Fixes for CUDA use-case

First fixes to revive CUDA. Still working on it. Shouldn't have broken anything out of CUDA. But CUDA might not work, yet.
tlorach · Dec 1, 2015 · 49d2936 · 49d2936
1 parent 45f901c
commit 49d2936
Show file tree

Hide file tree

Showing 25 changed files with 189 additions and 35 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -206,7 +206,7 @@ if(USE_OPENGL)
   # or even allow to use CUDA without any Gfx API...
   if(USE_CUDA AND CUDA_FOUND)
     add_subdirectory( FxLibCUDA )
-    list(APPEND NVFX_LIBS FxLibCUDA)
+    list(APPEND NVFX_LIBS FxLibCUDA) # FxLib refers to it when cmake allowed CUDA... so we need it for all :-(
   endif()
   # for now, Optix makes sense only for OpenGL, too
   if(USE_OPTIX AND OPTIX_FOUND)

diff --git a/FxLib/Container.cpp b/FxLib/Container.cpp
@@ -469,7 +469,7 @@ IShader* Container::createCUDAShader(const char *name, bool bGlobal)
             pShdRep->addShader(p);
         }
     } else {
-        p = new_ShaderCUDA(name, this);
+        p = new_ShaderCUDA(name);
         p->addUser(this);
         if(!p)
             return NULL;

diff --git a/FxLibCUDA/CUDAShader.cpp b/FxLibCUDA/CUDAShader.cpp
@@ -47,7 +47,7 @@ CUDAShader::~CUDAShader()
  ** 
  ** 
  **/ /*************************************************************************/ 
-CUDAShader::CUDAShader(const char * name, IContainer* pCont) : Shader(name, pCont)
+CUDAShader::CUDAShader(const char * name) : Shader(name)
 {
     m_targetType = TCUDA;
 }

diff --git a/FxLibCUDA/FxCUDA.h b/FxLibCUDA/FxCUDA.h
@@ -133,7 +133,7 @@ namespace nvFX
     public:
         CstBufferCUDA(const char* name) : CstBuffer(name) {}
         ~CstBufferCUDA() {}
-        virtual CstBuffer*  update(Pass *pass, int layerID, bool bBindProgram, bool bCreateIfNeeded, bool bCreateBufferIfNeeded)
+        virtual CstBuffer*  update(Pass *pass, int layerID, bool bCreateIfNeeded, bool bCreateBufferIfNeeded)
         { /*TODO if needed*/return this; }
     };
     /*************************************************************************/ /**
@@ -146,9 +146,9 @@ namespace nvFX
         virtual ~UniformCUDA();
         UniformCUDA(const char* name = NULL, const char* groupname = NULL, const char* semantic = NULL);
 
-        Uniform*            updateTextures(ShadowedData* data, Pass *pass, int layerID, bool bBindProgram, bool bCreateIfNeeded);
-        virtual Uniform*    update(ShadowedData* pData, Pass *pass, int ppID, bool bBindProgram, bool bCreateIfNeeded);
-        virtual Uniform*    updateForTarget(ShadowedData* pData, STarget &t, bool bBindProgram = false);
+        Uniform*            updateTextures(ShadowedData* data, Pass *pass, int layerID, bool bCreateIfNeeded);
+        virtual Uniform*    update(ShadowedData* pData, Pass *pass, int ppID, bool bCreateIfNeeded);
+        virtual Uniform*    updateForTarget(ShadowedData* pData, int target);
     };
 
     /*************************************************************************/ /**
@@ -157,7 +157,7 @@ namespace nvFX
     class CUDAShader : public Shader
     {
     public:
-        CUDAShader(const char *name = NULL, IContainer* pCont = NULL);
+        CUDAShader(const char *name = NULL);
         ~CUDAShader();
 
         virtual void    cleanupShader()                     {}

diff --git a/FxLibCUDA/FxLibCUDA.cpp b/FxLibCUDA/FxLibCUDA.cpp
@@ -60,9 +60,9 @@ void delete_ProgramCUDA(IProgram *pProg)
     delete p;
 }
 
-Shader* new_ShaderCUDA(const char* name, Container *pCont)
+Shader* new_ShaderCUDA(const char* name)
 {
-    return new CUDAShader(name, pCont);
+    return new CUDAShader(name);
 }
 void delete_ShaderCUDA(IShader *pShd)
 {

diff --git a/FxLibCUDA/UniformCUDA.cpp b/FxLibCUDA/UniformCUDA.cpp
@@ -62,7 +62,7 @@ UniformCUDA::~UniformCUDA()
     }
 }
 
-Uniform*  UniformCUDA::update(ShadowedData* data, Pass *pass, int layerID, bool bBindProgram, bool bCreateIfNeeded)
+Uniform*  UniformCUDA::update(ShadowedData* data, Pass *pass, int layerID, bool bCreateIfNeeded)
 {
     CUresult res;
     CUarray cuArray = NULL;
@@ -74,13 +74,13 @@ Uniform*  UniformCUDA::update(ShadowedData* data, Pass *pass, int layerID, bool
     case IUniform::TTexture3D:
     case IUniform::TTextureCube:
         //if(data && data->tex.pRes)
-            return updateTextures(data, pass, layerID, bBindProgram, bCreateIfNeeded);
+            return updateTextures(data, pass, layerID, bCreateIfNeeded);
         break;
         // TODO whenever necessary
     }
     return this;
 }
-Uniform*  UniformCUDA::updateTextures(ShadowedData* data, Pass *pass, int layerID, bool bBindProgram, bool bCreateIfNeeded)
+Uniform*  UniformCUDA::updateTextures(ShadowedData* data, Pass *pass, int layerID, bool bCreateIfNeeded)
 {
     CUresult res;
     CUarray cuArray = NULL;
@@ -156,17 +156,19 @@ Uniform*  UniformCUDA::updateTextures(ShadowedData* data, Pass *pass, int layerI
                 if(data && data->tex.pRes)
                     static_cast<ResourceCUDA*>(data->tex.pRes)->setupAsCUDATexture();
                 // update again now we found it
-                return UniformCUDA::updateForTarget(data, t);
+                return UniformCUDA::updateForTarget(data, i);
             }
         }
     }
     return this;
 }
 
-Uniform*  UniformCUDA::updateForTarget(ShadowedData* data, STarget &t, bool bBindProgram)
+Uniform*  UniformCUDA::updateForTarget(ShadowedData* data, int target)
 {
     CUresult res;
     CUarray cuArray = NULL;
+	UniformCUDA::STarget &t = m_targets[target];
+	m_activeTarget = target;
     if(data && data->tex.pRes)
     {
         cuArray = static_cast<ResourceCUDA*>(data->tex.pRes)->mapResource();

diff --git a/cmake/FindCuda.cmake b/cmake/FindCuda.cmake
@@ -0,0 +1,146 @@
+# Try to find OptiX project dll and include file
+#
+unset(CUDA_DLL CACHE)
+unset(CUDA_ROOT_DIR CACHE)
+unset(CUDA_LIB CACHE)
+unset(CUDA_VERSION CACHE)
+unset(CUDA_FOUND CACHE)
+unset(RESULT CACHE)
+
+if ( NOT DEFINED ${CUDA_LOCATION} )
+	SET(CUDA_LOCATION "$ENV{CUDA_PATH}" CACHE PATH "path")
+endif()
+
+find_path( CUDA_ROOT_DIR include/cuda.h
+  ${CUDA_LOCATION}  
+)
+
+macro(_find_files targetVar incDir dllName dllName64 folder result)
+  unset ( result )
+  unset ( fileList )
+  if(ARCH STREQUAL "x86")
+      file(GLOB fileList "${${incDir}}/../${folder}${dllName}")
+      list(LENGTH fileList NUMLIST)
+      if(NUMLIST EQUAL 0)
+        file(GLOB fileList "${${incDir}}/${folder}${dllName}")
+      endif()
+  else()
+      file(GLOB fileList "${${incDir}}/../${folder}${dllName64}")
+      list(LENGTH fileList NUMLIST)
+      if(NUMLIST EQUAL 0)
+        file(GLOB fileList "${${incDir}}/${folder}${dllName64}")
+      endif()
+  endif()  
+  list(LENGTH fileList NUMLIST)
+  if(NUMLIST EQUAL 0)
+    set (${result} "NOTFOUND" )
+  else()
+    set (${result} "OK" )
+  endif()
+  list(APPEND ${targetVar} ${fileList} )
+
+endmacro()
+
+if(CUDA_ROOT_DIR)
+    #---------Retrieve VERSION
+	set( CUDA_VERSION_MAX "0.0" )
+    string( REGEX REPLACE "(^.*)/[^/]*$" "\\1" CUDA_GENERAL_DIR ${CUDA_ROOT_DIR} )
+	file( GLOB CUDA_VERSION_PATHS ${CUDA_GENERAL_DIR}/* )
+	foreach( CUDA_VERSION_PATH ${CUDA_VERSION_PATHS} )
+	  if( IS_DIRECTORY ${CUDA_VERSION_PATH} )
+        string( REGEX REPLACE "^.*/v([^/]*)$" "\\1" CUDA_VERSION_CURRENT ${CUDA_VERSION_PATH} )	
+	      if( CUDA_VERSION_CURRENT MATCHES "[0-9]+[.]+[0-9]+" )	
+		    if( ${CUDA_VERSION_MAX} VERSION_LESS ${CUDA_VERSION_CURRENT} )
+		      set( CUDA_VERSION_MAX ${CUDA_VERSION_CURRENT} ) 
+		    endif()
+		  endif()
+	  endif()
+	endforeach()
+
+    SET(CUDA_VERSION ${CUDA_VERSION_MAX} CACHE STRING "string")
+    string( REGEX REPLACE "(^[^.]*)[.]+.*$" "\\1" CUDA_VERSION_MAJOR ${CUDA_VERSION} )
+    string( REGEX REPLACE "^[^.]*[.]+(.*$)" "\\1" CUDA_VERSION_MINOR ${CUDA_VERSION} )
+
+	if( CUDA_VERSION_MAX STREQUAL "0.0" )
+	  message( "CUDA directory is not valid" )
+	else()
+	  set( CUDA_ROOT_DIR ${CUDA_GENERAL_DIR}/v${CUDA_VERSION} )
+	  set( CUDA_NVCC_EXECUTABLE ${CUDA_ROOT_DIR}/bin/nvcc )
+	endif()
+
+
+	#-------- Locate DLLS
+    _find_files( CUDA_DLL CUDA_ROOT_DIR "cudart32_${CUDA_VERSION_MAJOR}${CUDA_VERSION_MINOR}.dll" "cudart64_${CUDA_VERSION_MAJOR}${CUDA_VERSION_MINOR}.dll" "bin/" RESULT)	
+	if ( ${RESULT} STREQUAL "NOTFOUND" )
+	   message( "CUDA is required for this sample (${PROJECT_NAME})." )
+	   message( "CUDA binaries have not been found. Please set the CUDA_LOCATION variable above to the root location of the desired CUDA version." )
+	endif()
+
+	#-------- Locate LIBS
+    _find_files( CUDA_LIB CUDA_ROOT_DIR "win32/cudart.lib" "x64/cudart.lib" "lib/" RESULT)
+    _find_files( CUDA_LIB CUDA_ROOT_DIR "win32/cublas.lib" "x64/cublas.lib" "lib/" RESULT)
+    _find_files( CUDA_LIB CUDA_ROOT_DIR "win32/cuda.lib" "x64/cuda.lib" "lib/" RESULT)
+    _find_files( CUDA_LIB CUDA_ROOT_DIR "win32/OpenCL.lib" "x64/OpenCL.lib" "lib/" RESULT)
+
+	#-------- Locate HEADERS
+	_find_files( CUDA_HEADERS CUDA_ROOT_DIR "cuda.h" "cuda.h" "include/" RESULT)
+
+    if(CUDA_DLL)
+	   set( CUDA_FOUND "YES" )      
+    endif(CUDA_DLL)
+else(CUDA_ROOT_DIR)
+
+  message(INFO "CUDA not found. Please set the CUDA_LOCATION variable above to the root location of CUDA 5.5.")
+
+endif(CUDA_ROOT_DIR)
+
+include(FindPackageHandleStandardArgs)
+
+SET(CUDA_DLL ${CUDA_DLL} CACHE PATH "path")
+SET(CUDA_LIB ${CUDA_LIB} CACHE PATH "path")
+SET(CUDA_INCLUDE_DIR "${CUDA_ROOT_DIR}/include")
+
+find_package_handle_standard_args(CUDA DEFAULT_MSG
+    CUDA_INCLUDE_DIR
+    CUDA_DLL
+)
+
+# Generate PTX files
+# NVCUDA_COMPILE_PTX( SOURCES file1.cu file2.cu TARGET_PATH <path where ptxs should be stored> GENERATED_FILES ptx_sources NVCC_OPTIONS -arch=sm_20)
+# Generates ptx files for the given source files. ptx_sources will contain the list of generated files.
+FUNCTION(NVCUDA_COMPILE_PTX)
+  set(options "")
+  set(oneValueArgs TARGET_PATH GENERATED_FILES)
+  set(multiValueArgs NVCC_OPTIONS SOURCES)
+  CMAKE_PARSE_ARGUMENTS(NVCUDA_COMPILE_PTX "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+
+  # Match the bitness of the ptx to the bitness of the application
+  set( MACHINE "--machine=32" )
+  if( CMAKE_SIZEOF_VOID_P EQUAL 8)
+    set( MACHINE "--machine=64" )
+  endif()
+
+  # Custom build rule to generate ptx files from cuda files
+  FOREACH( input ${NVCUDA_COMPILE_PTX_SOURCES} )
+    get_filename_component( input_we ${input} NAME_WE )
+
+    # generate the *.ptx files inside "ptx" folder inside the executable's output directory.
+    set( output "${CMAKE_CURRENT_SOURCE_DIR}/${input_we}.ptx" )
+
+    LIST( APPEND PTX_FILES  ${output} )
+
+    message("${CUDA_NVCC_EXECUTABLE} ${MACHINE} --ptx ${NVCUDA_COMPILE_PTX_NVCC_OPTIONS} ${input} -o ${output} WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}")
+
+    add_custom_command(
+      OUTPUT  ${output}
+      DEPENDS ${input}
+      COMMAND ${CUDA_NVCC_EXECUTABLE} ${MACHINE} --ptx ${NVCUDA_COMPILE_PTX_NVCC_OPTIONS} ${input} -o ${output} WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}"
+      COMMAND ${CMAKE_COMMAND} -E echo ${NVCUDA_COMPILE_PTX_TARGET_PATH}
+      COMMAND ${CMAKE_COMMAND} -E copy ${output} "${NVCUDA_COMPILE_PTX_TARGET_PATH}/${input_we}.ptx"
+    )
+  ENDFOREACH( )
+
+  set(${NVCUDA_COMPILE_PTX_GENERATED_FILES} ${PTX_FILES} PARENT_SCOPE)
+ENDFUNCTION()
+
+mark_as_advanced( CUDA_FOUND )
diff --git a/nvFxcc/CMakeLists.txt b/nvFxcc/CMakeLists.txt
@@ -3,7 +3,7 @@ project(nvFxcc)
 set(PLATFORM_LIBRARIES
     ${NVFX_LIBS}
     ${OPENGL_LIBRARY}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 

diff --git a/samples/D3DSkinnedModelEffect/CMakeLists.txt b/samples/D3DSkinnedModelEffect/CMakeLists.txt
@@ -4,7 +4,7 @@ set(PLATFORM_LIBRARIES
     ${NVFX_LIBS}
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${ZLIB_LIBRARIES}
     ${NSIGHT_LIB}
 )

diff --git a/samples/OpenGLFluitSimEffect/CMakeLists.txt b/samples/OpenGLFluitSimEffect/CMakeLists.txt
@@ -5,7 +5,7 @@ set(PLATFORM_LIBRARIES
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
     ${ZLIB_LIBRARIES}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(

diff --git a/samples/OpenGLLigthingEffect/CMakeLists.txt b/samples/OpenGLLigthingEffect/CMakeLists.txt
@@ -5,7 +5,7 @@ set(PLATFORM_LIBRARIES
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
     ${ZLIB_LIBRARIES}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(

diff --git a/samples/OpenGLSimpleEffect/CMakeLists.txt b/samples/OpenGLSimpleEffect/CMakeLists.txt
@@ -5,7 +5,7 @@ set(PLATFORM_LIBRARIES
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
     ${ZLIB_LIBRARIES}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(

diff --git a/samples/OpenGLSimpleEffect_PNTriangles/CMakeLists.txt b/samples/OpenGLSimpleEffect_PNTriangles/CMakeLists.txt
@@ -5,7 +5,7 @@ set(PLATFORM_LIBRARIES
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
     ${ZLIB_LIBRARIES}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(

diff --git a/samples/OpenGLSimpleEffect_Subroutine/CMakeLists.txt b/samples/OpenGLSimpleEffect_Subroutine/CMakeLists.txt
@@ -5,7 +5,7 @@ set(PLATFORM_LIBRARIES
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
     ${ZLIB_LIBRARIES}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(

diff --git a/samples/OpenGLSimpleEffect_skinning/CMakeLists.txt b/samples/OpenGLSimpleEffect_skinning/CMakeLists.txt
@@ -5,7 +5,7 @@ set(PLATFORM_LIBRARIES
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
     ${ZLIB_LIBRARIES}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(

diff --git a/samples/OpenGLSimpleSceneEffect/CMakeLists.txt b/samples/OpenGLSimpleSceneEffect/CMakeLists.txt
@@ -5,7 +5,7 @@ set(PLATFORM_LIBRARIES
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
     ${ZLIB_LIBRARIES}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(

diff --git a/samples/OpenGLSkinnedModelEffect/CMakeLists.txt b/samples/OpenGLSkinnedModelEffect/CMakeLists.txt
@@ -4,7 +4,7 @@ set(PLATFORM_LIBRARIES
     ${NVFX_LIBS}
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${ZLIB_LIBRARIES}
     ${NSIGHT_LIB}
 )

diff --git a/samples/OpenGLUniformUpdateExample/CMakeLists.txt b/samples/OpenGLUniformUpdateExample/CMakeLists.txt
@@ -4,7 +4,7 @@ set(PLATFORM_LIBRARIES
     ${NVFX_LIBS}
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(

diff --git a/samples/OpenGL_CUDAConvolution/CMakeLists.txt b/samples/OpenGL_CUDAConvolution/CMakeLists.txt
@@ -5,7 +5,7 @@ set(PLATFORM_LIBRARIES
     ${OPENGL_LIBRARY}
     ${GLEW_LIBRARY}
     ${ZLIB_LIBRARIES}
-    ${CUDA_LIBRARIES}
+    ${CUDA_LIB}
     ${NSIGHT_LIB}
 )
 include_directories(
@@ -16,6 +16,7 @@ include_directories(
     ${NV_MATH_DIR}
     ${BK3D_DIR}
     ${NV_IMAGE_DIR}
+    ${CUDA_INCLUDE_DIR}
 )
 set(SOURCE_FILES 
   OpenGL_CUDAConvolution.cpp