Replace broken compressor with the latest version from trunk.

Delete images more efficiently?

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

@@ -1,7 +1,42 @@
-NVIDIA Texture Tools version 2.1.0
- * CTX1 CUDA compressor.
- * DXT1n CUDA compressor.
- * Support alpha premultiplication by Charles Nicholson. See issue 30.
+NVIDIA Texture Tools version 2.0.7
+ * Output correct exit codes. Fixes issue 92.
+ * Fix thread-safety errors. Fixes issue 90.
+ * Add SIMD power method. Fixes issue 94.
+ * Interact better with applications that already use CUDA.
+ * Faster CPU compression.
+
+NVIDIA Texture Tools version 2.0.6
+ * Fix dll version checking.
+ * Detect CUDA 2.1 and future CUDA versions correctly.
+ * Print CUDA detection message in nvcompress.
+ * Select the fastest CUDA device.
+ * Compile squish with -fPIC. Fixes issue 74.
+ * Fix warnings under gcc 4.3.2.
+ * Fix nvzoom option typo by Frank Richter. Fixes issue 81.
+ * Do not use CUDA to compress small mipmaps. Fixes issue 76.
+ * Compute mipmaps of semi-transparent images correctly.
+ * Shutdown CUDA properly. Fixes issue 83.
+ * Fix pixel format converions. Fixes issue 87.
+ * Update single color compression tables. Fixes issue 85.
+
+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.
+ * Prevent infinite loop in indexMirror when width equal 1. Fixes issue 65.
+ * Implement general scale filter, including upsampling.
+
+NVIDIA Texture Tools version 2.0.3
+ * More accurate DXT3 compressor. Fixes issue 38.
+ * Remove legacy compressors. Fix issue 34.
+ * Check for single color in all compressors. Fixes issue 43.
+ * Fix error in fast downsample filter, reported by Noel Llopis.
 
 NVIDIA Texture Tools version 2.0.2
  * Fix copy ctor error reported by Richard Sim.
@@ -10,7 +45,6 @@ NVIDIA Texture Tools version 2.0.2
  * Fix RGBA modes with less than 32 bpp by Viktor Linder.
  * Fix alpha decompression by Amorilia. See issue 40.
  * Avoid default-initialized constructors for POD types, reported by Jim Tilander.
- * Improved decompressor tool submitted by Amorilia. See issue 41.
  * Add single color compresor for DXT1a.
  * Set swizzle code to ATI2 files. See issue 41.
 

NVIDIA_Texture_Tools_README.txt

@@ -2,7 +2,7 @@
 --------------------------------------------------------------------------------
 NVIDIA Texture Tools
 README.txt
-Version 2.1
+Version 2.0
 --------------------------------------------------------------------------------
 --------------------------------------------------------------------------------
 

VERSION

@@ -1 +1 @@
-2.1.0
+2.0.7

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/nvassemble/nvassemble.vcproj

@@ -278,6 +278,7 @@
 				AdditionalDependencies="libpng.lib jpeg.lib tiff.lib"
 				OutputFile="$(SolutionDir)\$(ConfigurationName).$(PlatformName)\bin\$(ProjectName).exe"
 				AdditionalLibraryDirectories="..\..\..\gnuwin32\lib"
+				LinkTimeCodeGeneration="1"
 				TargetMachine="17"
 			/>
 			<Tool

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/nvddsinfo/nvddsinfo.vcproj

@@ -277,6 +277,7 @@
 				AdditionalDependencies="libpng.lib jpeg.lib tiff.lib"
 				OutputFile="$(SolutionDir)\$(ConfigurationName).$(PlatformName)\bin\$(ProjectName).exe"
 				AdditionalLibraryDirectories="..\..\..\gnuwin32\lib"
+				LinkTimeCodeGeneration="1"
 				TargetMachine="17"
 			/>
 			<Tool

project/vc8/nvimage/nvimage.vcproj

@@ -355,6 +355,10 @@
 				RelativePath="..\..\..\src\nvimage\nvimage.h"
 				>
 			</File>
+			<File
+				RelativePath="..\..\..\src\nvimage\PixelFormat.h"
+				>
+			</File>
 			<File
 				RelativePath="..\..\..\src\nvimage\PsdFile.h"
 				>

project/vc8/nvmath/nvmath.vcproj

@@ -277,6 +277,10 @@
 			Filter="cpp;c;cc;cxx;def;odl;idl;hpj;bat;asm;asmx"
 			UniqueIdentifier="{4FC737F1-C7A5-4376-A066-2A32D752A2FF}"
 			>
+			<File
+				RelativePath="..\..\..\src\nvmath\Plane.cpp"
+				>
+			</File>
 		</Filter>
 		<Filter
 			Name="Header Files"
@@ -295,6 +299,10 @@
 				RelativePath="..\..\..\src\nvmath\Matrix.h"
 				>
 			</File>
+			<File
+				RelativePath="..\..\..\src\nvmath\Plane.h"
+				>
+			</File>
 			<File
 				RelativePath="..\..\..\src\nvmath\Vector.h"
 				>

project/vc8/nvtt.sln

@@ -3,18 +3,18 @@ Microsoft Visual Studio Solution File, Format Version 9.00
 # Visual Studio 2005
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvtt", "nvtt\nvtt.vcproj", "{1AEB7681-57D8-48EE-813D-5C41CC38B647}"
 	ProjectSection(ProjectDependencies) = postProject
-		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
-		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
-		{F143D180-D4C4-4037-B3DE-BE89A21C8D1D} = {F143D180-D4C4-4037-B3DE-BE89A21C8D1D}
 		{CE017322-01FC-4851-9C8B-64E9A8E26C38} = {CE017322-01FC-4851-9C8B-64E9A8E26C38}
+		{F143D180-D4C4-4037-B3DE-BE89A21C8D1D} = {F143D180-D4C4-4037-B3DE-BE89A21C8D1D}
+		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
+		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
 	EndProjectSection
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvcompress", "nvcompress\nvcompress.vcproj", "{88079E38-83AA-4E8A-B18A-66A78D1B058B}"
 	ProjectSection(ProjectDependencies) = postProject
-		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
-		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
-		{1AEB7681-57D8-48EE-813D-5C41CC38B647} = {1AEB7681-57D8-48EE-813D-5C41CC38B647}
 		{F143D180-D4C4-4037-B3DE-BE89A21C8D1D} = {F143D180-D4C4-4037-B3DE-BE89A21C8D1D}
+		{1AEB7681-57D8-48EE-813D-5C41CC38B647} = {1AEB7681-57D8-48EE-813D-5C41CC38B647}
+		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
+		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
 	EndProjectSection
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvimage", "nvimage\nvimage.vcproj", "{4046F392-A18B-4C66-9639-3EABFFF5D531}"
@@ -34,16 +34,16 @@ Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvddsinfo", "nvddsinfo\nvdd
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvdecompress", "nvdecompress\nvdecompress.vcproj", "{75A0527D-BFC9-49C3-B46B-CD1A901D5927}"
 	ProjectSection(ProjectDependencies) = postProject
-		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
-		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
 		{F143D180-D4C4-4037-B3DE-BE89A21C8D1D} = {F143D180-D4C4-4037-B3DE-BE89A21C8D1D}
+		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
+		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
 	EndProjectSection
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvimgdiff", "nvimgdiff\nvimgdiff.vcproj", "{05A59E8B-EA70-4F22-89E8-E0927BA13064}"
 	ProjectSection(ProjectDependencies) = postProject
+		{F143D180-D4C4-4037-B3DE-BE89A21C8D1D} = {F143D180-D4C4-4037-B3DE-BE89A21C8D1D}
 		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
 		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
-		{F143D180-D4C4-4037-B3DE-BE89A21C8D1D} = {F143D180-D4C4-4037-B3DE-BE89A21C8D1D}
 	EndProjectSection
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvassemble", "nvassemble\nvassemble.vcproj", "{3BC6D760-91E8-4FFB-BD0E-F86F367AD8EA}"
@@ -55,18 +55,13 @@ Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvassemble", "nvassemble\nv
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "nvzoom", "nvzoom\nvzoom.vcproj", "{51999D3E-EF22-4BDD-965F-4201034D3DCE}"
 	ProjectSection(ProjectDependencies) = postProject
-		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
-		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
 		{F143D180-D4C4-4037-B3DE-BE89A21C8D1D} = {F143D180-D4C4-4037-B3DE-BE89A21C8D1D}
+		{4046F392-A18B-4C66-9639-3EABFFF5D531} = {4046F392-A18B-4C66-9639-3EABFFF5D531}
+		{50C465FE-B308-42BC-894D-89484482AF06} = {50C465FE-B308-42BC-894D-89484482AF06}
 	EndProjectSection
 EndProject
 Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "Nvidia.TextureTools", "Nvidia.TextureTools\Nvidia.TextureTools.csproj", "{CAB55C39-8FA9-4912-98D9-E52669C8911D}"
 EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "stress", "stress\stress.vcproj", "{317B694E-B5C1-42A6-956F-FC12B69175A6}"
-	ProjectSection(ProjectDependencies) = postProject
-		{1AEB7681-57D8-48EE-813D-5C41CC38B647} = {1AEB7681-57D8-48EE-813D-5C41CC38B647}
-	EndProjectSection
-EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug (no cuda)|Any CPU = Debug (no cuda)|Any CPU
@@ -319,22 +314,6 @@ Global
 		{CAB55C39-8FA9-4912-98D9-E52669C8911D}.Release|Any CPU.Build.0 = Release|Any CPU
 		{CAB55C39-8FA9-4912-98D9-E52669C8911D}.Release|Win32.ActiveCfg = Release|Any CPU
 		{CAB55C39-8FA9-4912-98D9-E52669C8911D}.Release|x64.ActiveCfg = Release|Any CPU
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Debug (no cuda)|Any CPU.ActiveCfg = Debug|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Debug (no cuda)|Win32.ActiveCfg = Debug|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Debug (no cuda)|Win32.Build.0 = Debug|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Debug (no cuda)|x64.ActiveCfg = Debug|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Debug|Any CPU.ActiveCfg = Debug|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Debug|Win32.ActiveCfg = Debug|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Debug|Win32.Build.0 = Debug|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Debug|x64.ActiveCfg = Debug|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Release (no cuda)|Any CPU.ActiveCfg = Release|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Release (no cuda)|Win32.ActiveCfg = Release|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Release (no cuda)|Win32.Build.0 = Release|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Release (no cuda)|x64.ActiveCfg = Release|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Release|Any CPU.ActiveCfg = Release|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Release|Win32.ActiveCfg = Release|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Release|Win32.Build.0 = Release|Win32
-		{317B694E-B5C1-42A6-956F-FC12B69175A6}.Release|x64.ActiveCfg = Release|Win32
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE

project/vc8/nvtt/nvtt.rc

@@ -53,8 +53,8 @@ END
 //
 
 VS_VERSION_INFO VERSIONINFO
- FILEVERSION 2,1,0,0
- PRODUCTVERSION 2,1,0,0
+ FILEVERSION 2,0,6,0
+ PRODUCTVERSION 2,0,6,0
  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, 1, 0, 0"
+            VALUE "FileVersion", "2, 0, 6, 0"
             VALUE "InternalName", "nvtt"
             VALUE "LegalCopyright", "Copyright (C) 2007"
             VALUE "OriginalFilename", "nvtt.dll"
             VALUE "ProductName", "NVIDIA Texture Tools Dynamic Link Library"
-            VALUE "ProductVersion", "2, 1, 0, 0"
+            VALUE "ProductVersion", "2, 0, 6, 0"
         END
     END
     BLOCK "VarFileInfo"

project/vc8/nvtt/nvtt.vcproj

@@ -691,7 +691,7 @@
 					>
 					<Tool
 						Name="VCCustomBuildTool"
-						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -keep -ccbin &quot;$(VCInstallDir)bin&quot; -c -D_DEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/Od,/Zi,/RTC1,/MDd -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
+						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -m32 -ccbin &quot;$(VCInstallDir)bin&quot; -c -D_DEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/Od,/Zi,/RTC1,/MDd -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
 						AdditionalDependencies="CudaMath.h"
 						Outputs="$(IntDir)\$(InputName).obj"
 					/>
@@ -701,7 +701,7 @@
 					>
 					<Tool
 						Name="VCCustomBuildTool"
-						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -keep -ccbin &quot;$(VCInstallDir)bin&quot; -c -D_DEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/Od,/Zi,/RTC1,/MDd -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
+						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -m64 -ccbin &quot;$(VCInstallDir)bin&quot; -c -D_DEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/Od,/Zi,/RTC1,/MDd -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
 						AdditionalDependencies="CudaMath.h"
 						Outputs="$(IntDir)\$(InputName).obj"
 					/>
@@ -711,7 +711,7 @@
 					>
 					<Tool
 						Name="VCCustomBuildTool"
-						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -keep -ccbin &quot;$(VCInstallDir)bin&quot; -c -DNDEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/O2,/Zi,/MD -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
+						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -m32 -ccbin &quot;$(VCInstallDir)bin&quot; -c -DNDEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/O2,/Zi,/MD -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
 						AdditionalDependencies="CudaMath.h"
 						Outputs="$(IntDir)\$(InputName).obj"
 					/>
@@ -721,7 +721,7 @@
 					>
 					<Tool
 						Name="VCCustomBuildTool"
-						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -keep -ccbin &quot;$(VCInstallDir)bin&quot; -c -DNDEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/O2,/Zi,/MD -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
+						CommandLine="&quot;$(CUDA_BIN_PATH)\nvcc.exe&quot; -m64 -ccbin &quot;$(VCInstallDir)bin&quot; -c -DNDEBUG -DWIN32 -D_CONSOLE -D_MBCS -Xcompiler /EHsc,/W3,/nologo,/Wp64,/O2,/Zi,/MD -I&quot;$(CUDA_INC_PATH)&quot; -I./ -o $(IntDir)\$(InputName).obj ..\\..\\..\\src\\nvtt\\cuda\\CompressKernel.cu&#x0D;&#x0A;"
 						AdditionalDependencies="CudaMath.h"
 						Outputs="$(IntDir)\$(InputName).obj"
 					/>
@@ -849,10 +849,6 @@
 				RelativePath="..\..\..\src\nvtt\cuda\CudaUtils.cpp"
 				>
 			</File>
-			<File
-				RelativePath="..\..\..\src\nvtt\FastCompressDXT.cpp"
-				>
-			</File>
 			<File
 				RelativePath="..\..\..\src\nvtt\InputOptions.cpp"
 				>
@@ -865,6 +861,10 @@
 				RelativePath="..\..\..\src\nvtt\nvtt_wrapper.cpp"
 				>
 			</File>
+			<File
+				RelativePath="..\..\..\src\nvtt\OptimalCompressDXT.cpp"
+				>
+			</File>
 			<File
 				RelativePath="..\..\..\src\nvtt\OutputOptions.cpp"
 				>
@@ -911,10 +911,6 @@
 				RelativePath="..\..\..\src\nvtt\cuda\CudaUtils.h"
 				>
 			</File>
-			<File
-				RelativePath="..\..\..\src\nvtt\FastCompressDXT.h"
-				>
-			</File>
 			<File
 				RelativePath="..\..\..\src\nvtt\InputOptions.h"
 				>
@@ -927,6 +923,10 @@
 				RelativePath="..\..\..\src\nvtt\nvtt_wrapper.h"
 				>
 			</File>
+			<File
+				RelativePath="..\..\..\src\nvtt\OptimalCompressDXT.h"
+				>
+			</File>
 			<File
 				RelativePath="..\..\..\src\nvtt\OutputOptions.h"
 				>

project/vc8/stress/stress.vcproj

@@ -1,201 +0,0 @@
-<?xml version="1.0" encoding="Windows-1252"?>
-<VisualStudioProject
-	ProjectType="Visual C++"
-	Version="8.00"
-	Name="stress"
-	ProjectGUID="{317B694E-B5C1-42A6-956F-FC12B69175A6}"
-	RootNamespace="stress"
-	Keyword="Win32Proj"
-	>
-	<Platforms>
-		<Platform
-			Name="Win32"
-		/>
-	</Platforms>
-	<ToolFiles>
-	</ToolFiles>
-	<Configurations>
-		<Configuration
-			Name="Debug|Win32"
-			OutputDirectory="$(ConfigurationName)\$(PlatformName)"
-			IntermediateDirectory="$(ConfigurationName)\$(PlatformName)"
-			ConfigurationType="1"
-			CharacterSet="1"
-			>
-			<Tool
-				Name="VCPreBuildEventTool"
-			/>
-			<Tool
-				Name="VCCustomBuildTool"
-			/>
-			<Tool
-				Name="VCXMLDataGeneratorTool"
-			/>
-			<Tool
-				Name="VCWebServiceProxyGeneratorTool"
-			/>
-			<Tool
-				Name="VCMIDLTool"
-			/>
-			<Tool
-				Name="VCCLCompilerTool"
-				Optimization="0"
-				AdditionalIncludeDirectories="..;..\..\..\src"
-				PreprocessorDefinitions="WIN32;_DEBUG;_CONSOLE"
-				MinimalRebuild="true"
-				BasicRuntimeChecks="3"
-				RuntimeLibrary="3"
-				UsePrecompiledHeader="0"
-				WarningLevel="3"
-				Detect64BitPortabilityProblems="true"
-				DebugInformationFormat="4"
-			/>
-			<Tool
-				Name="VCManagedResourceCompilerTool"
-			/>
-			<Tool
-				Name="VCResourceCompilerTool"
-			/>
-			<Tool
-				Name="VCPreLinkEventTool"
-			/>
-			<Tool
-				Name="VCLinkerTool"
-				OutputFile="$(SolutionDir)\$(ConfigurationName).$(PlatformName)\bin\$(ProjectName).exe"
-				LinkIncremental="2"
-				GenerateDebugInformation="true"
-				SubSystem="1"
-				TargetMachine="1"
-			/>
-			<Tool
-				Name="VCALinkTool"
-			/>
-			<Tool
-				Name="VCManifestTool"
-			/>
-			<Tool
-				Name="VCXDCMakeTool"
-			/>
-			<Tool
-				Name="VCBscMakeTool"
-			/>
-			<Tool
-				Name="VCFxCopTool"
-			/>
-			<Tool
-				Name="VCAppVerifierTool"
-			/>
-			<Tool
-				Name="VCWebDeploymentTool"
-			/>
-			<Tool
-				Name="VCPostBuildEventTool"
-			/>
-		</Configuration>
-		<Configuration
-			Name="Release|Win32"
-			OutputDirectory="$(ConfigurationName)\$(PlatformName)"
-			IntermediateDirectory="$(ConfigurationName)\$(PlatformName)"
-			ConfigurationType="1"
-			CharacterSet="1"
-			WholeProgramOptimization="1"
-			>
-			<Tool
-				Name="VCPreBuildEventTool"
-			/>
-			<Tool
-				Name="VCCustomBuildTool"
-			/>
-			<Tool
-				Name="VCXMLDataGeneratorTool"
-			/>
-			<Tool
-				Name="VCWebServiceProxyGeneratorTool"
-			/>
-			<Tool
-				Name="VCMIDLTool"
-			/>
-			<Tool
-				Name="VCCLCompilerTool"
-				AdditionalIncludeDirectories="..;..\..\..\src"
-				PreprocessorDefinitions="WIN32;NDEBUG;_CONSOLE"
-				RuntimeLibrary="2"
-				UsePrecompiledHeader="0"
-				WarningLevel="3"
-				Detect64BitPortabilityProblems="true"
-				DebugInformationFormat="3"
-			/>
-			<Tool
-				Name="VCManagedResourceCompilerTool"
-			/>
-			<Tool
-				Name="VCResourceCompilerTool"
-			/>
-			<Tool
-				Name="VCPreLinkEventTool"
-			/>
-			<Tool
-				Name="VCLinkerTool"
-				OutputFile="$(SolutionDir)\$(ConfigurationName).$(PlatformName)\bin\$(ProjectName).exe"
-				LinkIncremental="1"
-				GenerateDebugInformation="true"
-				SubSystem="1"
-				OptimizeReferences="2"
-				EnableCOMDATFolding="2"
-				TargetMachine="1"
-			/>
-			<Tool
-				Name="VCALinkTool"
-			/>
-			<Tool
-				Name="VCManifestTool"
-			/>
-			<Tool
-				Name="VCXDCMakeTool"
-			/>
-			<Tool
-				Name="VCBscMakeTool"
-			/>
-			<Tool
-				Name="VCFxCopTool"
-			/>
-			<Tool
-				Name="VCAppVerifierTool"
-			/>
-			<Tool
-				Name="VCWebDeploymentTool"
-			/>
-			<Tool
-				Name="VCPostBuildEventTool"
-			/>
-		</Configuration>
-	</Configurations>
-	<References>
-	</References>
-	<Files>
-		<Filter
-			Name="Source Files"
-			Filter="cpp;c;cc;cxx;def;odl;idl;hpj;bat;asm;asmx"
-			UniqueIdentifier="{4FC737F1-C7A5-4376-A066-2A32D752A2FF}"
-			>
-			<File
-				RelativePath="..\..\..\src\nvtt\tests\stress.cpp"
-				>
-			</File>
-		</Filter>
-		<Filter
-			Name="Header Files"
-			Filter="h;hpp;hxx;hm;inl;inc;xsd"
-			UniqueIdentifier="{93995380-89BD-4b04-88EB-625FBE52EBFB}"
-			>
-		</Filter>
-		<Filter
-			Name="Resource Files"
-			Filter="rc;ico;cur;bmp;dlg;rc2;rct;bin;rgs;gif;jpg;jpeg;jpe;resx;tiff;tif;png;wav"
-			UniqueIdentifier="{67DA6AB6-F800-4c08-8B7A-83BB121AAD01}"
-			>
-		</Filter>
-	</Files>
-	<Globals>
-	</Globals>
-</VisualStudioProject>

src/CMakeLists.txt

@@ -1,11 +1,11 @@
 
-INCLUDE_DIRECTORIES(${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR})
-
 SUBDIRS(nvcore)
 SUBDIRS(nvmath)
 SUBDIRS(nvimage)
 SUBDIRS(nvtt)
 
+INCLUDE_DIRECTORIES(${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR})
+
 # OpenGL
 INCLUDE(FindOpenGL)
 IF(OPENGL_FOUND)
@@ -84,7 +84,6 @@ ELSE(PNG_FOUND)
 ENDIF(PNG_FOUND)
 
 # TIFF
-SET(TIFF_NAMES libtiff)
 INCLUDE(FindTIFF)
 IF(TIFF_FOUND)
 	SET(HAVE_TIFF ${TIFF_FOUND} CACHE BOOL "Set to TRUE if TIFF is found, FALSE otherwise")
@@ -106,7 +105,8 @@ ENDIF(OPENEXR_FOUND)
 FIND_PACKAGE(Qt4)
 
 # Threads
-FIND_PACKAGE(Threads)
+FIND_PACKAGE(Threads REQUIRED)
+MESSAGE(STATUS "Use thread library: ${CMAKE_THREAD_LIBS_INIT}")
 
 # configuration file
 INCLUDE(CheckIncludeFiles)

src/nvcore/Algorithms.h

@@ -1,147 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#ifndef NV_CORE_ALGORITHMS_H
-#define NV_CORE_ALGORITHMS_H
-
-namespace nv
-{
-
-	/// Return the maximum of two values.
-	template <typename T> 
-	inline const T & max(const T & a, const T & b)
-	{
-		//return std::max(a, b);
-		if( a < b ) {
-			return b; 
-		}
-		return a;
-	}
-	
-	/// Return the minimum of two values.
-	template <typename T> 
-	inline const T & min(const T & a, const T & b)
-	{
-		//return std::min(a, b);
-		if( b < a ) {
-			return b; 
-		}
-		return a;
-	}
-
-	/// Clamp between two values.
-	template <typename T> 
-	inline const T & clamp(const T & x, const T & a, const T & b)
-	{
-		return min(max(x, a), b);
-	}
-	
-	/// Delete all the elements of a container.
-	template <typename T>
-	void deleteAll(T & container)
-	{
-		for (typename T::PseudoIndex i = container.start(); !container.isDone(i); container.advance(i))
-		{
-			delete container[i];
-		}
-	}
-	
-	
-	// @@ Swap should be implemented here.
-	
-	
-#if 0
-	// This does not use swap, but copies, in some cases swaps are much faster than copies!
-	// Container should implement operator[], and size()
-	template <class Container, class T>
-	void insertionSort(Container<T> & container)
-	{
-		const uint n = container.size();
-		for (uint i=1; i < n; ++i)
-		{
-			T value = container[i];
-			uint j = i;
-			while (j > 0 && container[j-1] > value)
-			{
-				container[j] = container[j-1];
-				--j;
-			}
-			if (i != j)
-			{
-				container[j] = value;
-			}
-		}
-	}
-
-	template <class Container, class T>
-	void quickSort(Container<T> & container)
-	{
-		quickSort(container, 0, container.count());
-	}
-	
-	{
-		/* threshhold for transitioning to insertion sort */
-		while (n > 12) {
-			int c01,c12,c,m,i,j;
-
-			/* compute median of three */
-			m = n >> 1;
-			c = p[0] > p[m];
-			c01 = c;
-			c = &p[m] > &p[n-1];
-			c12 = c;
-			/* if 0 >= mid >= end, or 0 < mid < end, then use mid */
-			if (c01 != c12) {
-				/* otherwise, we'll need to swap something else to middle */
-				int z;
-				c = p[0] < p[n-1];
-				/* 0>mid && mid<n:  0>n => n; 0<n => 0 */
-				/* 0<mid && mid>n:  0>n => 0; 0<n => n */
-				z = (c == c12) ? 0 : n-1;
-				swap(p[z], p[m]);
-			}
-			/* now p[m] is the median-of-three */
-			/* swap it to the beginning so it won't move around */
-			swap(p[0], p[m]);
-
-			/* partition loop */
-			i=1;
-			j=n-1;
-			for(;;) {
-				/* handling of equality is crucial here */
-				/* for sentinels & efficiency with duplicates */
-				for (;;++i) {
-					c = p[i] > p[0];
-					if (!c) break;
-				}
-				a = &p[0];
-				for (;;--j) {
-					b=&p[j];
-					c = p[j] > p[0]
-					if (!c) break;
-				}
-				/* make sure we haven't crossed */
-				if (i >= j) break;
-				swap(p[i], p[j]);
-
-				++i;
-				--j;
-			}
-			/* recurse on smaller side, iterate on larger */
-			if (j < (n-i)) {
-				quickSort(p, j);
-				p = p+i;
-				n = n-i;
-			} 
-			else {
-				quickSort(p+i, n-i);
-				n = j;
-			}
-		}
-		
-		insertionSort();
-	}
-#endif // 0
-
-} // nv namespace
-
-#endif // NV_CORE_ALGORITHMS_H

src/nvcore/BitArray.h

@@ -80,13 +80,13 @@ public:
 	/// Clear all the bits.
 	void clearAll()
 	{
-		memset(m_bitArray.mutableBuffer(), 0, m_bitArray.size());
+		memset(m_bitArray.unsecureBuffer(), 0, m_bitArray.size());
 	}
 
 	/// Set all the bits.
 	void setAll()
 	{
-		memset(m_bitArray.mutableBuffer(), 0xFF, m_bitArray.size());
+		memset(m_bitArray.unsecureBuffer(), 0xFF, m_bitArray.size());
 	}
 
 	/// Toggle all the bits.

src/nvcore/CMakeLists.txt

@@ -3,13 +3,7 @@ ADD_SUBDIRECTORY(poshlib)
 
 SET(CORE_SRCS
 	nvcore.h
-	DefsGnucDarwin.h
-	DefsGnucLinux.h
-	DefsGnucWin32.h
-	DefsVcWin32.h
 	Ptr.h
-	RefCounted.h
-	RefCounted.cpp
 	BitArray.h
 	Memory.h
 	Memory.cpp
@@ -26,15 +20,18 @@ SET(CORE_SRCS
 	TextWriter.cpp
 	Radix.h
 	Radix.cpp
-	CpuInfo.h
-	CpuInfo.cpp
-	Algorithms.h)
+	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/Containers.h

@@ -19,7 +19,6 @@ Do not use memmove in insert & remove, use copy ctors instead.
 #include <nvcore/nvcore.h>
 #include <nvcore/Memory.h>
 #include <nvcore/Debug.h>
-//#include <nvcore/Stream.h>
 
 #include <string.h>	// memmove
 #include <new>		// for placement new
@@ -71,10 +70,40 @@ namespace nv
 {
 	// Templates
 
+	/// Return the maximum of two values.
+	template <typename T> 
+	inline const T & max(const T & a, const T & b)
+	{
+		//return std::max(a, b);
+		if( a < b ) {
+			return b; 
+		}
+		return a;
+	}
+	
+	/// Return the minimum of two values.
+	template <typename T> 
+	inline const T & min(const T & a, const T & b)
+	{
+		//return std::min(a, b);
+		if( b < a ) {
+			return b; 
+		}
+		return a;
+	}
+	
+	/// Clamp between two values.
+	template <typename T> 
+	inline const T & clamp(const T & x, const T & a, const T & b)
+	{
+		return min(max(x, a), b);
+	}
+	
 	/// Swap two values.
 	template <typename T> 
 	inline void swap(T & a, T & b)
 	{
+		//return std::swap(a, b);
 		T temp = a; 
 		a = b; 
 		b = temp;
@@ -105,6 +134,16 @@ namespace nv
 		uint operator()(uint x) const { return x; }
 	};
 	
+	/// Delete all the elements of a container.
+	template <typename T>
+	void deleteAll(T & container)
+	{
+		for(typename T::PseudoIndex i = container.start(); !container.isDone(i); container.advance(i))
+		{
+			delete container[i];
+		}
+	}
+
 	
 	/** Return the next power of two. 
 	* @see http://graphics.stanford.edu/~seander/bithacks.html
@@ -115,7 +154,7 @@ namespace nv
 	inline uint nextPowerOfTwo( uint x )
 	{
 		nvDebugCheck( x != 0 );
-	#if 1	// On modern CPUs this is supposed to be as fast as using the bsr instruction.
+	#if 1	// On modern CPUs this is as fast as using the bsr instruction.
 		x--;
 		x |= x >> 1;
 		x |= x >> 2;
@@ -138,6 +177,15 @@ namespace nv
 		return (n & (n-1)) == 0;
 	}
 
+	/// Simple iterator interface.
+	template <typename T>
+	struct Iterator
+	{
+		virtual void advance();
+		virtual bool isDone();
+		virtual T current();
+	};
+
 
 	/**
 	* Replacement for std::vector that is easier to debug and provides
@@ -204,7 +252,7 @@ namespace nv
 		const T * buffer() const { return m_buffer; }
 	
 		/// Get vector pointer.
-		T * mutableBuffer() { return m_buffer; }
+		T * unsecureBuffer() { return m_buffer; }
 	
 		/// Is vector empty.
 		bool isEmpty() const { return m_size == 0; }
@@ -776,13 +824,13 @@ namespace nv
 		}
 	
 		/// Number of entries in the hash.
-		int	size()
+		int	size() const
 		{
 			return entry_count;
 		}
 	
 		/// Number of entries in the hash.
-		int	count()
+		int	count() const
 		{
 			return size();
 		}

src/nvcore/CpuInfo.cpp

@@ -1,88 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#include <nvcore/CpuInfo.h>
-#include <nvcore/Debug.h>
-
-using namespace nv;
-
-#if NV_OS_WIN32
-
-#define _WIN32_WINNT 0x0501
-#define WIN32_LEAN_AND_MEAN
-#include <windows.h>
-
-typedef BOOL (WINAPI *LPFN_ISWOW64PROCESS) (HANDLE, PBOOL);
-
-static bool isWow64()
-{
-	LPFN_ISWOW64PROCESS fnIsWow64Process = (LPFN_ISWOW64PROCESS)GetProcAddress(GetModuleHandle(TEXT("kernel32")), "IsWow64Process");
-
-	BOOL bIsWow64 = FALSE;
-
-	if (NULL != fnIsWow64Process)
-	{
-		if (!fnIsWow64Process(GetCurrentProcess(), &bIsWow64))
-		{
-			return false;
-		}
-	}
-
-	return bIsWow64 == TRUE;
-}
-
-#endif // NV_OS_WIN32
-
-
-
-uint CpuInfo::processorCount()
-{
-#if NV_OS_WIN32
-	SYSTEM_INFO sysInfo;
-
-	typedef BOOL (WINAPI *LPFN_ISWOW64PROCESS) (HANDLE, PBOOL);
-
-	if (isWow64())
-	{
-		GetNativeSystemInfo(&sysInfo);
-	}
-	else
-	{
-		GetSystemInfo(&sysInfo);
-	}
-
-	uint count = (uint)sysInfo.dwNumberOfProcessors;
-	nvDebugCheck(count >= 1);
-
-	return count;
-#else
-	return 1;
-#endif
-}
-
-uint CpuInfo::coreCount()
-{
-	return 1;
-}
-
-bool CpuInfo::hasMMX()
-{
-	return false;
-}
-
-bool CpuInfo::hasSSE()
-{
-	return false;
-}
-
-bool CpuInfo::hasSSE2()
-{
-	return false;
-}
-
-bool CpuInfo::hasSSE3()
-{
-	return false;
-}
-
-
-

src/nvcore/CpuInfo.h

@@ -1,88 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#ifndef NV_CORE_CPUINFO_H
-#define NV_CORE_CPUINFO_H
-
-#include <nvcore/nvcore.h>
-
-#if NV_CC_MSVC
-#	include <intrin.h> // __rdtsc
-#endif
-
-
-namespace nv
-{
-
-	// CPU Information.
-	class CpuInfo
-	{
-		static uint processorCount();
-		static uint coreCount();
-
-		static bool hasMMX();
-		static bool hasSSE();
-		static bool hasSSE2();
-		static bool hasSSE3();
-
-	};
-
-#if NV_CC_MSVC
-	#pragma intrinsic(__rdtsc)
-
-	inline uint64 rdtsc()
-	{
-		return __rdtsc();
-	}
-#endif
-
-#if NV_CC_GNUC
-
-#if defined(__i386__)
-
-	inline /*volatile*/ uint64 rdtsc()
-	{
-		uint64 x;
-		//__asm__ volatile ("rdtsc" : "=A" (x));
-		__asm__ volatile (".byte 0x0f, 0x31" : "=A" (x));
-		return x;
-	} 
-
-#elif defined(__x86_64__)
-
-	static __inline__ uint64 rdtsc(void)
-	{
-		unsigned int hi, lo;
-		__asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi));
-		return ( (unsigned long long)lo)|( ((unsigned long long)hi)<<32 );
-	}
-
-#elif defined(__powerpc__)
-
-	static __inline__ uint64 rdtsc(void)
-	{
-		uint64 result=0;
-		unsigned long int upper, lower, tmp;
-		__asm__ volatile(
-					"0:                  \n"
-					"\tmftbu   %0           \n"
-					"\tmftb    %1           \n"
-					"\tmftbu   %2           \n"
-					"\tcmpw    %2,%0        \n"
-					"\tbne     0b         \n"
-					: "=r"(upper),"=r"(lower),"=r"(tmp)
-					);
-		result = upper;
-		result = result<<32;
-		result = result|lower;
-
-		return(result);
-	}
-
-#endif
-
-#endif // NV_CC_GNUC
-
-
-} // nv namespace
-
-#endif // NV_CORE_CPUINFO_H

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
@@ -38,7 +38,14 @@
 #	include <unistd.h>	// getpid
 #	include <sys/types.h>
 #	include <sys/sysctl.h>	// sysctl
-#	include <ucontext.h>
+#	include <sys/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 +135,14 @@ namespace
 
 #if defined(HAVE_EXECINFO_H) // NV_OS_LINUX
 
+	static bool nvHasStackTrace() {
+#if NV_OS_DARWIN
+		return backtrace != NULL;
+#else
+		return true;
+#endif
+	}
+
 	static void nvPrintStackTrace(void * trace[], int size, int start=0) {
 		char ** string_array = backtrace_symbols(trace, size);
 	
@@ -166,24 +181,36 @@ 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 +255,18 @@ 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 +401,12 @@ 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, 2);
+			}
 #		endif
 
 			// Exit cleanly.
@@ -422,9 +453,12 @@ 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/Debug.h

@@ -115,6 +115,7 @@ namespace nv
 	{
 		NVCORE_API void dumpInfo();
 	
+		// These functions are not thread safe.
 		NVCORE_API void setMessageHandler( MessageHandler * messageHandler );
 		NVCORE_API void resetMessageHandler();
 	

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

@@ -8,11 +8,6 @@
 
 #include <stdio.h>	// NULL
 
-#define NV_DECLARE_PTR(Class) \
-	typedef SmartPtr<class Class> Class ## Ptr; \
-	typedef SmartPtr<const class Class> ClassConst ## Ptr
-
-
 namespace nv
 {
 	
@@ -48,8 +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. */
@@ -98,23 +96,125 @@ private:
 	T * m_ptr;
 };
 
+#if 0
+/** Reference counted base class to be used with Pointer.
+ *
+ * The only requirement of the Pointer class is that the RefCounted class implements the 
+ * addRef and release methods.
+ */
+class RefCounted
+{
+	NV_FORBID_COPY(RefCounted);
+public:
+
+	/// Ctor.
+	RefCounted() : m_count(0), m_weak_proxy(NULL)
+	{
+		s_total_obj_count++;
+	}
+
+	/// Virtual dtor.
+	virtual ~RefCounted()
+	{
+		nvCheck( m_count == 0 );
+		nvCheck( s_total_obj_count > 0 );
+		s_total_obj_count--;
+	}
+
+
+	/// Increase reference count.
+	uint addRef() const
+	{
+		s_total_ref_count++;
+		m_count++;
+		return m_count;
+	}
+
+
+	/// Decrease reference count and remove when 0.
+	uint release() const
+	{
+		nvCheck( m_count > 0 );
+		
+		s_total_ref_count--;
+		m_count--;
+		if( m_count == 0 ) {
+			releaseWeakProxy();
+			delete this;
+			return 0;
+		}
+		return m_count;
+	}
+
+	/// Get weak proxy.
+	WeakProxy * getWeakProxy() const
+	{
+		if (m_weak_proxy == NULL) {
+			m_weak_proxy = new WeakProxy;
+			m_weak_proxy->AddRef();
+		}
+		return m_weak_proxy;
+	}
+
+	/// Release the weak proxy.	
+	void releaseWeakProxy() const
+	{
+		if (m_weak_proxy != NULL) {
+			m_weak_proxy->NotifyObjectDied();
+			m_weak_proxy->Release();
+			m_weak_proxy = NULL;
+		}
+	}
+
+	/** @name Debug methods: */
+	//@{
+		/// Get reference count.
+		int refCount() const
+		{
+			return m_count;
+		}
+
+		/// Get total number of objects.
+		static int totalObjectCount()
+		{
+			return s_total_obj_count;
+		}
+
+		/// Get total number of references.
+		static int totalReferenceCount()
+		{
+			return s_total_ref_count;
+		}
+	//@}
+
+
+private:
+
+	NVCORE_API static int s_total_ref_count;
+	NVCORE_API static int s_total_obj_count;
+
+	mutable int m_count;
+	mutable WeakProxy * weak_proxy;
+
+};
+#endif
 
 /// Smart pointer template class.
 template <class BaseClass>
-class SmartPtr {
+class Pointer {
 public:
 
 	// BaseClass must implement addRef() and release().
-	typedef SmartPtr<BaseClass>	ThisType;
+	typedef Pointer<BaseClass>	ThisType;
 
 	/// Default ctor.
-	SmartPtr() : m_ptr(NULL) 
+	Pointer() : m_ptr(NULL) 
 	{
 	}
 
 	/** Other type assignment. */
 	template <class OtherBase>
-	SmartPtr( const SmartPtr<OtherBase> & tc )
+	Pointer( const Pointer<OtherBase> & tc )
 	{
 		m_ptr = static_cast<BaseClass *>( tc.ptr() );
 		if( m_ptr ) {
@@ -123,7 +223,7 @@ public:
 	}
 
 	/** Copy ctor. */
-	SmartPtr( const ThisType & bc )
+	Pointer( const ThisType & bc )
 	{
 		m_ptr = bc.ptr();
 		if( m_ptr ) {
@@ -131,8 +231,8 @@ public:
 		}
 	}
 
-	/** Copy cast ctor. SmartPtr(NULL) is valid. */
-	explicit SmartPtr( BaseClass * bc )
+	/** Copy cast ctor. Pointer(NULL) is valid. */
+	explicit Pointer( BaseClass * bc )
 	{
 		m_ptr = bc;
 		if( m_ptr ) {
@@ -141,7 +241,7 @@ public:
 	}
 
 	/** Dtor. */
-	~SmartPtr()
+	~Pointer()
 	{
 		set(NULL);
 	}
@@ -175,7 +275,7 @@ public:
 	//@{
 		/** Other type assignment. */
 		template <class OtherBase>
-		void operator = ( const SmartPtr<OtherBase> & tc )
+		void operator = ( const Pointer<OtherBase> & tc )
 		{
 			set( static_cast<BaseClass *>(tc.ptr()) );
 		}
@@ -198,7 +298,7 @@ public:
 	//@{
 		/** Other type equal comparation. */
 		template <class OtherBase>
-		bool operator == ( const SmartPtr<OtherBase> & other ) const
+		bool operator == ( const Pointer<OtherBase> & other ) const
 		{
 			return m_ptr == other.ptr();
 		}
@@ -217,7 +317,7 @@ public:
 
 		/** Other type not equal comparation. */
 		template <class OtherBase>
-		bool operator != ( const SmartPtr<OtherBase> & other ) const
+		bool operator != ( const Pointer<OtherBase> & other ) const
 		{
 			return m_ptr != other.ptr();
 		}

src/nvcore/RefCounted.cpp

@@ -1,9 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#include "RefCounted.h"
-
-using namespace nv;
-
-int nv::RefCounted::s_total_ref_count = 0;
-int nv::RefCounted::s_total_obj_count = 0;
-

src/nvcore/RefCounted.h

@@ -1,114 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#ifndef NV_CORE_REFCOUNTED_H
-#define NV_CORE_REFCOUNTED_H
-
-#include <nvcore/nvcore.h>
-#include <nvcore/Debug.h>
-
-
-namespace nv
-{
-
-	/// Reference counted base class to be used with SmartPtr and WeakPtr.
-	class RefCounted
-	{
-		NV_FORBID_COPY(RefCounted);
-	public:
-
-		/// Ctor.
-		RefCounted() : m_count(0)/*, m_weak_proxy(NULL)*/
-		{
-			s_total_obj_count++;
-		}
-
-		/// Virtual dtor.
-		virtual ~RefCounted()
-		{
-			nvCheck( m_count == 0 );
-			nvCheck( s_total_obj_count > 0 );
-			s_total_obj_count--;
-		}
-
-
-		/// Increase reference count.
-		uint addRef() const
-		{
-			s_total_ref_count++;
-			m_count++;
-			return m_count;
-		}
-
-
-		/// Decrease reference count and remove when 0.
-		uint release() const
-		{
-			nvCheck( m_count > 0 );
-			
-			s_total_ref_count--;
-			m_count--;
-			if( m_count == 0 ) {
-			//	releaseWeakProxy();
-				delete this;
-				return 0;
-			}
-			return m_count;
-		}
-	/*
-		/// Get weak proxy.
-		WeakProxy * getWeakProxy() const
-		{
-			if (m_weak_proxy == NULL) {
-				m_weak_proxy = new WeakProxy;
-				m_weak_proxy->AddRef();
-			}
-			return m_weak_proxy;
-		}
-
-		/// Release the weak proxy.	
-		void releaseWeakProxy() const
-		{
-			if (m_weak_proxy != NULL) {
-				m_weak_proxy->NotifyObjectDied();
-				m_weak_proxy->Release();
-				m_weak_proxy = NULL;
-			}
-		}
-	*/
-		/** @name Debug methods: */
-		//@{
-			/// Get reference count.
-			int refCount() const
-			{
-				return m_count;
-			}
-
-			/// Get total number of objects.
-			static int totalObjectCount()
-			{
-				return s_total_obj_count;
-			}
-
-			/// Get total number of references.
-			static int totalReferenceCount()
-			{
-				return s_total_ref_count;
-			}
-		//@}
-
-
-	private:
-
-		NVCORE_API static int s_total_ref_count;
-		NVCORE_API static int s_total_obj_count;
-
-		mutable int m_count;
-	//	mutable WeakProxy * weak_proxy;
-
-	};
-
-
-} // nv namespace
-
-
-#endif // NV_CORE_REFCOUNTED_H

src/nvcore/StdStream.h

@@ -47,25 +47,25 @@ public:
 
 	/** @name Stream implementation. */
 	//@{
-		virtual void seek( int pos )
+		virtual void seek( uint pos )
 		{
 			nvDebugCheck(m_fp != NULL);
-			nvDebugCheck(pos >= 0 && pos < size());
+			nvDebugCheck(pos < size());
 			fseek(m_fp, pos, SEEK_SET);
 		}
 		
-		virtual int tell() const
+		virtual uint tell() const
 		{
 			nvDebugCheck(m_fp != NULL);
 			return ftell(m_fp);
 		}
 		
-		virtual int size() const
+		virtual uint size() const
 		{
 			nvDebugCheck(m_fp != NULL);
-			int pos = ftell(m_fp);
+			uint pos = ftell(m_fp);
 			fseek(m_fp, 0, SEEK_END);
-			int end = ftell(m_fp);
+			uint end = ftell(m_fp);
 			fseek(m_fp, pos, SEEK_SET);
 			return end;
 		}
@@ -117,11 +117,11 @@ public:
 	/** @name Stream implementation. */
 	//@{
 		/// Write data.
-		virtual void serialize( void * data, int len )
+		virtual uint serialize( void * data, uint len )
 		{
 			nvDebugCheck(data != NULL);
 			nvDebugCheck(m_fp != NULL);
-			fwrite(data, len, 1, m_fp);
+			return (uint)fwrite(data, 1, len, m_fp);
 		}
 		
 		virtual bool isLoading() const
@@ -156,11 +156,11 @@ public:
 	/** @name Stream implementation. */
 	//@{
 		/// Read data.
-		virtual void serialize( void * data, int len )
+		virtual uint serialize( void * data, uint len )
 		{
 			nvDebugCheck(data != NULL);
 			nvDebugCheck(m_fp != NULL);
-			fread(data, len, 1, m_fp);
+			return (uint)fread(data, 1, len, m_fp);
 		}
 		
 		virtual bool isLoading() const
@@ -184,33 +184,40 @@ class NVCORE_CLASS MemoryInputStream : public Stream
 public:
 
 	/// Ctor.
-	MemoryInputStream( const uint8 * mem, int size ) : 
+	MemoryInputStream( const uint8 * mem, uint size ) : 
 		m_mem(mem), m_ptr(mem), m_size(size) { }
 
 	/** @name Stream implementation. */
 	//@{
 		/// Read data.
-		virtual void serialize( void * data, int len )
+		virtual uint serialize( void * data, uint len )
 		{
 			nvDebugCheck(data != NULL);
 			nvDebugCheck(!isError());
+
+			uint left = m_size - tell();
+			if (len > left) len = left;
+			
 			memcpy( data, m_ptr, len );
 			m_ptr += len;
+				
+			return len;
 		}
 		
-		virtual void seek( int pos )
+		virtual void seek( uint pos )
 		{
 			nvDebugCheck(!isError());
 			m_ptr = m_mem + pos;
 			nvDebugCheck(!isError());
 		}
 		
-		virtual int tell() const
+		virtual uint tell() const
 		{
-			return int(m_ptr - m_mem);
+			nvDebugCheck(m_ptr >= m_mem);
+			return uint(m_ptr - m_mem);
 		}
 		
-		virtual int size() const
+		virtual uint size() const
 		{
 			return m_size;
 		}
@@ -252,7 +259,7 @@ private:
 
 	const uint8 * m_mem;
 	const uint8 * m_ptr;
-	int m_size;
+	uint m_size;
 
 };
 
@@ -286,17 +293,19 @@ public:
 	/** @name Stream implementation. */
 	//@{
 		/// Read data.
-		virtual void serialize( void * data, int len )
+		virtual uint serialize( void * data, uint len )
 		{
 			nvDebugCheck(data != NULL);
-			m_s->serialize( data, len );
+			len = m_s->serialize( data, len );
 			
 			if( m_s->isError() ) {
 				throw std::exception();
 			}
+
+			return len;
 		}
 		
-		virtual void seek( int pos )
+		virtual void seek( uint pos )
 		{
 			m_s->seek( pos );
 			
@@ -305,12 +314,12 @@ public:
 			}
 		}
 		
-		virtual int tell() const
+		virtual uint tell() const
 		{
 			return m_s->tell();
 		}
 		
-		virtual int size() const
+		virtual uint size() const
 		{
 			return m_s->size();
 		}

src/nvcore/StrLib.cpp

@@ -209,48 +209,11 @@ StringBuilder::StringBuilder( const StringBuilder & s ) : m_size(0), m_str(NULL)
 }
 
 /** Copy string. */
-StringBuilder::StringBuilder( const char * s )
+StringBuilder::StringBuilder( const char * s ) : m_size(0), m_str(NULL)
 {
 	copy(s);
 }
 
-/** Allocate and copy string. */
-StringBuilder::StringBuilder( int size_hint, const StringBuilder & s) : m_size(size_hint), m_str(NULL)
-{
-	nvDebugCheck(m_size > 0);
-	m_str = strAlloc(m_size);
-	copy(s);
-}
-
-/** Allocate and format string. */
-StringBuilder::StringBuilder( const char * fmt, ... ) : m_size(0), m_str(NULL)
-{
-	nvDebugCheck(fmt != NULL);
-	va_list arg;
-	va_start( arg, fmt );
-
-	format( fmt, arg );
-
-	va_end( arg );
-}
-
-/** Allocate and format string. */
-StringBuilder::StringBuilder( int size_hint, const char * fmt, ... ) : m_size(size_hint), m_str(NULL)
-{
-	nvDebugCheck(m_size > 0);	
-	nvDebugCheck(fmt != NULL);
-	
-	m_str = strAlloc(m_size);
-
-	va_list arg;
-	va_start( arg, fmt );
-
-	format( fmt, arg );
-
-	va_end( arg );
-}
-
-
 /** Delete the string. */
 StringBuilder::~StringBuilder()
 {
@@ -278,8 +241,7 @@ StringBuilder & StringBuilder::format( const char * fmt, ... )
 /** Format a string safely. */
 StringBuilder & StringBuilder::format( const char * fmt, va_list arg )
 {
-	nvCheck(fmt != NULL);
-	nvCheck(m_size >= 0);
+	nvDebugCheck(fmt != NULL);
 
 	if( m_size == 0 ) {
 		m_size = 64;
@@ -327,8 +289,7 @@ StringBuilder & StringBuilder::format( const char * fmt, va_list arg )
 /** Append a string. */
 StringBuilder & StringBuilder::append( const char * s )
 {
-	nvCheck(s != NULL);
-	nvCheck(m_size >= 0);
+	nvDebugCheck(s != NULL);
 
 	const uint slen = uint(strlen( s ));
 
@@ -475,31 +436,6 @@ void StringBuilder::reset()
 }
 
 
-Path::Path(const char * fmt, ...)
-{
-	nvDebugCheck( fmt != NULL );
-
-	va_list arg;
-	va_start( arg, fmt );
-
-	format( fmt, arg );
-
-	va_end( arg );
-}
-
-Path::Path(int size_hint, const char * fmt, ...) : StringBuilder(size_hint)
-{
-	nvDebugCheck( fmt != NULL );
-
-	va_list arg;
-	va_start( arg, fmt );
-
-	format( fmt, arg );
-
-	va_end( arg );
-}
-
-
 /// Get the file name from a path.
 const char * Path::fileName() const
 {
@@ -609,8 +545,6 @@ const char * Path::extension(const char * str)
 }
 
 
-// static
-String String::s_null(String::null);
 
 /// Clone this string
 String String::clone() const
@@ -621,13 +555,13 @@ String String::clone() const
 
 void String::setString(const char * str)
 {
-	if( str == NULL ) {
-		data = s_null.data;
+	if (str == NULL) {
+		data = NULL;
 	}
 	else {
 		allocString( str );
+		addRef();
 	}
-	addRef();
 }
 
 void String::setString(const char * str, int length)
@@ -640,11 +574,11 @@ void String::setString(const char * str, int length)
 
 void String::setString(const StringBuilder & str)
 {
-	if( str.str() == NULL ) {
-		data = s_null.data;
+	if (str.str() == NULL) {
+		data =	NULL;
 	}
 	else {
 		allocString(str);
+		addRef();
 	}
-	addRef();
 }	

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;
@@ -47,9 +47,6 @@ namespace nv
 		explicit StringBuilder( int size_hint );
 		StringBuilder( const char * str );
 		StringBuilder( const StringBuilder & );
-		StringBuilder( int size_hint, const StringBuilder & );	
-		StringBuilder( const char * format, ... ) __attribute__((format (printf, 2, 3)));
-		StringBuilder( int size_hint, const char * format, ... ) __attribute__((format (printf, 3, 4)));
 	
 		~StringBuilder();
 	
@@ -120,18 +117,16 @@ namespace nv
 		char * m_str;
 		
 	};
-	
 
-	/// Path string.
+
+	/// Path string. @@ This should be called PathBuilder.
 	class NVCORE_CLASS Path : public StringBuilder
 	{
 	public:
 		Path() : StringBuilder() {}
 		explicit Path(int size_hint) : StringBuilder(size_hint) {}
-		Path(const StringBuilder & str) : StringBuilder(str) {}
-		Path(int size_hint, const StringBuilder & str) : StringBuilder(size_hint, str) {}	
-		Path(const char * format, ...) __attribute__((format (printf, 2, 3)));
-		Path(int size_hint, const char * format, ...) __attribute__((format (printf, 3, 4)));
+		Path(const char * str) : StringBuilder(str) {}
+		Path(const Path & path) : StringBuilder(path) {}
 		
 		const char * fileName() const;
 		const char * extension() const;
@@ -140,7 +135,7 @@ namespace nv
 		
 		void stripFileName();
 		void stripExtension();
-		
+
 		// statics
 		NVCORE_API static char separator();
 		NVCORE_API static const char * fileName(const char *);
@@ -156,15 +151,14 @@ namespace nv
 		/// Constructs a null string. @sa isNull()
 		String()
 		{
-			data = s_null.data;
-			addRef();
+			data = NULL;
 		}
 
 		/// Constructs a shared copy of str.
 		String(const String & str)
 		{
 			data = str.data;
-			addRef();
+			if (data != NULL) addRef();
 		}
 
 		/// Constructs a shared string from a standard string.
@@ -188,7 +182,6 @@ namespace nv
 		/// Dtor.
 		~String()
 		{
-			nvDebugCheck(data != NULL);
 			release();
 		}
 
@@ -213,52 +206,61 @@ 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;
 		}
 
 		/// Equal operator.
 		bool operator==( const String & str ) const
 		{
-			nvDebugCheck(data != NULL);
-			nvDebugCheck(str.data != NULL);
 			if( str.data == data ) {
 				return true;
 			}
+			if ((data == NULL) != (str.data == NULL)) {
+				return false;
+			}
 			return strcmp(data, str.data) == 0;
 		}
 
 		/// Equal operator.
 		bool operator==( const char * str ) const
 		{
-			nvDebugCheck(data != NULL);
 			nvCheck(str != NULL);	// Use isNull!
+			if (data == NULL) {
+				return false;
+			}
 			return strcmp(data, str) == 0;
 		}
 
 		/// Not equal operator.
 		bool operator!=( const String & str ) const
 		{
-			nvDebugCheck(data != NULL);
-			nvDebugCheck(str.data != NULL);
 			if( str.data == data ) {
 				return false;
 			}
+			if ((data == NULL) != (str.data == NULL)) {
+				return true;
+			}
 			return strcmp(data, str.data) != 0;
 		}
 	
 		/// Not equal operator.
 		bool operator!=( const char * str ) const
 		{
-			nvDebugCheck(data != NULL);
 			nvCheck(str != NULL);	// Use isNull!
+			if (data == NULL) {
+				return false;
+			}
 			return strcmp(data, str) != 0;
 		}
 	
 		/// Returns true if this string is the null string.
-		bool isNull() const { nvDebugCheck(data != NULL); return data == s_null.data; }
+		bool isNull() const { return data == NULL; }
 	
 		/// Return the exact length.
 		uint length() const { nvDebugCheck(data != NULL); return uint(strlen(data)); }
@@ -267,44 +269,45 @@ namespace nv
 		uint hash() const { nvDebugCheck(data != NULL); return strHash(data); }
 	
 		/// const char * cast operator.
-		operator const char * () const { nvDebugCheck(data != NULL); return data; }
+		operator const char * () const { return data; }
 	
 		/// Get string pointer.
-		const char * str() const { nvDebugCheck(data != NULL); return data; }
+		const char * str() const { return data; }
 	
 
 	private:
 
-		enum null_t { null };
-		
-		// Private constructor for null string.
-		String(null_t) {
-			setString("");
-		}
-
 		// Add reference count.
-		void addRef() {
-			nvDebugCheck(data != NULL);
-			setRefCount(getRefCount() + 1);
-		}
-		
-		// Decrease reference count.
-		void release() {
-			nvDebugCheck(data != NULL);
-
-			const uint16 count = getRefCount();
-			setRefCount(count - 1);
-			if( count - 1 == 0 ) {
-				mem::free(data - 2);
-				data = NULL;
+		void addRef()
+		{
+			if (data != NULL)
+			{
+				setRefCount(getRefCount() + 1);
 			}
 		}
 		
-		uint16 getRefCount() const {
+		// Decrease reference count.
+		void release()
+		{
+			if (data != NULL)
+			{
+				const uint16 count = getRefCount();
+				setRefCount(count - 1);
+				if (count - 1 == 0) {
+					mem::free(data - 2);
+					data = NULL;
+				}
+			}
+		}
+		
+		uint16 getRefCount() const
+		{
+			nvDebugCheck(data != NULL);
 			return *reinterpret_cast<const uint16 *>(data - 2);
 		}
 		
 		void setRefCount(uint16 count) {
+			nvDebugCheck(data != NULL);
 			nvCheck(count < 0xFFFF);
 			*reinterpret_cast<uint16 *>(const_cast<char *>(data - 2)) = uint16(count);
 		}
@@ -343,8 +346,6 @@ namespace nv
 	
 	private:
 
-		NVCORE_API static String s_null;
-
 		const char * data;
 		
 	};

src/nvcore/Stream.h

@@ -41,17 +41,17 @@ public:
 	ByteOrder byteOrder() const { return m_byteOrder; }
 
 	
-	/// Serialize the given data. @@ Should return bytes serialized?
-	virtual void serialize( void * data, int len ) = 0;
+	/// Serialize the given data.
+	virtual uint serialize( void * data, uint len ) = 0;
 
 	/// Move to the given position in the archive.
-	virtual void seek( int pos ) = 0;
+	virtual void seek( uint pos ) = 0;
 
 	/// Return the current position in the archive.
-	virtual int tell() const = 0;
+	virtual uint tell() const = 0;
 
 	/// Return the current size of the archive.
-	virtual int size() const = 0;
+	virtual uint size() const = 0;
 
 	/// Determine if there has been any error.
 	virtual bool isError() const = 0;
@@ -136,13 +136,13 @@ public:
 protected:
 
 	/// Serialize in the stream byte order.
-	Stream & byteOrderSerialize( void * v, int len ) {
+	Stream & byteOrderSerialize( void * v, uint len ) {
 		if( m_byteOrder == getSystemByteOrder() ) {
 			serialize( v, len );
 		}
 		else {
-			for( int i=len-1; i>=0; i-- ) {
-				serialize( (uint8 *)v + i, 1 );
+			for( uint i = len; i > 0; i-- ) {
+				serialize( (uint8 *)v + i - 1, 1 );
 			}
 		}
 		return *this;

src/nvcore/TextReader.cpp

@@ -48,7 +48,7 @@ const char * TextReader::readToEnd()
 	m_text.reserve(size + 1);
 	m_text.resize(size);
 	
-	m_stream->serialize(m_text.mutableBuffer(), size);
+	m_stream->serialize(m_text.unsecureBuffer(), size);
 	m_text.pushBack('\0');
 	
 	return m_text.buffer();

src/nvimage/BlockDXT.cpp

@@ -22,7 +22,6 @@
 // OTHER DEALINGS IN THE SOFTWARE.
 
 #include <nvcore/Stream.h>
-#include <nvcore/Containers.h> // swap
 
 #include "ColorBlock.h"
 #include "BlockDXT.h"

src/nvimage/ColorBlock.cpp

@@ -1,6 +1,5 @@
 // This code is in the public domain -- castanyo@yahoo.es
 
-#include <nvcore/Containers.h> // swap
 #include <nvmath/Box.h>
 #include <nvimage/ColorBlock.h>
 #include <nvimage/Image.h>

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] = (0 << 16) | (9 << 8) | (5);	// major.minor.revision
+	this->reserved[10] = (2 << 16) | (0 << 8) | (7);	// major.minor.revision
 
 	this->pf.size = 32;
 	this->pf.flags = 0;
@@ -989,10 +989,10 @@ void DirectDrawSurface::readLinearImage(Image * img)
 			stream->serialize(&c, byteCount);
 
 			Color32 pixel(0, 0, 0, 0xFF);
-			pixel.r = PixelFormat::convert(c >> rshift, rsize, 8);
-			pixel.g = PixelFormat::convert(c >> gshift, gsize, 8);
-			pixel.b = PixelFormat::convert(c >> bshift, bsize, 8);
-			pixel.a = PixelFormat::convert(c >> ashift, asize, 8);
+			pixel.r = PixelFormat::convert((c & header.pf.rmask) >> rshift, rsize, 8);
+			pixel.g = PixelFormat::convert((c & header.pf.gmask) >> gshift, gsize, 8);
+			pixel.b = PixelFormat::convert((c & header.pf.bmask) >> bshift, bsize, 8);
+			pixel.a = PixelFormat::convert((c & header.pf.amask) >> ashift, asize, 8);
 
 			img->pixel(x, y) = pixel;
 		}

src/nvimage/Filter.cpp

@@ -26,18 +26,17 @@
  * http://www.xmission.com/~legalize/zoom.html
  * 
  * Reconstruction Filters in Computer Graphics
- * http://www.mentallandscape.com/Papers_siggraph88.pdf 
+ * http://www.mentallandscape.com/Papers_siggraph88.pdf
  *
  * More references:
- * http://www.worldserver.com/turk/computergraphics/ResamplingFilters.pdf 
+ * http://www.worldserver.com/turk/computergraphics/ResamplingFilters.pdf
  * http://www.dspguide.com/ch16.htm
  */
 
+#include "Filter.h"
 
-#include <nvcore/Containers.h>	// swap
-#include <nvmath/nvmath.h>	// fabs
 #include <nvmath/Vector.h>	// Vector4
-#include <nvimage/Filter.h>
+#include <nvcore/Containers.h>	// swap
 
 using namespace nv;
 
@@ -244,7 +243,7 @@ SincFilter::SincFilter(float w) : Filter(w) {}
 
 float SincFilter::evaluate(float x) const
 {
-	return 0.0f;
+	return sincf(PI * x);
 }
 
 
@@ -504,7 +503,7 @@ void Kernel2::initBlendedSobel(const Vector4 & scale)
 
 		for (int i = 0; i < 7; i++) {
 			for (int e = 0; e < 7; e++) {
-				m_data[i * 9 + e + 1] += elements[i * 7 + e] * scale.z();
+				m_data[(i + 1) * 9 + e + 1] += elements[i * 7 + e] * scale.z();
 			}
 		}
 	}
@@ -519,7 +518,7 @@ void Kernel2::initBlendedSobel(const Vector4 & scale)
 
 		for (int i = 0; i < 5; i++) {
 			for (int e = 0; e < 5; e++) {
-				m_data[i * 9 + e + 2] += elements[i * 5 + e] * scale.y();
+				m_data[(i + 2) * 9 + e + 2] += elements[i * 5 + e] * scale.y();
 			}
 		}
 	}
@@ -532,7 +531,7 @@ void Kernel2::initBlendedSobel(const Vector4 & scale)
 
 		for (int i = 0; i < 3; i++) {
 			for (int e = 0; e < 3; e++) {
-				m_data[i * 9 + e + 3] += elements[i * 3 + e] * scale.x();
+				m_data[(i + 3) * 9 + e + 3] += elements[i * 3 + e] * scale.x();
 			}
 		}
 	}
@@ -541,12 +540,17 @@ void Kernel2::initBlendedSobel(const Vector4 & scale)
 
 PolyphaseKernel::PolyphaseKernel(const Filter & f, uint srcLength, uint dstLength, int samples/*= 32*/)
 {
-	nvCheck(srcLength >= dstLength);	// @@ Upsampling not implemented!
 	nvDebugCheck(samples > 0);
-	
-	const float scale = float(dstLength) / float(srcLength);
+
+	float scale = float(dstLength) / float(srcLength);
 	const float iscale = 1.0f / scale;
 
+	if (scale > 1) {
+		// Upsampling.
+		samples = 1;
+		scale = 1;
+	}
+
 	m_length = dstLength;
 	m_width = f.width() * iscale;
 	m_windowSize = (int)ceilf(m_width * 2) + 1;

src/nvimage/Filter.h

@@ -78,7 +78,7 @@ namespace nv
 		MitchellFilter();
 		virtual float evaluate(float x) const;
 
-		void setParameters(float a, float b);
+		void setParameters(float b, float c);
 
 	private:
 		float p0, p2, p3;

src/nvimage/FloatImage.cpp

@@ -1,16 +1,18 @@
 // This code is in the public domain -- castanyo@yahoo.es
 
-#include <nvcore/Containers.h>
-#include <nvcore/Ptr.h>
-
-#include <nvmath/Color.h>
-
 #include "FloatImage.h"
 #include "Filter.h"
 #include "Image.h"
 
+#include <nvmath/Color.h>
+#include <nvmath/Matrix.h>
+
+#include <nvcore/Containers.h>
+#include <nvcore/Ptr.h>
+
 #include <math.h>
 
+
 using namespace nv;
 
 namespace 
@@ -140,7 +142,8 @@ Image * FloatImage::createImageGammaCorrect(float gamma/*= 2.2f*/) const
 /// Allocate a 2d float image of the given format and the given extents.
 void FloatImage::allocate(uint c, uint w, uint h)
 {
-	nvCheck(m_mem == NULL);
+	free();
+
 	m_width = w;
 	m_height = h;
 	m_componentNum = c;
@@ -151,7 +154,6 @@ void FloatImage::allocate(uint c, uint w, uint h)
 /// Free the image, but don't clear the members.
 void FloatImage::free()
 {
-	nvCheck(m_mem != NULL);
 	nv::mem::free( reinterpret_cast<void *>(m_mem) );
 	m_mem = NULL;
 }
@@ -376,7 +378,7 @@ FloatImage * FloatImage::fastDownSample() const
 	{
 		const uint n = w * h;
 		
-		if (n & 1)
+		if ((m_width * m_height) & 1)
 		{
 			const float scale = 1.0f / (2 * n + 1);
 			
@@ -540,73 +542,27 @@ FloatImage * FloatImage::fastDownSample() const
 	return dst_image.release();
 }
 
-/*
-/// Downsample applying a 1D kernel separately in each dimension.
-FloatImage * FloatImage::downSample(const Kernel1 & kernel, WrapMode wm) const
-{
-	const uint w = max(1, m_width / 2);
-	const uint h = max(1, m_height / 2);
-	
-	return downSample(kernel, w, h, wm);
-}
-
-
-/// Downsample applying a 1D kernel separately in each dimension.
-FloatImage * FloatImage::downSample(const Kernel1 & kernel, uint w, uint h, WrapMode wm) const
-{
-	nvCheck(!(kernel.windowSize() & 1));	// Make sure that kernel m_width is even.
-
-	AutoPtr<FloatImage> tmp_image( new FloatImage() );
-	tmp_image->allocate(m_componentNum, w, m_height);
-	
-	AutoPtr<FloatImage> dst_image( new FloatImage() );	
-	dst_image->allocate(m_componentNum, w, h);
-	
-	const float xscale = float(m_width) / float(w);
-	const float yscale = float(m_height) / float(h);
-	
-	for(uint c = 0; c < m_componentNum; c++) {
-		float * tmp_channel = tmp_image->channel(c);
-		
-		for(uint y = 0; y < m_height; y++) {
-			for(uint x = 0; x < w; x++) {
-				
-				float sum = this->applyKernelHorizontal(&kernel, uint(x*xscale), y, c, wm);
-				
-				const uint tmp_index = tmp_image->index(x, y);
-				tmp_channel[tmp_index] = sum;
-			}
-		}
-		
-		float * dst_channel = dst_image->channel(c);
-		
-		for(uint y = 0; y < h; y++) {
-			for(uint x = 0; x < w; x++) {
-				
-				float sum = tmp_image->applyKernelVertical(&kernel, uint(x*xscale), uint(y*yscale), c, wm);
-				
-				const uint dst_index = dst_image->index(x, y);
-				dst_channel[dst_index] = sum;
-			}
-		}
-	}
-	
-	return dst_image.release();
-}
-*/
-
 /// Downsample applying a 1D kernel separately in each dimension.
 FloatImage * FloatImage::downSample(const Filter & filter, WrapMode wm) const
 {
 	const uint w = max(1, m_width / 2);
 	const uint h = max(1, m_height / 2);
 
-	return downSample(filter, w, h, wm);
+	return resize(filter, w, h, wm);
+}
+
+/// Downsample applying a 1D kernel separately in each dimension.
+FloatImage * FloatImage::downSample(const Filter & filter, WrapMode wm, uint alpha) const
+{
+	const uint w = max(1, m_width / 2);
+	const uint h = max(1, m_height / 2);
+
+	return resize(filter, w, h, wm, alpha);
 }
 
 
 /// Downsample applying a 1D kernel separately in each dimension.
-FloatImage * FloatImage::downSample(const Filter & filter, uint w, uint h, WrapMode wm) const
+FloatImage * FloatImage::resize(const Filter & filter, uint w, uint h, WrapMode wm) const
 {
 	// @@ Use monophase filters when frac(m_width / w) == 0
 
@@ -636,7 +592,7 @@ FloatImage * FloatImage::downSample(const Filter & filter, uint w, uint h, WrapM
 			float * dst_channel = dst_image->channel(c);
 			
 			for (uint x = 0; x < w; x++) {
-				tmp_image->applyKernelVertical(ykernel, x, c, wm, tmp_column.mutableBuffer());
+				tmp_image->applyKernelVertical(ykernel, x, c, wm, tmp_column.unsecureBuffer());
 				
 				for (uint y = 0; y < h; y++) {
 					dst_channel[y * w + x] = tmp_column[y];
@@ -657,7 +613,7 @@ FloatImage * FloatImage::downSample(const Filter & filter, uint w, uint h, WrapM
 			float * tmp_channel = tmp_image->channel(c);
 
 			for (uint x = 0; x < w; x++) {
-				tmp_image->applyKernelVertical(ykernel, x, c, wm, tmp_column.mutableBuffer());
+				tmp_image->applyKernelVertical(ykernel, x, c, wm, tmp_column.unsecureBuffer());
 				
 				for (uint y = 0; y < h; y++) {
 					tmp_channel[y * w + x] = tmp_column[y];
@@ -675,10 +631,56 @@ FloatImage * FloatImage::downSample(const Filter & filter, uint w, uint h, WrapM
 	return dst_image.release();
 }
 
+/// Downsample applying a 1D kernel separately in each dimension.
+FloatImage * FloatImage::resize(const Filter & filter, uint w, uint h, WrapMode wm, uint alpha) const
+{
+	nvCheck(alpha < m_componentNum);
+
+	AutoPtr<FloatImage> tmp_image( new FloatImage() );
+	AutoPtr<FloatImage> dst_image( new FloatImage() );	
+	
+	PolyphaseKernel xkernel(filter, m_width, w, 32);
+	PolyphaseKernel ykernel(filter, m_height, h, 32);
+	
+	{
+		tmp_image->allocate(m_componentNum, w, m_height);
+		dst_image->allocate(m_componentNum, w, h);
+		
+		Array<float> tmp_column(h);
+		tmp_column.resize(h);
+		
+		for (uint c = 0; c < m_componentNum; c++)
+		{
+			float * tmp_channel = tmp_image->channel(c);
+			
+			for (uint y = 0; y < m_height; y++) {
+				this->applyKernelHorizontal(xkernel, y, c, alpha, wm, tmp_channel + y * w);
+			}
+		}
+
+		// Process all channels before applying vertical kernel to make sure alpha has been computed.
+
+		for (uint c = 0; c < m_componentNum; c++)
+		{
+			float * dst_channel = dst_image->channel(c);
+			
+			for (uint x = 0; x < w; x++) {
+				tmp_image->applyKernelVertical(ykernel, x, c, alpha, wm, tmp_column.unsecureBuffer());
+				
+				for (uint y = 0; y < h; y++) {
+					dst_channel[y * w + x] = tmp_column[y];
+				}
+			}
+		}
+	}
+	
+	return dst_image.release();
+}
+
 
 
 /// Apply 2D kernel at the given coordinates and return result.
-float FloatImage::applyKernel(const Kernel2 * k, int x, int y, int c, WrapMode wm) const
+float FloatImage::applyKernel(const Kernel2 * k, int x, int y, uint c, WrapMode wm) const
 {
 	nvDebugCheck(k != NULL);
 	
@@ -707,7 +709,7 @@ float FloatImage::applyKernel(const Kernel2 * k, int x, int y, int c, WrapMode w
 
 
 /// Apply 1D vertical kernel at the given coordinates and return result.
-float FloatImage::applyKernelVertical(const Kernel1 * k, int x, int y, int c, WrapMode wm) const
+float FloatImage::applyKernelVertical(const Kernel1 * k, int x, int y, uint c, WrapMode wm) const
 {
 	nvDebugCheck(k != NULL);
 	
@@ -729,7 +731,7 @@ float FloatImage::applyKernelVertical(const Kernel1 * k, int x, int y, int c, Wr
 }
 
 /// Apply 1D horizontal kernel at the given coordinates and return result.
-float FloatImage::applyKernelHorizontal(const Kernel1 * k, int x, int y, int c, WrapMode wm) const
+float FloatImage::applyKernelHorizontal(const Kernel1 * k, int x, int y, uint c, WrapMode wm) const
 {
 	nvDebugCheck(k != NULL);
 	
@@ -752,7 +754,7 @@ float FloatImage::applyKernelHorizontal(const Kernel1 * k, int x, int y, int c,
 
 
 /// Apply 1D vertical kernel at the given coordinates and return result.
-void FloatImage::applyKernelVertical(const PolyphaseKernel & k, int x, int c, WrapMode wm, float * output) const
+void FloatImage::applyKernelVertical(const PolyphaseKernel & k, int x, uint c, WrapMode wm, float * __restrict output) const
 {
 	const uint length = k.length();
 	const float scale = float(length) / float(m_height);
@@ -784,7 +786,7 @@ void FloatImage::applyKernelVertical(const PolyphaseKernel & k, int x, int c, Wr
 }
 
 /// Apply 1D horizontal kernel at the given coordinates and return result.
-void FloatImage::applyKernelHorizontal(const PolyphaseKernel & k, int y, int c, WrapMode wm, float * output) const
+void FloatImage::applyKernelHorizontal(const PolyphaseKernel & k, int y, uint c, WrapMode wm, float * __restrict output) const
 {
 	const uint length = k.length();
 	const float scale = float(length) / float(m_width);
@@ -815,3 +817,93 @@ void FloatImage::applyKernelHorizontal(const PolyphaseKernel & k, int y, int c,
 	}
 }
 
+
+/// Apply 1D vertical kernel at the given coordinates and return result.
+void FloatImage::applyKernelVertical(const PolyphaseKernel & k, int x, uint c, uint a, WrapMode wm, float * __restrict output) const
+{
+	const uint length = k.length();
+	const float scale = float(length) / float(m_height);
+	const float iscale = 1.0f / scale;
+
+	const float width = k.width();
+	const int windowSize = k.windowSize();
+
+	const float * channel = this->channel(c);
+	const float * alpha = this->channel(a);
+
+	for (uint i = 0; i < length; i++)
+	{
+		const float center = (0.5f + i) * iscale;
+		
+		const int left = (int)floorf(center - width);
+		const int right = (int)ceilf(center + width);
+		nvCheck(right - left <= windowSize);
+		
+		float norm = 0;
+		float sum = 0;
+		for (int j = 0; j < windowSize; ++j)
+		{
+			const int idx = this->index(x, j+left, wm);
+			
+			float w = k.valueAt(i, j) * (alpha[idx] + (1.0f / 256.0f));
+			norm += w;
+			sum += w * channel[idx];
+		}
+		
+		output[i] = sum / norm;
+	}
+}
+
+/// Apply 1D horizontal kernel at the given coordinates and return result.
+void FloatImage::applyKernelHorizontal(const PolyphaseKernel & k, int y, uint c, uint a, WrapMode wm, float * __restrict output) const
+{
+	const uint length = k.length();
+	const float scale = float(length) / float(m_width);
+	const float iscale = 1.0f / scale;
+
+	const float width = k.width();
+	const int windowSize = k.windowSize();
+
+	const float * channel = this->channel(c);
+	const float * alpha = this->channel(a);
+
+	for (uint i = 0; i < length; i++)
+	{
+		const float center = (0.5f + i) * iscale;
+		
+		const int left = (int)floorf(center - width);
+		const int right = (int)ceilf(center + width);
+		nvDebugCheck(right - left <= windowSize);
+		
+		float norm = 0.0f;
+		float sum = 0;
+		for (int j = 0; j < windowSize; ++j)
+		{
+			const int idx = this->index(left + j, y, wm);
+
+			float w = k.valueAt(i, j) * (alpha[idx] + (1.0f / 256.0f));
+			norm += w;
+			sum += w * channel[idx];
+		}
+		
+		output[i] = sum / norm;
+	}
+}
+
+FloatImage* FloatImage::clone() const
+{
+	FloatImage* copy = new FloatImage();
+	copy->m_width = m_width;
+	copy->m_height = m_height;
+	copy->m_componentNum = m_componentNum;
+	copy->m_count = m_count;
+	
+	if(m_mem)
+	{
+		copy->allocate(m_componentNum, m_width, m_height);
+		memcpy(copy->m_mem, m_mem, m_count * sizeof(float));
+	}
+		
+	return copy;
+}
+

src/nvimage/FloatImage.h

@@ -3,14 +3,20 @@
 #ifndef NV_IMAGE_FLOATIMAGE_H
 #define NV_IMAGE_FLOATIMAGE_H
 
-#include <stdlib.h> // abs
+#include <nvimage/nvimage.h>
+
+#include <nvmath/Vector.h>
 
 #include <nvcore/Debug.h>
-#include <nvcore/Algorithms.h> // clamp
-#include <nvimage/nvimage.h>
+#include <nvcore/Containers.h> // clamp
+
+#include <stdlib.h> // abs
+
 
 namespace nv
 {
+class Vector4;
+class Matrix;
 class Image;
 class Filter;
 class Kernel1;
@@ -62,20 +68,22 @@ public:
 	NVIMAGE_API void toGamma(uint base_component, uint num, float gamma = 2.2f);
 	NVIMAGE_API void exponentiate(uint base_component, uint num, float power);
 	
-	
+
 	NVIMAGE_API FloatImage * fastDownSample() const;
 	NVIMAGE_API FloatImage * downSample(const Filter & filter, WrapMode wm) const;
-	NVIMAGE_API FloatImage * downSample(const Filter & filter, uint w, uint h, WrapMode wm) const;
+	NVIMAGE_API FloatImage * downSample(const Filter & filter, WrapMode wm, uint alpha) const;
+	NVIMAGE_API FloatImage * resize(const Filter & filter, uint w, uint h, WrapMode wm) const;
 
-	//NVIMAGE_API FloatImage * downSample(const Kernel1 & filter, WrapMode wm) const;
-	//NVIMAGE_API FloatImage * downSample(const Kernel1 & filter, uint w, uint h, WrapMode wm) const;
+	NVIMAGE_API FloatImage * resize(const Filter & filter, uint w, uint h, WrapMode wm, uint alpha) const;
 	//@}
 
-	NVIMAGE_API float applyKernel(const Kernel2 * k, int x, int y, int c, WrapMode wm) const;
-	NVIMAGE_API float applyKernelVertical(const Kernel1 * k, int x, int y, int c, WrapMode wm) const;
-	NVIMAGE_API float applyKernelHorizontal(const Kernel1 * k, int x, int y, int c, WrapMode wm) const;
-	NVIMAGE_API void applyKernelVertical(const PolyphaseKernel & k, int x, int c, WrapMode wm, float * output) const;
-	NVIMAGE_API void applyKernelHorizontal(const PolyphaseKernel & k, int y, int c, WrapMode wm, float * output) const;
+	NVIMAGE_API float applyKernel(const Kernel2 * k, int x, int y, uint c, WrapMode wm) const;
+	NVIMAGE_API float applyKernelVertical(const Kernel1 * k, int x, int y, uint c, WrapMode wm) const;
+	NVIMAGE_API float applyKernelHorizontal(const Kernel1 * k, int x, int y, uint c, WrapMode wm) const;
+	NVIMAGE_API void applyKernelVertical(const PolyphaseKernel & k, int x, uint c, WrapMode wm, float * output) const;
+	NVIMAGE_API void applyKernelHorizontal(const PolyphaseKernel & k, int y, uint c, WrapMode wm, float * output) const;
+	NVIMAGE_API void applyKernelVertical(const PolyphaseKernel & k, int x, uint c, uint a, WrapMode wm, float * output) const;
+	NVIMAGE_API void applyKernelHorizontal(const PolyphaseKernel & k, int y, uint c, uint a, WrapMode wm, float * output) const;
 	
 	
 	uint width() const { return m_width; }
@@ -111,6 +119,9 @@ public:
 	float sampleLinearMirror(float x, float y, int c) const;
 	//@}
 	
+	
+	FloatImage* clone() const;
+	
 public:
 	
 	uint index(uint x, uint y) const;
@@ -228,11 +239,15 @@ inline uint FloatImage::indexRepeat(int x, int y) const
 
 inline uint FloatImage::indexMirror(int x, int y) const
 {
+	if (m_width == 1) x = 0;
+
 	x = abs(x);
 	while (x >= m_width) {
 		x = abs(m_width + m_width - x - 2);
 	}
 
+	if (m_height == 1) y = 0;
+
 	y = abs(y);
 	while (y >= m_height) {
 		y = abs(m_height + m_height - y - 2);

src/nvimage/Image.cpp

@@ -2,7 +2,6 @@
 
 #include <nvcore/Debug.h>
 #include <nvcore/Ptr.h>
-#include <nvcore/Containers.h> // swap
 
 #include <nvmath/Color.h>
 
@@ -79,7 +78,7 @@ void Image::unwrap()
 
 void Image::free()
 {
-	::free(m_data);
+	nv::mem::free(m_data);
 	m_data = NULL;
 }
 

src/nvimage/ImageIO.cpp

@@ -162,9 +162,6 @@ FloatImage * nv::ImageIO::loadFloat(const char * fileName, Stream & s)
 	if (strCaseCmp(extension, ".pfm") == 0) {
 		return loadFloatPFM(fileName, s);
 	}
-	if (strCaseCmp(extension, ".hdr") == 0) {
-		return loadGridFloat(fileName, s);
-	}
 */
 
 	return NULL;
@@ -797,7 +794,7 @@ Image * nv::ImageIO::loadJPG(Stream & s)
 	// Read the entire file.
 	Array<uint8> byte_array;
 	byte_array.resize(s.size());
-	s.serialize(byte_array.mutableBuffer(), s.size());
+	s.serialize(byte_array.unsecureBuffer(), s.size());
 	
 	jpeg_decompress_struct cinfo;
 	jpeg_error_mgr jerr;
@@ -1092,8 +1089,7 @@ namespace
 		
 		virtual void seekg(Imf::Int64 pos)
 		{
-			nvDebugCheck(pos >= 0 && pos < UINT_MAX);
-			m_stream.seek((uint)pos);
+			m_stream.seek(pos);
 		}
 		
 		virtual void clear()
@@ -1302,77 +1298,6 @@ bool nv::ImageIO::saveFloatPFM(const char * fileName, const FloatImage * fimage,
 	return true;
 }
 
-#pragma warning(disable : 4996)
-
-NVIMAGE_API FloatImage * nv::ImageIO::loadGridFloat(const char * fileName, Stream & s)
-{
-	nvCheck(s.isLoading());
-	nvCheck(!s.isError());
-
-	Tokenizer parser(&s);
-
-	parser.nextLine();
-
-	if (parser.token() != "ncols")
-	{
-		nvDebug("Failed to find 'ncols' token in file '%s'.\n", fileName);
-		return NULL;
-	}
-
-	parser.nextToken(true);
-	const int nCols = parser.token().toInt(); 
-
-	parser.nextToken(true);
-	if (parser.token() != "nrows")
-	{
-		nvDebug("Failed to find 'nrows' token in file '%s'.\n", fileName);
-		return NULL;
-	}
-
-	parser.nextToken(true);
-	const int nRows = parser.token().toInt();
-
-	/* There's a byte order defined in the header.  We could read it.  However, here we 
-	   just assume that it matches the platform's byte order.
-	// There is then a bunch of data that we don't care about (lat, long definitions, etc).
-	for (int i=0; i!=9; ++i)
-		parser.nextToken(true);
-
-	if (parser.token() != "byteorder")
-		return NULL;
-
-	parser.nextToken(true);
-
-	const Stream::ByteOrder byteOrder = (parser.token() == "LSBFIRST")? Stream::LittleEndian: Stream::BigEndian;
-	*/
-
-	// GridFloat comes in two files: an ASCII header which was parsed above (.hdr) and a big blob
-	// of binary data in a .flt file.
-	Path dataPath(fileName);
-	dataPath.stripExtension();
-	dataPath.append(".flt");
-
-	// Open the binary data.
-	FILE* file = fopen(dataPath.fileName(), "rb");
-	if (!file)
-	{
-		nvDebug("Failed to find GridFloat blob file '%s' corresponding to '%s'.\n", dataPath.fileName(), fileName);
-		return NULL;
-	}
-
-	// Allocate image.
-	AutoPtr<FloatImage> fimage(new FloatImage());
-	fimage->allocate(1, nCols, nRows);
-
-	float * channel = fimage->channel(0);
-
-	// The binary blob is defined to be in row-major order, containing IEEE floats.
-	// So we can just slurp it in.  Theoretically, we ought to use the byte order.
-	const size_t nRead = fread((void*) channel, sizeof(float), nRows * nCols, file);
-	fclose(file);
-
-	return fimage.release();
-}
 #endif
 
 #if 0

src/nvimage/ImageIO.h

@@ -47,15 +47,10 @@ namespace nv
 		
 		NVIMAGE_API bool saveFloatEXR(const char * fileName, const FloatImage * fimage, uint base_component, uint num_components);
 #endif
-/*
-		NVIMAGE_API FloatImage * loadFloatPFM(const char * fileName, Stream & s);
-		NVIMAGE_API bool saveFloatPFM(const char * fileName, const FloatImage * fimage, uint base_component, uint num_components);
 
-		// GridFloat is a simple, open format for terrain elevation data.  See http://ned.usgs.gov/Ned/faq.asp.
-		// Expects: 1) fileName will be an ".hdr" header file, 2) there will also exist a corresponding float data
-		// blob in a ".flt" file.  (This is what USGS gives you.)
-		NVIMAGE_API FloatImage * loadGridFloat(const char * fileName, Stream & s);
-*/
+	//	NVIMAGE_API FloatImage * loadFloatPFM(const char * fileName, Stream & s);
+	//	NVIMAGE_API bool saveFloatPFM(const char * fileName, const FloatImage * fimage, uint base_component, uint num_components);
+
 	} // ImageIO namespace
 	
 } // nv namespace

src/nvimage/NormalMap.cpp

@@ -21,15 +21,16 @@
 // FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 // OTHER DEALINGS IN THE SOFTWARE.
 
-#include <nvcore/Ptr.h>
-
-#include <nvmath/Color.h>
-
 #include <nvimage/NormalMap.h>
 #include <nvimage/Filter.h>
 #include <nvimage/FloatImage.h>
 #include <nvimage/Image.h>
 
+#include <nvmath/Color.h>
+
+#include <nvcore/Ptr.h>
+
+
 using namespace nv;
 
 // Create normal map using the given kernels.

src/nvimage/PsdFile.h

@@ -39,7 +39,7 @@ namespace nv
 		bool isSupported() const
 		{
 			if (version != 1) {
-				printf("*** bad version number %u\n", version);
+				nvDebug("*** bad version number %u\n", version);
 				return false;
 			}
 			if (channel_count > 4) {

src/nvimage/Quantize.cpp

@@ -12,14 +12,14 @@ http://www.efg2.com/Lab/Library/ImageProcessing/DHALF.TXT
 @@ This code needs to be reviewed, I'm not sure it's correct.
 */
 
-#include <string.h> // memset
-
-#include <nvcore/Containers.h> // swap
+#include <nvimage/Quantize.h>
+#include <nvimage/Image.h>
+#include <nvimage/PixelFormat.h>
 
 #include <nvmath/Color.h>
 
-#include <nvimage/Image.h>
-#include <nvimage/Quantize.h>
+#include <nvcore/Containers.h> // swap
+
 
 using namespace nv;
 
@@ -51,94 +51,20 @@ void nv::Quantize::BinaryAlpha( Image * image, int alpha_threshold /*= 127*/ )
 // Simple quantization.
 void nv::Quantize::RGB16( Image * image )
 {
-	nvCheck(image != NULL);
-	
-	const uint w = image->width();
-	const uint h = image->height();
-	
-	for(uint y = 0; y < h; y++) {
-		for(uint x = 0; x < w; x++) {
-			
-			Color32 pixel32 = image->pixel(x, y);
-			
-			// Convert to 16 bit and back to 32 using regular bit expansion.
-			Color32 pixel16 = toColor32( toColor16(pixel32) );
-			
-			// Store color.
-			image->pixel(x, y) = pixel16;
-		}
-	}
+	Truncate(image, 5, 6, 5, 8);
 }
 
 // Alpha quantization.
 void nv::Quantize::Alpha4( Image * image )
 {
-	nvCheck(image != NULL);
-	
-	const uint w = image->width();
-	const uint h = image->height();
-	
-	for(uint y = 0; y < h; y++) {
-		for(uint x = 0; x < w; x++) {
-			
-			Color32 pixel = image->pixel(x, y);
-			
-			// Convert to 4 bit using regular bit expansion.
-			pixel.a = (pixel.a & 0xF0) | ((pixel.a & 0xF0) >> 4);
-			
-			// Store color.
-			image->pixel(x, y) = pixel;
-		}
-	}
+	Truncate(image, 8, 8, 8, 4);
 }
 
 
 // Error diffusion. Floyd Steinberg.
 void nv::Quantize::FloydSteinberg_RGB16( Image * image )
 {
-	nvCheck(image != NULL);
-	
-	const uint w = image->width();
-	const uint h = image->height();
-	
-	// @@ Use fixed point?
-	Vector3 * row0 = new Vector3[w+2];
-	Vector3 * row1 = new Vector3[w+2];
-	memset(row0, 0, sizeof(Vector3)*(w+2));
-	memset(row1, 0, sizeof(Vector3)*(w+2));
-	
-	for(uint y = 0; y < h; y++) {
-		for(uint x = 0; x < w; x++) {
-			
-			Color32 pixel32 = image->pixel(x, y);
-			
-			// Add error.	// @@ We shouldn't clamp here!
-			pixel32.r = clamp(int(pixel32.r) + int(row0[1+x].x()), 0, 255);
-			pixel32.g = clamp(int(pixel32.g) + int(row0[1+x].y()), 0, 255);
-			pixel32.b = clamp(int(pixel32.b) + int(row0[1+x].z()), 0, 255);
-			
-			// Convert to 16 bit. @@ Use regular clamp?
-			Color32 pixel16 = toColor32( toColor16(pixel32) );
-			
-			// Store color.
-			image->pixel(x, y) = pixel16;
-			
-			// Compute new error.
-			Vector3 diff(float(pixel32.r - pixel16.r), float(pixel32.g - pixel16.g), float(pixel32.b - pixel16.b));
-			
-			// Propagate new error.
-			row0[1+x+1] += 7.0f / 16.0f * diff;
-			row1[1+x-1] += 3.0f / 16.0f * diff;
-			row1[1+x+0] += 5.0f / 16.0f * diff;
-			row1[1+x+1] += 1.0f / 16.0f * diff;
-		}
-		
-		swap(row0, row1);
-		memset(row1, 0, sizeof(Vector3)*(w+2));
-	}
-	
-	delete [] row0;
-	delete [] row1;
+	FloydSteinberg(image, 5, 6, 5, 8);
 }
 
 
@@ -192,34 +118,90 @@ void nv::Quantize::FloydSteinberg_BinaryAlpha( Image * image, int alpha_threshol
 
 // Error diffusion. Floyd Steinberg.
 void nv::Quantize::FloydSteinberg_Alpha4( Image * image )
+{
+	FloydSteinberg(image, 8, 8, 8, 4);
+}
+
+
+void nv::Quantize::Truncate(Image * image, uint rsize, uint gsize, uint bsize, uint asize)
 {
 	nvCheck(image != NULL);
 	
 	const uint w = image->width();
 	const uint h = image->height();
 	
-	// @@ Use fixed point?
-	float * row0 = new float[(w+2)];
-	float * row1 = new float[(w+2)];
-	memset(row0, 0, sizeof(float)*(w+2));
-	memset(row1, 0, sizeof(float)*(w+2));
-	
 	for(uint y = 0; y < h; y++) {
 		for(uint x = 0; x < w; x++) {
 			
 			Color32 pixel = image->pixel(x, y);
-			
-			// Add error.
-			int alpha = int(pixel.a) + int(row0[1+x]);
-			
-			// Convert to 4 bit using regular bit expansion.
-			pixel.a = (pixel.a & 0xF0) | ((pixel.a & 0xF0) >> 4);
-			
+
+			// Convert to our desired size, and reconstruct.
+			pixel.r = PixelFormat::convert(pixel.r, 8, rsize);
+			pixel.r = PixelFormat::convert(pixel.r, rsize, 8);
+
+			pixel.g = PixelFormat::convert(pixel.g, 8, gsize);
+			pixel.g = PixelFormat::convert(pixel.g, gsize, 8);
+
+			pixel.b = PixelFormat::convert(pixel.b, 8, bsize);
+			pixel.b = PixelFormat::convert(pixel.b, bsize, 8);
+
+			pixel.a = PixelFormat::convert(pixel.a, 8, asize);
+			pixel.a = PixelFormat::convert(pixel.a, asize, 8);
+
 			// Store color.
 			image->pixel(x, y) = pixel;
+		}
+	}
+}
+
+
+// Error diffusion. Floyd Steinberg.
+void nv::Quantize::FloydSteinberg(Image * image, uint rsize, uint gsize, uint bsize, uint asize)
+{
+	nvCheck(image != NULL);
+	
+	const uint w = image->width();
+	const uint h = image->height();
+	
+	Vector4 * row0 = new Vector4[w+2];
+	Vector4 * row1 = new Vector4[w+2];
+	memset(row0, 0, sizeof(Vector4)*(w+2));
+	memset(row1, 0, sizeof(Vector4)*(w+2));
+	
+	for (uint y = 0; y < h; y++) {
+		for (uint x = 0; x < w; x++) {
+			
+			Color32 pixel = image->pixel(x, y);
+
+			// Add error.
+			pixel.r = clamp(int(pixel.r) + int(row0[1+x].x()), 0, 255);
+			pixel.g = clamp(int(pixel.g) + int(row0[1+x].y()), 0, 255);
+			pixel.b = clamp(int(pixel.b) + int(row0[1+x].z()), 0, 255);
+			pixel.a = clamp(int(pixel.a) + int(row0[1+x].w()), 0, 255);
+			
+			int r = pixel.r;
+			int g = pixel.g;
+			int b = pixel.b;
+			int a = pixel.a;
+
+			// Convert to our desired size, and reconstruct.
+			r = PixelFormat::convert(r, 8, rsize);
+			r = PixelFormat::convert(r, rsize, 8);
+
+			g = PixelFormat::convert(g, 8, gsize);
+			g = PixelFormat::convert(g, gsize, 8);
+
+			b = PixelFormat::convert(b, 8, bsize);
+			b = PixelFormat::convert(b, bsize, 8);
+
+			a = PixelFormat::convert(a, 8, asize);
+			a = PixelFormat::convert(a, asize, 8);
+
+			// Store color.
+			image->pixel(x, y) = Color32(r, g, b, a);
 			
 			// Compute new error.
-			float diff = float(alpha - pixel.a);
+			Vector4 diff(float(int(pixel.r) - r), float(int(pixel.g) - g), float(int(pixel.b) - b), float(int(pixel.a) - a));
 			
 			// Propagate new error.
 			row0[1+x+1] += 7.0f / 16.0f * diff;
@@ -229,10 +211,9 @@ void nv::Quantize::FloydSteinberg_Alpha4( Image * image )
 		}
 		
 		swap(row0, row1);
-		memset(row1, 0, sizeof(float)*(w+2));
+		memset(row1, 0, sizeof(Vector4)*(w+2));
 	}
 	
 	delete [] row0;
 	delete [] row1;
 }
-

src/nvimage/Quantize.h

@@ -3,6 +3,9 @@
 #ifndef NV_IMAGE_QUANTIZE_H
 #define NV_IMAGE_QUANTIZE_H
 
+#include <nvimage/nvimage.h>
+
+
 namespace nv
 {
 	class Image;
@@ -17,6 +20,9 @@ namespace nv
 		void FloydSteinberg_BinaryAlpha(Image * img, int alpha_threshold = 127);
 		void FloydSteinberg_Alpha4(Image * img);
 
+		void Truncate(Image * image, uint rsize, uint gsize, uint bsize, uint asize);
+		void FloydSteinberg(Image * image, uint rsize, uint gsize, uint bsize, uint asize);
+
 		// @@ Add palette quantization algorithms!
 	}
 }

src/nvmath/Basis.cpp

@@ -49,7 +49,7 @@ void Basis::robustOrthonormalize(float epsilon /*= NV_EPSILON*/)
 			return;
 		}
 	}
-	normal = nv::normalize(normal, epsilon);
+	normal = ::normalize(normal, epsilon);
 	
 	tangent -= normal * dot(normal, tangent);
 	bitangent -= normal * dot(normal, bitangent);
@@ -68,8 +68,7 @@ void Basis::robustOrthonormalize(float epsilon /*= NV_EPSILON*/)
 	}
 	else
 	{
-#if 0
-		tangent = nv::normalize(tangent, epsilon);
+		tangent = ::normalize(tangent, epsilon);
 		bitangent -= tangent * dot(tangent, bitangent);
 		
 		if (length(bitangent) < epsilon)
@@ -79,47 +78,11 @@ void Basis::robustOrthonormalize(float epsilon /*= NV_EPSILON*/)
 		}
 		else
 		{
-			bitangent = nv::normalize(bitangent, epsilon);
+			tangent = ::normalize(tangent, epsilon);
 		}
-#else
-		if (length(bitangent) < epsilon)
-		{
-			bitangent = cross(tangent, normal);
-			nvCheck(isNormalized(bitangent));
-		}
-		else
-		{
-			tangent = nv::normalize(tangent);
-			bitangent = nv::normalize(bitangent);
-			
-			Vector3 bisector = nv::normalize(tangent + bitangent);
-			Vector3 axis = cross(bisector, normal);
-			
-			nvDebugCheck(isNormalized(axis, epsilon));
-			nvDebugCheck(equal(dot(axis, tangent), -dot(axis, bitangent), epsilon));
-			
-			if (dot(axis, tangent) > 0)
-			{
-				tangent = nv::normalize(bisector + axis);
-				bitangent = nv::normalize(bisector - axis);
-			}
-			else
-			{
-				tangent = nv::normalize(bisector - axis);
-				bitangent = nv::normalize(bisector + axis);
-			}
-		}
-#endif
 	}
 	
-	/*// Check vector lengths.
-	if (!isNormalized(normal, epsilon))
-	{
-		nvDebug("%f %f %f\n", normal.x(), normal.y(), normal.z());
-		nvDebug("%f %f %f\n", tangent.x(), tangent.y(), tangent.z());
-		nvDebug("%f %f %f\n", bitangent.x(), bitangent.y(), bitangent.z());
-	}*/
-	
+	// Check vector lengths.
 	nvCheck(isNormalized(normal, epsilon));
 	nvCheck(isNormalized(tangent, epsilon));
 	nvCheck(isNormalized(bitangent, epsilon));
@@ -162,18 +125,9 @@ void Basis::buildFrameForDirection(Vector3::Arg d)
 	bitangent = cross(normal, tangent);
 }
 
-bool Basis::isValid() const
-{
-	if (equal(normal, Vector3(zero))) return false;
-	if (equal(tangent, Vector3(zero))) return false;
-	if (equal(bitangent, Vector3(zero))) return false;
-
-	if (equal(determinant(), 0.0f)) return false;
-	
-	return true;
-}
 
 
+/*
 /// Transform by this basis. (From this basis to object space).
 Vector3 Basis::transform(Vector3::Arg v) const
 {
@@ -190,31 +144,30 @@ Vector3 Basis::transformT(Vector3::Arg v)
 }
 
 /// Transform by the inverse. (From object space to this basis).
-/// @note Uses Cramer's rule so the inverse is not accurate if the basis is ill-conditioned.
+/// @note Uses Kramer's rule so the inverse is not accurate if the basis is ill-conditioned.
 Vector3 Basis::transformI(Vector3::Arg v) const
 {
 	const float det = determinant();
-	nvDebugCheck(!equal(det, 0.0f, 0.0f));
+	nvCheck(!equalf(det, 0.0f));
 	
 	const float idet = 1.0f / det;
 
 	// Rows of the inverse matrix.
-	Vector3 r0(
-		 (bitangent.y() * normal.z() - bitangent.z() * normal.y()),
-		-(bitangent.x() * normal.z() - bitangent.z() * normal.x()),
-		 (bitangent.x() * normal.y() - bitangent.y() * normal.x()));
+	Vector3 r0, r1, r2;
+	r0.x =  (bitangent.y() * normal.z() - bitangent.z() * normal.y()) * idet;
+	r0.y = -(bitangent.x() * normal.z() - bitangent.z() * normal.x()) * idet;
+	r0.z =  (bitangent.x() * normal.y() - bitangent.y() * normal.x()) * idet;
 
-	Vector3 r1(
-		-(tangent.y() * normal.z() - tangent.z() * normal.y()),
-		 (tangent.x() * normal.z() - tangent.z() * normal.x()),
-		-(tangent.x() * normal.y() - tangent.y() * normal.x()));
+	r1.x = -(tangent.y() * normal.z() - tangent.z() * normal.y()) * idet;
+	r1.y =  (tangent.x() * normal.z() - tangent.z() * normal.x()) * idet;
+	r1.z = -(tangent.x() * normal.y() - tangent.y() * normal.x()) * idet;
 
-	Vector3 r2(
-		 (tangent.y() * bitangent.z() - tangent.z() * bitangent.y()),
-		-(tangent.x() * bitangent.z() - tangent.z() * bitangent.x()),
-		 (tangent.x() * bitangent.y() - tangent.y() * bitangent.x()));
+	r2.x =  (tangent.y() * bitangent.z() - tangent.z() * bitangent.y()) * idet;
+	r2.y = -(tangent.x() * bitangent.z() - tangent.z() * bitangent.x()) * idet;
+	r2.z =  (tangent.x() * bitangent.y() - tangent.y() * bitangent.x()) * idet;
 
-	return Vector3(dot(v, r0), dot(v, r1), dot(v, r2)) * idet;
+	return Vector3(dot(v, r0), dot(v, r1), dot(v, r2));
 }	
+*/
 
 

src/nvmath/Basis.h

@@ -54,8 +54,7 @@ namespace nv
 				tangent.z() * bitangent.x() * normal.y() - tangent.x() * bitangent.z() * normal.y();
 		}
 		
-		bool isValid() const;
-		
+		/*
 		// Get transform matrix for this basis.
 		NVMATH_API Matrix matrix() const;
 		
@@ -67,7 +66,7 @@ namespace nv
 
 		// Transform by the inverse. (From object space to this basis).
 		NVMATH_API Vector3 transformI(Vector3::Arg v) const;
-
+		*/
 		
 		Vector3 tangent;
 		Vector3 bitangent;

src/nvmath/Box.h

@@ -9,7 +9,6 @@
 
 namespace nv
 {
-class Stream;
 
 /// Axis Aligned Bounding Box.
 class Box
@@ -28,13 +27,11 @@ public:
 	// Cast operators.
 	operator const float * () const { return reinterpret_cast<const float *>(this); }
 
-	// Min corner of the box.
-	Vector3 minCorner() const { return m_mins; }
-	Vector3 & minCorner() { return m_mins; }
+	/// Min corner of the box.
+	Vector3 mins() const { return m_mins; }
 
-	// Max corner of the box.
-	Vector3 maxCorner() const { return m_maxs; }
-	Vector3 & maxCorner() { return m_maxs; }
+	/// Max corner of the box.
+	Vector3 maxs() const { return m_maxs; }
 
 	/// Clear the bounds.
 	void clearBounds()
@@ -129,8 +126,6 @@ public:
 			m_maxs.x() > p.x() && m_maxs.y() > p.y() && m_maxs.z() > p.z();
 	}
 
-	friend Stream & operator<< (Stream & s, Box & box);
-
 private:
 
 	Vector3 m_mins;

src/nvmath/CMakeLists.txt

@@ -5,19 +5,13 @@ SET(MATH_SRCS
 	Vector.h
 	Matrix.h
 	Quaternion.h
-	Plane.h Plane.cpp
 	Box.h
 	Color.h
 	Montecarlo.h Montecarlo.cpp
 	Random.h Random.cpp
 	SphericalHarmonic.h SphericalHarmonic.cpp
 	Basis.h Basis.cpp
-	Triangle.h Triangle.cpp TriBox.cpp
-	Polygon.h Polygon.cpp
-	TypeSerialization.h TypeSerialization.cpp
-	Sparse.h Sparse.cpp
-	Solver.h Solver.cpp
-	KahanSum.h)
+	Triangle.h Triangle.cpp TriBox.cpp)
 
 INCLUDE_DIRECTORIES(${CMAKE_CURRENT_SOURCE_DIR})
 

src/nvmath/KahanSum.h

@@ -1,38 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#ifndef NV_MATH_KAHANSUM_H
-#define NV_MATH_KAHANSUM_H
-
-#include <nvmath/nvmath.h>
-
-namespace nv
-{
-
-	class KahanSum
-	{
-	public:
-		KahanSum() : accum(0.0f), err(0) {};
-		
-		void add(float f)
-		{
-			float compensated = f + err;
-			float tmp = accum + compensated;
-			err = accum - tmp;
-			err += compensated;
-			accum = tmp;
-		}
-		
-		float sum() const
-		{
-			return accum;
-		}
-		
-	private:
-		float accum;
-		float err;
-	};
-	
-} // nv namespace
-
-
-#endif // NV_MATH_KAHANSUM_H

src/nvmath/Matrix.h

@@ -332,7 +332,7 @@ inline Matrix transpose(Matrix::Arg m)
 	Matrix r;
 	for (int i = 0; i < 4; i++)
 	{
-		for (int j = 0; j < 4; i++)
+		for (int j = 0; j < 4; j++)
 		{
 			r(i, j) = m(j, i);
 		}

src/nvmath/Polygon.cpp

@@ -1,168 +0,0 @@
-// This code is in the public domain -- Ignacio Casta<74>o <castanyo@yahoo.es>
-
-#include <nvmath/Polygon.h>
-
-#include <nvmath/Triangle.h>
-#include <nvmath/Plane.h>
-
-using namespace nv;
-
-
-Polygon::Polygon()
-{
-}
-
-Polygon::Polygon(const Triangle & t)
-{
-	pointArray.resize(3);
-	pointArray[0] = t.v[0];
-	pointArray[1] = t.v[1];
-	pointArray[2] = t.v[2];
-}
-
-Polygon::Polygon(const Vector3 * points, uint vertexCount)
-{
-	pointArray.resize(vertexCount);
-	
-	for (uint i = 0; i < vertexCount; i++)
-	{
-		pointArray[i] = points[i];
-	}
-}
-
-
-/// Compute polygon area.
-float Polygon::area() const
-{
-	float total = 0;
-	
-	const uint pointCount = pointArray.count();
-    for (uint i = 2; i < pointCount; i++)
-    {
-		Vector3 v1 = pointArray[i-1] - pointArray[0];
-		Vector3 v2 = pointArray[i] - pointArray[0];
-		
-		total += 0.5f * length(cross(v1, v2));
-    }
-	
-    return total;	
-}
-
-/// Get the bounds of the polygon.
-Box Polygon::bounds() const
-{
-	Box bounds;
-	bounds.clearBounds();
-	foreach(p, pointArray)
-	{
-		bounds.addPointToBounds(pointArray[p]);
-	}
-	return bounds;
-}
-
-
-/// Get the plane of the polygon.
-Plane Polygon::plane() const
-{
-	// @@ Do something better than this?
-	Vector3 n = cross(pointArray[1] - pointArray[0], pointArray[2] - pointArray[0]);
-	return Vector4(n, dot(n, pointArray[0]));
-}
-
-
-/// Clip polygon to box.
-uint Polygon::clipTo(const Box & box)
-{
-	const Plane posX( 1, 0, 0, box.maxCorner().x());
-	const Plane negX(-1, 0, 0,-box.minCorner().x());
-	const Plane posY( 0, 1, 0, box.maxCorner().y());
-	const Plane negY( 0,-1, 0,-box.minCorner().y());
-	const Plane posZ( 0, 0, 1, box.maxCorner().z());
-	const Plane negZ( 0, 0,-1,-box.minCorner().z());
-
-	if (clipTo(posX) == 0) return 0;
-	if (clipTo(negX) == 0) return 0;
-	if (clipTo(posY) == 0) return 0;
-	if (clipTo(negY) == 0) return 0;
-	if (clipTo(posZ) == 0) return 0;
-	if (clipTo(negZ) == 0) return 0;
-	
-	return pointArray.count();
-}
-
-
-/// Clip polygon to plane.
-uint Polygon::clipTo(const Plane & plane)
-{
-	int count = 0;
-
-	const uint pointCount = pointArray.count();
-	
-	Array<Vector3> newPointArray(pointCount + 1);	// @@ Do not create copy every time.
-	
-	Vector3 prevPoint = pointArray[pointCount - 1];
-	float prevDist = dot(plane.vector(), prevPoint) - plane.offset();
-	
-	for (uint i = 0; i < pointCount; i++)
-	{
-		const Vector3 point = pointArray[i];
-		float dist = dot(plane.vector(), point) - plane.offset();
-		
-		// @@ Handle points on plane better.
-		
-		if (dist <= 0) // interior.
-		{
-			if (prevDist > 0) // exterior
-			{
-				// Add segment intersection point.
-				Vector3 dp = point - prevPoint;
-				
-				float t = dist / prevDist;
-				newPointArray.append(point - dp * t);
-			}
-			
-			// Add interior point.
-			newPointArray.append(point);
-		}
-		else if (dist > 0 && prevDist < 0)
-		{
-			// Add segment intersection point.
-			Vector3 dp = point - prevPoint;
-			
-			float t = dist / prevDist;
-			newPointArray.append(point - dp * t);
-		}
-		
-		prevPoint = point;
-		prevDist = dist;
-	}
-
-	swap(pointArray, newPointArray);
-
-	return count;
-}
-
-
-void Polygon::removeColinearPoints()
-{
-	const uint pointCount = pointArray.count();
-	
-	Array<Vector3> newPointArray(pointCount);
-	
-    for (uint i = 0 ; i < pointCount; i++)
-    {
-        int j = (i + 1) % pointCount;
-        int k = (i + pointCount - 1) % pointCount;
-		
-		Vector3 v1 = normalize(pointArray[j] - pointArray[i]);
-		Vector3 v2 = normalize(pointArray[i] - pointArray[k]);
-		
-        if (dot(v1, v2) < 0.999)
-        {
-			newPointArray.append(pointArray[i]);
-        }
-    }
-
-	swap(pointArray, newPointArray);
-}
-

src/nvmath/Polygon.h

@@ -1,45 +0,0 @@
-// This code is in the public domain -- Ignacio Casta<74>o <castanyo@yahoo.es>
-
-#ifndef NV_MATH_POLYGON_H
-#define NV_MATH_POLYGON_H
-
-#include <nvcore/Containers.h>
-
-#include <nvmath/nvmath.h>
-#include <nvmath/Vector.h>
-#include <nvmath/Box.h>
-
-namespace nv
-{
-	class Box;
-	class Plane;
-	class Triangle;
-	
-
-	class Polygon
-	{
-		NV_FORBID_COPY(Polygon);
-	public:
-
-		Polygon();
-		Polygon(const Triangle & t);
-		Polygon(const Vector3 * points, uint vertexCount);
-
-		float area() const;
-		Box bounds() const;
-		Plane plane() const;
-
-		uint clipTo(const Box & box);
-		uint clipTo(const Plane & plane);
-
-		void removeColinearPoints();
-		
-	private:
-
-		Array<Vector3> pointArray;
-	};
-
-
-} // nv namespace
-
-#endif	// NV_MATH_POLYGON_H

src/nvmath/Quaternion.h

@@ -51,6 +51,7 @@ namespace nv
 
 	inline Quaternion mul(Quaternion::Arg a, Quaternion::Arg b)
 	{
+		// @@ Efficient SIMD implementation?
 		return Quaternion(
 			+ a.x() * b.w() + a.y()*b.z() - a.z()*b.y() + a.w()*b.x(),
 			- a.x() * b.z() + a.y()*b.w() + a.z()*b.x() + a.w()*b.y(),
@@ -58,40 +59,6 @@ namespace nv
 			- a.x() * b.x() - a.y()*b.y() - a.z()*b.z() + a.w()*b.w());
 	}
 
-	inline Quaternion mul(Quaternion::Arg a, Vector3::Arg b)
-	{
-		return Quaternion(
-			                + a.y()*b.z() - a.z()*b.y() + a.w()*b.x(),
-			- a.x() * b.z()               + a.z()*b.x() + a.w()*b.y(),
-			+ a.x() * b.y() - a.y()*b.x()               + a.w()*b.z(),
-			- a.x() * b.x() - a.y()*b.y() - a.z()*b.z() );
-	}
-
-	inline Quaternion mul(Vector3::Arg a, Quaternion::Arg b)
-	{
-		return Quaternion(
-			+ a.x() * b.w() + a.y()*b.z() - a.z()*b.y(),
-			- a.x() * b.z() + a.y()*b.w() + a.z()*b.x(),
-			+ a.x() * b.y() - a.y()*b.x() + a.z()*b.w(),
-			- a.x() * b.x() - a.y()*b.y() - a.z()*b.z());
-	}
-
-	inline Quaternion operator *(Quaternion::Arg a, Quaternion::Arg b)
-	{
-		return mul(a, b);
-	}
-
-	inline Quaternion operator *(Quaternion::Arg a, Vector3::Arg b)
-	{
-		return mul(a, b);
-	}
-
-	inline Quaternion operator *(Vector3::Arg a, Quaternion::Arg b)
-	{
-		return mul(a, b);
-	}
-
-
 	inline Quaternion scale(Quaternion::Arg q, float s)
 	{
 		return scale(q.asVector(), s);
@@ -155,24 +122,6 @@ namespace nv
 		return Quaternion(Vector4(v * s, c));
 	}
 
-	inline Vector3 imag(Quaternion::Arg q)
-	{
-		return q.asVector().xyz();
-	}
-
-	inline float real(Quaternion::Arg q)
-	{
-		return q.w();
-	}
-
-
-	/// Transform vector.
-	inline Vector3 transform(Quaternion::Arg q, Vector3::Arg v)
-	{
-		Quaternion t = q * v * conjugate(q);
-		return imag(t);
-	}
-
 
 } // nv namespace
 

src/nvmath/Solver.cpp

@@ -1,726 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#include <nvmath/Solver.h>
-
-using namespace nv;
-
-namespace
-{
-	class Preconditioner
-	{
-	public:
-		// Virtual dtor.
-		virtual ~Preconditioner() { }
-	 
-		// Apply preconditioning step.
-		virtual void apply(const FullVector & x, FullVector & y) const = 0;
-	};
-
-
-	// Jacobi preconditioner.
-	class JacobiPreconditioner : public Preconditioner
-	{
-	public:
-
-		JacobiPreconditioner(const SparseMatrix & M, bool symmetric) : m_inverseDiagonal(M.width())
-		{
-			nvCheck(M.isSquare());
-			
-			for(uint x = 0; x < M.width(); x++)
-			{
-				float elem = M.getCoefficient(x, x);
-				nvDebugCheck( elem != 0.0f );
-
-				if (symmetric) 
-				{
-					m_inverseDiagonal[x] = 1.0f / sqrt(fabs(elem));
-				}
-				else 
-				{
-					m_inverseDiagonal[x] = 1.0f / elem;
-				}
-			}
-		}
-
-		void apply(const FullVector & x, FullVector & y) const
-		{
-			y *= x;
-		}
-
-	private:
-
-		FullVector m_inverseDiagonal;
-
-	};
-
-} // namespace
-
-
-static int ConjugateGradientSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon);
-static int ConjugateGradientSolver(const Preconditioner & preconditioner, const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon);
-
-
-// Solve the symmetric system: At<41>A<EFBFBD>x = At<41>b
-void nv::LeastSquaresSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon/*1e-5f*/)
-{
-	nvDebugCheck(A.width() == x.dimension());
-	nvDebugCheck(A.height() == b.dimension());
-	nvDebugCheck(A.height() >= A.width()); // @@ If height == width we could solve it directly...
-
-	const uint D = A.width();
-
-	FullVector Atb(D);
-	mult(Transposed, A, b, Atb);
-
-	SparseMatrix AtA(D);
-	mult(Transposed, A, NoTransposed, A, AtA);
-
-	SymmetricSolver(AtA, Atb, x, epsilon);
-}
-
-
-// See section 10.4.3 in: Mesh Parameterization: Theory and Practice, Siggraph Course Notes, August 2007
-void nv::LeastSquaresSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, const uint * lockedParameters, uint lockedCount, float epsilon/*= 1e-5f*/)
-{
-	nvDebugCheck(A.width() == x.dimension());
-	nvDebugCheck(A.height() == b.dimension());
-	nvDebugCheck(A.height() >= A.width() - lockedCount);
-
-	// @@ This is not the most efficient way of building a system with reduced degrees of freedom. It would be faster to do it on the fly.
-
-	const uint D = A.width() - lockedCount;
-	nvDebugCheck(D > 0);
-
-	// Compute: b - Al * xl
-	FullVector b_Alxl(b);
-
-	for (uint y = 0; y < A.height(); y++)
-	{
-		const uint count = A.getRow(y).count();
-		for (uint e = 0; e < count; e++)
-		{
-			uint column = A.getRow(y)[e].x;
-
-			bool isFree = true;
-			for (uint i = 0; i < lockedCount; i++) 
-			{
-				isFree &= (lockedParameters[i] != column);
-			}
-
-			if (!isFree)
-			{
-				b_Alxl[y] -= x[column] * A.getRow(y)[e].v;
-			}
-		}
-	}
-
-	// Remove locked columns from A.
-	SparseMatrix Af(D, A.height());
-
-	for (uint y = 0; y < A.height(); y++)
-	{
-		const uint count = A.getRow(y).count();
-		for (uint e = 0; e < count; e++)
-		{
-			uint column = A.getRow(y)[e].x;
-			uint ix = column;
-
-			bool isFree = true;
-			for (uint i = 0; i < lockedCount; i++) 
-			{
-				isFree &= (lockedParameters[i] != column);
-				if (column > lockedParameters[i]) ix--; // shift columns
-			}
-
-			if (isFree)
-			{
-				Af.setCoefficient(ix, y, A.getRow(y)[e].v);
-			}
-		}
-	}
-	
-	// Remove elements from x
-	FullVector xf(D);
-
-	for (uint i = 0, j = 0; i < A.width(); i++)
-	{
-		bool isFree = true;
-		for (uint l = 0; l < lockedCount; l++) 
-		{
-			isFree &= (lockedParameters[l] != i);
-		}
-
-		if (isFree)
-		{
-			xf[j++] = x[i];
-		}
-	}
-
-	// Solve reduced system.
-	LeastSquaresSolver(Af, b_Alxl, xf, epsilon);
-
-	// Copy results back to x.
-	for (uint i = 0, j = 0; i < A.width(); i++)
-	{
-		bool isFree = true;
-		for (uint l = 0; l < lockedCount; l++) 
-		{
-			isFree &= (lockedParameters[l] != i);
-		}
-
-		if (isFree)
-		{
-			x[i] = xf[j++];
-		}
-	}
-}
-
-
-void nv::SymmetricSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon/*1e-5f*/)
-{
-	nvDebugCheck(A.height() == A.width());
-	nvDebugCheck(A.height() == b.dimension());
-	nvDebugCheck(b.dimension() == x.dimension());
-
-//	JacobiPreconditioner jacobi(A, true);
-
-//	ConjugateGradientSolver(jacobi, A, b, x, epsilon);
-	ConjugateGradientSolver(A, b, x, epsilon);
-}
-
-
-/**
- * Compute the solution of the sparse linear system Ab=x using the Conjugate
- * Gradient method.
- *
- * Solving sparse linear systems:
- * (1)		A<>x = b
- * 
- * The conjugate gradient algorithm solves (1) only in the case that A is 
- * symmetric and positive definite. It is based on the idea of minimizing the 
- * function
- * 
- * (2)		f(x) = 1/2<>x<EFBFBD>A<EFBFBD>x - b<>x
- * 
- * This function is minimized when its gradient
- * 
- * (3)		df = A<>x - b
- * 
- * is zero, which is equivalent to (1). The minimization is carried out by 
- * generating a succession of search directions p.k and improved minimizers x.k. 
- * At each stage a quantity alfa.k is found that minimizes f(x.k + alfa.k<>p.k), 
- * and x.k+1 is set equal to the new point x.k + alfa.k<>p.k. The p.k and x.k are 
- * built up in such a way that x.k+1 is also the minimizer of f over the whole
- * vector space of directions already taken, {p.1, p.2, . . . , p.k}. After N 
- * iterations you arrive at the minimizer over the entire vector space, i.e., the 
- * solution to (1).
- *
- * For a really good explanation of the method see:
- *
- * "An Introduction to the Conjugate Gradient Method Without the Agonizing Pain",
- * Jonhathan Richard Shewchuk.
- *
-**/
-/*static*/ int ConjugateGradientSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon)
-{
-	nvDebugCheck( A.isSquare() );
-	nvDebugCheck( A.width() == b.dimension() );
-	nvDebugCheck( A.width() == x.dimension() );
-
-	int i = 0;
-	const int D = A.width();
-	const int i_max = 4 * D;   // Convergence should be linear, but in some cases, it's not.
-
-	FullVector r(D);   // residual
-	FullVector p(D);   // search direction
-	FullVector q(D);   // 
-	float delta_0;
-	float delta_old;
-	float delta_new;
-	float alpha;
-	float beta;
-
-	// r = b - A<>x;
-	copy(b, r);
-	sgemv(-1, A, x, 1, r);
-
-	// p = r;
-	copy(r, p);
-	
-	delta_new = dot( r, r );
-	delta_0 = delta_new;
-
-	while (i < i_max && delta_new > epsilon*epsilon*delta_0)
-	{
-		i++;
-
-		// q = A<>p
-		mult(A, p, q);
-
-		// alpha = delta_new / p<>q
-		alpha = delta_new / dot( p, q );
-
-		// x = alfa<66>p + x
-		saxpy(alpha, p, x);
-
-		if ((i & 31) == 0) // recompute r after 32 steps
-		{
-			// r = b - A<>x
-			copy(b, r);
-			sgemv(-1, A, x, 1, r);
-		}
-		else
-		{
-			// r = r - alpha<68>q
-			saxpy(-alpha, q, r);
-		}
-
-		delta_old = delta_new;
-		delta_new = dot( r, r );
-
-		beta = delta_new / delta_old;
-
-		// p = r + beta<74>p
-		copy(r, p);
-		saxpy(beta, p, r);
-	}
-
-	return i;
-}
-
-
-// Conjugate gradient with preconditioner.
-/*static*/ int ConjugateGradientSolver(const Preconditioner & preconditioner, const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon)
-{
-	nvDebugCheck( A.isSquare() );
-	nvDebugCheck( A.width() == b.dimension() );
-	nvDebugCheck( A.width() == x.dimension() );
-
-	int i = 0;
-	const int D = A.width();
-	const int i_max = 4 * D;   // Convergence should be linear, but in some cases, it's not.
-
-	FullVector r(D);    // residual
-	FullVector p(D);    // search direction
-	FullVector q(D);    // 
-	FullVector s(D);    // preconditioned
-	float delta_0;
-	float delta_old;
-	float delta_new;
-	float alpha;
-	float beta;
-
-	// r = b - A<>x
-	copy(b, r);
-	sgemv(-1, A, x, 1, r);
-
-
-	// p = M^-1 <20> r
-	preconditioner.apply(r, p);
-	//copy(r, p);
-
-
-	delta_new = dot(r, p);
-	delta_0 = delta_new;
-
-	while (i < i_max && delta_new > epsilon*epsilon*delta_0)
-	{
-		i++;
-
-		// q = A<>p
-		mult(A, p, q);
-
-		// alpha = delta_new / p<>q
-		alpha = delta_new / dot(p, q);
-
-		// x = alfa<66>p + x
-		saxpy(alpha, p, x);
-
-		if ((i & 31) == 0)  // recompute r after 32 steps
-		{			
-			// r = b - A<>x
-			copy(b, r);
-			sgemv(-1, A, x, 1, r);
-		}
-		else
-		{
-			// r = r - alfa<66>q
-			saxpy(-alpha, q, r);
-		}
-
-		// s = M^-1 <20> r
-		preconditioner.apply(r, s);
-		//copy(r, s);
-
-		delta_old = delta_new;
-		delta_new = dot( r, s );
-
-		beta = delta_new / delta_old;
-
-		// p = s + beta<74>p
-		copy(s, p);
-		saxpy(beta, p, s);
-	}
-
-	return i;
-}
-
-
-#if 0 // Nonsymmetric solvers
-
-/** Bi-conjugate gradient method.  */
-MATHLIB_API int BiConjugateGradientSolve( const SparseMatrix &A, const DenseVector &b, DenseVector &x, float epsilon ) {
-	piDebugCheck( A.IsSquare() );
-	piDebugCheck( A.Width() == b.Dim() );
-	piDebugCheck( A.Width() == x.Dim() );
-
-	int i = 0;
-	const int D = A.Width();
-	const int i_max = 4 * D;
-
-	float resid;
-	float rho_1 = 0;
-	float rho_2 = 0;
-	float alpha;
-	float beta;
-
-	DenseVector r(D);
-	DenseVector rtilde(D);
-	DenseVector p(D);
-	DenseVector ptilde(D);
-	DenseVector q(D);
-	DenseVector qtilde(D);
-	DenseVector tmp(D);	// temporal vector.
-
-	// r = b - A<>x;
-	A.Product( x, tmp );
-	r.Sub( b, tmp );
-
-	// rtilde = r
-	rtilde.Set( r );
-
-	// p = r;
-	p.Set( r );
-
-	// ptilde = rtilde
-	ptilde.Set( rtilde );
-
-
-
-	float normb = b.Norm();
-  	if( normb == 0.0 ) normb = 1;
-
-	// test convergence
-	resid = r.Norm() / normb;
-	if( resid < epsilon ) {
-		// method converges?
-		return 0;
-	}
-
-
-	while( i < i_max ) {
-
-		i++;
-		
-		rho_1 = DenseVectorDotProduct( r, rtilde );
-
-		if( rho_1 == 0 ) {
-			// method fails.
-			return -i;
-		}
-
-		if (i == 1) {
-			p.Set( r );
-			ptilde.Set( rtilde );
-		} 
-		else {
-			beta = rho_1 / rho_2;
-			
-			// p = r + beta * p;
-			p.Mad( r, p, beta );
-			
-			// ptilde = ztilde + beta * ptilde;
-			ptilde.Mad( rtilde, ptilde, beta );
-		}
-
-		// q = A * p;
-		A.Product( p, q );
-
-		// qtilde = A^t * ptilde;
-		A.TransProduct( ptilde, qtilde );
-
-		alpha = rho_1 / DenseVectorDotProduct( ptilde, q );
-
-		// x += alpha * p;
-		x.Mad( x, p, alpha );
-
-		// r -= alpha * q;
-		r.Mad( r, q, -alpha );
-		
-		// rtilde -= alpha * qtilde;
-		rtilde.Mad( rtilde, qtilde, -alpha );
-
-		rho_2 = rho_1;
-
-		// test convergence
-		resid = r.Norm() / normb;
-		if( resid < epsilon ) {
-			// method converges
-			return i;
-		}
-	}
-
-	return i;
-}
-
-
-/** Bi-conjugate gradient stabilized method. */
-int BiCGSTABSolve( const SparseMatrix &A, const DenseVector &b, DenseVector &x, float epsilon ) {
-	piDebugCheck( A.IsSquare() );
-	piDebugCheck( A.Width() == b.Dim() );
-	piDebugCheck( A.Width() == x.Dim() );
-
-	int i = 0;
-	const int D = A.Width();
-	const int i_max = 2 * D;
-
-
-	float resid;
-	float rho_1 = 0;
-	float rho_2 = 0;
-	float alpha = 0;
-	float beta = 0;
-	float omega = 0;
-
-	DenseVector p(D);
-	DenseVector phat(D);
-	DenseVector s(D);
-	DenseVector shat(D);
-	DenseVector t(D);
-	DenseVector v(D);
-
-	DenseVector r(D);
-	DenseVector rtilde(D);
-
-	DenseVector tmp(D);
-
-	// r = b - A<>x;
-	A.Product( x, tmp );
-	r.Sub( b, tmp );
-
-	// rtilde = r
-	rtilde.Set( r );
-
-
-	float normb = b.Norm();
-	if( normb == 0.0 ) normb = 1;
-  
-	// test convergence
-	resid = r.Norm() / normb;
-	if( resid < epsilon ) {
-		// method converges?
-		return 0;
-	}
-
-
-	while( i<i_max ) {
-
-		i++;
-	
-		rho_1 = DenseVectorDotProduct( rtilde, r );
-		if( rho_1 == 0 ) {
-			// method fails
-			return -i;
-		}
-
-
-		if( i == 1 ) {
-			p.Set( r );
-		}
-		else {
-			beta = (rho_1 / rho_2) * (alpha / omega);
-
-			// p = r + beta * (p - omega * v);
-			p.Mad( p, v, -omega );
-			p.Mad( r, p, beta );
-		}
-
-		//phat = M.solve(p);
-		phat.Set( p );
-		//Precond( &phat, p );
-
-		//v = A * phat;
-		A.Product( phat, v );
-
-		alpha = rho_1 / DenseVectorDotProduct( rtilde, v );
-
-		// s = r - alpha * v;
-		s.Mad( r, v, -alpha );
-
-
-		resid = s.Norm() / normb;
-		if( resid < epsilon ) {
-			// x += alpha * phat;
-			x.Mad( x, phat, alpha );
-			return i;
-		}
-
-		//shat = M.solve(s);
-		shat.Set( s );
-		//Precond( &shat, s );
-
-		//t = A * shat;
-		A.Product( shat, t );
-
-		omega = DenseVectorDotProduct( t, s ) / DenseVectorDotProduct( t, t );
-
-		// x += alpha * phat + omega * shat;
-		x.Mad( x, shat, omega );
-		x.Mad( x, phat, alpha );
-
-		//r = s - omega * t;
-		r.Mad( s, t, -omega );
-
-		rho_2 = rho_1;
-		
-		resid = r.Norm() / normb;
-		if( resid < epsilon ) {
-			return i;
-		}
-
-		if( omega == 0 ) {
-			return -i;	// ???
-		}
-	}
-
-	return i;
-}
-
-
-/** Bi-conjugate gradient stabilized method. */
-int BiCGSTABPrecondSolve( const SparseMatrix &A, const DenseVector &b, DenseVector &x, const IPreconditioner &M, float epsilon ) {
-	piDebugCheck( A.IsSquare() );
-	piDebugCheck( A.Width() == b.Dim() );
-	piDebugCheck( A.Width() == x.Dim() );
-
-	int i = 0;
-	const int D = A.Width();
-	const int i_max = D;
-//	const int i_max = 1000;
-
-
-	float resid;
-	float rho_1 = 0;
-	float rho_2 = 0;
-	float alpha = 0;
-	float beta = 0;
-	float omega = 0;
-
-	DenseVector p(D);
-	DenseVector phat(D);
-	DenseVector s(D);
-	DenseVector shat(D);
-	DenseVector t(D);
-	DenseVector v(D);
-
-	DenseVector r(D);
-	DenseVector rtilde(D);
-
-	DenseVector tmp(D);
-
-	// r = b - A<>x;
-	A.Product( x, tmp );
-	r.Sub( b, tmp );
-
-	// rtilde = r
-	rtilde.Set( r );
-
-
-	float normb = b.Norm();
-	if( normb == 0.0 ) normb = 1;
-  
-	// test convergence
-	resid = r.Norm() / normb;
-	if( resid < epsilon ) {
-		// method converges?
-		return 0;
-	}
-
-
-	while( i<i_max ) {
-
-		i++;
-	
-		rho_1 = DenseVectorDotProduct( rtilde, r );
-		if( rho_1 == 0 ) {
-			// method fails
-			return -i;
-		}
-
-
-		if( i == 1 ) {
-			p.Set( r );
-		}
-		else {
-			beta = (rho_1 / rho_2) * (alpha / omega);
-
-			// p = r + beta * (p - omega * v);
-			p.Mad( p, v, -omega );
-			p.Mad( r, p, beta );
-		}
-
-		//phat = M.solve(p);
-		//phat.Set( p );
-		M.Precond( &phat, p );
-
-		//v = A * phat;
-		A.Product( phat, v );
-
-		alpha = rho_1 / DenseVectorDotProduct( rtilde, v );
-
-		// s = r - alpha * v;
-		s.Mad( r, v, -alpha );
-
-
-		resid = s.Norm() / normb;
-
-		//printf( "--- Iteration %d: residual = %f\n", i, resid );
-
-		if( resid < epsilon ) {
-			// x += alpha * phat;
-			x.Mad( x, phat, alpha );
-			return i;
-		}
-
-		//shat = M.solve(s);
-		//shat.Set( s );
-		M.Precond( &shat, s );
-
-		//t = A * shat;
-		A.Product( shat, t );
-
-		omega = DenseVectorDotProduct( t, s ) / DenseVectorDotProduct( t, t );
-
-		// x += alpha * phat + omega * shat;
-		x.Mad( x, shat, omega );
-		x.Mad( x, phat, alpha );
-
-		//r = s - omega * t;
-		r.Mad( s, t, -omega );
-
-		rho_2 = rho_1;
-		
-		resid = r.Norm() / normb;
-		if( resid < epsilon ) {
-			return i;
-		}
-
-		if( omega == 0 ) {
-			return -i;	// ???
-		}
-	}
-
-	return i;
-}
-
-#endif

src/nvmath/Solver.h

@@ -1,20 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#ifndef NV_MATH_SOLVER_H
-#define NV_MATH_SOLVER_H
-
-#include <nvmath/Sparse.h>
-
-namespace nv
-{
-
-	// Linear solvers.
-	NVMATH_API void LeastSquaresSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon = 1e-5f);
-	NVMATH_API void LeastSquaresSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, const uint * lockedParameters, uint lockedCount, float epsilon = 1e-5f);
-	NVMATH_API void SymmetricSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon = 1e-5f);
-//	NVMATH_API void NonSymmetricSolver(const SparseMatrix & A, const FullVector & b, FullVector & x, float epsilon = 1e-5f);
-
-} // nv namespace
-
-
-#endif // NV_MATH_SOLVER_H

src/nvmath/Sparse.cpp

@@ -1,831 +0,0 @@
-// This code is in the public domain -- Ignacio Casta<74>o <castanyo@yahoo.es>
-
-#include <nvmath/Sparse.h>
-#include <nvmath/KahanSum.h>
-
-using namespace nv;
-
-
-/// Ctor.
-FullVector::FullVector(uint dim)
-{ 
-	m_array.resize(dim); 
-}
-
-/// Copy ctor.
-FullVector::FullVector(const FullVector & v) : m_array(v.m_array)
-{
-}
-
-/// Copy operator
-const FullVector & FullVector::operator=(const FullVector & v)
-{
-	nvCheck(dimension() == v.dimension());
-	
-	m_array = v.m_array;
-
-	return *this;
-}
-		
-
-void FullVector::fill(float f)
-{
-	const uint dim = dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		m_array[i] = f;
-	}
-}
-
-void FullVector::operator+= (const FullVector & v)
-{
-	nvDebugCheck(dimension() == v.dimension());
-
-	const uint dim = dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		m_array[i] += v.m_array[i];
-	}
-}
-
-void FullVector::operator-= (const FullVector & v)
-{
-	nvDebugCheck(dimension() == v.dimension());
-
-	const uint dim = dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		m_array[i] -= v.m_array[i];
-	}
-}
-
-void FullVector::operator*= (const FullVector & v)
-{
-	nvDebugCheck(dimension() == v.dimension());
-
-	const uint dim = dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		m_array[i] *= v.m_array[i];
-	}
-}
-
-void FullVector::operator+= (float f)
-{
-	const uint dim = dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		m_array[i] += f;
-	}
-}
-
-void FullVector::operator-= (float f)
-{
-	const uint dim = dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		m_array[i] -= f;
-	}
-}
-
-void FullVector::operator*= (float f)
-{
-	const uint dim = dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		m_array[i] *= f;
-	}
-}
-
-
-void nv::saxpy(float a, const FullVector & x, FullVector & y)
-{
-	nvDebugCheck(x.dimension() == y.dimension());
-
-	const uint dim = x.dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		y[i] += a * x[i];
-	}
-}
-
-void nv::copy(const FullVector & x, FullVector & y)
-{
-	nvDebugCheck(x.dimension() == y.dimension());
-
-	const uint dim = x.dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		y[i] = x[i];
-	}
-}
-
-void nv::scal(float a, FullVector & x)
-{
-	const uint dim = x.dimension();
-	for (uint i = 0; i < dim; i++)
-	{
-		x[i] *= a;
-	}
-}
-
-float nv::dot(const FullVector & x, const FullVector & y)
-{
-	nvDebugCheck(x.dimension() == y.dimension());
-
-	const uint dim = x.dimension();
-
-	/*float sum = 0;
-	for (uint i = 0; i < dim; i++)
-	{
-		sum += x[i] * y[i];
-	}
-	return sum;*/
-
-	KahanSum kahan;
-
-	for (uint i = 0; i < dim; i++)
-	{
-		kahan.add(x[i] * y[i]);
-	}
-
-	return kahan.sum();
-}
-
-
-FullMatrix::FullMatrix(uint d) : m_width(d), m_height(d)
-{
-	m_array.resize(d*d, 0.0f);
-}
-
-FullMatrix::FullMatrix(uint w, uint h) : m_width(w), m_height(h)
-{
-	m_array.resize(w*h, 0.0f);
-}
-
-FullMatrix::FullMatrix(const FullMatrix & m) : m_width(m.m_width), m_height(m.m_height)
-{
-	m_array = m.m_array;
-}
-
-const FullMatrix & FullMatrix::operator=(const FullMatrix & m)
-{
-	nvCheck(width() == m.width());
-	nvCheck(height() == m.height());
-
-	m_array = m.m_array;
-	
-	return *this;
-}
-
-
-float FullMatrix::getCoefficient(uint x, uint y) const
-{
-	nvDebugCheck( x < width() );
-	nvDebugCheck( y < height() );
-	
-	return m_array[y * width() + x];
-}
-
-void FullMatrix::setCoefficient(uint x, uint y, float f)
-{
-	nvDebugCheck( x < width() );
-	nvDebugCheck( y < height() );
-	
-	m_array[y * width() + x] = f;
-}
-
-void FullMatrix::addCoefficient(uint x, uint y, float f)
-{
-	nvDebugCheck( x < width() );
-	nvDebugCheck( y < height() );
-	
-	m_array[y * width() + x] += f;
-}
-
-void FullMatrix::mulCoefficient(uint x, uint y, float f)
-{
-	nvDebugCheck( x < width() );
-	nvDebugCheck( y < height() );
-	
-	m_array[y * width() + x] *= f;
-}
-
-float FullMatrix::dotRow(uint y, const FullVector & v) const
-{
-	nvDebugCheck( v.dimension() == width() );
-	nvDebugCheck( y < height() );
-
-	float sum = 0;
-
-	const uint count = v.dimension();
-	for (uint i = 0; i < count; i++)
-	{
-		sum += m_array[y * count + i] * v[i];
-	}
-
-	return sum;
-}
-
-void FullMatrix::madRow(uint y, float alpha, FullVector & v) const
-{
-	nvDebugCheck( v.dimension() == width() );
-	nvDebugCheck( y < height() );
-
-	const uint count = v.dimension();
-	for (uint i = 0; i < count; i++)
-	{
-		v[i] += m_array[y * count + i];
-	}
-}
-
-
-// y = M * x
-void nv::mult(const FullMatrix & M, const FullVector & x, FullVector & y)
-{
-	mult(NoTransposed, M, x, y);
-}
-
-void nv::mult(Transpose TM, const FullMatrix & M, const FullVector & x, FullVector & y)
-{
-	const uint w = M.width();
-	const uint h = M.height();
-
-	if (TM == Transposed)
-	{
-		nvDebugCheck( h == x.dimension() );
-		nvDebugCheck( w == y.dimension() );
-
-		y.fill(0.0f);
-
-		for (uint i = 0; i < h; i++)
-		{
-			M.madRow(i, x[i], y);
-		}
-	}
-	else
-	{
-		nvDebugCheck( w == x.dimension() );
-		nvDebugCheck( h == y.dimension() );
-
-		for (uint i = 0; i < h; i++)
-		{
-			y[i] = M.dotRow(i, x);
-		}
-	}
-}
-
-// y = alpha*A*x + beta*y
-void nv::sgemv(float alpha, const FullMatrix & A, const FullVector & x, float beta, FullVector & y)
-{
-	sgemv(alpha, NoTransposed, A, x, beta, y);
-}
-
-void nv::sgemv(float alpha, Transpose TA, const FullMatrix & A, const FullVector & x, float beta, FullVector & y)
-{
-	const uint w = A.width();
-	const uint h = A.height();
-
-	if (TA == Transposed)
-	{
-		nvDebugCheck( h == x.dimension() );
-		nvDebugCheck( w == y.dimension() );
-
-		for (uint i = 0; i < h; i++)
-		{
-			A.madRow(i, alpha * x[i], y);
-		}
-	}
-	else
-	{
-		nvDebugCheck( w == x.dimension() );
-		nvDebugCheck( h == y.dimension() );
-
-		for (uint i = 0; i < h; i++)
-		{
-			y[i] = alpha * A.dotRow(i, x) + beta * y[i];
-		}
-	}
-}
-
-
-// Multiply a row of A by a column of B.
-static float dot(uint j, Transpose TA, const FullMatrix & A, uint i, Transpose TB, const FullMatrix & B)
-{
-	const uint w = (TA == NoTransposed) ? A.width() : A.height();
-	nvDebugCheck(w == (TB == NoTransposed) ? B.height() : A.width());
-
-	float sum = 0.0f;
-
-	for (uint k = 0; k < w; k++)
-	{
-		const float a = (TA == NoTransposed) ? A.getCoefficient(k, j) : A.getCoefficient(j, k);	// @@ Move branches out of the loop?
-		const float b = (TB == NoTransposed) ? B.getCoefficient(i, k) : A.getCoefficient(k, i);
-		sum += a * b;
-	}
-
-	return sum;
-}
-
-
-// C = A * B
-void nv::mult(const FullMatrix & A, const FullMatrix & B, FullMatrix & C)
-{
-	mult(NoTransposed, A, NoTransposed, B, C);
-}
-
-void nv::mult(Transpose TA, const FullMatrix & A, Transpose TB, const FullMatrix & B, FullMatrix & C)
-{
-	sgemm(1.0f, TA, A, TB, B, 0.0f, C);
-}
-
-// C = alpha*A*B + beta*C
-void nv::sgemm(float alpha, const FullMatrix & A, const FullMatrix & B, float beta, FullMatrix & C)
-{
-	sgemm(alpha, NoTransposed, A, NoTransposed, B, beta, C);
-}
-
-void nv::sgemm(float alpha, Transpose TA, const FullMatrix & A, Transpose TB, const FullMatrix & B, float beta, FullMatrix & C)
-{
-	const uint w = C.width();
-	const uint h = C.height();
-
-	uint aw = (TA == NoTransposed) ? A.width() : A.height();
-	uint ah = (TA == NoTransposed) ? A.height() : A.width();
-	uint bw = (TB == NoTransposed) ? B.width() : B.height();
-	uint bh = (TB == NoTransposed) ? B.height() : B.width();
-
-	nvDebugCheck(aw == bh);
-	nvDebugCheck(bw == ah);
-	nvDebugCheck(w == bw);
-	nvDebugCheck(h == ah);
-
-	for (uint y = 0; y < h; y++)
-	{
-		for (uint x = 0; x < w; x++)
-		{
-			float c = alpha * ::dot(x, TA, A, y, TB, B) + beta * C.getCoefficient(x, y);
-			C.setCoefficient(x, y, c);
-		}
-	}
-}
-
-
-
-
-
-/// Ctor. Init the size of the sparse matrix.
-SparseMatrix::SparseMatrix(uint d) : m_width(d)
-{
-	m_array.resize(d);
-}
-
-/// Ctor. Init the size of the sparse matrix.
-SparseMatrix::SparseMatrix(uint w, uint h) : m_width(w)
-{
-	m_array.resize(h);
-}
-
-SparseMatrix::SparseMatrix(const SparseMatrix & m) : m_width(m.m_width)
-{
-	m_array = m.m_array;
-}
-
-const SparseMatrix & SparseMatrix::operator=(const SparseMatrix & m)
-{
-	nvCheck(width() == m.width());
-	nvCheck(height() == m.height());
-
-	m_array = m.m_array;
-
-	return *this;
-}
-
-
-// x is column, y is row
-float SparseMatrix::getCoefficient(uint x, uint y) const
-{
-	nvDebugCheck( x < width() );
-	nvDebugCheck( y < height() );
-
-	const uint count = m_array[y].count();
-	for (uint i = 0; i < count; i++)
-	{
-		if (m_array[y][i].x == x) return m_array[y][i].v;
-	}
-
-	return 0.0f;
-}
-
-void SparseMatrix::setCoefficient(uint x, uint y, float f)
-{
-	nvDebugCheck( x < width() );
-	nvDebugCheck( y < height() );
-
-	const uint count = m_array[y].count();
-	for (uint i = 0; i < count; i++)
-	{
-		if (m_array[y][i].x == x) 
-		{
-			m_array[y][i].v = f;
-			return;
-		}
-	}
-
-	Coefficient c = { x, f };
-	m_array[y].append( c );
-}
-
-void SparseMatrix::addCoefficient(uint x, uint y, float f)
-{
-	nvDebugCheck( x < width() );
-	nvDebugCheck( y < height() );
-
-	const uint count = m_array[y].count();
-	for (uint i = 0; i < count; i++)
-	{
-		if (m_array[y][i].x == x) 
-		{
-			m_array[y][i].v += f;
-			return;
-		}
-	}
-
-	Coefficient c = { x, f };
-	m_array[y].append( c );
-}
-
-void SparseMatrix::mulCoefficient(uint x, uint y, float f)
-{
-	nvDebugCheck( x < width() );
-	nvDebugCheck( y < height() );
-
-	const uint count = m_array[y].count();
-	for (uint i = 0; i < count; i++)
-	{
-		if (m_array[y][i].x == x) 
-		{
-			m_array[y][i].v *= f;
-			return;
-		}
-	}
-
-	Coefficient c = { x, f };
-	m_array[y].append( c );
-}
-
-
-float SparseMatrix::sumRow(uint y) const
-{
-	nvDebugCheck( y < height() );
-
-	const uint count = m_array[y].count();
-
-	/*float sum = 0;
-	for (uint i = 0; i < count; i++)
-	{
-		sum += m_array[y][i].v;
-	}
-	return sum;*/
-
-	KahanSum kahan;
-
-	for (uint i = 0; i < count; i++)
-	{
-		kahan.add(m_array[y][i].v);
-	}
-
-	return kahan.sum();
-}
-
-float SparseMatrix::dotRow(uint y, const FullVector & v) const
-{
-	nvDebugCheck( y < height() );
-
-	const uint count = m_array[y].count();
-
-	/*float sum = 0;
-	for (uint i = 0; i < count; i++)
-	{
-		sum += m_array[y][i].v * v[m_array[y][i].x];
-	}
-	return sum;*/
-
-	KahanSum kahan;
-
-	for (uint i = 0; i < count; i++)
-	{
-		kahan.add(m_array[y][i].v * v[m_array[y][i].x]);
-	}
-
-	return kahan.sum();
-}
-
-void SparseMatrix::madRow(uint y, float alpha, FullVector & v) const
-{
-	nvDebugCheck(y < height());
-
-	const uint count = m_array[y].count();
-	for (uint i = 0; i < count; i++)
-	{
-		v[m_array[y][i].x] += alpha * m_array[y][i].v;
-	}
-}
-
-
-void SparseMatrix::clearRow(uint y)
-{
-	nvDebugCheck( y < height() );
-
-	m_array[y].clear();
-}
-
-void SparseMatrix::scaleRow(uint y, float f)
-{
-	nvDebugCheck( y < height() );
-
-	const uint count = m_array[y].count();
-	for (uint i = 0; i < count; i++)
-	{
-		m_array[y][i].v *= f;
-	}
-}
-
-void SparseMatrix::normalizeRow(uint y)
-{
-	nvDebugCheck( y < height() );
-
-	float norm = 0.0f;
-
-	const uint count = m_array[y].count();
-	for (uint i = 0; i < count; i++)
-	{
-		float f = m_array[y][i].v;
-		norm += f * f;
-	}
-
-	scaleRow(y, 1.0f / sqrtf(norm));
-}
-
-
-void SparseMatrix::clearColumn(uint x)
-{
-	nvDebugCheck(x < width());
-
-	for (uint y = 0; y < height(); y++)
-	{
-		const uint count = m_array[y].count();
-		for (uint e = 0; e < count; e++)
-		{
-			if (m_array[y][e].x == x)
-			{
-				m_array[y][e].v = 0.0f;
-				break;
-			}
-		}
-	}
-}
-
-void SparseMatrix::scaleColumn(uint x, float f)
-{
-	nvDebugCheck(x < width());
-
-	for (uint y = 0; y < height(); y++)
-	{
-		const uint count = m_array[y].count();
-		for (uint e = 0; e < count; e++)
-		{
-			if (m_array[y][e].x == x)
-			{
-				m_array[y][e].v *= f;
-				break;
-			}
-		}
-	}
-}
-
-const Array<SparseMatrix::Coefficient> & SparseMatrix::getRow(uint y) const
-{
-	return m_array[y];
-}
-
-
-// y = M * x
-void nv::mult(const SparseMatrix & M, const FullVector & x, FullVector & y)
-{
-	mult(NoTransposed, M, x, y);
-}
-
-void nv::mult(Transpose TM, const SparseMatrix & M, const FullVector & x, FullVector & y)
-{
-	const uint w = M.width();
-	const uint h = M.height();
-
-	if (TM == Transposed)
-	{
-		nvDebugCheck( h == x.dimension() );
-		nvDebugCheck( w == y.dimension() );
-
-		y.fill(0.0f);
-
-		for (uint i = 0; i < h; i++)
-		{
-			M.madRow(i, x[i], y);
-		}
-	}
-	else
-	{
-		nvDebugCheck( w == x.dimension() );
-		nvDebugCheck( h == y.dimension() );
-
-		for (uint i = 0; i < h; i++)
-		{
-			y[i] = M.dotRow(i, x);
-		}
-	}
-}
-
-// y = alpha*A*x + beta*y
-void nv::sgemv(float alpha, const SparseMatrix & A, const FullVector & x, float beta, FullVector & y)
-{
-	sgemv(alpha, NoTransposed, A, x, beta, y);
-}
-
-void nv::sgemv(float alpha, Transpose TA, const SparseMatrix & A, const FullVector & x, float beta, FullVector & y)
-{
-	const uint w = A.width();
-	const uint h = A.height();
-
-	if (TA == Transposed)
-	{
-		nvDebugCheck( h == x.dimension() );
-		nvDebugCheck( w == y.dimension() );
-
-		for (uint i = 0; i < h; i++)
-		{
-			A.madRow(i, alpha * x[i], y);
-		}
-	}
-	else
-	{
-		nvDebugCheck( w == x.dimension() );
-		nvDebugCheck( h == y.dimension() );
-
-		for (uint i = 0; i < h; i++)
-		{
-			y[i] = alpha * A.dotRow(i, x) + beta * y[i];
-		}
-	}
-}
-
-
-// dot y-row of A by x-column of B
-static float dotRowColumn(int y, const SparseMatrix & A, int x, const SparseMatrix & B)
-{
-	const Array<SparseMatrix::Coefficient> & row = A.getRow(y);
-
-	const uint count = row.count();
-
-	/*float sum = 0.0f;
-	for (uint i = 0; i < count; i++)
-	{
-		const SparseMatrix::Coefficient & c = row[i];
-		sum += c.v * B.getCoefficient(x, c.x);
-	}
-	return sum;*/
-
-	KahanSum kahan;
-	for (uint i = 0; i < count; i++)
-	{
-		const SparseMatrix::Coefficient & c = row[i];
-		kahan.add(c.v * B.getCoefficient(x, c.x));
-	}
-
-	return kahan.sum();
-}
-
-// dot y-row of A by x-row of B
-static float dotRowRow(int y, const SparseMatrix & A, int x, const SparseMatrix & B)
-{
-	const Array<SparseMatrix::Coefficient> & row = A.getRow(y);
-
-	const uint count = row.count();
-
-	/*float sum = 0.0f;
-	for (uint i = 0; i < count; i++)
-	{
-		const SparseMatrix::Coefficient & c = row[i];
-		sum += c.v * B.getCoefficient(c.x, x);
-	}
-	//return sum;*/
-
-	KahanSum kahan;
-	for (uint i = 0; i < count; i++)
-	{
-		const SparseMatrix::Coefficient & c = row[i];
-		kahan.add(c.v * B.getCoefficient(c.x, x));
-	}
-
-	return kahan.sum();
-}
-
-// dot y-column of A by x-column of B
-static float dotColumnColumn(int y, const SparseMatrix & A, int x, const SparseMatrix & B)
-{
-	nvDebugCheck(A.height() == B.height());
-
-	const uint h = A.height();
-
-	/*float sum = 0.0f;
-	for (uint i = 0; i < h; i++)
-	{
-		sum += A.getCoefficient(y, i) * B.getCoefficient(x, i);
-	}
-	//return sum;*/
-
-	KahanSum kahan;
-	for (uint i = 0; i < h; i++)
-	{
-		kahan.add(A.getCoefficient(y, i) * B.getCoefficient(x, i));
-	}
-
-	return kahan.sum();
-}
-
-
-
-// C = A * B
-void nv::mult(const SparseMatrix & A, const SparseMatrix & B, SparseMatrix & C)
-{
-	mult(NoTransposed, A, NoTransposed, B, C);
-}
-
-void nv::mult(Transpose TA, const SparseMatrix & A, Transpose TB, const SparseMatrix & B, SparseMatrix & C)
-{
-	sgemm(1.0f, TA, A, TB, B, 0.0f, C);
-}
-
-// C = alpha*A*B + beta*C
-void nv::sgemm(float alpha, const SparseMatrix & A, const SparseMatrix & B, float beta, SparseMatrix & C)
-{
-	sgemm(alpha, NoTransposed, A, NoTransposed, B, beta, C);
-}
-
-void nv::sgemm(float alpha, Transpose TA, const SparseMatrix & A, Transpose TB, const SparseMatrix & B, float beta, SparseMatrix & C)
-{
-	const uint w = C.width();
-	const uint h = C.height();
-
-	uint aw = (TA == NoTransposed) ? A.width() : A.height();
-	uint ah = (TA == NoTransposed) ? A.height() : A.width();
-	uint bw = (TB == NoTransposed) ? B.width() : B.height();
-	uint bh = (TB == NoTransposed) ? B.height() : B.width();
-
-	nvDebugCheck(aw == bh);
-	nvDebugCheck(bw == ah);
-	nvDebugCheck(w == bw);
-	nvDebugCheck(h == ah);
-
-
-	for (uint y = 0; y < h; y++)
-	{
-		for (uint x = 0; x < w; x++)
-		{
-			float c = beta * C.getCoefficient(x, y);
-
-			if (TA == NoTransposed && TB == NoTransposed)
-			{
-				// dot y-row of A by x-column of B.
-				c += alpha * dotRowColumn(y, A, x, B);
-			}
-			else if (TA == Transposed && TB == Transposed)
-			{
-				// dot y-column of A by x-row of B.
-				c += alpha * dotRowColumn(x, B, y, A);
-			}
-			else if (TA == Transposed && TB == NoTransposed)
-			{
-				// dot y-column of A by x-column of B.
-				c += alpha * dotColumnColumn(y, A, x, B);
-			}
-			else if (TA == NoTransposed && TB == Transposed)
-			{
-				// dot y-row of A by x-row of B.
-				c += alpha * dotRowRow(y, A, x, B);
-			}
-
-			if (c != 0.0f)
-			{
-				C.setCoefficient(x, y, c);
-			}
-		}
-	}
-
-}
-
-// C = At * A
-void nv::sqm(const SparseMatrix & A, SparseMatrix & C)
-{
-	// This is quite expensive...
-	mult(Transposed, A, NoTransposed, A, C);
-}

src/nvmath/Sparse.h

@@ -1,198 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#ifndef NV_MATH_SPARSE_H
-#define NV_MATH_SPARSE_H
-
-#include <nvcore/Containers.h>
-#include <nvmath/nvmath.h>
-
-// Full and sparse vector and matrix classes. BLAS subset.
-
-namespace nv
-{
-	class FullVector;
-	class FullMatrix;
-	class SparseMatrix;
-
-
-	/// Fixed size vector class.
-	class FullVector
-	{
-	public:
-
-		FullVector(uint dim);
-		FullVector(const FullVector & v);
-
-		const FullVector & operator=(const FullVector & v);
-		
-		uint dimension() const { return m_array.count(); }
-
-		const float & operator[]( uint index ) const { return m_array[index]; }
-		float & operator[] ( uint index ) { return m_array[index]; }
-
-		void fill(float f);
-		
-		void operator+= (const FullVector & v);
-		void operator-= (const FullVector & v);
-		void operator*= (const FullVector & v);
-
-		void operator+= (float f);
-		void operator-= (float f);
-		void operator*= (float f);
-
-
-	private:
-
-		Array<float> m_array;
-
-	};
-
-	// Pseudo-BLAS interface.
-	NVMATH_API void saxpy(float a, const FullVector & x, FullVector & y);	// y = a * x + y
-	NVMATH_API void copy(const FullVector & x, FullVector & y);
-	NVMATH_API void scal(float a, FullVector & x);
-	NVMATH_API float dot(const FullVector & x, const FullVector & y);
-
-
-	enum Transpose
-	{
-		NoTransposed = 0,
-		Transposed = 1
-	};
-
-	/// Full matrix class.
-	class FullMatrix
-	{
-	public:
-
-		FullMatrix(uint d);
-		FullMatrix(uint w, uint h);
-		FullMatrix(const FullMatrix & m);
-
-		const FullMatrix & operator=(const FullMatrix & m);
-
-		uint width() const { return m_width; }
-		uint height() const { return m_height; }
-		bool isSquare() const { return m_width == m_height; }
-		
-		float getCoefficient(uint x, uint y) const;
-
-		void setCoefficient(uint x, uint y, float f);
-		void addCoefficient(uint x, uint y, float f);
-		void mulCoefficient(uint x, uint y, float f);
-		
-		float dotRow(uint y, const FullVector & v) const;
-		void madRow(uint y, float alpha, FullVector & v) const;
-
-	protected:
-
-		bool isValid() const {
-			return m_array.size() == (m_width * m_height);
-		}
-
-	private:
-
-		const uint m_width;
-		const uint m_height;
-		Array<float> m_array;
-
-	};
-
-	NVMATH_API void mult(const FullMatrix & M, const FullVector & x, FullVector & y);
-	NVMATH_API void mult(Transpose TM, const FullMatrix & M, const FullVector & x, FullVector & y);
-
-	// y = alpha*A*x + beta*y
-	NVMATH_API void sgemv(float alpha, const FullMatrix & A, const FullVector & x, float beta, FullVector & y);
-	NVMATH_API void sgemv(float alpha, Transpose TA, const FullMatrix & A, const FullVector & x, float beta, FullVector & y);
-
-	NVMATH_API void mult(const FullMatrix & A, const FullMatrix & B, FullMatrix & C);
-	NVMATH_API void mult(Transpose TA, const FullMatrix & A, Transpose TB, const FullMatrix & B, FullMatrix & C);
-
-	// C = alpha*A*B + beta*C
-	NVMATH_API void sgemm(float alpha, const FullMatrix & A, const FullMatrix & B, float beta, FullMatrix & C);
-	NVMATH_API void sgemm(float alpha, Transpose TA, const FullMatrix & A, Transpose TB, const FullMatrix & B, float beta, FullMatrix & C);
-
-
-	/**
-	 * Sparse matrix class. The matrix is assumed to be sparse and to have
-	 * very few non-zero elements, for this reason it's stored in indexed 
-	 * format. To multiply column vectors efficiently, the matrix stores 
-	 * the elements in indexed-column order, there is a list of indexed 
-	 * elements for each row of the matrix. As with the FullVector the 
-	 * dimension of the matrix is constant.
-	**/
-	class SparseMatrix
-	{
-		friend class FullMatrix;
-	public:
-
-		// An element of the sparse array.
-		struct Coefficient {
-			uint x;  // column
-			float v; // value
-		};
-
-
-	public:
-
-		SparseMatrix(uint d);
-		SparseMatrix(uint w, uint h);
-		SparseMatrix(const SparseMatrix & m);
-
-		const SparseMatrix & operator=(const SparseMatrix & m);
-
-
-		uint width() const { return m_width; }
-		uint height() const { return m_array.count(); }
-		bool isSquare() const { return width() == height(); }
-
-		float getCoefficient(uint x, uint y) const; // x is column, y is row
-
-		void setCoefficient(uint x, uint y, float f);
-		void addCoefficient(uint x, uint y, float f);
-		void mulCoefficient(uint x, uint y, float f);
-
-		float sumRow(uint y) const;
-		float dotRow(uint y, const FullVector & v) const;
-		void madRow(uint y, float alpha, FullVector & v) const;
-
-		void clearRow(uint y);
-		void scaleRow(uint y, float f);
-		void normalizeRow(uint y);
-
-		void clearColumn(uint x);
-		void scaleColumn(uint x, float f);
-
-		const Array<Coefficient> & getRow(uint y) const;
-
-	private:
-
-		/// Number of columns.
-		const uint m_width;
-		
-		/// Array of matrix elements.
-		Array< Array<Coefficient> > m_array;
-
-	};
-
-	NVMATH_API void mult(const SparseMatrix & M, const FullVector & x, FullVector & y);
-	NVMATH_API void mult(Transpose TM, const SparseMatrix & M, const FullVector & x, FullVector & y);
-
-	// y = alpha*A*x + beta*y
-	NVMATH_API void sgemv(float alpha, const SparseMatrix & A, const FullVector & x, float beta, FullVector & y);
-	NVMATH_API void sgemv(float alpha, Transpose TA, const SparseMatrix & A, const FullVector & x, float beta, FullVector & y);
-
-	NVMATH_API void mult(const SparseMatrix & A, const SparseMatrix & B, SparseMatrix & C);
-	NVMATH_API void mult(Transpose TA, const SparseMatrix & A, Transpose TB, const SparseMatrix & B, SparseMatrix & C);
-
-	// C = alpha*A*B + beta*C
-	NVMATH_API void sgemm(float alpha, const SparseMatrix & A, const SparseMatrix & B, float beta, SparseMatrix & C);
-	NVMATH_API void sgemm(float alpha, Transpose TA, const SparseMatrix & A, Transpose TB, const SparseMatrix & B, float beta, SparseMatrix & C);
-
-	// C = At * A
-	NVMATH_API void sqm(const SparseMatrix & A, SparseMatrix & C);
-
-} // nv namespace
-
-
-#endif // NV_MATH_SPARSE_H

src/nvmath/SphericalHarmonic.h

@@ -3,7 +3,6 @@
 #ifndef NV_MATH_SPHERICALHARMONIC_H
 #define NV_MATH_SPHERICALHARMONIC_H
 
-#include <string.h> // memcpy
 #include <nvmath/Vector.h>
 
 namespace nv

src/nvmath/TriBox.cpp

@@ -96,7 +96,7 @@ static bool planeBoxOverlap(Vector3::Arg normal, Vector3::Arg vert, Vector3::Arg
 	if(min>rad || max<-rad) return false;
 
 
-bool nv::triBoxOverlap(Vector3::Arg boxcenter, Vector3::Arg boxhalfsize, const Triangle & tri)
+bool triBoxOverlap(Vector3::Arg boxcenter, Vector3::Arg boxhalfsize, const Triangle & tri)
 {
 	// use separating axis theorem to test overlap between triangle and box
 	// need to test for overlap in these directions:
@@ -170,7 +170,7 @@ bool nv::triBoxOverlap(Vector3::Arg boxcenter, Vector3::Arg boxhalfsize, const T
 }
 
 
-bool nv::triBoxOverlapNoBounds(Vector3::Arg boxcenter, Vector3::Arg boxhalfsize, const Triangle & tri)
+bool triBoxOverlapNoBounds(Vector3::Arg boxcenter, Vector3::Arg boxhalfsize, const Triangle & tri)
 {
 	// use separating axis theorem to test overlap between triangle and box
 	// need to test for overlap in these directions:

src/nvmath/TypeSerialization.cpp

@@ -1,82 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#include <nvcore/Stream.h>
-
-#include <nvmath/Vector.h>
-#include <nvmath/Matrix.h>
-#include <nvmath/Quaternion.h>
-#include <nvmath/Basis.h>
-#include <nvmath/Box.h>
-
-#include <nvmath/TypeSerialization.h>
-
-using namespace nv;
-
-Stream & nv::operator<< (Stream & s, Vector2 & v)
-{
-	float x = v.x();
-	float y = v.y();
-
-	s << x << y;
-
-	if (s.isLoading())
-	{
-		v.set(x, y);
-	}
-
-	return s;
-}
-
-Stream & nv::operator<< (Stream & s, Vector3 & v)
-{
-	float x = v.x();
-	float y = v.y();
-	float z = v.z();
-
-	s << x << y << z;
-
-	if (s.isLoading())
-	{
-		v.set(x, y, z);
-	}
-
-	return s;
-}
-
-Stream & nv::operator<< (Stream & s, Vector4 & v)
-{
-	float x = v.x();
-	float y = v.y();
-	float z = v.z();
-	float w = v.w();
-
-	s << x << y << z << w;
-
-	if (s.isLoading())
-	{
-		v.set(x, y, z, w);
-	}
-
-	return s;
-}
-
-Stream & nv::operator<< (Stream & s, Matrix & m)
-{
-	return s;
-}
-
-Stream & nv::operator<< (Stream & s, Quaternion & q)
-{
-	return s << q.asVector();
-}
-
-Stream & nv::operator<< (Stream & s, Basis & basis)
-{
-	return s << basis.tangent << basis.bitangent << basis.normal;
-}
-
-Stream & nv::operator<< (Stream & s, Box & box)
-{
-	return s << box.m_mins << box.m_maxs;
-}
-

src/nvmath/TypeSerialization.h

@@ -1,32 +0,0 @@
-// This code is in the public domain -- castanyo@yahoo.es
-
-#ifndef NV_MATH_TYPESERIALIZATION_H
-#define NV_MATH_TYPESERIALIZATION_H
-
-#include <nvmath/nvmath.h>
-
-namespace nv
-{
-	class Stream;
-
-	class Vector2;
-	class Vector3;
-	class Vector4;
-
-	class Matrix;
-	class Quaternion;
-	struct Basis;
-	class Box;
-	
-	NVMATH_API Stream & operator<< (Stream & s, Vector2 & obj);
-	NVMATH_API Stream & operator<< (Stream & s, Vector3 & obj);
-	NVMATH_API Stream & operator<< (Stream & s, Vector4 & obj);
-
-	NVMATH_API Stream & operator<< (Stream & s, Matrix & obj);
-	NVMATH_API Stream & operator<< (Stream & s, Quaternion & obj);
-	NVMATH_API Stream & operator<< (Stream & s, Basis & obj);
-	NVMATH_API Stream & operator<< (Stream & s, Box & obj);
-
-} // nv namespace
-
-#endif // NV_MATH_TYPESERIALIZATION_H

src/nvmath/Vector.h

@@ -4,7 +4,7 @@
 #define NV_MATH_VECTOR_H
 
 #include <nvmath/nvmath.h>
-#include <nvcore/Algorithms.h> // min, max
+#include <nvcore/Containers.h> // min, max
 
 namespace nv
 {
@@ -71,7 +71,6 @@ public:
 	const Vector2 & xy() const;
 
 	scalar component(uint idx) const;
-	void setComponent(uint idx, scalar f);
 
 	const scalar * ptr() const;
 
@@ -240,21 +239,13 @@ inline const Vector2 & Vector3::xy() const
 inline scalar Vector3::component(uint idx) const
 {
 	nvDebugCheck(idx < 3);
-	if (idx == 0) return m_x;
-	if (idx == 1) return m_y;
-	if (idx == 2) return m_z;
+	if (idx == 0) return x();
+	if (idx == 1) return y();
+	if (idx == 2) return z();
 	nvAssume(false);
 	return 0.0f;
 }
 
-inline void Vector3::setComponent(uint idx, float f)
-{
-	nvDebugCheck(idx < 3);
-	if (idx == 0) m_x = f;
-	else if (idx == 1) m_y = f;
-	else if (idx == 2) m_z = f;
-}
-
 inline const scalar * Vector3::ptr() const
 {
 	return &m_x;
@@ -486,35 +477,6 @@ inline scalar length(Vector2::Arg v)
 	return sqrtf(length_squared(v));
 }
 
-inline scalar inverse_length(Vector2::Arg v)
-{
-	return 1.0f / sqrtf(length_squared(v));
-}
-
-inline bool isNormalized(Vector2::Arg v, float epsilon = NV_NORMAL_EPSILON)
-{
-	return equal(length(v), 1, epsilon);
-}
-
-inline Vector2 normalize(Vector2::Arg v, float epsilon = NV_EPSILON)
-{
-	float l = length(v);
-	nvDebugCheck(!isZero(l, epsilon));
-	Vector2 n = scale(v, 1.0f / l);
-	nvDebugCheck(isNormalized(n));
-	return n;
-}
-
-inline Vector2 normalizeSafe(Vector2::Arg v, Vector2::Arg fallback, float epsilon = NV_EPSILON)
-{
-	float l = length(v);
-	if (isZero(l, epsilon)) {
-		return fallback;
-	}
-	return scale(v, 1.0f / l);
-}
-
-
 inline bool equal(Vector2::Arg v1, Vector2::Arg v2, float epsilon = NV_EPSILON)
 {
 	return equal(v1.x(), v2.x(), epsilon) && equal(v1.y(), v2.y(), epsilon);
@@ -633,11 +595,6 @@ inline scalar length(Vector3::Arg v)
 	return sqrtf(length_squared(v));
 }
 
-inline scalar inverse_length(Vector3::Arg v)
-{
-	return 1.0f / sqrtf(length_squared(v));
-}
-
 inline bool isNormalized(Vector3::Arg v, float epsilon = NV_NORMAL_EPSILON)
 {
 	return equal(length(v), 1, epsilon);
@@ -759,11 +716,6 @@ inline scalar length(Vector4::Arg v)
 	return sqrtf(length_squared(v));
 }
 
-inline scalar inverse_length(Vector4::Arg v)
-{
-	return 1.0f / sqrtf(length_squared(v));
-}
-
 inline bool isNormalized(Vector4::Arg v, float epsilon = NV_NORMAL_EPSILON)
 {
 	return equal(length(v), 1, epsilon);

src/nvmath/nvmath.h

@@ -48,19 +48,37 @@
 #define IS_NEGATIVE_FLOAT(x)	(IR(x)&SIGN_BITMASK)
 */
 
-inline float sqrt_assert(const float f)
+inline double sqrt_assert(const double f)
+{
+	nvDebugCheck(f >= 0.0f);
+	return sqrt(f);
+}
+
+inline float sqrtf_assert(const float f)
 {
 	nvDebugCheck(f >= 0.0f);
 	return sqrtf(f);
 }
 
-inline float acos_assert(const float f)
+inline double acos_assert(const double f)
+{
+	nvDebugCheck(f >= -1.0f && f <= 1.0f);
+	return acos(f);
+}
+
+inline float acosf_assert(const float f)
 {
 	nvDebugCheck(f >= -1.0f && f <= 1.0f);
 	return acosf(f);
 }
 
-inline float asin_assert(const float f)
+inline double asin_assert(const double f)
+{
+	nvDebugCheck(f >= -1.0f && f <= 1.0f);
+	return asin(f);
+}
+
+inline float asinf_assert(const float f)
 {
 	nvDebugCheck(f >= -1.0f && f <= 1.0f);
 	return asinf(f);
@@ -68,11 +86,11 @@ inline float asin_assert(const float f)
 
 // Replace default functions with asserting ones.
 #define sqrt sqrt_assert
-#define sqrtf sqrt_assert
+#define sqrtf sqrtf_assert
 #define acos acos_assert
-#define acosf acos_assert
+#define acosf acosf_assert
 #define asin asin_assert
-#define asinf asin_assert
+#define asinf asinf_assert
 
 #if NV_OS_WIN32
 #include <float.h>

src/nvtt/CMakeLists.txt

@@ -13,10 +13,10 @@ SET(NVTT_SRCS
 	CompressDXT.cpp
 	CompressRGB.h
 	CompressRGB.cpp
-	FastCompressDXT.h
-	FastCompressDXT.cpp
 	QuickCompressDXT.h
 	QuickCompressDXT.cpp
+	OptimalCompressDXT.h
+	OptimalCompressDXT.cpp
 	SingleColorLookup.h
 	CompressionOptions.h
 	CompressionOptions.cpp
@@ -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)
@@ -79,15 +80,12 @@ TARGET_LINK_LIBRARIES(nvassemble nvcore nvmath nvimage)
 ADD_EXECUTABLE(filtertest tests/filtertest.cpp tools/cmdline.h)
 TARGET_LINK_LIBRARIES(filtertest nvcore nvmath nvimage)
 
-ADD_EXECUTABLE(stress tests/stress.cpp tools/cmdline.h)
-TARGET_LINK_LIBRARIES(stress nvcore nvmath nvimage nvtt)
-
 ADD_EXECUTABLE(nvzoom tools/resize.cpp tools/cmdline.h)
 TARGET_LINK_LIBRARIES(nvzoom nvcore nvmath nvimage)
 
 INSTALL(TARGETS nvcompress nvdecompress nvddsinfo nvimgdiff nvassemble nvzoom DESTINATION bin)
 
-# UI tools
+# UI tools
 IF(QT4_FOUND AND NOT MSVC)
 	SET(QT_USE_QTOPENGL TRUE)
 	INCLUDE_DIRECTORIES(${QT_INCLUDE_DIR} ${CMAKE_CURRENT_BINARY_DIR})
@@ -109,7 +107,7 @@ IF(QT4_FOUND AND NOT MSVC)
 
 	ADD_EXECUTABLE(nvcompressui MACOSX_BUNDLE ${SRCS} ${UICS} ${MOCS})
 	TARGET_LINK_LIBRARIES(nvcompressui ${LIBS})
-
+
 ENDIF(QT4_FOUND AND NOT MSVC)
 
 

src/nvtt/CompressDXT.cpp

@@ -29,8 +29,8 @@
 
 #include "nvtt.h"
 #include "CompressDXT.h"
-#include "FastCompressDXT.h"
 #include "QuickCompressDXT.h"
+#include "OptimalCompressDXT.h"
 #include "CompressionOptions.h"
 #include "OutputOptions.h"
 
@@ -57,26 +57,33 @@ using namespace nv;
 using namespace nvtt;
 
 
-void nv::fastCompressDXT1(const Image * image, const OutputOptions::Private & outputOptions)
+nv::FastCompressor::FastCompressor() : m_image(NULL), m_alphaMode(AlphaMode_None)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+}
+
+nv::FastCompressor::~FastCompressor()
+{
+}
+
+void nv::FastCompressor::setImage(const Image * image, nvtt::AlphaMode alphaMode)
+{
+	m_image = image;
+	m_alphaMode = alphaMode;
+}
+
+void nv::FastCompressor::compressDXT1(const OutputOptions::Private & outputOptions)
+{
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT1 block;
 
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 			
-			if (rgba.isSingleColor())
-			{
-				QuickCompress::compressDXT1(rgba.color(0), &block);
-			}
-			else
-			{
-				QuickCompress::compressDXT1(rgba, &block);
-			}
+			QuickCompress::compressDXT1(rgba, &block);
 			
 			if (outputOptions.outputHandler != NULL) {
 				outputOptions.outputHandler->writeData(&block, sizeof(block));
@@ -86,27 +93,19 @@ void nv::fastCompressDXT1(const Image * image, const OutputOptions::Private & ou
 }
 
 
-void nv::fastCompressDXT1a(const Image * image, const OutputOptions::Private & outputOptions)
+void nv::FastCompressor::compressDXT1a(const OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT1 block;
 
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 			
-			// @@ We could do better here: check for single RGB, but varying alpha.
-			if (rgba.isSingleColor())
-			{
-				QuickCompress::compressDXT1a(rgba.color(0), &block);
-			}
-			else
-			{
-				QuickCompress::compressDXT1a(rgba, &block);
-			}
+			QuickCompress::compressDXT1a(rgba, &block);
 			
 			if (outputOptions.outputHandler != NULL) {
 				outputOptions.outputHandler->writeData(&block, sizeof(block));
@@ -116,17 +115,18 @@ void nv::fastCompressDXT1a(const Image * image, const OutputOptions::Private & o
 }
 
 
-void nv::fastCompressDXT3(const Image * image, const nvtt::OutputOptions::Private & outputOptions)
+void nv::FastCompressor::compressDXT3(const nvtt::OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT3 block;
 
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
+
 			QuickCompress::compressDXT3(rgba, &block);
 			
 			if (outputOptions.outputHandler != NULL) {
@@ -137,19 +137,19 @@ void nv::fastCompressDXT3(const Image * image, const nvtt::OutputOptions::Privat
 }
 
 
-void nv::fastCompressDXT5(const Image * image, const nvtt::OutputOptions::Private & outputOptions)
+void nv::FastCompressor::compressDXT5(const nvtt::OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT5 block;
 
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
-			rgba.init(image, x, y);
-			//QuickCompress::compressDXT5(rgba, &block);	// @@ Use fast version!!
-			nv::compressBlock_BoundsRange(rgba, &block);
+			rgba.init(m_image, x, y);
+			
+			QuickCompress::compressDXT5(rgba, &block, 0);
 			
 			if (outputOptions.outputHandler != NULL) {
 				outputOptions.outputHandler->writeData(&block, sizeof(block));
@@ -159,23 +159,21 @@ void nv::fastCompressDXT5(const Image * image, const nvtt::OutputOptions::Privat
 }
 
 
-void nv::fastCompressDXT5n(const Image * image, const nvtt::OutputOptions::Private & outputOptions)
+void nv::FastCompressor::compressDXT5n(const nvtt::OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT5 block;
 
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 			
-			// copy X coordinate to alpha channel and Y coordinate to green channel.
 			rgba.swizzleDXT5n();
 
-			//QuickCompress::compressDXT5(rgba, &block);	// @@ Use fast version!!
-			nv::compressBlock_BoundsRange(rgba, &block);
+			QuickCompress::compressDXT5(rgba, &block, 0);
 			
 			if (outputOptions.outputHandler != NULL) {
 				outputOptions.outputHandler->writeData(&block, sizeof(block));
@@ -185,59 +183,45 @@ void nv::fastCompressDXT5n(const Image * image, const nvtt::OutputOptions::Priva
 }
 
 
-void nv::fastCompressBC4(const Image * image, const nvtt::OutputOptions::Private & outputOptions)
+nv::SlowCompressor::SlowCompressor() : m_image(NULL), m_alphaMode(AlphaMode_None)
 {
-	// @@ TODO
-	// compress red channel (X)
 }
 
-
-void nv::fastCompressBC5(const Image * image, const nvtt::OutputOptions::Private & outputOptions)
+nv::SlowCompressor::~SlowCompressor()
 {
-	// @@ TODO
-	// compress red, green channels (X,Y)
 }
 
-
-void nv::doPrecomputation()
+void nv::SlowCompressor::setImage(const Image * image, nvtt::AlphaMode alphaMode)
 {
-	static bool done = false;	// @@ Stop using statics for reentrancy. Although the worst that could happen is that this stuff is precomputed multiple times.
-	
-	if (!done)
-	{
-		done = true;
-		squish::FastClusterFit::DoPrecomputation();
-	}
+	m_image = image;
+	m_alphaMode = alphaMode;
 }
 
-
-void nv::compressDXT1(const Image * image, const OutputOptions::Private & outputOptions, const CompressionOptions::Private & compressionOptions)
+void nv::SlowCompressor::compressDXT1(const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT1 block;
 
-	doPrecomputation();
-
-	//squish::WeightedClusterFit fit;
+	squish::WeightedClusterFit fit;
 	//squish::ClusterFit fit;
-	squish::FastClusterFit fit;
+	//squish::FastClusterFit fit;
 	fit.SetMetric(compressionOptions.colorWeight.x(), compressionOptions.colorWeight.y(), compressionOptions.colorWeight.z());
 
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
 			
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 			
 			if (rgba.isSingleColor())
 			{
-				QuickCompress::compressDXT1(rgba.color(0), &block);
+				OptimalCompress::compressDXT1(rgba.color(0), &block);
 			}
 			else
 			{
-				squish::ColourSet colours((uint8 *)rgba.colors(), 0);
+				squish::ColourSet colours((uint8 *)rgba.colors(), 0, true);
 				fit.SetColourSet(&colours, squish::kDxt1);
 				fit.Compress(&block);
 			}
@@ -250,10 +234,10 @@ void nv::compressDXT1(const Image * image, const OutputOptions::Private & output
 }
 
 
-void nv::compressDXT1a(const Image * image, const OutputOptions::Private & outputOptions, const CompressionOptions::Private & compressionOptions)
+void nv::SlowCompressor::compressDXT1a(const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT1 block;
@@ -264,11 +248,20 @@ void nv::compressDXT1a(const Image * image, const OutputOptions::Private & outpu
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
 			
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 			
-			if (rgba.isSingleColor())
+			bool anyAlpha = false;
+			bool allAlpha = true;
+			
+			for (uint i = 0; i < 16; i++)
 			{
-				QuickCompress::compressDXT1a(rgba.color(0), &block);
+				if (rgba.color(i).a < 128) anyAlpha = true;
+				else allAlpha = false;
+			}
+			
+			if ((!anyAlpha && rgba.isSingleColor() || allAlpha))
+			{
+				OptimalCompress::compressDXT1a(rgba.color(0), &block);
 			}
 			else
 			{
@@ -285,29 +278,37 @@ void nv::compressDXT1a(const Image * image, const OutputOptions::Private & outpu
 }
 
 
-void nv::compressDXT3(const Image * image, const OutputOptions::Private & outputOptions, const CompressionOptions::Private & compressionOptions)
+void nv::SlowCompressor::compressDXT3(const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT3 block;
 	
 	squish::WeightedClusterFit fit;
+	//squish::FastClusterFit fit;
 	fit.SetMetric(compressionOptions.colorWeight.x(), compressionOptions.colorWeight.y(), compressionOptions.colorWeight.z());
 
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
 			
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 			
 			// Compress explicit alpha.
-			QuickCompress::compressDXT3A(rgba, &block.alpha);
-			
+			OptimalCompress::compressDXT3A(rgba, &block.alpha);
+
 			// Compress color.
-			squish::ColourSet colours((uint8 *)rgba.colors(), squish::kWeightColourByAlpha);
-			fit.SetColourSet(&colours, 0);
-			fit.Compress(&block.color);
+			if (rgba.isSingleColor())
+			{
+				OptimalCompress::compressDXT1(rgba.color(0), &block.color);
+			}
+			else
+			{
+				squish::ColourSet colours((uint8 *)rgba.colors(), squish::kWeightColourByAlpha);
+				fit.SetColourSet(&colours, 0);
+				fit.Compress(&block.color);
+			}
 			
 			if (outputOptions.outputHandler != NULL) {
 				outputOptions.outputHandler->writeData(&block, sizeof(block));
@@ -316,10 +317,10 @@ void nv::compressDXT3(const Image * image, const OutputOptions::Private & output
 	}
 }
 
-void nv::compressDXT5(const Image * image, const OutputOptions::Private & outputOptions, const CompressionOptions::Private & compressionOptions)
+void nv::SlowCompressor::compressDXT5(const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT5 block;
@@ -330,22 +331,29 @@ void nv::compressDXT5(const Image * image, const OutputOptions::Private & output
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
 			
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 
 			// Compress alpha.
 			if (compressionOptions.quality == Quality_Highest)
 			{
-				compressBlock_BruteForce(rgba, &block.alpha);
+				OptimalCompress::compressDXT5A(rgba, &block.alpha);
 			}
 			else
 			{
 				QuickCompress::compressDXT5A(rgba, &block.alpha);
 			}
-
+		
 			// Compress color.
-			squish::ColourSet colours((uint8 *)rgba.colors(), squish::kWeightColourByAlpha);
-			fit.SetColourSet(&colours, 0);
-			fit.Compress(&block.color);
+			if (rgba.isSingleColor())
+			{
+				OptimalCompress::compressDXT1(rgba.color(0), &block.color);
+			}
+			else
+			{
+				squish::ColourSet colours((uint8 *)rgba.colors(), squish::kWeightColourByAlpha);
+				fit.SetColourSet(&colours, 0);
+				fit.Compress(&block.color);
+			}
 			
 			if (outputOptions.outputHandler != NULL) {
 				outputOptions.outputHandler->writeData(&block, sizeof(block));
@@ -355,28 +363,25 @@ void nv::compressDXT5(const Image * image, const OutputOptions::Private & output
 }
 
 
-void nv::compressDXT5n(const Image * image, const OutputOptions::Private & outputOptions, const CompressionOptions::Private & compressionOptions)
+void nv::SlowCompressor::compressDXT5n(const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	BlockDXT5 block;
 	
-	doPrecomputation();
-
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
 			
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 			
-			// copy X coordinate to green channel and Y coordinate to alpha channel.
 			rgba.swizzleDXT5n();			
 			
 			// Compress X.
 			if (compressionOptions.quality == Quality_Highest)
 			{
-				compressBlock_BruteForce(rgba, &block.alpha);
+				OptimalCompress::compressDXT5A(rgba, &block.alpha);
 			}
 			else
 			{
@@ -384,7 +389,7 @@ void nv::compressDXT5n(const Image * image, const OutputOptions::Private & outpu
 			}
 			
 			// Compress Y.
-			QuickCompress::compressDXT1G(rgba, &block.color);
+			OptimalCompress::compressDXT1G(rgba, &block.color);
 			
 			if (outputOptions.outputHandler != NULL) {
 				outputOptions.outputHandler->writeData(&block, sizeof(block));
@@ -394,10 +399,10 @@ void nv::compressDXT5n(const Image * image, const OutputOptions::Private & outpu
 }
 
 
-void nv::compressBC4(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const CompressionOptions::Private & compressionOptions)
+void nv::SlowCompressor::compressBC4(const CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 	
 	ColorBlock rgba;
 	AlphaBlockDXT5 block;
@@ -405,11 +410,11 @@ void nv::compressBC4(const Image * image, const nvtt::OutputOptions::Private & o
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
 			
-			rgba.init(image, x, y);
+			rgba.init(m_image, x, y);
 
 			if (compressionOptions.quality == Quality_Highest)
 			{
-				compressBlock_BruteForce(rgba, &block);
+				OptimalCompress::compressDXT5A(rgba, &block);
 			}
 			else
 			{
@@ -424,10 +429,10 @@ void nv::compressBC4(const Image * image, const nvtt::OutputOptions::Private & o
 }
 
 
-void nv::compressBC5(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const CompressionOptions::Private & compressionOptions)
+void nv::SlowCompressor::compressBC5(const CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions)
 {
-	const uint w = image->width();
-	const uint h = image->height();
+	const uint w = m_image->width();
+	const uint h = m_image->height();
 
 	ColorBlock xcolor;
 	ColorBlock ycolor;
@@ -437,16 +442,16 @@ void nv::compressBC5(const Image * image, const nvtt::OutputOptions::Private & o
 	for (uint y = 0; y < h; y += 4) {
 		for (uint x = 0; x < w; x += 4) {
 			
-			xcolor.init(image, x, y);
+			xcolor.init(m_image, x, y);
 			xcolor.splatX();
 			
-			ycolor.init(image, x, y);
+			ycolor.init(m_image, x, y);
 			ycolor.splatY();
 
 			if (compressionOptions.quality == Quality_Highest)
 			{
-				compressBlock_BruteForce(xcolor, &block.x);
-				compressBlock_BruteForce(ycolor, &block.y);
+				OptimalCompress::compressDXT5A(xcolor, &block.x);
+				OptimalCompress::compressDXT5A(ycolor, &block.y);
 			}
 			else
 			{

src/nvtt/CompressDXT.h

@@ -32,26 +32,46 @@ namespace nv
 	class Image;
 	class FloatImage;
 
-	void doPrecomputation();
-	
-	// Fast compressors.
-	void fastCompressDXT1(const Image * image, const nvtt::OutputOptions::Private & outputOptions);
-	void fastCompressDXT1a(const Image * image, const nvtt::OutputOptions::Private & outputOptions);
-	void fastCompressDXT3(const Image * image, const nvtt::OutputOptions::Private & outputOptions);
-	void fastCompressDXT5(const Image * image, const nvtt::OutputOptions::Private & outputOptions);
-	void fastCompressDXT5n(const Image * image, const nvtt::OutputOptions::Private & outputOptions);
-	void fastCompressBC4(const Image * image, const nvtt::OutputOptions::Private & outputOptions);
-	void fastCompressBC5(const Image * image, const nvtt::OutputOptions::Private & outputOptions);
+	class FastCompressor
+	{
+	public:
+		FastCompressor();
+		~FastCompressor();
+
+		void setImage(const Image * image, nvtt::AlphaMode alphaMode);
+
+		void compressDXT1(const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT1a(const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT3(const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT5(const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT5n(const nvtt::OutputOptions::Private & outputOptions);
+
+	private:
+		const Image * m_image;
+		nvtt::AlphaMode m_alphaMode;
+	};
+
+	class SlowCompressor
+	{
+	public:
+		SlowCompressor();
+		~SlowCompressor();
+
+		void setImage(const Image * image, nvtt::AlphaMode alphaMode);
+
+		void compressDXT1(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT1a(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT3(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT5(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT5n(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void compressBC4(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void compressBC5(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+
+	private:
+		const Image * m_image;
+		nvtt::AlphaMode m_alphaMode;
+	};
 
-	// Normal compressors.
-	void compressDXT1(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
-	void compressDXT1a(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
-	void compressDXT3(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
-	void compressDXT5(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
-	void compressDXT5n(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
-	void compressBC4(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
-	void compressBC5(const Image * image, const nvtt::OutputOptions::Private & outputOptions, const nvtt::CompressionOptions::Private & compressionOptions);
-	
 	// External compressors.
 #if defined(HAVE_S3QUANT)
 	void s3CompressDXT1(const Image * image, const nvtt::OutputOptions::Private & outputOptions);

src/nvtt/CompressRGB.cpp

@@ -123,7 +123,7 @@ void nv::compressRGB(const Image * image, const OutputOptions::Private & outputO
 			}
 			
 			// Zero padding.
-			for (uint x = w; x < pitch; x++)
+			for (uint x = w * byteCount; x < pitch; x++)
 			{
 				*(dst + x) = 0;
 			}

src/nvtt/Compressor.cpp

@@ -34,6 +34,7 @@
 #include <nvimage/Filter.h>
 #include <nvimage/Quantize.h>
 #include <nvimage/NormalMap.h>
+#include <nvimage/PixelFormat.h>
 
 #include "Compressor.h"
 #include "InputOptions.h"
@@ -41,7 +42,6 @@
 #include "OutputOptions.h"
 
 #include "CompressDXT.h"
-#include "FastCompressDXT.h"
 #include "CompressRGB.h"
 #include "cuda/CudaUtils.h"
 #include "cuda/CudaCompressDXT.h"
@@ -53,10 +53,10 @@ using namespace nvtt;
 
 namespace
 {
-	
+
 	static int blockSize(Format format)
 	{
-		if (format == Format_DXT1 || format == Format_DXT1a || format == Format_DXT1n) {
+		if (format == Format_DXT1 || format == Format_DXT1a) {
 			return 8;
 		}
 		else if (format == Format_DXT3) {
@@ -71,9 +71,6 @@ namespace
 		else if (format == Format_BC5) {
 			return 16;
 		}
-		else if (format == Format_CTX1) {
-			return 8;
-		}
 		return 0;
 	}
 
@@ -124,7 +121,7 @@ namespace nvtt
 			m_fixedImage = NULL;
 			m_floatImage = image;
 		}
-		
+
 
 		// Convert linear float image to fixed image ready for compression.
 		void toFixedImage(const InputOptions::Private & inputOptions)
@@ -156,7 +153,7 @@ namespace nvtt
 				if (inputOptions.isNormalMap)
 				{
 					// Expand normals to [-1, 1] range.
-				//	floatImage->expandNormals(0);
+					//	floatImage->expandNormals(0);
 				}
 				else if (inputOptions.inputGamma != 1.0f)
 				{
@@ -196,36 +193,29 @@ namespace nvtt
 			return m_fixedImage.ptr();
 		}
 
-		
+
 	private:
 		const Image * m_inputImage;
 		AutoPtr<Image> m_fixedImage;
 		AutoPtr<FloatImage> m_floatImage;
 	};
 
-}
+} // nvtt namespace
 
 
 Compressor::Compressor() : m(*new Compressor::Private())
 {
 	// CUDA initialization.
 	m.cudaSupported = cuda::isHardwarePresent();
-	m.cudaEnabled = m.cudaSupported;
+	m.cudaEnabled = false;
+	m.cudaDevice = -1;
 
-	if (m.cudaEnabled)
-	{
-		m.cuda = new CudaCompressor();
-
-		if (!m.cuda->isValid())
-		{
-			m.cudaEnabled = false;
-			m.cuda = NULL;
-		}
-	}
+	enableCudaAcceleration(m.cudaSupported);
 }
 
 Compressor::~Compressor()
 {
+	enableCudaAcceleration(false);
 	delete &m;
 }
 
@@ -235,17 +225,33 @@ void Compressor::enableCudaAcceleration(bool enable)
 {
 	if (m.cudaSupported)
 	{
-		m.cudaEnabled = enable;
-	}
-
-	if (m.cudaEnabled && m.cuda == NULL)
-	{
-		m.cuda = new CudaCompressor();
-		
-		if (!m.cuda->isValid())
+		if (m.cudaEnabled && !enable)
 		{
 			m.cudaEnabled = false;
 			m.cuda = NULL;
+
+			if (m.cudaDevice != -1)
+			{
+				// Exit device.
+				cuda::exitDevice();
+			}
+		}
+		else if (!m.cudaEnabled && enable)
+		{
+			// Init the CUDA device. This may return -1 if CUDA was already initialized by the app.
+			m.cudaEnabled = cuda::initDevice(&m.cudaDevice);
+
+			if (m.cudaEnabled)
+			{
+				// Create compressor if initialization succeeds.
+				m.cuda = new CudaCompressor();
+
+				// But cleanup if failed.
+				if (!m.cuda->isValid())
+				{
+					enableCudaAcceleration(false);
+				}
+			}
 		}
 	}
 }
@@ -286,9 +292,9 @@ bool Compressor::Private::compress(const InputOptions::Private & inputOptions, c
 		if (outputOptions.errorHandler) outputOptions.errorHandler->error(Error_FileOpen);
 		return false;
 	}
-	
+
 	inputOptions.computeTargetExtents();
-	
+
 	// Output DDS header.
 	if (!outputHeader(inputOptions, compressionOptions, outputOptions))
 	{
@@ -304,7 +310,7 @@ bool Compressor::Private::compress(const InputOptions::Private & inputOptions, c
 	}
 
 	outputOptions.closeFile();
-	
+
 	return true;
 }
 
@@ -319,15 +325,15 @@ bool Compressor::Private::outputHeader(const InputOptions::Private & inputOption
 	}
 
 	DDSHeader header;
-	
+
 	header.setWidth(inputOptions.targetWidth);
 	header.setHeight(inputOptions.targetHeight);
-	
+
 	int mipmapCount = inputOptions.realMipmapCount();
 	nvDebugCheck(mipmapCount > 0);
-	
+
 	header.setMipmapCount(mipmapCount);
-	
+
 	if (inputOptions.textureType == TextureType_2D) {
 		header.setTexture2D();
 	}
@@ -335,10 +341,10 @@ bool Compressor::Private::outputHeader(const InputOptions::Private & inputOption
 		header.setTextureCube();
 	}		
 	/*else if (inputOptions.textureType == TextureType_3D) {
-		header.setTexture3D();
-		header.setDepth(inputOptions.targetDepth);
+	header.setTexture3D();
+	header.setDepth(inputOptions.targetDepth);
 	}*/
-	
+
 	if (compressionOptions.format == Format_RGBA)
 	{
 		header.setPitch(computePitch(inputOptions.targetWidth, compressionOptions.bitcount));
@@ -347,8 +353,8 @@ bool Compressor::Private::outputHeader(const InputOptions::Private & inputOption
 	else
 	{
 		header.setLinearSize(computeImageSize(inputOptions.targetWidth, inputOptions.targetHeight, inputOptions.targetDepth, compressionOptions.bitcount, compressionOptions.format));
-		
-		if (compressionOptions.format == Format_DXT1 || compressionOptions.format == Format_DXT1a || compressionOptions.format == Format_DXT1n) {
+
+		if (compressionOptions.format == Format_DXT1 || compressionOptions.format == Format_DXT1a) {
 			header.setFourCC('D', 'X', 'T', '1');
 			if (inputOptions.isNormalMap) header.setNormalFlag(true);
 		}
@@ -369,15 +375,11 @@ bool Compressor::Private::outputHeader(const InputOptions::Private & inputOption
 			header.setFourCC('A', 'T', 'I', '2');
 			if (inputOptions.isNormalMap) header.setNormalFlag(true);
 		}
-		else if (compressionOptions.format == Format_CTX1) {
-			header.setFourCC('C', 'T', 'X', '1');
-			if (inputOptions.isNormalMap) header.setNormalFlag(true);
-		}
 	}
-	
+
 	// Swap bytes if necessary.
 	header.swapBytes();
-	
+
 	uint headerSize = 128;
 	if (header.hasDX10Header())
 	{
@@ -390,7 +392,7 @@ bool Compressor::Private::outputHeader(const InputOptions::Private & inputOption
 	{
 		outputOptions.errorHandler->error(Error_FileWrite);
 	}
-	
+
 	return writeSucceed;
 }
 
@@ -426,7 +428,7 @@ bool Compressor::Private::compressMipmaps(uint f, const InputOptions::Private &
 				return false;
 			}
 		}
-		
+
 		quantizeMipmap(mipmap, compressionOptions);
 
 		compressMipmap(mipmap, inputOptions, compressionOptions, outputOptions);
@@ -436,7 +438,7 @@ bool Compressor::Private::compressMipmaps(uint f, const InputOptions::Private &
 		h = max(1U, h / 2);
 		d = max(1U, d / 2);
 	}
-	
+
 	return true;
 }
 
@@ -478,11 +480,6 @@ bool Compressor::Private::initMipmap(Mipmap & mipmap, const InputOptions::Privat
 	// Convert linear float image to fixed image ready for compression.
 	mipmap.toFixedImage(inputOptions);
 
-	if (inputOptions.premultiplyAlpha)
-	{
-		premultiplyAlphaMipmap(mipmap, inputOptions);
-	}
-
 	return true;
 }
 
@@ -492,7 +489,7 @@ int Compressor::Private::findExactMipmap(const InputOptions::Private & inputOpti
 	{
 		int idx = f * inputOptions.mipmapCount + m;
 		const InputOptions::Private::InputImage & inputImage = inputOptions.images[idx];
-		
+
 		if (inputImage.width == int(w) && inputImage.height == int(h) && inputImage.depth == int(d))
 		{
 			if (inputImage.data != NULL)
@@ -547,7 +544,7 @@ void Compressor::Private::downsampleMipmap(Mipmap & mipmap, const InputOptions::
 	mipmap.toFloatImage(inputOptions);
 
 	const FloatImage * floatImage = mipmap.asFloatImage();
-	
+
 	if (inputOptions.mipmapFilter == MipmapFilter_Box)
 	{
 		// Use fast downsample.
@@ -565,7 +562,7 @@ void Compressor::Private::downsampleMipmap(Mipmap & mipmap, const InputOptions::
 		filter.setParameters(inputOptions.kaiserAlpha, inputOptions.kaiserStretch);
 		mipmap.setImage(floatImage->downSample(filter, (FloatImage::WrapMode)inputOptions.wrapMode));
 	}
-	
+
 	// Normalize mipmap.
 	if ((inputOptions.isNormalMap || inputOptions.convertToNormalMap) && inputOptions.normalizeMipmaps)
 	{
@@ -583,40 +580,17 @@ void Compressor::Private::scaleMipmap(Mipmap & mipmap, const InputOptions::Priva
 
 	// Resize image. 
 	BoxFilter boxFilter;
-	mipmap.setImage(mipmap.asFloatImage()->downSample(boxFilter, w, h, (FloatImage::WrapMode)inputOptions.wrapMode));
+	mipmap.setImage(mipmap.asFloatImage()->resize(boxFilter, w, h, (FloatImage::WrapMode)inputOptions.wrapMode));
 }
 
 
-void Compressor::Private::premultiplyAlphaMipmap(Mipmap & mipmap, const InputOptions::Private & inputOptions) const
-{
-	nvDebugCheck(mipmap.asFixedImage() != NULL);
-
-	Image * image = mipmap.asMutableFixedImage();
-
-	const uint w = image->width();
-	const uint h = image->height();
-
-	const uint count = w * h;
-
-	for (uint i = 0; i < count; ++i)
-	{
-		Color32 c = image->pixel(i);
-
-		c.r = (uint(c.r) * uint(c.a)) >> 8;
-		c.g = (uint(c.g) * uint(c.a)) >> 8;
-		c.b = (uint(c.b) * uint(c.a)) >> 8;
-
-		image->pixel(i) = c;
-	}
-}
-
 // Process an input image: Convert to normal map, normalize, or convert to linear space.
 void Compressor::Private::processInputImage(Mipmap & mipmap, const InputOptions::Private & inputOptions) const
 {
 	if (inputOptions.convertToNormalMap)
 	{
 		mipmap.toFixedImage(inputOptions);
-		
+
 		Vector4 heightScale = inputOptions.heightFactors;
 		mipmap.setImage(createNormalMap(mipmap.asFixedImage(), (FloatImage::WrapMode)inputOptions.wrapMode, heightScale, inputOptions.bumpFrequencyScale));
 	}
@@ -653,13 +627,6 @@ void Compressor::Private::quantizeMipmap(Mipmap & mipmap, const CompressionOptio
 {
 	nvDebugCheck(mipmap.asFixedImage() != NULL);
 
-	if (compressionOptions.enableColorDithering)
-	{
-		if (compressionOptions.format >= Format_DXT1 && compressionOptions.format <= Format_DXT5)
-		{
-			Quantize::FloydSteinberg_RGB16(mipmap.asMutableFixedImage());
-		}
-	}
 	if (compressionOptions.binaryAlpha)
 	{
 		if (compressionOptions.enableAlphaDithering)
@@ -671,19 +638,50 @@ void Compressor::Private::quantizeMipmap(Mipmap & mipmap, const CompressionOptio
 			Quantize::BinaryAlpha(mipmap.asMutableFixedImage(), compressionOptions.alphaThreshold);
 		}
 	}
-	else
+
+	if (compressionOptions.enableColorDithering || compressionOptions.enableAlphaDithering)
 	{
+		uint rsize = 8;
+		uint gsize = 8;
+		uint bsize = 8;
+		uint asize = 8;
+
+		if (compressionOptions.enableColorDithering)
+		{
+			if (compressionOptions.format >= Format_DXT1 && compressionOptions.format <= Format_DXT5)
+			{
+				rsize = 5;
+				gsize = 6;
+				bsize = 5;
+			}
+			else if (compressionOptions.format == Format_RGB)
+			{
+				uint rshift, gshift, bshift;
+				PixelFormat::maskShiftAndSize(compressionOptions.rmask, &rshift, &rsize);
+				PixelFormat::maskShiftAndSize(compressionOptions.gmask, &gshift, &gsize);
+				PixelFormat::maskShiftAndSize(compressionOptions.bmask, &bshift, &bsize);
+			}
+		}
+
 		if (compressionOptions.enableAlphaDithering)
 		{
 			if (compressionOptions.format == Format_DXT3)
 			{
-				Quantize::Alpha4(mipmap.asMutableFixedImage());
+				asize = 4;
 			}
-			else if (compressionOptions.format == Format_DXT1a)
+			else if (compressionOptions.format == Format_RGB)
 			{
-				Quantize::BinaryAlpha(mipmap.asMutableFixedImage(), compressionOptions.alphaThreshold);
+				uint ashift;
+				PixelFormat::maskShiftAndSize(compressionOptions.amask, &ashift, &asize);
 			}
 		}
+
+		if (compressionOptions.binaryAlpha)
+		{
+			asize = 8; // Already quantized.
+		}
+
+		Quantize::FloydSteinberg(mipmap.asMutableFixedImage(), rsize, gsize, bsize, asize);
 	}
 }
 
@@ -692,9 +690,16 @@ void Compressor::Private::quantizeMipmap(Mipmap & mipmap, const CompressionOptio
 bool Compressor::Private::compressMipmap(const Mipmap & mipmap, const InputOptions::Private & inputOptions, const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions) const
 {
 	const Image * image = mipmap.asFixedImage();
-
 	nvDebugCheck(image != NULL);
 
+	FastCompressor fast;
+	fast.setImage(image, inputOptions.alphaMode);
+
+	SlowCompressor slow;
+	slow.setImage(image, inputOptions.alphaMode);
+
+	const bool useCuda = cudaEnabled && image->width() * image->height() >= 512;
+
 	if (compressionOptions.format == Format_RGBA || compressionOptions.format == Format_RGB)
 	{
 		compressRGB(image, outputOptions, compressionOptions);
@@ -710,76 +715,66 @@ bool Compressor::Private::compressMipmap(const Mipmap & mipmap, const InputOptio
 #endif
 
 #if defined(HAVE_ATITC)
-		if (compressionOptions.externalCompressor == "ati")
-		{
-			atiCompressDXT1(image, outputOptions);
-		}
-		else
-#endif
-		if (compressionOptions.quality == Quality_Fastest)
-		{
-			fastCompressDXT1(image, outputOptions);
-		}
-		else
-		{
-			if (cudaEnabled)
+			if (compressionOptions.externalCompressor == "ati")
 			{
-				nvDebugCheck(cudaSupported);
-				cuda->compressDXT1(image, compressionOptions, outputOptions);
+				atiCompressDXT1(image, outputOptions);
 			}
 			else
-			{
-				compressDXT1(image, outputOptions, compressionOptions);
-			}
-		}
+#endif
+				if (compressionOptions.quality == Quality_Fastest)
+				{
+					fast.compressDXT1(outputOptions);
+				}
+				else
+				{
+					if (useCuda)
+					{
+						nvDebugCheck(cudaSupported);
+						cuda->setImage(image, inputOptions.alphaMode);
+						cuda->compressDXT1(compressionOptions, outputOptions);
+					}
+					else
+					{
+						slow.compressDXT1(compressionOptions, outputOptions);
+					}
+				}
 	}
 	else if (compressionOptions.format == Format_DXT1a)
 	{
 		if (compressionOptions.quality == Quality_Fastest)
 		{
-			fastCompressDXT1a(image, outputOptions);
+			fast.compressDXT1a(outputOptions);
 		}
 		else
 		{
-			if (cudaEnabled)
+			if (useCuda)
 			{
 				nvDebugCheck(cudaSupported);
-				/*cuda*/compressDXT1a(image, outputOptions, compressionOptions);
+				/*cuda*/slow.compressDXT1a(compressionOptions, outputOptions);
 			}
 			else
 			{
-				compressDXT1a(image, outputOptions, compressionOptions);
+				slow.compressDXT1a(compressionOptions, outputOptions);
 			}
 		}
 	}
-	else if (compressionOptions.format == Format_DXT1n)
-	{
-		if (cudaEnabled)
-		{
-			nvDebugCheck(cudaSupported);
-			cuda->compressDXT1n(image, compressionOptions, outputOptions);
-		}
-		else
-		{
-			if (outputOptions.errorHandler) outputOptions.errorHandler->error(Error_UnsupportedFeature);
-		}
-	}
 	else if (compressionOptions.format == Format_DXT3)
 	{
 		if (compressionOptions.quality == Quality_Fastest)
 		{
-			fastCompressDXT3(image, outputOptions);
+			fast.compressDXT3(outputOptions);
 		}
 		else
 		{
-			if (cudaEnabled)
+			if (useCuda)
 			{
 				nvDebugCheck(cudaSupported);
-				cuda->compressDXT3(image, inputOptions, compressionOptions, outputOptions);
+				cuda->setImage(image, inputOptions.alphaMode);
+				cuda->compressDXT3(compressionOptions, outputOptions);
 			}
 			else
 			{
-				compressDXT3(image, outputOptions, compressionOptions);
+				slow.compressDXT3(compressionOptions, outputOptions);
 			}
 		}
 	}
@@ -787,18 +782,19 @@ bool Compressor::Private::compressMipmap(const Mipmap & mipmap, const InputOptio
 	{
 		if (compressionOptions.quality == Quality_Fastest)
 		{
-			fastCompressDXT5(image, outputOptions);
+			fast.compressDXT5(outputOptions);
 		}
 		else
 		{
-			if (cudaEnabled)
+			if (useCuda)
 			{
 				nvDebugCheck(cudaSupported);
-				cuda->compressDXT5(image, inputOptions, compressionOptions, outputOptions);
+				cuda->setImage(image, inputOptions.alphaMode);
+				cuda->compressDXT5(compressionOptions, outputOptions);
 			}
 			else
 			{
-				compressDXT5(image, outputOptions, compressionOptions);
+				slow.compressDXT5(compressionOptions, outputOptions);
 			}
 		}
 	}
@@ -806,32 +802,20 @@ bool Compressor::Private::compressMipmap(const Mipmap & mipmap, const InputOptio
 	{
 		if (compressionOptions.quality == Quality_Fastest)
 		{
-			fastCompressDXT5n(image, outputOptions);
+			fast.compressDXT5n(outputOptions);
 		}
 		else
 		{
-			compressDXT5n(image, outputOptions, compressionOptions);
+			slow.compressDXT5n(compressionOptions, outputOptions);
 		}
 	}
 	else if (compressionOptions.format == Format_BC4)
 	{
-		compressBC4(image, outputOptions, compressionOptions);
+		slow.compressBC4(compressionOptions, outputOptions);
 	}
 	else if (compressionOptions.format == Format_BC5)
 	{
-		compressBC5(image, outputOptions, compressionOptions);
-	}
-	else if (compressionOptions.format == Format_CTX1)
-	{
-		if (cudaEnabled)
-		{
-			nvDebugCheck(cudaSupported);
-			cuda->compressCTX1(image, compressionOptions, outputOptions);
-		}
-		else
-		{
-			if (outputOptions.errorHandler) outputOptions.errorHandler->error(Error_UnsupportedFeature);
-		}
+		slow.compressBC5(compressionOptions, outputOptions);
 	}
 
 	return true;
@@ -844,27 +828,27 @@ int Compressor::Private::estimateSize(const InputOptions::Private & inputOptions
 	const uint bitCount = compressionOptions.bitcount;
 
 	inputOptions.computeTargetExtents();
-	
+
 	uint mipmapCount = inputOptions.realMipmapCount();
-	
+
 	int size = 0;
-	
+
 	for (uint f = 0; f < inputOptions.faceCount; f++)
 	{
 		uint w = inputOptions.targetWidth;
 		uint h = inputOptions.targetHeight;
 		uint d = inputOptions.targetDepth;
-		
+
 		for (uint m = 0; m < mipmapCount; m++)
 		{
 			size += computeImageSize(w, h, d, bitCount, format);
-			
+
 			// Compute extents of next mipmap:
 			w = max(1U, w / 2);
 			h = max(1U, h / 2);
 			d = max(1U, d / 2);
 		}
 	}
-	
+
 	return size;
 }

src/nvtt/Compressor.h

@@ -58,16 +58,17 @@ namespace nvtt
 
 		void downsampleMipmap(Mipmap & mipmap, const InputOptions::Private & inputOptions) const;
 		void scaleMipmap(Mipmap & mipmap, const InputOptions::Private & inputOptions, uint w, uint h, uint d) const;
-		void premultiplyAlphaMipmap(Mipmap & mipmap, const InputOptions::Private & inputOptions) const;
 		void processInputImage(Mipmap & mipmap, const InputOptions::Private & inputOptions) const;
 		void quantizeMipmap(Mipmap & mipmap, const CompressionOptions::Private & compressionOptions) const;
 		bool compressMipmap(const Mipmap & mipmap, const InputOptions::Private & inputOptions, const CompressionOptions::Private & compressionOptions, const OutputOptions::Private & outputOptions) const;
 
 
+
 	public:
 
 		bool cudaSupported;
 		bool cudaEnabled;
+		int cudaDevice;
 
 		nv::AutoPtr<nv::CudaCompressor> cuda;
 

src/nvtt/FastCompressDXT.cpp

@@ -1,456 +0,0 @@
-// Copyright NVIDIA Corporation 2007 -- Ignacio Castano <icastano@nvidia.com>
-// 
-// Permission is hereby granted, free of charge, to any person
-// obtaining a copy of this software and associated documentation
-// files (the "Software"), to deal in the Software without
-// restriction, including without limitation the rights to use,
-// copy, modify, merge, publish, distribute, sublicense, and/or sell
-// copies of the Software, and to permit persons to whom the
-// Software is furnished to do so, subject to the following
-// conditions:
-// 
-// The above copyright notice and this permission notice shall be
-// included in all copies or substantial portions of the Software.
-// 
-// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
-// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
-// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
-// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
-// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
-// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
-// OTHER DEALINGS IN THE SOFTWARE.
-
-#include <nvmath/Color.h>
-#include <nvimage/ColorBlock.h>
-#include <nvimage/BlockDXT.h>
-
-#include "FastCompressDXT.h"
-
-#if defined(__SSE2__)
-#include <emmintrin.h>
-#endif
-
-#if defined(__SSE__)
-#include <xmmintrin.h>
-#endif
-
-#if defined(__MMX__)
-#include <mmintrin.h>
-#endif
-
-#undef __VEC__
-#if defined(__VEC__)
-#include <altivec.h>
-#undef bool
-#endif
-// Online Resources:
-// - http://www.jasondorie.com/ImageLib.zip
-// - http://homepage.hispeed.ch/rscheidegger/dri_experimental/s3tc_index.html
-// - http://www.sjbrown.co.uk/?article=dxt
-
-using namespace nv;
-
-
-#if defined(__SSE2__) && 0
-
-// @@ TODO
-
-typedef __m128i VectorColor;
-
-inline static __m128i loadColor(Color32 c)
-{
-	return ...;
-}
-
-inline static __m128i absoluteDifference(__m128i a, __m128i b)
-{
-	return ...;
-}
-	
-inline uint colorDistance(__m128i a, __m128i b)
-{
-	return 0;
-}
-
-#elif defined(__MMX__) && 0
-
-typedef __m64 VectorColor;
-
-inline static __m64 loadColor(Color32 c)
-{
-	return _mm_unpacklo_pi8(_mm_cvtsi32_si64(c), _mm_setzero_si64());
-}
-
-inline static __m64 absoluteDifference(__m64 a, __m64 b)
-{
-	// = |a-b| or |b-a|
-	return _mm_or_si64(_mm_subs_pu16(a, b), _mm_subs_pu16(b, a));
-}
-	
-inline uint colorDistance(__m64 a, __m64 b)
-{
-	union {
-		__m64 v;
-		uint16 part[4];
-	} s;
-	
-	s.v = absoluteDifference(a, b);
-		
-	// @@ This is very slow!
-	return s.part[0] + s.part[1] + s.part[2] + s.part[3];
-}
-
-#define vectorEnd	_mm_empty
-
-#elif defined(__VEC__)
-
-typedef vector signed int VectorColor;
-
-inline static vector signed int loadColor(Color32 c)
-{
-	return (vector signed int) (c.r, c.g, c.b, c.a);
-}
-
-// Get the absolute distance between the given colors.
-inline static uint colorDistance(vector signed int c0, vector signed int c1)
-{
-	int result;
-	vector signed int v = vec_sums(vec_abs(vec_sub(c0, c1)), (vector signed int)0);
-	vec_ste(vec_splat(v, 3), 0, &result);
-	return result;
-}
-
-inline void vectorEnd()
-{
-}
-
-#else
-
-typedef Color32 VectorColor;
-
-inline static Color32 loadColor(Color32 c)
-{
-	return c;
-}
-
-inline static Color32 premultiplyAlpha(Color32 c)
-{
-	Color32 pm;
-	pm.r = (c.r * c.a) >> 8;
-	pm.g = (c.g * c.a) >> 8;
-	pm.b = (c.b * c.a) >> 8;
-	pm.a = c.a;
-	return pm;
-}
-
-inline static uint sqr(uint s)
-{
-	return s*s;
-}
-
-// Get the absolute distance between the given colors.
-inline static uint colorDistance(Color32 c0, Color32 c1)
-{
-	return sqr(c0.r - c1.r) + sqr(c0.g - c1.g) + sqr(c0.b - c1.b);
-	//return abs(c0.r - c1.r) + abs(c0.g - c1.g) + abs(c0.b - c1.b);
-}
-
-inline void vectorEnd()
-{
-}
-
-#endif
-
-
-inline static uint computeIndices(const ColorBlock & rgba, const Color32 palette[4])
-{
-	const VectorColor vcolor0 = loadColor(palette[0]);
-	const VectorColor vcolor1 = loadColor(palette[1]);
-	const VectorColor vcolor2 = loadColor(palette[2]);
-	const VectorColor vcolor3 = loadColor(palette[3]);
-	
-	uint indices = 0;
-	for(int i = 0; i < 16; i++) {
-		const VectorColor vcolor = loadColor(rgba.color(i));
-		
-		uint d0 = colorDistance(vcolor0, vcolor);
-		uint d1 = colorDistance(vcolor1, vcolor);
-		uint d2 = colorDistance(vcolor2, vcolor);
-		uint d3 = colorDistance(vcolor3, vcolor);
-		
-		uint b0 = d0 > d3;
-		uint b1 = d1 > d2;
-		uint b2 = d0 > d2;
-		uint b3 = d1 > d3;
-		uint b4 = d2 > d3;
-		
-		uint x0 = b1 & b2;
-		uint x1 = b0 & b3;
-		uint x2 = b0 & b4;
-		
-		indices |= (x2 | ((x0 | x1) << 1)) << (2 * i);
-	}
-
-	vectorEnd();
-	return indices;
-}
-
-
-// Compressor that uses bounding box.
-void nv::compressBlock_BoundsRange(const ColorBlock & rgba, BlockDXT1 * block)
-{
-	Color32 c0, c1;
-	rgba.boundsRange(&c1, &c0);
-	
-	block->col0 = toColor16(c0);
-	block->col1 = toColor16(c1);
-	
-	nvDebugCheck(block->col0.u > block->col1.u);
-	
-	// Use 4 color mode only.
-	//if (block->col0.u < block->col1.u) {
-	//	swap(block->col0.u, block->col1.u);
-	//}
-	
-	Color32 palette[4];
-	block->evaluatePalette4(palette);
-	
-	block->indices = computeIndices(rgba, palette);
-}
-
-
-
-
-
-
-
-
-// Encode DXT3 block.
-void nv::compressBlock_BoundsRange(const ColorBlock & rgba, BlockDXT3 * block)
-{
-	compressBlock_BoundsRange(rgba, &block->color);
-	compressBlock(rgba, &block->alpha);
-}
-
-// Encode DXT3 alpha block.
-void nv::compressBlock(const ColorBlock & rgba, AlphaBlockDXT3 * block)
-{
-	block->alpha0 = rgba.color(0).a >> 4;
-	block->alpha1 = rgba.color(1).a >> 4;
-	block->alpha2 = rgba.color(2).a >> 4;
-	block->alpha3 = rgba.color(3).a >> 4;
-	block->alpha4 = rgba.color(4).a >> 4;
-	block->alpha5 = rgba.color(5).a >> 4;
-	block->alpha6 = rgba.color(6).a >> 4;
-	block->alpha7 = rgba.color(7).a >> 4;
-	block->alpha8 = rgba.color(8).a >> 4;
-	block->alpha9 = rgba.color(9).a >> 4;
-	block->alphaA = rgba.color(10).a >> 4;
-	block->alphaB = rgba.color(11).a >> 4;
-	block->alphaC = rgba.color(12).a >> 4;
-	block->alphaD = rgba.color(13).a >> 4;
-	block->alphaE = rgba.color(14).a >> 4;
-	block->alphaF = rgba.color(15).a >> 4;
-}
-
-
-
-static uint computeAlphaIndices(const ColorBlock & rgba, AlphaBlockDXT5 * block)
-{
-	uint8 alphas[8];
-	block->evaluatePalette(alphas);
-
-	uint totalError = 0;
-
-	for (uint i = 0; i < 16; i++)
-	{
-		uint8 alpha = rgba.color(i).a;
-
-		uint besterror = 256*256;
-		uint best = 8;
-		for(uint p = 0; p < 8; p++)
-		{
-			int d = alphas[p] - alpha;
-			uint error = d * d;
-
-			if (error < besterror)
-			{
-				besterror = error;
-				best = p;
-			}
-		}
-		nvDebugCheck(best < 8);
-
-		totalError += besterror;
-		block->setIndex(i, best);
-	}
-
-	return totalError;
-}
-
-static uint computeAlphaError(const ColorBlock & rgba, const AlphaBlockDXT5 * block)
-{
-	uint8 alphas[8];
-	block->evaluatePalette(alphas);
-
-	uint totalError = 0;
-
-	for (uint i = 0; i < 16; i++)
-	{
-		uint8 alpha = rgba.color(i).a;
-
-		uint besterror = 256*256;
-		uint best;
-		for(uint p = 0; p < 8; p++)
-		{
-			int d = alphas[p] - alpha;
-			uint error = d * d;
-
-			if (error < besterror)
-			{
-				besterror = error;
-				best = p;
-			}
-		}
-
-		totalError += besterror;
-	}
-
-	return totalError;
-}
-
-
-void nv::compressBlock_BoundsRange(const ColorBlock & rgba, BlockDXT5 * block)
-{
-	Color32 c0, c1;
-	rgba.boundsRangeAlpha(&c1, &c0);
-	
-	block->color.col0 = toColor16(c0);
-	block->color.col1 = toColor16(c1);
-	
-	nvDebugCheck(block->color.col0.u > block->color.col1.u);
-	
-	Color32 palette[4];
-	block->color.evaluatePalette4(palette);
-	
-	block->color.indices = computeIndices(rgba, palette);
-	
-	nvDebugCheck(c0.a <= c1.a);
-	
-	block->alpha.alpha0 = c0.a;
-	block->alpha.alpha1 = c1.a;
-	
-	computeAlphaIndices(rgba, &block->alpha);
-}
-
-
-uint nv::compressBlock_BoundsRange(const ColorBlock & rgba, AlphaBlockDXT5 * block)
-{
-	uint8 alpha0 = 0;
-	uint8 alpha1 = 255;
-
-	// Get min/max alpha.
-	for (uint i = 0; i < 16; i++)
-	{
-		uint8 alpha = rgba.color(i).a;
-		alpha0 = max(alpha0, alpha);
-		alpha1 = min(alpha1, alpha);
-	}
-
-	alpha0 = alpha0 - (alpha0 - alpha1) / 32;
-	alpha1 = alpha1 + (alpha0 - alpha1) / 32;
-
-	AlphaBlockDXT5 block0;
-	block0.alpha0 = alpha0;
-	block0.alpha1 = alpha1;
-	uint error0 = computeAlphaIndices(rgba, &block0);
-
-	AlphaBlockDXT5 block1;
-	block1.alpha0 = alpha1;
-	block1.alpha1 = alpha0;
-	uint error1 = computeAlphaIndices(rgba, &block1);
-
-	if (error0 < error1)
-	{
-		*block = block0;
-		return error0;
-	}
-	else
-	{
-		*block = block1;
-		return error1;
-	}
-}
-
-uint nv::compressBlock_BruteForce(const ColorBlock & rgba, AlphaBlockDXT5 * block)
-{
-	uint8 mina = 255;
-	uint8 maxa = 0;
-
-	// Get min/max alpha.
-	for (uint i = 0; i < 16; i++)
-	{
-		uint8 alpha = rgba.color(i).a;
-		mina = min(mina, alpha);
-		maxa = max(maxa, alpha);
-	}
-
-	block->alpha0 = maxa;
-	block->alpha1 = mina;
-
-	/*int centroidDist = 256;
-	int centroid;
-
-	// Get the closest to the centroid.
-	for (uint i = 0; i < 16; i++)
-	{
-		uint8 alpha = rgba.color(i).a;
-		int dist = abs(alpha - (maxa + mina) / 2);
-		if (dist < centroidDist)
-		{
-			centroidDist = dist;
-			centroid = alpha;
-		}
-	}*/
-
-	if (maxa - mina > 8)
-	{
-		int besterror = computeAlphaError(rgba, block);
-		int besta0 = maxa;
-		int besta1 = mina;
-
-		for (int a0 = mina+9; a0 < maxa; a0++)
-		{
-			for (int a1 = mina; a1 < a0-8; a1++)
-			//for (int a1 = mina; a1 < maxa; a1++)
-			{
-				//nvCheck(abs(a1-a0) > 8);
-
-				//if (abs(a0 - a1) < 8) continue;
-				//if ((maxa-a0) + (a1-mina) + min(abs(centroid-a0), abs(centroid-a1)) > besterror)
-				if ((maxa-a0) + (a1-mina) > besterror)
-					continue;
-
-				block->alpha0 = a0;
-				block->alpha1 = a1;
-				int error = computeAlphaError(rgba, block);
-
-				if (error < besterror)
-				{
-					besterror = error;
-					besta0 = a0;
-					besta1 = a1;
-				}
-			}
-		}
-
-		block->alpha0 = besta0;
-		block->alpha1 = besta1;
-	}
-
-	return computeAlphaIndices(rgba, block);
-}
-
-
-

src/nvtt/FastCompressDXT.h

@@ -1,84 +0,0 @@
-// Copyright NVIDIA Corporation 2007 -- Ignacio Castano <icastano@nvidia.com>
-// 
-// Permission is hereby granted, free of charge, to any person
-// obtaining a copy of this software and associated documentation
-// files (the "Software"), to deal in the Software without
-// restriction, including without limitation the rights to use,
-// copy, modify, merge, publish, distribute, sublicense, and/or sell
-// copies of the Software, and to permit persons to whom the
-// Software is furnished to do so, subject to the following
-// conditions:
-// 
-// The above copyright notice and this permission notice shall be
-// included in all copies or substantial portions of the Software.
-// 
-// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
-// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
-// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
-// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
-// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
-// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
-// OTHER DEALINGS IN THE SOFTWARE.
-
-#ifndef NV_TT_FASTCOMPRESSDXT_H
-#define NV_TT_FASTCOMPRESSDXT_H
-
-#include <nvimage/nvimage.h>
-
-namespace nv
-{
-	struct ColorBlock;
-	struct BlockDXT1;
-	struct BlockDXT3;
-	struct BlockDXT5;
-	struct AlphaBlockDXT3;
-	struct AlphaBlockDXT5;
-
-	// Color compression:
-
-	// Compressor that uses the extremes of the luminance axis.
-//	void compressBlock_DiameterAxis(const ColorBlock & rgba, BlockDXT1 * block);
-
-	// Compressor that uses the extremes of the luminance axis.
-//	void compressBlock_LuminanceAxis(const ColorBlock & rgba, BlockDXT1 * block);
-
-	// Compressor that uses bounding box.
-	void compressBlock_BoundsRange(const ColorBlock & rgba, BlockDXT1 * block);
-
-	// Compressor that uses bounding box and takes alpha into account.
-//	void compressBlock_BoundsRangeAlpha(const ColorBlock & rgba, BlockDXT1 * block);
-
-
-	// Simple, but slow compressor that tests all color pairs.
-//	void compressBlock_TestAllPairs(const ColorBlock & rgba, BlockDXT1 * block);
-	
-	// Brute force 6d search along the best fit axis.
-//	void compressBlock_AnalyzeBestFitAxis(const ColorBlock & rgba, BlockDXT1 * block);
-
-	// Spatial greedy search.
-//	void refineSolution_1dSearch(const ColorBlock & rgba, BlockDXT1 * block);
-//	void refineSolution_3dSearch(const ColorBlock & rgba, BlockDXT1 * block);
-//	void refineSolution_6dSearch(const ColorBlock & rgba, BlockDXT1 * block);
-	
-	// Brute force compressor for DXT5n
-//	void compressGreenBlock_BruteForce(const ColorBlock & rgba, BlockDXT1 * block);
-
-	// Minimize error of the endpoints.
-//	void optimizeEndPoints(const ColorBlock & rgba, BlockDXT1 * block);
-	
-//	uint blockError(const ColorBlock & rgba, const BlockDXT1 & block);
-//	uint blockError(const ColorBlock & rgba, const AlphaBlockDXT5 & block);
-
-	// Alpha compression:
-	void compressBlock(const ColorBlock & rgba, AlphaBlockDXT3 * block);
-	void compressBlock_BoundsRange(const ColorBlock & rgba, BlockDXT3 * block);
-	void compressBlock_BoundsRange(const ColorBlock & rgba, BlockDXT5 * block);
-
-	uint compressBlock_BoundsRange(const ColorBlock & rgba, AlphaBlockDXT5 * block);
-	uint compressBlock_BruteForce(const ColorBlock & rgba, AlphaBlockDXT5 * block);
-//	uint compressBlock_Iterative(const ColorBlock & rgba, AlphaBlockDXT5 * block);
-
-} // nv namespace
-
-#endif // NV_TT_FASTCOMPRESSDXT_H

src/nvtt/InputOptions.cpp

@@ -23,8 +23,6 @@
 
 #include <string.h> // memcpy
 
-#include <nvcore/Containers.h> // nextPowerOfTwo
-
 #include <nvcore/Memory.h>
 
 #include "nvtt.h"
@@ -96,7 +94,7 @@ void InputOptions::reset()
 	m.textureType = TextureType_2D;
 	m.inputFormat = InputFormat_BGRA_8UB;
 
-	m.alphaMode = AlphaMode_Transparency;
+	m.alphaMode = AlphaMode_None;
 
 	m.inputGamma = 2.2f;
 	m.outputGamma = 2.2f;
@@ -120,8 +118,6 @@ void InputOptions::reset()
 	
 	m.maxExtent = 0;
 	m.roundMode = RoundMode_None;
-
-	m.premultiplyAlpha = false;
 }
 
 
@@ -309,19 +305,6 @@ void InputOptions::setLinearTransform(int channel, float w0, float w1, float w2,
 	//m.linearTransform.setRow(channel, w);
 }
 
-void InputOptions::setSwizzleTransform(int x, int y, int z, int w)
-{
-	nvCheck(x >= 0 && x < 3);
-	nvCheck(y >= 0 && y < 3);
-	nvCheck(z >= 0 && z < 3);
-	nvCheck(w >= 0 && w < 3);
-	
-	// m.xswizzle = x;
-	// m.yswizzle = y;
-	// m.zswizzle = z;
-	// m.wswizzle = w;
-}
-
 void InputOptions::setMaxExtents(int e)
 {
 	nvDebugCheck(e > 0);
@@ -333,10 +316,6 @@ void InputOptions::setRoundMode(RoundMode mode)
 	m.roundMode = mode;
 }
 
-void InputOptions::setPremultiplyAlpha(bool b)
-{
-	m.premultiplyAlpha = b;
-}
 
 void InputOptions::Private::computeTargetExtents() const
 {

src/nvtt/InputOptions.h

@@ -78,8 +78,6 @@ namespace nvtt
 		uint maxExtent;
 		RoundMode roundMode;
 		
-		bool premultiplyAlpha;
-
 		// @@ These are computed in nvtt::compress, so they should be mutable or stored elsewhere...
 		mutable uint targetWidth;
 		mutable uint targetHeight;
@@ -91,7 +89,7 @@ namespace nvtt
 		int realMipmapCount() const;
 		
 		const nv::Image * image(uint face, uint mipmap) const;
-		const nv::Image * image(uint idx) const;
+		const nv::Image * image(uint idx) const;
 
 	};
 

src/nvtt/OptimalCompressDXT.cpp

@@ -0,0 +1,368 @@
+// Copyright NVIDIA Corporation 2007 -- Ignacio Castano <icastano@nvidia.com>
+// 
+// Permission is hereby granted, free of charge, to any person
+// obtaining a copy of this software and associated documentation
+// files (the "Software"), to deal in the Software without
+// restriction, including without limitation the rights to use,
+// copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the
+// Software is furnished to do so, subject to the following
+// conditions:
+// 
+// The above copyright notice and this permission notice shall be
+// included in all copies or substantial portions of the Software.
+// 
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+// OTHER DEALINGS IN THE SOFTWARE.
+
+#include <nvcore/Containers.h> // swap
+
+#include <nvmath/Color.h>
+
+#include <nvimage/ColorBlock.h>
+#include <nvimage/BlockDXT.h>
+
+#include "OptimalCompressDXT.h"
+#include "SingleColorLookup.h"
+
+
+using namespace nv;
+using namespace OptimalCompress;
+
+
+
+namespace
+{
+	static int computeGreenError(const ColorBlock & rgba, const BlockDXT1 * block)
+	{
+		nvDebugCheck(block != NULL);
+
+		int palette[4];
+		palette[0] = (block->col0.g << 2) | (block->col0.g >> 4);
+		palette[1] = (block->col1.g << 2) | (block->col1.g >> 4);
+		palette[2] = (2 * palette[0] + palette[1]) / 3;
+		palette[3] = (2 * palette[1] + palette[0]) / 3;
+
+		int totalError = 0;
+
+		for (int i = 0; i < 16; i++)
+		{
+			const int green = rgba.color(i).g;
+			
+			int error = abs(green - palette[0]);
+			error = min(error, abs(green - palette[1]));
+			error = min(error, abs(green - palette[2]));
+			error = min(error, abs(green - palette[3]));
+			
+			totalError += error;
+		}
+
+		return totalError;
+	}
+
+	static uint computeGreenIndices(const ColorBlock & rgba, const Color32 palette[4])
+	{
+		const int color0 = palette[0].g;
+		const int color1 = palette[1].g;
+		const int color2 = palette[2].g;
+		const int color3 = palette[3].g;
+		
+		uint indices = 0;
+		for (int i = 0; i < 16; i++)
+		{
+			const int color = rgba.color(i).g;
+			
+			uint d0 = abs(color0 - color);
+			uint d1 = abs(color1 - color);
+			uint d2 = abs(color2 - color);
+			uint d3 = abs(color3 - color);
+			
+			uint b0 = d0 > d3;
+			uint b1 = d1 > d2;
+			uint b2 = d0 > d2;
+			uint b3 = d1 > d3;
+			uint b4 = d2 > d3;
+			
+			uint x0 = b1 & b2;
+			uint x1 = b0 & b3;
+			uint x2 = b0 & b4;
+			
+			indices |= (x2 | ((x0 | x1) << 1)) << (2 * i);
+		}
+
+		return indices;
+	}
+
+	// Choose quantized color that produces less error. Used by DXT3 compressor.
+	inline static uint quantize4(uint8 a)
+	{
+		int q0 = (a >> 4) - 1;
+		int q1 = (a >> 4);
+		int q2 = (a >> 4) + 1;
+		
+		q0 = (q0 << 4) | q0;
+		q1 = (q1 << 4) | q1;
+		q2 = (q2 << 4) | q2;
+		
+		int d0 = abs(q0 - a);
+		int d1 = abs(q1 - a);
+		int d2 = abs(q2 - a);
+
+		if (d0 < d1 && d0 < d2) return q0 >> 4;
+		if (d1 < d2) return q1 >> 4;
+		return q2 >> 4;
+	}
+	
+	static uint computeAlphaError(const ColorBlock & rgba, const AlphaBlockDXT5 * block)
+	{
+		uint8 alphas[8];
+		block->evaluatePalette(alphas);
+
+		uint totalError = 0;
+
+		for (uint i = 0; i < 16; i++)
+		{
+			uint8 alpha = rgba.color(i).a;
+
+			uint besterror = 256*256;
+			uint best;
+			for (uint p = 0; p < 8; p++)
+			{
+				int d = alphas[p] - alpha;
+				uint error = d * d;
+
+				if (error < besterror)
+				{
+					besterror = error;
+					best = p;
+				}
+			}
+
+			totalError += besterror;
+		}
+
+		return totalError;
+	}
+	
+	static void computeAlphaIndices(const ColorBlock & rgba, AlphaBlockDXT5 * block)
+	{
+		uint8 alphas[8];
+		block->evaluatePalette(alphas);
+
+		for (uint i = 0; i < 16; i++)
+		{
+			uint8 alpha = rgba.color(i).a;
+
+			uint besterror = 256*256;
+			uint best = 8;
+			for(uint p = 0; p < 8; p++)
+			{
+				int d = alphas[p] - alpha;
+				uint error = d * d;
+
+				if (error < besterror)
+				{
+					besterror = error;
+					best = p;
+				}
+			}
+			nvDebugCheck(best < 8);
+
+			block->setIndex(i, best);
+		}
+	}
+
+} // namespace
+
+
+
+
+
+// Single color compressor, based on:
+// https://mollyrocket.com/forums/viewtopic.php?t=392
+void OptimalCompress::compressDXT1(Color32 c, BlockDXT1 * dxtBlock)
+{
+	dxtBlock->col0.r = OMatch5[c.r][0];
+	dxtBlock->col0.g = OMatch6[c.g][0];
+	dxtBlock->col0.b = OMatch5[c.b][0];
+	dxtBlock->col1.r = OMatch5[c.r][1];
+	dxtBlock->col1.g = OMatch6[c.g][1];
+	dxtBlock->col1.b = OMatch5[c.b][1];
+	dxtBlock->indices = 0xaaaaaaaa;
+
+	if (dxtBlock->col0.u < dxtBlock->col1.u)
+	{
+		swap(dxtBlock->col0.u, dxtBlock->col1.u);
+		dxtBlock->indices ^= 0x55555555;
+	}
+}
+
+void OptimalCompress::compressDXT1a(Color32 rgba, BlockDXT1 * dxtBlock)
+{
+	if (rgba.a < 128)
+	{
+		dxtBlock->col0.u = 0;
+		dxtBlock->col1.u = 0;
+		dxtBlock->indices = 0xFFFFFFFF;
+	}
+	else
+	{
+		compressDXT1(rgba, dxtBlock);
+	}
+}
+
+
+// Brute force green channel compressor
+void OptimalCompress::compressDXT1G(const ColorBlock & rgba, BlockDXT1 * block)
+{
+	nvDebugCheck(block != NULL);
+	
+	uint8 ming = 63;
+	uint8 maxg = 0;
+	
+	// Get min/max green.
+	for (uint i = 0; i < 16; i++)
+	{
+		uint8 green = rgba.color(i).g >> 2;
+		ming = min(ming, green);
+		maxg = max(maxg, green);
+	}
+
+	block->col0.r = 31;
+	block->col1.r = 31;
+	block->col0.g = maxg;
+	block->col1.g = ming;
+	block->col0.b = 0;
+	block->col1.b = 0;
+
+	if (maxg - ming > 4)
+	{
+		int besterror = computeGreenError(rgba, block);
+		int bestg0 = maxg;
+		int bestg1 = ming;
+		
+		for (int g0 = ming+5; g0 < maxg; g0++)
+		{
+			for (int g1 = ming; g1 < g0-4; g1++)
+			{
+				if ((maxg-g0) + (g1-ming) > besterror)
+					continue;
+				
+				block->col0.g = g0;
+				block->col1.g = g1;
+				int error = computeGreenError(rgba, block);
+				
+				if (error < besterror)
+				{
+					besterror = error;
+					bestg0 = g0;
+					bestg1 = g1;
+				}
+			}
+		}
+		
+		block->col0.g = bestg0;
+		block->col1.g = bestg1;
+	}
+	
+	Color32 palette[4];
+	block->evaluatePalette(palette);
+	block->indices = computeGreenIndices(rgba, palette);
+}
+
+void OptimalCompress::compressDXT3A(const ColorBlock & rgba, AlphaBlockDXT3 * dxtBlock)
+{
+	dxtBlock->alpha0 = quantize4(rgba.color(0).a);
+	dxtBlock->alpha1 = quantize4(rgba.color(1).a);
+	dxtBlock->alpha2 = quantize4(rgba.color(2).a);
+	dxtBlock->alpha3 = quantize4(rgba.color(3).a);
+	dxtBlock->alpha4 = quantize4(rgba.color(4).a);
+	dxtBlock->alpha5 = quantize4(rgba.color(5).a);
+	dxtBlock->alpha6 = quantize4(rgba.color(6).a);
+	dxtBlock->alpha7 = quantize4(rgba.color(7).a);
+	dxtBlock->alpha8 = quantize4(rgba.color(8).a);
+	dxtBlock->alpha9 = quantize4(rgba.color(9).a);
+	dxtBlock->alphaA = quantize4(rgba.color(10).a);
+	dxtBlock->alphaB = quantize4(rgba.color(11).a);
+	dxtBlock->alphaC = quantize4(rgba.color(12).a);
+	dxtBlock->alphaD = quantize4(rgba.color(13).a);
+	dxtBlock->alphaE = quantize4(rgba.color(14).a);
+	dxtBlock->alphaF = quantize4(rgba.color(15).a);
+}
+
+
+void OptimalCompress::compressDXT5A(const ColorBlock & rgba, AlphaBlockDXT5 * dxtBlock)
+{
+	uint8 mina = 255;
+	uint8 maxa = 0;
+
+	// Get min/max alpha.
+	for (uint i = 0; i < 16; i++)
+	{
+		uint8 alpha = rgba.color(i).a;
+		mina = min(mina, alpha);
+		maxa = max(maxa, alpha);
+	}
+
+	dxtBlock->alpha0 = maxa;
+	dxtBlock->alpha1 = mina;
+
+	/*int centroidDist = 256;
+	int centroid;
+
+	// Get the closest to the centroid.
+	for (uint i = 0; i < 16; i++)
+	{
+		uint8 alpha = rgba.color(i).a;
+		int dist = abs(alpha - (maxa + mina) / 2);
+		if (dist < centroidDist)
+		{
+			centroidDist = dist;
+			centroid = alpha;
+		}
+	}*/
+
+	if (maxa - mina > 8)
+	{
+		int besterror = computeAlphaError(rgba, dxtBlock);
+		int besta0 = maxa;
+		int besta1 = mina;
+
+		for (int a0 = mina+9; a0 < maxa; a0++)
+		{
+			for (int a1 = mina; a1 < a0-8; a1++)
+			//for (int a1 = mina; a1 < maxa; a1++)
+			{
+				//nvCheck(abs(a1-a0) > 8);
+
+				//if (abs(a0 - a1) < 8) continue;
+				//if ((maxa-a0) + (a1-mina) + min(abs(centroid-a0), abs(centroid-a1)) > besterror)
+				if ((maxa-a0) + (a1-mina) > besterror)
+					continue;
+
+				dxtBlock->alpha0 = a0;
+				dxtBlock->alpha1 = a1;
+				int error = computeAlphaError(rgba, dxtBlock);
+
+				if (error < besterror)
+				{
+					besterror = error;
+					besta0 = a0;
+					besta1 = a1;
+				}
+			}
+		}
+
+		dxtBlock->alpha0 = besta0;
+		dxtBlock->alpha1 = besta1;
+	}
+
+	computeAlphaIndices(rgba, dxtBlock);
+}
+

src/nvtt/OptimalCompressDXT.h

@@ -0,0 +1,49 @@
+// Copyright NVIDIA Corporation 2008 -- Ignacio Castano <icastano@nvidia.com>
+// 
+// Permission is hereby granted, free of charge, to any person
+// obtaining a copy of this software and associated documentation
+// files (the "Software"), to deal in the Software without
+// restriction, including without limitation the rights to use,
+// copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the
+// Software is furnished to do so, subject to the following
+// conditions:
+// 
+// The above copyright notice and this permission notice shall be
+// included in all copies or substantial portions of the Software.
+// 
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+// OTHER DEALINGS IN THE SOFTWARE.
+
+#ifndef NV_TT_OPTIMALCOMPRESSDXT_H
+#define NV_TT_OPTIMALCOMPRESSDXT_H
+
+#include <nvimage/nvimage.h>
+
+namespace nv
+{
+	struct ColorBlock;
+	struct BlockDXT1;
+	struct BlockDXT3;
+	struct BlockDXT5;
+	struct AlphaBlockDXT3;
+	struct AlphaBlockDXT5;
+
+	namespace OptimalCompress
+	{
+		void compressDXT1(Color32 rgba, BlockDXT1 * dxtBlock);
+		void compressDXT1a(Color32 rgba, BlockDXT1 * dxtBlock);
+		
+		void compressDXT1G(const ColorBlock & rgba, BlockDXT1 * block);
+		void compressDXT3A(const ColorBlock & rgba, AlphaBlockDXT3 * dxtBlock);
+		void compressDXT5A(const ColorBlock & rgba, AlphaBlockDXT5 * dxtBlock);
+	}
+} // nv namespace
+
+#endif // NV_TT_OPTIMALCOMPRESSDXT_H

src/nvtt/QuickCompressDXT.cpp

@@ -21,15 +21,13 @@
 // FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 // OTHER DEALINGS IN THE SOFTWARE.
 
-#include <nvcore/Containers.h> // swap
-
 #include <nvmath/Color.h>
 
 #include <nvimage/ColorBlock.h>
 #include <nvimage/BlockDXT.h>
 
 #include "QuickCompressDXT.h"
-#include "SingleColorLookup.h"
+#include "OptimalCompressDXT.h"
 
 
 using namespace nv;
@@ -290,70 +288,6 @@ static void optimizeEndPoints4(Vector3 block[16], BlockDXT1 * dxtBlock)
 	dxtBlock->indices = computeIndices3(block, a, b);
 }*/
 
-namespace
-{
-	static int computeGreenError(const ColorBlock & rgba, const BlockDXT1 * block)
-	{
-		nvDebugCheck(block != NULL);
-
-		int palette[4];
-		palette[0] = (block->col0.g << 2) | (block->col0.g >> 4);
-		palette[1] = (block->col1.g << 2) | (block->col1.g >> 4);
-		palette[2] = (2 * palette[0] + palette[1]) / 3;
-		palette[3] = (2 * palette[1] + palette[0]) / 3;
-
-		int totalError = 0;
-
-		for (int i = 0; i < 16; i++)
-		{
-			const int green = rgba.color(i).g;
-			
-			int error = abs(green - palette[0]);
-			error = min(error, abs(green - palette[1]));
-			error = min(error, abs(green - palette[2]));
-			error = min(error, abs(green - palette[3]));
-			
-			totalError += error;
-		}
-
-		return totalError;
-	}
-
-	static uint computeGreenIndices(const ColorBlock & rgba, const Color32 palette[4])
-	{
-		const int color0 = palette[0].g;
-		const int color1 = palette[1].g;
-		const int color2 = palette[2].g;
-		const int color3 = palette[3].g;
-		
-		uint indices = 0;
-		for (int i = 0; i < 16; i++)
-		{
-			const int color = rgba.color(i).g;
-			
-			uint d0 = abs(color0 - color);
-			uint d1 = abs(color1 - color);
-			uint d2 = abs(color2 - color);
-			uint d3 = abs(color3 - color);
-			
-			uint b0 = d0 > d3;
-			uint b1 = d1 > d2;
-			uint b2 = d0 > d2;
-			uint b3 = d1 > d3;
-			uint b4 = d2 > d3;
-			
-			uint x0 = b1 & b2;
-			uint x1 = b0 & b3;
-			uint x2 = b0 & b4;
-			
-			indices |= (x2 | ((x0 | x1) << 1)) << (2 * i);
-		}
-
-		return indices;
-	}
-
-} // namespace
-
 namespace
 {
 
@@ -505,79 +439,63 @@ namespace
 
 
 
-
-
-// Single color compressor, based on:
-// https://mollyrocket.com/forums/viewtopic.php?t=392
-void QuickCompress::compressDXT1(Color32 c, BlockDXT1 * dxtBlock)
-{
-	dxtBlock->col0.r = OMatch5[c.r][0];
-	dxtBlock->col0.g = OMatch6[c.g][0];
-	dxtBlock->col0.b = OMatch5[c.b][0];
-	dxtBlock->col1.r = OMatch5[c.r][1];
-	dxtBlock->col1.g = OMatch6[c.g][1];
-	dxtBlock->col1.b = OMatch5[c.b][1];
-	dxtBlock->indices = 0xaaaaaaaa;
-
-	if (dxtBlock->col0.u < dxtBlock->col1.u)
-	{
-		swap(dxtBlock->col0.u, dxtBlock->col1.u);
-		dxtBlock->indices ^= 0x55555555;
-	}
-}
-
 void QuickCompress::compressDXT1(const ColorBlock & rgba, BlockDXT1 * dxtBlock)
 {
-	// read block
-	Vector3 block[16];
-	extractColorBlockRGB(rgba, block);
-	
-	// find min and max colors
-	Vector3 maxColor, minColor;
-	findMinMaxColorsBox(block, 16, &maxColor, &minColor);
-	
-	selectDiagonal(block, 16, &maxColor, &minColor);
-	
-	insetBBox(&maxColor, &minColor);
-	
-	uint16 color0 = roundAndExpand(&maxColor);
-	uint16 color1 = roundAndExpand(&minColor);
-
-	if (color0 < color1)
+	if (rgba.isSingleColor())
 	{
-		swap(maxColor, minColor);
-		swap(color0, color1);
-	}
-
-	dxtBlock->col0 = Color16(color0);
-	dxtBlock->col1 = Color16(color1);
-	dxtBlock->indices = computeIndices4(block, maxColor, minColor);
-
-	optimizeEndPoints4(block, dxtBlock);
-}
-
-
-void QuickCompress::compressDXT1a(Color32 rgba, BlockDXT1 * dxtBlock)
-{
-	if (rgba.a == 0)
-	{
-		dxtBlock->col0.u = 0;
-		dxtBlock->col1.u = 0;
-		dxtBlock->indices = 0xFFFFFFFF;
+		OptimalCompress::compressDXT1(rgba.color(0), dxtBlock);
 	}
 	else
 	{
-		compressDXT1(rgba, dxtBlock);
+		// read block
+		Vector3 block[16];
+		extractColorBlockRGB(rgba, block);
+		
+		// find min and max colors
+		Vector3 maxColor, minColor;
+		findMinMaxColorsBox(block, 16, &maxColor, &minColor);
+		
+		selectDiagonal(block, 16, &maxColor, &minColor);
+		
+		insetBBox(&maxColor, &minColor);
+		
+		uint16 color0 = roundAndExpand(&maxColor);
+		uint16 color1 = roundAndExpand(&minColor);
+
+		if (color0 < color1)
+		{
+			swap(maxColor, minColor);
+			swap(color0, color1);
+		}
+
+		dxtBlock->col0 = Color16(color0);
+		dxtBlock->col1 = Color16(color1);
+		dxtBlock->indices = computeIndices4(block, maxColor, minColor);
+
+		optimizeEndPoints4(block, dxtBlock);
 	}
 }
 
+
 void QuickCompress::compressDXT1a(const ColorBlock & rgba, BlockDXT1 * dxtBlock)
 {
-	if (!rgba.hasAlpha())
+	bool hasAlpha = false;
+	
+	for (uint i = 0; i < 16; i++)
+	{
+		if (rgba.color(i).a < 128) {
+			hasAlpha = true;
+			break;
+		}
+	}
+	
+	if (!hasAlpha)
 	{
 		compressDXT1(rgba, dxtBlock);
 	}
-	else
+	// @@ Handle single RGB, with varying alpha? We need tables for single color compressor in 3 color mode.
+	//else if (rgba.isSingleColorNoAlpha()) { ... }
+	else 
 	{
 		// read block
 		Vector3 block[16];
@@ -609,95 +527,14 @@ void QuickCompress::compressDXT1a(const ColorBlock & rgba, BlockDXT1 * dxtBlock)
 }
 
 
-// Brute force green channel compressor
-void QuickCompress::compressDXT1G(const ColorBlock & rgba, BlockDXT1 * block)
-{
-	nvDebugCheck(block != NULL);
-	
-	uint8 ming = 63;
-	uint8 maxg = 0;
-	
-	// Get min/max green.
-	for (uint i = 0; i < 16; i++)
-	{
-		uint8 green = rgba.color(i).g >> 2;
-		ming = min(ming, green);
-		maxg = max(maxg, green);
-	}
-
-	block->col0.r = 31;
-	block->col1.r = 31;
-	block->col0.g = maxg;
-	block->col1.g = ming;
-	block->col0.b = 0;
-	block->col1.b = 0;
-
-	if (maxg - ming > 4)
-	{
-		int besterror = computeGreenError(rgba, block);
-		int bestg0 = maxg;
-		int bestg1 = ming;
-		
-		for (int g0 = ming+5; g0 < maxg; g0++)
-		{
-			for (int g1 = ming; g1 < g0-4; g1++)
-			{
-				if ((maxg-g0) + (g1-ming) > besterror)
-					continue;
-				
-				block->col0.g = g0;
-				block->col1.g = g1;
-				int error = computeGreenError(rgba, block);
-				
-				if (error < besterror)
-				{
-					besterror = error;
-					bestg0 = g0;
-					bestg1 = g1;
-				}
-			}
-		}
-		
-		block->col0.g = bestg0;
-		block->col1.g = bestg1;
-	}
-	
-	Color32 palette[4];
-	block->evaluatePalette(palette);
-	block->indices = computeGreenIndices(rgba, palette);
-}
-
-
-
-void QuickCompress::compressDXT3A(const ColorBlock & rgba, AlphaBlockDXT3 * dxtBlock)
-{
-	// @@ Round instead of truncate. When rounding take into account bit expansion.
-	dxtBlock->alpha0 = rgba.color(0).a >> 4;
-	dxtBlock->alpha1 = rgba.color(1).a >> 4;
-	dxtBlock->alpha2 = rgba.color(2).a >> 4;
-	dxtBlock->alpha3 = rgba.color(3).a >> 4;
-	dxtBlock->alpha4 = rgba.color(4).a >> 4;
-	dxtBlock->alpha5 = rgba.color(5).a >> 4;
-	dxtBlock->alpha6 = rgba.color(6).a >> 4;
-	dxtBlock->alpha7 = rgba.color(7).a >> 4;
-	dxtBlock->alpha8 = rgba.color(8).a >> 4;
-	dxtBlock->alpha9 = rgba.color(9).a >> 4;
-	dxtBlock->alphaA = rgba.color(10).a >> 4;
-	dxtBlock->alphaB = rgba.color(11).a >> 4;
-	dxtBlock->alphaC = rgba.color(12).a >> 4;
-	dxtBlock->alphaD = rgba.color(13).a >> 4;
-	dxtBlock->alphaE = rgba.color(14).a >> 4;
-	dxtBlock->alphaF = rgba.color(15).a >> 4;
-}
-
 void QuickCompress::compressDXT3(const ColorBlock & rgba, BlockDXT3 * dxtBlock)
 {
 	compressDXT1(rgba, &dxtBlock->color);
-	compressDXT3A(rgba, &dxtBlock->alpha);
+	OptimalCompress::compressDXT3A(rgba, &dxtBlock->alpha);
 }
 
 
-void QuickCompress::compressDXT5A(const ColorBlock & rgba, AlphaBlockDXT5 * dxtBlock)
+void QuickCompress::compressDXT5A(const ColorBlock & rgba, AlphaBlockDXT5 * dxtBlock, int iterationCount/*=8*/)
 {
 	uint8 alpha0 = 0;
 	uint8 alpha1 = 255;
@@ -716,8 +553,8 @@ void QuickCompress::compressDXT5A(const ColorBlock & rgba, AlphaBlockDXT5 * dxtB
 	uint besterror = computeAlphaIndices(rgba, &block);
 	
 	AlphaBlockDXT5 bestblock = block;
-	
-	while(true)
+
+	for (int i = 0; i < iterationCount; i++)
 	{
 		optimizeAlpha8(rgba, &block);
 		uint error = computeAlphaIndices(rgba, &block);
@@ -741,9 +578,8 @@ void QuickCompress::compressDXT5A(const ColorBlock & rgba, AlphaBlockDXT5 * dxtB
 	*dxtBlock = bestblock;
 }
 
-void QuickCompress::compressDXT5(const ColorBlock & rgba, BlockDXT5 * dxtBlock)
+void QuickCompress::compressDXT5(const ColorBlock & rgba, BlockDXT5 * dxtBlock, int iterationCount/*=8*/)
 {
 	compressDXT1(rgba, &dxtBlock->color);
-	compressDXT5A(rgba, &dxtBlock->alpha);
+	compressDXT5A(rgba, &dxtBlock->alpha, iterationCount);
 }
-

src/nvtt/QuickCompressDXT.h

@@ -37,17 +37,13 @@ namespace nv
 
 	namespace QuickCompress
 	{
-		void compressDXT1(Color32 rgba, BlockDXT1 * dxtBlock);
 		void compressDXT1(const ColorBlock & rgba, BlockDXT1 * dxtBlock);
-		void compressDXT1a(Color32 rgba, BlockDXT1 * dxtBlock);
 		void compressDXT1a(const ColorBlock & rgba, BlockDXT1 * dxtBlock);
-		void compressDXT1G(const ColorBlock & rgba, BlockDXT1 * block);
 		
-		void compressDXT3A(const ColorBlock & rgba, AlphaBlockDXT3 * dxtBlock);
 		void compressDXT3(const ColorBlock & rgba, BlockDXT3 * dxtBlock);
 		
-		void compressDXT5A(const ColorBlock & rgba, AlphaBlockDXT5 * dxtBlock);
-		void compressDXT5(const ColorBlock & rgba, BlockDXT5 * dxtBlock);
+		void compressDXT5A(const ColorBlock & rgba, AlphaBlockDXT5 * dxtBlock, int iterationCount=8);
+		void compressDXT5(const ColorBlock & rgba, BlockDXT5 * dxtBlock, int iterationCount=8);
 	}
 } // nv namespace
 

src/nvtt/cuda/CompressKernel.cu

@@ -159,7 +159,7 @@ __device__ void loadColorBlock(const uint * image, float3 colors[16], float3 sum
 	}
 }
 
-__device__ void loadColorBlock(const uint * image, float3 colors[16], float3 sums[16], float weights[16], int xrefs[16])
+__device__ void loadColorBlock(const uint * image, float3 colors[16], float3 sums[16], float weights[16], int xrefs[16], int * sameColor)
 {
 	const int bid = blockIdx.x;
 	const int idx = threadIdx.x;
@@ -189,6 +189,11 @@ __device__ void loadColorBlock(const uint * image, float3 colors[16], float3 sum
 		colorSums(colors, sums);
 		float3 axis = bestFitLine(colors, sums[0], kColorMetric);
 
+		*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__
@@ -205,43 +210,6 @@ __device__ void loadColorBlock(const uint * image, float3 colors[16], float3 sum
 	}
 }
 
-__device__ void loadColorBlock(const uint * image, float2 colors[16], float2 sums[16], int xrefs[16])
-{
-	const int bid = blockIdx.x;
-	const int idx = threadIdx.x;
-
-	__shared__ float dps[16];
-
-	if (idx < 16)
-	{
-		// Read color and copy to shared mem.
-		uint c = image[(bid) * 16 + idx];
-	
-		colors[idx].y = ((c >> 8) & 0xFF) * (1.0f / 255.0f);
-		colors[idx].x = ((c >> 16) & 0xFF) * (1.0f / 255.0f);
-		
-		// No need to synchronize, 16 < warp size.
-#if __DEVICE_EMULATION__
-		} __debugsync(); if (idx < 16) {
-#endif
-
-		// Sort colors along the best fit line.
-		colorSums(colors, sums);
-		float2 axis = bestFitLine(colors, sums[0]);
-		
-		dps[idx] = dot(colors[idx], axis);
-		
-#if __DEVICE_EMULATION__
-		} __debugsync(); if (idx < 16) {
-#endif
-		
-		sortColors(dps, xrefs);
-		
-		float2 tmp = colors[idx];
-		colors[xrefs[idx]] = tmp;
-	}
-}
-
 
 ////////////////////////////////////////////////////////////////////////////////
 // Round color to RGB565 and expand
@@ -258,26 +226,6 @@ inline __device__ float3 roundAndExpand565(float3 v, ushort * w)
 	return v;
 }
 
-inline __device__ float2 roundAndExpand56(float2 v, ushort * w)
-{
-	v.x = rintf(__saturatef(v.x) * 31.0f);
-	v.y = rintf(__saturatef(v.y) * 63.0f);
-	*w = ((ushort)v.x << 11) | ((ushort)v.y << 5);
-	v.x *= 0.03227752766457f; // approximate integer bit expansion.
-	v.y *= 0.01583151765563f;
-	return v;
-}
-
-inline __device__ float2 roundAndExpand88(float2 v, ushort * w)
-{
-	v.x = rintf(__saturatef(v.x) * 255.0f);
-	v.y = rintf(__saturatef(v.y) * 255.0f);
-	*w = ((ushort)v.x << 8) | ((ushort)v.y);
-	v.x *= 1.0f / 255.0f;
-	v.y *= 1.0f / 255.0f;
-	return v;
-}
-
 
 ////////////////////////////////////////////////////////////////////////////////
 // Evaluate permutations
@@ -521,114 +469,6 @@ __device__ float evalPermutation3(const float3 * colors, const float * weights,
 }
 */
 
-__device__ float evalPermutation4(const float2 * colors, float2 color_sum, uint permutation, ushort * start, ushort * end)
-{
-	// Compute endpoints using least squares.
-	float2 alphax_sum = make_float2(0.0f, 0.0f);
-	uint akku = 0;
-
-	// Compute alpha & beta for this permutation.
-	#pragma unroll
-	for (int i = 0; i < 16; i++)
-	{
-		const uint bits = permutation >> (2*i);
-		
-		alphax_sum += alphaTable4[bits & 3] * colors[i];
-		akku += prods4[bits & 3];
-	}
-
-	float alpha2_sum = float(akku >> 16);
-	float beta2_sum = float((akku >> 8) & 0xff);
-	float alphabeta_sum = float(akku & 0xff);
-	float2 betax_sum = 9.0f * color_sum - alphax_sum;
-
-	const float factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
-
-	float2 a = (alphax_sum * beta2_sum - betax_sum * alphabeta_sum) * factor;
-	float2 b = (betax_sum * alpha2_sum - alphax_sum * alphabeta_sum) * factor;
-	
-	// Round a, b to the closest 5-6 color and expand...
-	a = roundAndExpand56(a, start);
-	b = roundAndExpand56(b, end);
-
-	// compute the error
-	float2 e = a * a * alpha2_sum + b * b * beta2_sum + 2.0f * (a * b * alphabeta_sum - a * alphax_sum - b * betax_sum);
-
-	return (1.0f / 9.0f) * (e.x + e.y);
-}
-
-__device__ float evalPermutation3(const float2 * colors, float2 color_sum, uint permutation, ushort * start, ushort * end)
-{
-	// Compute endpoints using least squares.
-	float2 alphax_sum = make_float2(0.0f, 0.0f);
-	uint akku = 0;
-
-	// Compute alpha & beta for this permutation.
-	#pragma unroll
-	for (int i = 0; i < 16; i++)
-	{
-		const uint bits = permutation >> (2*i);
-
-		alphax_sum += alphaTable3[bits & 3] * colors[i];
-		akku += prods3[bits & 3];
-	}
-
-	float alpha2_sum = float(akku >> 16);
-	float beta2_sum = float((akku >> 8) & 0xff);
-	float alphabeta_sum = float(akku & 0xff);
-	float2 betax_sum = 4.0f * color_sum - alphax_sum;
-
-	const float factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
-
-	float2 a = (alphax_sum * beta2_sum - betax_sum * alphabeta_sum) * factor;
-	float2 b = (betax_sum * alpha2_sum - alphax_sum * alphabeta_sum) * factor;
-	
-	// Round a, b to the closest 5-6 color and expand...
-	a = roundAndExpand56(a, start);
-	b = roundAndExpand56(b, end);
-
-	// compute the error
-	float2 e = a * a * alpha2_sum + b * b * beta2_sum + 2.0f * (a * b * alphabeta_sum - a * alphax_sum - b * betax_sum);
-
-	return (1.0f / 4.0f) * (e.x + e.y);
-}
-
-__device__ float evalPermutationCTX(const float2 * colors, float2 color_sum, uint permutation, ushort * start, ushort * end)
-{
-	// Compute endpoints using least squares.
-	float2 alphax_sum = make_float2(0.0f, 0.0f);
-	uint akku = 0;
-
-	// Compute alpha & beta for this permutation.
-	#pragma unroll
-	for (int i = 0; i < 16; i++)
-	{
-		const uint bits = permutation >> (2*i);
-		
-		alphax_sum += alphaTable4[bits & 3] * colors[i];
-		akku += prods4[bits & 3];
-	}
-
-	float alpha2_sum = float(akku >> 16);
-	float beta2_sum = float((akku >> 8) & 0xff);
-	float alphabeta_sum = float(akku & 0xff);
-	float2 betax_sum = 9.0f * color_sum - alphax_sum;
-
-	const float factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
-
-	float2 a = (alphax_sum * beta2_sum - betax_sum * alphabeta_sum) * factor;
-	float2 b = (betax_sum * alpha2_sum - alphax_sum * alphabeta_sum) * factor;
-	
-	// Round a, b to the closest 8-8 color and expand...
-	a = roundAndExpand88(a, start);
-	b = roundAndExpand88(b, end);
-
-	// compute the error
-	float2 e = a * a * alpha2_sum + b * b * beta2_sum + 2.0f * (a * b * alphabeta_sum - a * alphax_sum - b * betax_sum);
-
-	return (1.0f / 9.0f) * (e.x + e.y);
-}
-
 
 ////////////////////////////////////////////////////////////////////////////////
 // Evaluate all permutations
@@ -757,14 +597,12 @@ __device__ void evalAllPermutations(const float3 * colors, const float * weights
 }
 */
 
-__device__ void evalAllPermutations(const float2 * colors, float2 colorSum, const uint * permutations, ushort & bestStart, ushort & bestEnd, uint & bestPermutation, float * errors)
+__device__ void evalLevel4Permutations(const float3 * colors, float3 colorSum, const uint * permutations, ushort & bestStart, ushort & bestEnd, uint & bestPermutation, float * errors)
 {
 	const int idx = threadIdx.x;
 	
 	float bestError = FLT_MAX;
 	
-	__shared__ uint s_permutations[160];
-
 	for(int i = 0; i < 16; i++)
 	{
 		int pidx = idx + NUM_THREADS * i;
@@ -772,8 +610,7 @@ __device__ void evalAllPermutations(const float2 * colors, float2 colorSum, cons
 		
 		ushort start, end;
 		uint permutation = permutations[pidx];
-		if (pidx < 160) s_permutations[pidx] = permutation;
-				
+
 		float error = evalPermutation4(colors, colorSum, permutation, &start, &end);
 		
 		if (error < bestError)
@@ -791,30 +628,6 @@ __device__ void evalAllPermutations(const float2 * colors, float2 colorSum, cons
 		bestPermutation ^= 0x55555555;	// Flip indices.
 	}
 
-	for(int i = 0; i < 3; i++)
-	{
-		int pidx = idx + NUM_THREADS * i;
-		if (pidx >= 160) break;
-		
-		ushort start, end;
-		uint permutation = s_permutations[pidx];
-		float error = evalPermutation3(colors, colorSum, permutation, &start, &end);
-		
-		if (error < bestError)
-		{
-			bestError = error;
-			bestPermutation = permutation;
-			bestStart = start;
-			bestEnd = end;
-			
-			if (bestStart > bestEnd)
-			{
-				swap(bestEnd, bestStart);
-				bestPermutation ^= (~bestPermutation >> 1) & 0x55555555;	// Flip indices.
-			}
-		}
-	}
-
 	errors[idx] = bestError;
 }
 
@@ -852,40 +665,6 @@ __device__ void evalLevel4Permutations(const float3 * colors, const float * weig
 	errors[idx] = bestError;
 }
 
-__device__ void evalAllPermutationsCTX(const float2 * colors, float2 colorSum, const uint * permutations, ushort & bestStart, ushort & bestEnd, uint & bestPermutation, float * errors)
-{
-	const int idx = threadIdx.x;
-	
-	float bestError = FLT_MAX;
-	
-	for(int i = 0; i < 16; i++)
-	{
-		int pidx = idx + NUM_THREADS * i;
-		if (pidx >= 992) break;
-		
-		ushort start, end;
-		uint permutation = permutations[pidx];
-
-		float error = evalPermutationCTX(colors, colorSum, permutation, &start, &end);
-		
-		if (error < bestError)
-		{
-			bestError = error;
-			bestPermutation = permutation;
-			bestStart = start;
-			bestEnd = end;
-		}
-	}
-
-	if (bestStart < bestEnd)
-	{
-		swap(bestEnd, bestStart);
-		bestPermutation ^= 0x55555555;	// Flip indices.
-	}
-
-	errors[idx] = bestError;
-}
-
 
 ////////////////////////////////////////////////////////////////////////////////
 // Find index with minimum error
@@ -996,11 +775,6 @@ __device__ void saveBlockDXT1(ushort start, ushort end, uint permutation, int xr
 	result[bid].y = indices;
 }
 
-__device__ void saveBlockCTX1(ushort start, ushort end, uint permutation, int xrefs[16], uint2 * result)
-{
-	saveBlockDXT1(start, end, permutation, xrefs, result);
-}
-
 __device__ void saveSingleColorBlockDXT1(float3 color, uint2 * result)
 {
 	const int bid = blockIdx.x;
@@ -1062,6 +836,39 @@ __global__ void compressDXT1(const uint * permutations, const uint * image, uint
 	}
 }
 
+__global__ void compressLevel4DXT1(const uint * permutations, const uint * image, uint2 * result)
+{
+	__shared__ float3 colors[16];
+	__shared__ float3 sums[16];
+	__shared__ int xrefs[16];
+	__shared__ int sameColor;
+	
+	loadColorBlock(image, colors, sums, xrefs, &sameColor);
+
+	__syncthreads();
+
+	if (sameColor)
+	{
+		if (threadIdx.x == 0) saveSingleColorBlockDXT1(colors[0], result);
+		return;
+	}
+
+	ushort bestStart, bestEnd;
+	uint bestPermutation;
+
+	__shared__ float errors[NUM_THREADS];
+
+	evalLevel4Permutations(colors, sums[0], permutations, bestStart, bestEnd, bestPermutation, errors);
+	
+	// Use a parallel reduction to find minimum error.
+	const int minIdx = findMinError(errors);
+	
+	// Only write the result of the winner thread.
+	if (threadIdx.x == minIdx)
+	{
+		saveBlockDXT1(bestStart, bestEnd, bestPermutation, xrefs, result);
+	}
+}
 
 __global__ void compressWeightedDXT1(const uint * permutations, const uint * image, uint2 * result)
 {
@@ -1069,11 +876,18 @@ __global__ void compressWeightedDXT1(const uint * permutations, const uint * ima
 	__shared__ float3 sums[16];
 	__shared__ float weights[16];
 	__shared__ int xrefs[16];
+	__shared__ int sameColor;
 	
-	loadColorBlock(image, colors, sums, weights, xrefs);
+	loadColorBlock(image, colors, sums, weights, xrefs, &sameColor);
 	
 	__syncthreads();
 
+	if (sameColor)
+	{
+		if (threadIdx.x == 0) saveSingleColorBlockDXT1(colors[0], result);
+		return;
+	}
+
 	ushort bestStart, bestEnd;
 	uint bestPermutation;
 
@@ -1092,61 +906,6 @@ __global__ void compressWeightedDXT1(const uint * permutations, const uint * ima
 }
 
 
-__global__ void compressNormalDXT1(const uint * permutations, const uint * image, uint2 * result)
-{
-	__shared__ float2 colors[16];
-	__shared__ float2 sums[16];
-	__shared__ int xrefs[16];
-	
-	loadColorBlock(image, colors, sums, xrefs);
-	
-	__syncthreads();
-
-	ushort bestStart, bestEnd;
-	uint bestPermutation;
-
-	__shared__ float errors[NUM_THREADS];
-
-	evalAllPermutations(colors, sums[0], permutations, bestStart, bestEnd, bestPermutation, errors);
-	
-	// Use a parallel reduction to find minimum error.
-	const int minIdx = findMinError(errors);
-	
-	// Only write the result of the winner thread.
-	if (threadIdx.x == minIdx)
-	{
-		saveBlockDXT1(bestStart, bestEnd, bestPermutation, xrefs, result);
-	}
-}
-
-__global__ void compressCTX1(const uint * permutations, const uint * image, uint2 * result)
-{
-	__shared__ float2 colors[16];
-	__shared__ float2 sums[16];
-	__shared__ int xrefs[16];
-	
-	loadColorBlock(image, colors, sums, xrefs);
-	
-	__syncthreads();
-
-	ushort bestStart, bestEnd;
-	uint bestPermutation;
-
-	__shared__ float errors[NUM_THREADS];
-
-	evalAllPermutationsCTX(colors, sums[0], permutations, bestStart, bestEnd, bestPermutation, errors);
-	
-	// Use a parallel reduction to find minimum error.
-	const int minIdx = findMinError(errors);
-	
-	// Only write the result of the winner thread.
-	if (threadIdx.x == minIdx)
-	{
-		saveBlockCTX1(bestStart, bestEnd, bestPermutation, xrefs, result);
-	}
-}
-
-
 /*
 __device__ float computeError(const float weights[16], uchar a0, uchar a1)
 {
@@ -1352,17 +1111,12 @@ extern "C" void compressKernelDXT1(uint blockNum, uint * d_data, uint * d_result
 	compressDXT1<<<blockNum, NUM_THREADS>>>(d_bitmaps, d_data, (uint2 *)d_result);
 }
 
+extern "C" void compressKernelDXT1_Level4(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps)
+{
+	compressLevel4DXT1<<<blockNum, NUM_THREADS>>>(d_bitmaps, d_data, (uint2 *)d_result);
+}
+
 extern "C" void compressWeightedKernelDXT1(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps)
 {
 	compressWeightedDXT1<<<blockNum, NUM_THREADS>>>(d_bitmaps, d_data, (uint2 *)d_result);
 }
-
-extern "C" void compressNormalKernelDXT1(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps)
-{
-	compressNormalDXT1<<<blockNum, NUM_THREADS>>>(d_bitmaps, d_data, (uint2 *)d_result);
-}
-
-extern "C" void compressKernelCTX1(uint blockNum, uint * d_data, uint * d_result, uint * d_bitmaps)
-{
-	compressCTX1<<<blockNum, NUM_THREADS>>>(d_bitmaps, d_data, (uint2 *)d_result);
-}

src/nvtt/cuda/CudaCompressDXT.h

@@ -39,17 +39,20 @@ namespace nv
 
 		bool isValid() const;
 
-		void compressDXT1(const Image * image, const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
-		void compressDXT3(const Image * image, const nvtt::InputOptions::Private & inputOptions, const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
-		void compressDXT5(const Image * image, const nvtt::InputOptions::Private & inputOptions, const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
-		void compressDXT1n(const Image * image, const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
-		void compressCTX1(const Image * image, const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void setImage(const Image * image, nvtt::AlphaMode alphaMode);
+
+		void compressDXT1(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT3(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
+		void compressDXT5(const nvtt::CompressionOptions::Private & compressionOptions, const nvtt::OutputOptions::Private & outputOptions);
 
 	private:
 
 		uint * m_bitmapTable;
 		uint * m_data;
 		uint * m_result;
+		
+		const Image * m_image;
+		nvtt::AlphaMode m_alphaMode;
 	};
 
 } // nv namespace

src/nvtt/cuda/CudaMath.h

@@ -87,64 +87,6 @@ inline __device__ __host__ bool operator ==(float3 a, float3 b)
 	return a.x == b.x && a.y == b.y && a.z == b.z;
 }
 
-
-// float2 operators
-inline __device__ __host__ float2 operator *(float2 a, float2 b)
-{
-    return make_float2(a.x*b.x, a.y*b.y);
-}
-
-inline __device__ __host__ float2 operator *(float f, float2 v)
-{
-    return make_float2(v.x*f, v.y*f);
-}
-
-inline __device__ __host__ float2 operator *(float2 v, float f)
-{
-    return make_float2(v.x*f, v.y*f);
-}
-
-inline __device__ __host__ float2 operator +(float2 a, float2 b)
-{
-    return make_float2(a.x+b.x, a.y+b.y);
-}
-
-inline __device__ __host__ void operator +=(float2 & b, float2 a)
-{
-    b.x += a.x;
-    b.y += a.y;
-}
-
-inline __device__ __host__ float2 operator -(float2 a, float2 b)
-{
-    return make_float2(a.x-b.x, a.y-b.y);
-}
-
-inline __device__ __host__ void operator -=(float2 & b, float2 a)
-{
-    b.x -= a.x;
-    b.y -= a.y;
-}
-
-inline __device__ __host__ float2 operator /(float2 v, float f)
-{
-    float inv = 1.0f / f;
-    return v * inv;
-}
-
-inline __device__ __host__ void operator /=(float2 & b, float f)
-{
-    float inv = 1.0f / f;
-    b.x *= inv;
-	b.y *= inv;
-}
-
-
-inline __device__ __host__ float dot(float2 a, float2 b)
-{
-    return a.x * b.x + a.y * b.y;
-}
-
 inline __device__ __host__ float dot(float3 a, float3 b)
 {
     return a.x * b.x + a.y * b.y + a.z * b.z;
@@ -206,7 +148,7 @@ inline __device__ bool singleColor(const float3 * colors)
 	bool sameColor = false;
 	for (int i = 0; i < 16; i++)
 	{
-		sameColor &= (colors[idx] == colors[0]);
+		sameColor &= (colors[i] == colors[0]);
 	}
 	return sameColor;
 #else
@@ -301,89 +243,5 @@ inline __device__ float3 bestFitLine(const float3 * colors, float3 color_sum, fl
 	return firstEigenVector(covariance);
 }
 
-// @@ For 2D this may not be the most efficient method. It's a quadratic equation, right?
-inline __device__ __host__ float2 firstEigenVector2D( float matrix[3] )
-{
-	// @@ 8 iterations is probably more than enough.
-
-	float2 v = make_float2(1.0f, 1.0f);
-	for(int i = 0; i < 8; i++) {
-		float x = v.x * matrix[0] + v.y * matrix[1];
-		float y = v.x * matrix[1] + v.y * matrix[2];
-		float m = max(x, y);        
-		float iv = 1.0f / m;
-		if (m == 0.0f) iv = 0.0f;
-		v = make_float2(x*iv, y*iv);
-	}
-
-	return v;
-}
-
-inline __device__ void colorSums(const float2 * colors, float2 * sums)
-{
-#if __DEVICE_EMULATION__
-	float2 color_sum = make_float2(0.0f, 0.0f, 0.0f);
-	for (int i = 0; i < 16; i++)
-	{
-		color_sum += colors[i];
-	}
-
-	for (int i = 0; i < 16; i++)
-	{
-		sums[i] = color_sum;
-	}
-#else
-
-	const int idx = threadIdx.x;
-
-	sums[idx] = colors[idx];
-	sums[idx] += sums[idx^8];
-	sums[idx] += sums[idx^4];
-	sums[idx] += sums[idx^2];
-	sums[idx] += sums[idx^1];
-
-#endif
-}
-
-inline __device__ float2 bestFitLine(const float2 * colors, float2 color_sum)
-{
-	// Compute covariance matrix of the given colors.
-#if __DEVICE_EMULATION__
-	float covariance[3] = {0, 0, 0};
-	for (int i = 0; i < 16; i++)
-	{
-		float2 a = (colors[i] - color_sum * (1.0f / 16.0f));
-		covariance[0] += a.x * a.x;
-		covariance[1] += a.x * a.y;
-		covariance[3] += a.y * a.y;
-	}
-#else
-
-	const int idx = threadIdx.x;
-
-	float2 diff = (colors[idx] - color_sum * (1.0f / 16.0f));
-
-	__shared__ float covariance[16*3];
-
-	covariance[3 * idx + 0] = diff.x * diff.x;
-	covariance[3 * idx + 1] = diff.x * diff.y;
-	covariance[3 * idx + 2] = diff.y * diff.y;
-
-	for(int d = 8; d > 0; d >>= 1)
-	{
-		if (idx < d)
-		{
-			covariance[3 * idx + 0] += covariance[3 * (idx+d) + 0];
-			covariance[3 * idx + 1] += covariance[3 * (idx+d) + 1];
-			covariance[3 * idx + 2] += covariance[3 * (idx+d) + 2];
-		}
-	}
-
-#endif
-
-	// Compute first eigen vector.
-	return firstEigenVector2D(covariance);
-}
-
 
 #endif // CUDAMATH_H

src/nvtt/cuda/CudaUtils.cpp

@@ -1,128 +1,300 @@
-// Copyright NVIDIA Corporation 2007 -- Ignacio Castano <icastano@nvidia.com>
-// 
-// Permission is hereby granted, free of charge, to any person
-// obtaining a copy of this software and associated documentation
-// files (the "Software"), to deal in the Software without
-// restriction, including without limitation the rights to use,
-// copy, modify, merge, publish, distribute, sublicense, and/or sell
-// copies of the Software, and to permit persons to whom the
-// Software is furnished to do so, subject to the following
-// conditions:
-// 
-// The above copyright notice and this permission notice shall be
-// included in all copies or substantial portions of the Software.
-// 
-// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
-// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
-// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
-// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
-// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
-// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
-// OTHER DEALINGS IN THE SOFTWARE.
-
-#include <nvcore/Debug.h>
-#include "CudaUtils.h"
-
-#if defined HAVE_CUDA
-#include <cuda_runtime.h>
-#endif
-
-using namespace nv;
-using namespace cuda;
-
-#if NV_OS_WIN32
-
-#define WINDOWS_LEAN_AND_MEAN
-#include <windows.h>
-
-static bool isWindowsVista()
-{
-	OSVERSIONINFO osvi;
-	osvi.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);
-
-	::GetVersionEx(&osvi);
-	return osvi.dwMajorVersion >= 6;
-}
-
-
-typedef BOOL (WINAPI *LPFN_ISWOW64PROCESS) (HANDLE, PBOOL);
-
-static bool isWow32()
-{
-	LPFN_ISWOW64PROCESS fnIsWow64Process = (LPFN_ISWOW64PROCESS)GetProcAddress(GetModuleHandle("kernel32"), "IsWow64Process");
-
-    BOOL bIsWow64 = FALSE;
- 
-    if (NULL != fnIsWow64Process)
-    {
-        if (!fnIsWow64Process(GetCurrentProcess(), &bIsWow64))
-        {
-			// Assume 32 bits.
-            return true;
-        }
-    }
-
-    return !bIsWow64;
-}
-
-#endif
-
-
-/// Determine if CUDA is available.
-bool nv::cuda::isHardwarePresent()
-{
-#if defined HAVE_CUDA
-#if NV_OS_WIN32
-	if (isWindowsVista()) return false;
-	//if (isWindowsVista() || !isWow32()) return false;
-#endif
-	int count = deviceCount();
-	if (count == 1)
-	{
-		// Make sure it's not an emulation device.
-		cudaDeviceProp deviceProp;
-		cudaGetDeviceProperties(&deviceProp, 0);
-
-		// deviceProp.name != Device Emulation (CPU)
-		if (deviceProp.major == -1 || deviceProp.minor == -1)
-		{
-			return false;
-		}
-
-		// @@ Make sure that warp size == 32
-	}
-
-	return count > 0;
-#else
-	return false;
-#endif
-}
-
-/// Get number of CUDA enabled devices.
-int nv::cuda::deviceCount()
-{
-#if defined HAVE_CUDA
-	int gpuCount = 0;
-
-	cudaError_t result = cudaGetDeviceCount(&gpuCount);
-
-	if (result == cudaSuccess)
-	{
-		return gpuCount;
-	}
-#endif
-	return 0;
-}
-
-/// Activate the given devices.
-bool nv::cuda::setDevice(int i)
-{
-	nvCheck(i < deviceCount());
-#if defined HAVE_CUDA
-	cudaError_t result = cudaSetDevice(i);
-	return result == cudaSuccess;
-#else
-	return false;
-#endif
-}
+// Copyright NVIDIA Corporation 2007 -- Ignacio Castano <icastano@nvidia.com>
+// 
+// Permission is hereby granted, free of charge, to any person
+// obtaining a copy of this software and associated documentation
+// files (the "Software"), to deal in the Software without
+// restriction, including without limitation the rights to use,
+// copy, modify, merge, publish, distribute, sublicense, and/or sell
+// copies of the Software, and to permit persons to whom the
+// Software is furnished to do so, subject to the following
+// conditions:
+// 
+// The above copyright notice and this permission notice shall be
+// included in all copies or substantial portions of the Software.
+// 
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+// OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+// WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+// OTHER DEALINGS IN THE SOFTWARE.
+
+#include <nvcore/Debug.h>
+#include <nvcore/Library.h>
+#include "CudaUtils.h"
+
+#if defined HAVE_CUDA
+#include <cuda.h>
+#include <cuda_runtime_api.h>
+#endif
+
+using namespace nv;
+using namespace cuda;
+
+/* @@ Move this to win32 utils or somewhere else.
+#if NV_OS_WIN32
+
+#define WINDOWS_LEAN_AND_MEAN
+#include <windows.h>
+
+static bool isWindowsVista()
+{
+OSVERSIONINFO osvi;
+osvi.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);
+
+::GetVersionEx(&osvi);
+return osvi.dwMajorVersion >= 6;
+}
+
+
+typedef BOOL (WINAPI *LPFN_ISWOW64PROCESS) (HANDLE, PBOOL);
+
+static bool isWow32()
+{
+LPFN_ISWOW64PROCESS fnIsWow64Process = (LPFN_ISWOW64PROCESS)GetProcAddress(GetModuleHandle("kernel32"), "IsWow64Process");
+
+BOOL bIsWow64 = FALSE;
+
+if (NULL != fnIsWow64Process)
+{
+if (!fnIsWow64Process(GetCurrentProcess(), &bIsWow64))
+{
+// Assume 32 bits.
+return true;
+}
+}
+
+return !bIsWow64;
+}
+
+#endif
+*/
+
+
+static bool isCudaDriverAvailable(int version)
+{
+#if defined HAVE_CUDA
+#if NV_OS_WIN32
+	Library nvcuda("nvcuda.dll");
+#else
+	Library nvcuda(NV_LIBRARY_NAME(cuda));
+#endif
+
+	if (!nvcuda.isValid())
+	{
+		nvDebug("*** CUDA driver not found.\n");
+		return false;
+	}
+
+	if (version >= 2000)
+	{
+		void * address = nvcuda.bindSymbol("cuStreamCreate");
+		if (address == NULL) {
+			nvDebug("*** CUDA driver version < 2.0.\n");
+			return false;
+		}
+	}
+
+	if (version >= 2010)
+	{
+		void * address = nvcuda.bindSymbol("cuModuleLoadDataEx");
+		if (address == NULL) {
+			nvDebug("*** CUDA driver version < 2.1.\n");
+			return false;
+		}
+	}
+
+	if (version >= 2020)
+	{
+		typedef CUresult (CUDAAPI * PFCU_DRIVERGETVERSION)(int * version);
+
+		PFCU_DRIVERGETVERSION driverGetVersion = (PFCU_DRIVERGETVERSION)nvcuda.bindSymbol("cuDriverGetVersion");
+		if (driverGetVersion == NULL) {
+			nvDebug("*** CUDA driver version < 2.2.\n");
+			return false;
+		}
+
+		int driverVersion;
+		CUresult err = driverGetVersion(&driverVersion);
+		if (err != CUDA_SUCCESS) {
+			nvDebug("*** Error querying driver version: '%s'.\n", cudaGetErrorString((cudaError_t)err));
+			return false;
+		}
+
+		return driverVersion >= version;
+	}
+#endif // HAVE_CUDA
+
+	return true;
+}
+
+
+/// Determine if CUDA is available.
+bool nv::cuda::isHardwarePresent()
+{
+#if defined HAVE_CUDA
+	// Make sure that CUDA driver matches CUDA runtime.
+	if (!isCudaDriverAvailable(CUDART_VERSION))
+	{
+		nvDebug("CUDA driver not available for CUDA runtime %d\n", CUDART_VERSION);
+		return false;
+	}
+
+	int count = deviceCount();
+	if (count == 1)
+	{
+		// Make sure it's not an emulation device.
+		cudaDeviceProp deviceProp;
+		cudaGetDeviceProperties(&deviceProp, 0);
+
+		// deviceProp.name != Device Emulation (CPU)
+		if (deviceProp.major == -1 || deviceProp.minor == -1)
+		{
+			return false;
+		}
+	}
+
+	// @@ Make sure that warp size == 32
+
+	// @@ Make sure available GPU is faster than the CPU.
+
+	return count > 0;
+#else
+	return false;
+#endif
+}
+
+/// Get number of CUDA enabled devices.
+int nv::cuda::deviceCount()
+{
+#if defined HAVE_CUDA
+	int gpuCount = 0;
+
+	cudaError_t result = cudaGetDeviceCount(&gpuCount);
+
+	if (result == cudaSuccess)
+	{
+		return gpuCount;
+	}
+#endif
+	return 0;
+}
+
+
+// Make sure device meets requirements:
+// - Not an emulation device.
+// - Not an integrated device?
+// - Faster than CPU.
+bool nv::cuda::isValidDevice(int i)
+{
+#if defined HAVE_CUDA
+
+	cudaDeviceProp device_properties;
+	cudaGetDeviceProperties(&device_properties, i);
+	int gflops = device_properties.multiProcessorCount * device_properties.clockRate;
+
+	if (device_properties.major == -1 || device_properties.minor == -1) {
+		// Emulation device.
+		return false;
+	}
+
+#if CUDART_VERSION >= 2030 // 2.3
+	/*if (device_properties.integrated)
+	{
+		// Integrated devices.
+		return false;
+	}*/
+#endif
+
+	return true;
+#else
+	return false;
+#endif
+}
+
+int nv::cuda::getFastestDevice()
+{
+	int max_gflops_device = -1;
+#if defined HAVE_CUDA
+	int max_gflops = 0;
+
+	const int device_count = deviceCount();
+	for (int i = 0; i < device_count; i++)
+	{
+		if (isValidDevice(i))
+		{
+			cudaDeviceProp device_properties;
+			cudaGetDeviceProperties(&device_properties, i);
+			int gflops = device_properties.multiProcessorCount * device_properties.clockRate;
+
+			if (gflops > max_gflops)
+			{
+				max_gflops = gflops;
+				max_gflops_device = i;
+			}
+		}
+	}
+#endif
+	return max_gflops_device;
+}
+
+
+/// Activate the given devices.
+bool nv::cuda::initDevice(int * device_ptr)
+{
+	nvDebugCheck(device_ptr != NULL);
+#if defined HAVE_CUDA
+
+#if CUDART_VERSION >= 2030 // 2.3
+
+	// Set device flags to yield in order to play nice with other threads and to find out if CUDA was already active.
+	cudaError_t resul = cudaSetDeviceFlags(cudaDeviceScheduleYield);
+
+#endif
+
+	int device = getFastestDevice();
+
+	if (device == -1)
+	{
+		// No device is fast enough.
+		*device_ptr = -1;
+		return false;
+	}
+
+	// Select CUDA device.
+	cudaError_t result = cudaSetDevice(device);
+
+	if (result == cudaErrorSetOnActiveProcess)
+	{
+		int device;
+		result = cudaGetDevice(&device);
+
+		*device_ptr = -1;  // No device to cleanup.
+		return isValidDevice(device); // Return true if device is valid.
+	}
+	else if (result != cudaSuccess)
+	{
+		nvDebug("*** CUDA Error: %s\n", cudaGetErrorString(result));
+		*device_ptr = -1;
+		return false;
+	}
+
+	*device_ptr = device;
+	return true;
+#else
+	return false;
+#endif
+}
+
+void nv::cuda::exitDevice()
+{
+#if defined HAVE_CUDA
+	cudaError_t result = cudaThreadExit();
+
+	if (result != cudaSuccess) {
+		nvDebug("*** CUDA Error: %s\n", cudaGetErrorString(result));
+	}
+#endif
+}

src/nvtt/cuda/CudaUtils.h

@@ -31,7 +31,11 @@ namespace nv
 	{
 		bool isHardwarePresent();
 		int deviceCount();
-		bool setDevice(int i);
+		int getFastestDevice();
+		bool isValidDevice(int i);
+
+		bool initDevice(int * device_ptr);
+		void exitDevice();
 	};
 	
 } // nv namespace

src/nvtt/experimental/nvtt_experimental.cpp

@@ -1,60 +0,0 @@
-
-#include "nvtt_experimental.h"
-
-struct NvttImage
-{
-	NvttImage() :
-		m_constant(false),
-		m_image(NULL),
-		m_floatImage(NULL)
-	{
-	}
-	
-	~NvttImage()
-	{
-		if (m_constant && m_image) m_image->unwrap();
-		delete m_image;
-		delete m_floatImage;
-	}
-	
-	bool m_constant;
-	Image * m_image;
-	FloatImage * m_floatImage;
-};
-
-NvttImage * nvttCreateImage() 
-{
-	return new NvttImage();
-}
-	
-void nvttDestroyImage(NvttImage * img)
-{
-	delete img;
-}
-
-void nvttSetImageData(NvttImage * img, NvttInputFormat format, uint w, uint h, void * data)
-{
-	nvCheck(img != NULL);
-	
-	if (format == NVTT_InputFormat_BGRA_8UB)
-	{
-		img->m_constant = false;
-		img->m_image->allocate(w, h);
-		memcpy(img->m_image->pixels(), data, w * h * 4);
-	}
-	else
-	{
-		nvCheck(false);
-	}
-}
-
-void nvttCompressImage(NvttImage * img, NvttFormat format)
-{
-	nvCheck(img != NULL);
-
-	// @@ Invoke appropriate compressor.
-}
-
-
-
-#endif // NVTT_EXPERIMENTAL_H

src/nvtt/experimental/nvtt_experimental.h

@@ -1,55 +0,0 @@
-
-#ifndef NVTT_EXPERIMENTAL_H
-#define NVTT_EXPERIMENTAL_H
-
-#include <nvtt/nvtt.h>
-
-typedef struct NvttImage NvttImage;
-
-NvttImage * nvttCreateImage();
-void nvttDestroyImage(NvttImage * img);
-
-void nvttSetImageData(NvttImage * img, NvttInputFormat format, uint w, uint h, void * data);
-
-void nvttCompressImage(NvttImage * img, NvttFormat format);
-
-// How to control the compression parameters?
-
-// Using many arguments:
-// void nvttCompressImage(img, format, quality, r, g, b, a, ...);
-
-// Using existing compression option class:
-// compressionOptions = nvttCreateCompressionOptions();
-// nvttSetCompressionOptionsFormat(compressionOptions, format);
-// nvttSetCompressionOptionsQuality(compressionOptions, quality);
-// nvttSetCompressionOptionsQuality(compressionOptions, quality);
-// nvttSetCompressionOptionsColorWeights(compressionOptions, r, g, b, a);
-// ...
-// nvttCompressImage(img, compressionOptions);
-
-// Using thread local context state:
-// void nvttSetCompressionFormat(format);
-// void nvttSetCompressionQuality(quality);
-// void nvttSetCompressionColorWeights(r, g, b, a);
-// ...
-// nvttCompressImage(img);
-
-// Using thread local context state, but with GL style function arguments:
-// nvttCompressorParameteri(NVTT_FORMAT, format);
-// nvttCompressorParameteri(NVTT_QUALITY, quality);
-// nvttCompressorParameterf(NVTT_COLOR_WEIGHT_RED, r);
-// nvttCompressorParameterf(NVTT_COLOR_WEIGHT_GREEN, g);
-// nvttCompressorParameterf(NVTT_COLOR_WEIGHT_BLUE, b);
-// nvttCompressorParameterf(NVTT_COLOR_WEIGHT_ALPHA, a);
-// or nvttCompressorParameter4f(NVTT_COLOR_WEIGHTS, r, g, b, a);
-// ...
-// nvttCompressImage(img);
-
-// How do we get the compressed output?
-// - Using callbacks. (via new entrypoints, or through outputOptions)
-// - Return it explicitely from nvttCompressImage.
-// - Store it along the image, retrieve later explicitely with 'nvttGetCompressedData(img, ...)'
-
-
-
-#endif // NVTT_EXPERIMENTAL_H

src/nvtt/nvtt.h

@@ -73,7 +73,7 @@ namespace nvtt
 		Format_DXT1a,   // DXT1 with binary alpha.
 		Format_DXT3,
 		Format_DXT5,
-		Format_DXT5n,   // Compressed HILO: R=0, G=x, B=0, A=y
+		Format_DXT5n,   // Compressed HILO: R=1, G=y, B=0, A=x
 		
 		// DX10 formats.
 		Format_BC1 = Format_DXT1,
@@ -83,9 +83,6 @@ namespace nvtt
 		Format_BC3n = Format_DXT5n,
 		Format_BC4,     // ATI1
 		Format_BC5,     // 3DC, ATI2
-
-		Format_DXT1n,
-		Format_CTX1,
 	};
 	
 	/// Quality modes.
@@ -157,7 +154,6 @@ namespace nvtt
 	{
 		ColorTransform_None,
 		ColorTransform_Linear,
-		ColorTransform_Swizzle
 	};
 	
 	/// Extents rounding mode.
@@ -198,7 +194,7 @@ namespace nvtt
 		// Describe the format of the input.
 		NVTT_API void setFormat(InputFormat format);
 		
-		// Set the way the input alpha channel is interpreted. @@ Not implemented!
+		// Set the way the input alpha channel is interpreted.
 		NVTT_API void setAlphaMode(AlphaMode alphaMode);
 		
 		// Set gamma settings.
@@ -222,14 +218,10 @@ namespace nvtt
 		// Set color transforms. @@ Not implemented!
 		NVTT_API void setColorTransform(ColorTransform t);
 		NVTT_API void setLinearTransform(int channel, float w0, float w1, float w2, float w3);
-		NVTT_API void setSwizzleTransform(int x, int y, int z, int w);
 		
 		// Set resizing options.
 		NVTT_API void setMaxExtents(int d);
 		NVTT_API void setRoundMode(RoundMode mode);
-
-		// Set whether or not to premultiply color by alpha
-		NVTT_API void setPremultiplyAlpha(bool b);
 	};
 	
 	

src/nvtt/squish/CMakeLists.txt

@@ -1,13 +1,8 @@
 PROJECT(squish)
-ENABLE_TESTING()
 
 INCLUDE_DIRECTORIES(${CMAKE_CURRENT_SOURCE_DIR})
 
 SET(SQUISH_SRCS
-#	alpha.cpp
-#	alpha.h
-#	clusterfit.cpp
-#	clusterfit.h
 	fastclusterfit.cpp
 	fastclusterfit.h
 	weightedclusterfit.cpp
@@ -21,32 +16,13 @@ SET(SQUISH_SRCS
 	config.h
 	maths.cpp
 	maths.h
-#	rangefit.cpp
-#	rangefit.h
-#	singlecolourfit.cpp
-#	singlecolourfit.h
-#	singlecolourlookup.inl
-#	squish.cpp
-#	squish.h
 	simd.h
 	simd_sse.h
 	simd_ve.h)
 
 ADD_LIBRARY(squish STATIC ${SQUISH_SRCS})
 
-# libpng
-#FIND_PACKAGE(PNG)
-
-#IF(PNG_FOUND)
-#	INCLUDE_DIRECTORIES(${PNG_INCLUDE_DIR})
-#	ADD_EXECUTABLE(squishpng extra/squishpng.cpp)
-#	TARGET_LINK_LIBRARIES(squishpng squish ${PNG_LIBRARY})
-#ENDIF(PNG_FOUND)
-
-##ADD_EXECUTABLE(squishgen extra/squishgen.cpp)
-
-#ADD_EXECUTABLE(squishtest extra/squishtest.cpp)
-#TARGET_LINK_LIBRARIES(squishtest squish)
-
-#ADD_TEST(SQUISHTEST squishtest)
+IF(CMAKE_COMPILER_IS_GNUCXX)
+	SET_TARGET_PROPERTIES(squish PROPERTIES COMPILE_FLAGS -fPIC)
+ENDIF(CMAKE_COMPILER_IS_GNUCXX)
 

src/nvtt/squish/fastclusterfit.cpp

@@ -29,6 +29,8 @@
 #include "colourblock.h"
 #include <cfloat>
 
+#include "fastclusterlookup.inl"
+
 namespace squish {
 
 FastClusterFit::FastClusterFit()
@@ -97,91 +99,6 @@ void FastClusterFit::SetColourSet( ColourSet const* colours, int flags )
 }
 
 
-struct Precomp {
-	float alpha2_sum;
-	float beta2_sum;
-	float alphabeta_sum;
-	float factor;
-};
-
-static SQUISH_ALIGN_16 Precomp s_threeElement[153];
-static SQUISH_ALIGN_16 Precomp s_fourElement[969];
-
-void FastClusterFit::DoPrecomputation()
-{
-	int i = 0;
-	
-	// Three element clusters:
-	for( int c0 = 0; c0 <= 16; c0++)	// At least two clusters.
-	{
-		for( int c1 = 0; c1 <=  16-c0; c1++)
-		{
-			int c2 = 16 - c0 - c1;
-
-			/*if (c2 == 16) {
-				// a = b = x2 / 16
-				s_threeElement[i].alpha2_sum = 0;
-				s_threeElement[i].beta2_sum = 16;
-				s_threeElement[i].alphabeta_sum = -16;
-				s_threeElement[i].factor = 1.0f / 256.0f;
-			}
-			else if (c0 == 16) {
-				// a = b = x0 / 16
-				s_threeElement[i].alpha2_sum = 16;
-				s_threeElement[i].beta2_sum = 0;
-				s_threeElement[i].alphabeta_sum = -16;
-				s_threeElement[i].factor = 1.0f / 256.0f;
-			}
-			else*/ {
-				s_threeElement[i].alpha2_sum = c0 + c1 * 0.25f;
-				s_threeElement[i].beta2_sum = c2 + c1 * 0.25f;
-				s_threeElement[i].alphabeta_sum = c1 * 0.25f;
-				s_threeElement[i].factor = 1.0f / (s_threeElement[i].alpha2_sum * s_threeElement[i].beta2_sum - s_threeElement[i].alphabeta_sum * s_threeElement[i].alphabeta_sum);
-			}
-			
-			i++;
-		}
-	}
-	//printf("%d three cluster elements\n", i);
-	
-	// Four element clusters:
-	i = 0;
-	for( int c0 = 0; c0 <= 16; c0++)
-	{
-		for( int c1 = 0; c1 <=  16-c0; c1++)
-		{
-			for( int c2 = 0; c2 <=  16-c0-c1; c2++)
-			{
-				int c3 = 16 - c0 - c1 - c2;
-				
-				/*if (c3 == 16) {
-					// a = b = x3 / 16
-					s_fourElement[i].alpha2_sum = 16.0f;
-					s_fourElement[i].beta2_sum = 0.0f;
-					s_fourElement[i].alphabeta_sum = -16.0f;
-					s_fourElement[i].factor = 1.0f / 256.0f;					
-				}
-				else if (c0 == 16) {
-					// a = b = x0 / 16
-					s_fourElement[i].alpha2_sum = 0.0f;
-					s_fourElement[i].beta2_sum = 16.0f;
-					s_fourElement[i].alphabeta_sum = -16.0f;
-					s_fourElement[i].factor = 1.0f / 256.0f;					
-				}
-				else*/ {
-					s_fourElement[i].alpha2_sum = c0 + c1 * (4.0f/9.0f) + c2 * (1.0f/9.0f);
-					s_fourElement[i].beta2_sum = c3 + c2 * (4.0f/9.0f) + c1 * (1.0f/9.0f);
-					s_fourElement[i].alphabeta_sum = (c1 + c2) * (2.0f/9.0f);
-					s_fourElement[i].factor = 1.0f / (s_fourElement[i].alpha2_sum * s_fourElement[i].beta2_sum - s_fourElement[i].alphabeta_sum * s_fourElement[i].alphabeta_sum);
-				}
-
-				i++;
-			}
-		}
-	}
-	//printf("%d four cluster elements\n", i);
-}
-
 void FastClusterFit::SetMetric(float r, float g, float b)
 {
 #if SQUISH_USE_SIMD

src/nvtt/squish/fastclusterfit.h

@@ -44,8 +44,6 @@ public:
 	void SetMetric(float r, float g, float b);
 	float GetBestError() const;
 
-	static void DoPrecomputation();
-
 	// Make them public
 	virtual void Compress3( void* block );
 	virtual void Compress4( void* block );