Create 2.0.5 release.

Rename our squish version to nvsquish.

CMakeLists.txt

@@ -1,4 +1,4 @@
-CMAKE_MINIMUM_REQUIRED(VERSION 2.4.0)
+CMAKE_MINIMUM_REQUIRED(VERSION 2.6.0)
 PROJECT(NV)
 ENABLE_TESTING()
 
@@ -16,6 +16,13 @@ MESSAGE(STATUS "Setting optimal options")
 MESSAGE(STATUS "  Processor: ${NV_SYSTEM_PROCESSOR}")
 MESSAGE(STATUS "  Compiler Flags: ${CMAKE_CXX_FLAGS}")
 
+IF(NVTT_SHARED)
+	SET(NVCORE_SHARED TRUE)
+	SET(NVMATH_SHARED TRUE)
+	SET(NVIMAGE_SHARED TRUE)
+ENDIF(NVTT_SHARED)
+
+
 ADD_SUBDIRECTORY(src)
 
 IF(WIN32)

ChangeLog

@@ -1,3 +1,10 @@
+NVIDIA Texture Tools version 2.0.5
+ * Fix error in single color compressor. Fixes issue 66.
+ * Detect mismatch between CUDA runtime and driver, and disable CUDA in that case.
+ * Fix cmake files when compiling NVTT as a shared library.
+ * When linking nvtt dynamically on unix, link all libraries dynamically.
+ * Select fastest CUDA device.
+
 NVIDIA Texture Tools version 2.0.4
  * Fix error in RGB format output; reported by jonsoh. See issue 49.
  * Added support RGB format dithering by jonsoh. Fixes issue 50 and 51.

VERSION

@@ -1 +1 @@
-2.0.4
+2.0.5

cmake/FindCUDA.cmake

@@ -57,7 +57,7 @@ MARK_AS_ADVANCED (CUDA_FOUND CUDA_COMPILER CUDA_RUNTIME_LIBRARY)
 
 
 #SET(CUDA_OPTIONS "-ncfe")
-SET(CUDA_OPTIONS "")
+SET(CUDA_OPTIONS "--host-compilation=C")
 
 IF (CUDA_EMULATION)
 	SET (CUDA_OPTIONS "${CUDA_OPTIONS} -deviceemu")

configure

@@ -53,7 +53,7 @@ echo "-- Configuring nvidia-texture-tools "`cat VERSION`
 
 mkdir -p ./build
 cd ./build
-$CMAKE .. -DCMAKE_BUILD_TYPE=$build -DCMAKE_INSTALL_PREFIX=$prefix -G "Unix Makefiles" || exit 1
+$CMAKE .. -DNVTT_SHARED=1 -DCMAKE_BUILD_TYPE=$build -DCMAKE_INSTALL_PREFIX=$prefix -G "Unix Makefiles" || exit 1
 cd ..
 
 echo ""

project/vc8/nvcore/nvcore.vcproj

@@ -281,6 +281,10 @@
 				RelativePath="..\..\..\src\nvcore\Debug.cpp"
 				>
 			</File>
+			<File
+				RelativePath="..\..\..\src\nvcore\Library.cpp"
+				>
+			</File>
 			<File
 				RelativePath="..\..\..\src\nvcore\Memory.cpp"
 				>
@@ -315,6 +319,10 @@
 				RelativePath="..\..\..\src\nvcore\DefsVcWin32.h"
 				>
 			</File>
+			<File
+				RelativePath="..\..\..\src\nvcore\Library.h"
+				>
+			</File>
 			<File
 				RelativePath="..\..\..\src\nvcore\Memory.h"
 				>

project/vc8/nvtt/nvtt.rc

@@ -53,8 +53,8 @@ END
 //
 
 VS_VERSION_INFO VERSIONINFO
- FILEVERSION 2,0,4,0
- PRODUCTVERSION 2,0,4,0
+ FILEVERSION 2,0,5,0
+ PRODUCTVERSION 2,0,5,0
  FILEFLAGSMASK 0x17L
 #ifdef _DEBUG
  FILEFLAGS 0x1L
@@ -71,12 +71,12 @@ BEGIN
         BEGIN
             VALUE "CompanyName", "NVIDIA Corporation"
             VALUE "FileDescription", "NVIDIA Texture Tools Dynamic Link Library"
-            VALUE "FileVersion", "2, 0, 4, 0"
+            VALUE "FileVersion", "2, 0, 5, 0"
             VALUE "InternalName", "nvtt"
             VALUE "LegalCopyright", "Copyright (C) 2007"
             VALUE "OriginalFilename", "nvtt.dll"
             VALUE "ProductName", "NVIDIA Texture Tools Dynamic Link Library"
-            VALUE "ProductVersion", "2, 0, 4, 0"
+            VALUE "ProductVersion", "2, 0, 5, 0"
         END
     END
     BLOCK "VarFileInfo"

src/nvcore/CMakeLists.txt

@@ -19,13 +19,19 @@ SET(CORE_SRCS
 	TextWriter.h
 	TextWriter.cpp
 	Radix.h
-	Radix.cpp)
+	Radix.cpp
+	Library.h
+	Library.cpp)
 
 INCLUDE_DIRECTORIES(${CMAKE_CURRENT_SOURCE_DIR})
 
 # targets
 ADD_DEFINITIONS(-DNVCORE_EXPORTS)
 
+IF(UNIX)
+	SET(LIBS ${LIBS} ${CMAKE_DL_LIBS})
+ENDIF(UNIX)
+
 IF(NVCORE_SHARED)
 	ADD_LIBRARY(nvcore SHARED ${CORE_SRCS})
 ELSE(NVCORE_SHARED)

src/nvcore/Debug.cpp

@@ -28,7 +28,7 @@
 #endif
 
 #if NV_OS_LINUX && defined(HAVE_EXECINFO_H)
-#	include <execinfo.h>
+#	include <execinfo.h> // backtrace
 #	if NV_CC_GNUC // defined(HAVE_CXXABI_H)
 #		include <cxxabi.h>
 #	endif
@@ -39,6 +39,13 @@
 #	include <sys/types.h>
 #	include <sys/sysctl.h>	// sysctl
 #	include <ucontext.h>
+#	undef HAVE_EXECINFO_H
+#	if defined(HAVE_EXECINFO_H) // only after OSX 10.5
+#		include <execinfo.h> // backtrace
+#		if NV_CC_GNUC // defined(HAVE_CXXABI_H)
+#			include <cxxabi.h>
+#		endif
+#	endif
 #endif
 
 #include <stdexcept> // std::runtime_error
@@ -128,6 +135,10 @@ namespace
 
 #if defined(HAVE_EXECINFO_H) // NV_OS_LINUX
 
+	static bool nvHasStackTrace() {
+		return backtrace != NULL;
+	}
+
 	static void nvPrintStackTrace(void * trace[], int size, int start=0) {
 		char ** string_array = backtrace_symbols(trace, size);
 	
@@ -166,24 +177,36 @@ namespace
 
 	static void * callerAddress(void * secret)
 	{
-#	if NV_OS_DARWIN && NV_CPU_PPC
-		ucontext_t * ucp = (ucontext_t *)secret;
-        return (void *) ucp->uc_mcontext->ss.srr0;
-#	elif NV_OS_DARWIN && NV_CPU_X86
-		ucontext_t * ucp = (ucontext_t *)secret;
-        return (void *) ucp->uc_mcontext->ss.eip;
-#	elif NV_CPU_X86_64
-		// #define REG_RIP REG_INDEX(rip) // seems to be 16
-		ucontext_t * ucp = (ucontext_t *)secret;
-		return (void *)ucp->uc_mcontext.gregs[REG_RIP];
-#	elif NV_CPU_X86
-		ucontext_t * ucp = (ucontext_t *)secret;
-		return (void *)ucp->uc_mcontext.gregs[14/*REG_EIP*/];
-#	elif NV_CPU_PPC
-		ucontext_t * ucp = (ucontext_t *)secret;
-		return (void *) ucp->uc_mcontext.regs->nip;
+#	if NV_OS_DARWIN
+#		if defined(_STRUCT_MCONTEXT)
+#			if NV_CPU_PPC
+				ucontext_t * ucp = (ucontext_t *)secret;
+				return (void *) ucp->uc_mcontext->__ss.__srr0;
+#			elif NV_CPU_X86
+				ucontext_t * ucp = (ucontext_t *)secret;
+				return (void *) ucp->uc_mcontext->__ss.__eip;
+#			endif
+#		else
+#			if NV_CPU_PPC
+				ucontext_t * ucp = (ucontext_t *)secret;
+				return (void *) ucp->uc_mcontext->ss.srr0;
+#			elif NV_CPU_X86
+				ucontext_t * ucp = (ucontext_t *)secret;
+				return (void *) ucp->uc_mcontext->ss.eip;
+#			endif
+#		endif
 #	else
-		return NULL;
+#		if NV_CPU_X86_64
+			// #define REG_RIP REG_INDEX(rip) // seems to be 16
+			ucontext_t * ucp = (ucontext_t *)secret;
+			return (void *)ucp->uc_mcontext.gregs[REG_RIP];
+#		elif NV_CPU_X86
+			ucontext_t * ucp = (ucontext_t *)secret;
+			return (void *)ucp->uc_mcontext.gregs[14/*REG_EIP*/];
+#		elif NV_CPU_PPC
+			ucontext_t * ucp = (ucontext_t *)secret;
+			return (void *) ucp->uc_mcontext.regs->nip;
+#		endif
 #	endif
 		
 		// How to obtain the instruction pointers in different platforms, from mlton's source code.
@@ -228,17 +251,18 @@ namespace
 		}
 		
 #	if defined(HAVE_EXECINFO_H)
+		if (nvHasStackTrace()) // in case of weak linking
+		{
+			void * trace[64];
+			int size = backtrace(trace, 64);
 		
-		void * trace[64];
-		int size = backtrace(trace, 64);
-		
-		if (pnt != NULL) {
-			// Overwrite sigaction with caller's address.
-			trace[1] = pnt;
+			if (pnt != NULL) {
+				// Overwrite sigaction with caller's address.
+				trace[1] = pnt;
+			}
+			
+			nvPrintStackTrace(trace, size, 1);
 		}
-		
-		nvPrintStackTrace(trace, size, 1);
-		
 #	endif // defined(HAVE_EXECINFO_H)
 		
 		exit(0);
@@ -373,9 +397,12 @@ namespace
 #		endif
 
 #		if defined(HAVE_EXECINFO_H)
-			void * trace[64];
-			int size = backtrace(trace, 64);
-			nvPrintStackTrace(trace, size, 3);
+			if (nvHasStackTrace())
+			{
+				void * trace[64];
+				int size = backtrace(trace, 64);
+				nvPrintStackTrace(trace, size, 3);
+			}
 #		endif
 
 			// Exit cleanly.
@@ -422,9 +449,12 @@ void NV_CDECL nvDebug(const char *msg, ...)
 void debug::dumpInfo()
 {
 #if !NV_OS_WIN32 && defined(HAVE_SIGNAL_H) && defined(HAVE_EXECINFO_H)
-	void * trace[64];
-	int size = backtrace(trace, 64);
-	nvPrintStackTrace(trace, size, 1);
+	if (nvHasStackTrace())
+	{
+		void * trace[64];
+		int size = backtrace(trace, 64);
+		nvPrintStackTrace(trace, size, 1);
+	}
 #endif
 }
 

src/nvcore/DefsGnucDarwin.h

@@ -2,8 +2,7 @@
 #error "Do not include this file directly."
 #endif
 
-#include <stdlib.h>	// uint8_t, int8_t, ...
-
+#include <stdint.h> // uint8_t, int8_t, ...
 
 // Function linkage
 #define DLL_IMPORT

src/nvcore/DefsVcWin32.h

@@ -19,7 +19,9 @@
 
 // Set standard function names.
 #define snprintf _snprintf
-#define vsnprintf _vsnprintf
+#if _MSC_VER < 1500
+#	define vsnprintf _vsnprintf
+#endif
 #define vsscanf _vsscanf
 #define chdir _chdir
 #define getcwd _getcwd 

src/nvcore/Library.cpp

@@ -0,0 +1,41 @@
+
+#include "Library.h"
+#include "Debug.h"
+
+#if NV_OS_WIN32
+#define WIN32_LEAN_AND_MEAN
+#define VC_EXTRALEAN
+#include <windows.h>
+#else
+#include <dlfcn.h>
+#endif
+
+
+
+void * nvLoadLibrary(const char * name)
+{
+#if NV_OS_WIN32
+	return (void *)LoadLibraryExA( name, NULL, 0 );
+#else
+	return dlopen(name, RTLD_LAZY);
+#endif
+}
+
+void nvUnloadLibrary(void * handle)
+{
+	nvDebugCheck(handle != NULL);
+#if NV_OS_WIN32
+	FreeLibrary((HMODULE)handle);
+#else
+	dlclose(handle);
+#endif
+}
+
+void * nvBindSymbol(void * handle, const char * symbol)
+{
+#if NV_OS_WIN32
+	return (void *)GetProcAddress((HMODULE)handle, symbol);
+#else
+	return (void *)dlsym(handle, symbol);
+#endif	
+}

src/nvcore/Library.h

@@ -0,0 +1,50 @@
+// This code is in the public domain -- castano@gmail.com
+
+#ifndef NV_CORE_LIBRARY_H
+#define NV_CORE_LIBRARY_H
+
+#include <nvcore/nvcore.h>
+
+#if NV_OS_WIN32
+#define LIBRARY_NAME(name)	#name ".dll"
+#elif NV_OS_DARWIN
+#define NV_LIBRARY_NAME(name)	"lib" #name ".dylib"
+#else
+#define NV_LIBRARY_NAME(name)	"lib" #name ".so"
+#endif
+
+NVCORE_API void * nvLoadLibrary(const char * name);
+NVCORE_API void nvUnloadLibrary(void * lib);
+NVCORE_API void * nvBindSymbol(void * lib, const char * symbol);
+
+class NVCORE_CLASS Library
+{
+public:
+	Library(const char * name)
+	{
+		handle = nvLoadLibrary(name);
+	}
+	~Library()
+	{
+		if (isValid())
+		{
+			nvUnloadLibrary(handle);
+		}
+	}
+	
+	bool isValid() const
+	{
+		return handle != NULL;
+	}
+	
+	void * bindSymbol(const char * symbol)
+	{
+		return nvBindSymbol(handle, symbol);
+	}
+	
+private:
+	void * handle;
+};
+
+
+#endif // NV_CORE_LIBRARY_H

src/nvcore/Prefetch.h

@@ -24,7 +24,7 @@ __forceinline void nvPrefetch(const void * mem)
 #else // NV_CC_MSVC
 
 // do nothing in other case.
-#define piPrefetch(ptr)
+#define nvPrefetch(ptr)
 
 #endif // NV_CC_MSVC
 

src/nvcore/Ptr.h

@@ -43,8 +43,11 @@ public:
 
 	/** Delete owned pointer and assign new one. */
 	void operator=( T * p ) {
-		delete m_ptr;
-		m_ptr = p;
+		if (p != m_ptr)
+		{
+			delete m_ptr;
+			m_ptr = p;
+		}
 	}
 
 	/** Member access. */
@@ -249,14 +252,14 @@ public:
 		/** -> operator. */
 		BaseClass * operator -> () const
 		{
-			piCheck( m_ptr != NULL );
+			nvCheck( m_ptr != NULL );
 			return m_ptr;
 		}
 
 		/** * operator. */
 		BaseClass & operator*() const
 		{
-			piCheck( m_ptr != NULL );
+			nvCheck( m_ptr != NULL );
 			return *m_ptr;
 		}
 

src/nvcore/StrLib.h

@@ -14,7 +14,7 @@ namespace nv
 
 	uint strHash(const char * str, uint h) NV_PURE;
 
-	/// String hash vased on Bernstein's hash.
+	/// String hash based on Bernstein's hash.
 	inline uint strHash(const char * data, uint h = 5381)
 	{
 		uint i;
@@ -213,9 +213,12 @@ namespace nv
 		/// Implement value semantics.
 		String & operator=( const String & str )
 		{
-			release();
-			data = str.data;
-			addRef();
+			if (str.data != data)
+			{
+				release();
+				data = str.data;
+				addRef();
+			}
 			return *this;
 		}
 

src/nvimage/DirectDrawSurface.cpp

@@ -532,7 +532,7 @@ DDSHeader::DDSHeader()
 
 	// Store version information on the reserved header attributes.
 	this->reserved[9] = MAKEFOURCC('N', 'V', 'T', 'T');
-	this->reserved[10] = (2 << 16) | (0 << 8) | (4);	// major.minor.revision
+	this->reserved[10] = (2 << 16) | (0 << 8) | (5);	// major.minor.revision
 
 	this->pf.size = 32;
 	this->pf.flags = 0;

src/nvmath/Box.h

@@ -108,7 +108,7 @@ public:
 	float area() const
 	{
 		const Vector3 d = extents();
-		return 4.0f * (d.x()*d.y() + d.x()*d.z() + d.y()*d.z());
+		return 8.0f * (d.x()*d.y() + d.x()*d.z() + d.y()*d.z());
 	}	
 
 	/// Get the volume of the box.
@@ -118,6 +118,14 @@ public:
 		return 8.0f * (d.x() * d.y() * d.z());
 	}
 	
+	/// Return true if the box contains the given point.
+	bool contains(Vector3::Arg p) const
+	{
+		return 
+			m_mins.x() < p.x() && m_mins.y() < p.y() && m_mins.z() < p.z() &&
+			m_maxs.x() > p.x() && m_maxs.y() > p.y() && m_maxs.z() > p.z();
+	}
+
 private:
 
 	Vector3 m_mins;
@@ -125,15 +133,6 @@ private:
 };
 
 
-/*
-/// Point inside box test.
-inline bool pointInsideBox(const Box & b, Vector3::Arg p) const
-{
-	return (m_mins.x() < p.x() && m_mins.y() < p.y() && m_mins.z() < p.z() &&
-		m_maxs.x() > p.x() && m_maxs.y() > p.y() && m_maxs.z() > p.z());
-}
-*/
-
 
 } // nv namespace
 

src/nvtt/CMakeLists.txt

@@ -43,8 +43,9 @@ INCLUDE_DIRECTORIES(${CMAKE_CURRENT_SOURCE_DIR})
 
 ADD_DEFINITIONS(-DNVTT_EXPORTS)
 
-IF(NVTT_SHARED)	
-	ADD_LIBRARY(nvtt SHARED ${DXT_SRCS})
+IF(NVTT_SHARED)
+	ADD_DEFINITIONS(-DNVTT_SHARED=1)
+	ADD_LIBRARY(nvtt SHARED ${NVTT_SRCS})
 ELSE(NVTT_SHARED)
 	ADD_LIBRARY(nvtt ${NVTT_SRCS})
 ENDIF(NVTT_SHARED)

src/nvtt/Compressor.cpp

@@ -211,6 +211,10 @@ Compressor::Compressor() : m(*new Compressor::Private())
 
 	if (m.cudaEnabled)
 	{
+		// Select fastest CUDA device.
+		int device = cuda::getFastestDevice();
+		cuda::setDevice(device);
+		
 		m.cuda = new CudaCompressor();
 
 		if (!m.cuda->isValid())
@@ -237,6 +241,10 @@ void Compressor::enableCudaAcceleration(bool enable)
 
 	if (m.cudaEnabled && m.cuda == NULL)
 	{
+		// Select fastest CUDA device.
+		int device = cuda::getFastestDevice();
+		cuda::setDevice(device);
+		
 		m.cuda = new CudaCompressor();
 		
 		if (!m.cuda->isValid())

src/nvtt/cuda/CompressKernel.cu

@@ -191,6 +191,9 @@ __device__ void loadColorBlock(const uint * image, float3 colors[16], float3 sum
 
 		*sameColor = (axis == make_float3(0, 0, 0));
 
+		// Single color compressor needs unweighted colors.
+		if (*sameColor) colors[idx] = rawColors[idx];
+
 		dps[idx] = dot(rawColors[idx], axis);
 		
 #if __DEVICE_EMULATION__

src/nvtt/cuda/CudaUtils.cpp

@@ -22,6 +22,7 @@
 // OTHER DEALINGS IN THE SOFTWARE.
 
 #include <nvcore/Debug.h>
+#include <nvcore/Library.h>
 #include "CudaUtils.h"
 
 #if defined HAVE_CUDA
@@ -52,23 +53,52 @@ static bool isWow32()
 {
 	LPFN_ISWOW64PROCESS fnIsWow64Process = (LPFN_ISWOW64PROCESS)GetProcAddress(GetModuleHandle("kernel32"), "IsWow64Process");
 
-    BOOL bIsWow64 = FALSE;
+	BOOL bIsWow64 = FALSE;
  
-    if (NULL != fnIsWow64Process)
-    {
-        if (!fnIsWow64Process(GetCurrentProcess(), &bIsWow64))
-        {
+	if (NULL != fnIsWow64Process)
+	{
+		if (!fnIsWow64Process(GetCurrentProcess(), &bIsWow64))
+		{
 			// Assume 32 bits.
-            return true;
-        }
-    }
+			return true;
+		}
+	}
 
-    return !bIsWow64;
+	return !bIsWow64;
 }
 
 #endif
 
 
+static bool isCudaDriverAvailable(uint version)
+{
+#if NV_OS_WIN32
+	Library nvcuda("nvcuda.dll");
+#else
+	Library nvcuda(NV_LIBRARY_NAME(cuda));
+#endif
+	
+	if (!nvcuda.isValid())
+	{
+		return false;
+	}
+	
+	if (version > 2000)
+	{
+		void * address = nvcuda.bindSymbol("cuStreamCreate");
+		if (address == NULL) return false;
+	}
+
+	if (version > 2010)
+	{
+		void * address = nvcuda.bindSymbol("cuLoadDataEx");
+		if (address == NULL) return false;
+	}
+	
+	return true;
+}
+
+
 /// Determine if CUDA is available.
 bool nv::cuda::isHardwarePresent()
 {
@@ -90,6 +120,12 @@ bool nv::cuda::isHardwarePresent()
 			return false;
 		}
 
+		// Make sure that CUDA driver matches CUDA runtime.
+		if (!isCudaDriverAvailable(CUDART_VERSION))
+		{
+			return false;
+		}
+
 		// @@ Make sure that warp size == 32
 	}
 
@@ -115,6 +151,35 @@ int nv::cuda::deviceCount()
 	return 0;
 }
 
+int nv::cuda::getFastestDevice()
+{
+	int max_gflops_device = 0;
+#if defined HAVE_CUDA	
+	int max_gflops = 0;
+
+	const int device_count = deviceCount();
+	int current_device = 0;
+	while (current_device < device_count)
+	{
+		cudaDeviceProp device_properties;
+		cudaGetDeviceProperties(&device_properties, current_device);
+		int gflops = device_properties.multiProcessorCount * device_properties.clockRate;
+
+		if (device_properties.major != -1 && device_properties.minor != -1)
+		{
+			if( gflops > max_gflops )
+			{
+				max_gflops = gflops;
+				max_gflops_device = current_device;
+			}
+		}
+		
+		current_device++;
+	}
+#endif
+	return max_gflops_device;
+}
+
 /// Activate the given devices.
 bool nv::cuda::setDevice(int i)
 {

src/nvtt/cuda/CudaUtils.h

@@ -31,6 +31,7 @@ namespace nv
 	{
 		bool isHardwarePresent();
 		int deviceCount();
+		int getFastestDevice();
 		bool setDevice(int i);
 	};