PROCESSOR_API DLL_EXPORT_API
+    #else
+        #define IMAGE_PROCESSOR_API DLL_IMPORT_API
+    #endif // IMAGE_PROCESSOR_EXPORT
+#else
+    #define IMAGE_PROCESSOR_API
+#endif // IMAGE_PROCESSOR_DLL
+
+
+#include <limits>
+#include <algorithm>
+
+#include "Vec.h"
+#include "ImageView.h"
+#include "CudaDeviceStats.h"
+
+#include "CWrapperAutogen.h"
+
+
+IMAGE_PROCESSOR_API void clearDevice();
+
+IMAGE_PROCESSOR_API int deviceCount();
+IMAGE_PROCESSOR_API int deviceStats(DevStats** stats);
+IMAGE_PROCESSOR_API int memoryStats(std::size_t** stats);
+
+
+/// Example wrapper header calls 
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<bool> imageIn, ImageView<bool> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<char> imageIn, ImageView<char> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<unsigned char> imageIn, ImageView<unsigned char> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<short> imageIn, ImageView<short> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<unsigned short> imageIn, ImageView<unsigned short> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<int> imageIn, ImageView<int> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<unsigned int> imageIn, ImageView<unsigned int> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<float> imageIn, ImageView<float> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
+//IMAGE_PROCESSOR_API void fooFilter(const ImageView<double> imageIn, ImageView<double> imageOut, ImageView<float> kernel, int numIterations = 1, int device = -1);
diff --git a/src/c/Mex/CMakeLists.txt b/src/c/Mex/CMakeLists.txt
@@ -14,7 +14,7 @@ if ( USE_PROCESS_MUTEX )
 endif()
 
 # Link against Matlab libraries
-target_link_libraries(HydraMex PRIVATE HydraCudaLib ${Matlab_MX_LIBRARY} ${Matlab_MEX_LIBRARY})
+target_link_libraries(HydraMex PRIVATE HydraCudaStatic ${Matlab_MX_LIBRARY} ${Matlab_MEX_LIBRARY})
 
 # Change output library name to <ModuleName>.mex<plaform>
 set_target_properties(HydraMex

diff --git a/src/c/Python/CMakeLists.txt b/src/c/Python/CMakeLists.txt
@@ -11,7 +11,7 @@ if ( USE_PROCESS_MUTEX )
 endif()
 
 # Link against Python and NumPy libraries
-target_link_libraries(HydraPy PRIVATE HydraCudaLib Python::Python Python::NumPy)
+target_link_libraries(HydraPy PRIVATE HydraCudaStatic Python::Python Python::NumPy)
 
 # Change output library name to Hydra.<dll/so/dylib>
 set_target_properties(HydraPy