288 lines
7.3 KiB
C++
288 lines
7.3 KiB
C++
// Copyright (C) 2017 Jérôme Leclercq
|
|
// This file is part of the "Nazara Engine - Utility module"
|
|
// For conditions of distribution and use, see copyright notice in Config.hpp
|
|
|
|
#include <Nazara/Core/Error.hpp>
|
|
#include <array>
|
|
#include <cstring>
|
|
#include <Nazara/Utility/Debug.hpp>
|
|
|
|
namespace Nz
|
|
{
|
|
inline PixelFormatInfo::PixelFormatInfo() :
|
|
content(PixelFormatContent_Undefined),
|
|
bitsPerPixel(0)
|
|
{
|
|
}
|
|
|
|
inline PixelFormatInfo::PixelFormatInfo(PixelFormatContent formatContent, UInt8 bpp, PixelFormatSubType subType) :
|
|
content(formatContent),
|
|
redType(subType),
|
|
greenType(subType),
|
|
blueType(subType),
|
|
alphaType(subType),
|
|
bitsPerPixel(bpp)
|
|
{
|
|
}
|
|
|
|
inline PixelFormatInfo::PixelFormatInfo(const String& formatName, PixelFormatContent formatContent, UInt8 bpp, PixelFormatSubType subType) :
|
|
content(formatContent),
|
|
redType(subType),
|
|
greenType(subType),
|
|
blueType(subType),
|
|
alphaType(subType),
|
|
name(formatName),
|
|
bitsPerPixel(bpp)
|
|
{
|
|
}
|
|
|
|
inline PixelFormatInfo::PixelFormatInfo(const String& formatName, PixelFormatContent formatContent, Bitset<> rMask, Bitset<> gMask, Bitset<> bMask, Bitset<> aMask, PixelFormatSubType subType) :
|
|
PixelFormatInfo(formatName, formatContent, subType, rMask, subType, gMask, subType, bMask, subType, aMask)
|
|
{
|
|
}
|
|
|
|
inline PixelFormatInfo::PixelFormatInfo(const String& formatName, PixelFormatContent formatContent, PixelFormatSubType rType, Bitset<> rMask, PixelFormatSubType gType, Bitset<> gMask, PixelFormatSubType bType, Bitset<> bMask, PixelFormatSubType aType, Bitset<> aMask, UInt8 bpp) :
|
|
redMask(rMask),
|
|
greenMask(gMask),
|
|
blueMask(bMask),
|
|
alphaMask(aMask),
|
|
content(formatContent),
|
|
redType(rType),
|
|
greenType(gType),
|
|
blueType(bType),
|
|
alphaType(aType),
|
|
name(formatName)
|
|
{
|
|
redMask.Reverse();
|
|
greenMask.Reverse();
|
|
blueMask.Reverse();
|
|
alphaMask.Reverse();
|
|
|
|
if (bpp == 0)
|
|
RecomputeBitsPerPixel();
|
|
}
|
|
|
|
inline void PixelFormatInfo::Clear()
|
|
{
|
|
bitsPerPixel = 0;
|
|
alphaMask.Clear();
|
|
blueMask.Clear();
|
|
greenMask.Clear();
|
|
redMask.Clear();
|
|
name.Clear();
|
|
}
|
|
|
|
inline bool PixelFormatInfo::IsCompressed() const
|
|
{
|
|
return redType == PixelFormatSubType_Compressed ||
|
|
greenType == PixelFormatSubType_Compressed ||
|
|
blueType == PixelFormatSubType_Compressed ||
|
|
alphaType == PixelFormatSubType_Compressed;
|
|
}
|
|
|
|
inline bool PixelFormatInfo::IsValid() const
|
|
{
|
|
return bitsPerPixel != 0;
|
|
}
|
|
|
|
inline void PixelFormatInfo::RecomputeBitsPerPixel()
|
|
{
|
|
Bitset<> counter;
|
|
counter |= redMask;
|
|
counter |= greenMask;
|
|
counter |= blueMask;
|
|
counter |= alphaMask;
|
|
|
|
bitsPerPixel = static_cast<UInt8>(counter.Count());
|
|
}
|
|
|
|
inline bool PixelFormatInfo::Validate() const
|
|
{
|
|
if (!IsValid())
|
|
return false;
|
|
|
|
if (content <= PixelFormatContent_Undefined || content > PixelFormatContent_Max)
|
|
return false;
|
|
|
|
std::array<const Nz::Bitset<>*, 4> masks = { {&redMask, &greenMask, &blueMask, &alphaMask} };
|
|
std::array<PixelFormatSubType, 4> types = { {redType, greenType, blueType, alphaType} };
|
|
|
|
for (unsigned int i = 0; i < 4; ++i)
|
|
{
|
|
UInt8 usedBits = static_cast<UInt8>(masks[i]->Count());
|
|
if (usedBits == 0)
|
|
continue;
|
|
|
|
if (usedBits > bitsPerPixel)
|
|
return false;
|
|
|
|
if (usedBits > 64) //< Currently, formats with over 64 bits per component are not supported
|
|
return false;
|
|
|
|
switch (types[i])
|
|
{
|
|
case PixelFormatSubType_Half:
|
|
if (usedBits != 16)
|
|
return false;
|
|
|
|
break;
|
|
|
|
case PixelFormatSubType_Float:
|
|
if (usedBits != 32)
|
|
return false;
|
|
|
|
break;
|
|
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
|
|
|
|
inline std::size_t PixelFormat::ComputeSize(PixelFormatType format, unsigned int width, unsigned int height, unsigned int depth)
|
|
{
|
|
if (IsCompressed(format))
|
|
{
|
|
switch (format)
|
|
{
|
|
case PixelFormatType_DXT1:
|
|
case PixelFormatType_DXT3:
|
|
case PixelFormatType_DXT5:
|
|
return (((width + 3) / 4) * ((height + 3) / 4) * ((format == PixelFormatType_DXT1) ? 8 : 16)) * depth;
|
|
|
|
default:
|
|
NazaraError("Unsupported format");
|
|
return 0;
|
|
}
|
|
}
|
|
else
|
|
return width * height * depth * GetBytesPerPixel(format);
|
|
}
|
|
|
|
inline bool PixelFormat::Convert(PixelFormatType srcFormat, PixelFormatType dstFormat, const void* src, void* dst)
|
|
{
|
|
if (srcFormat == dstFormat)
|
|
{
|
|
std::memcpy(dst, src, GetBytesPerPixel(srcFormat));
|
|
return true;
|
|
}
|
|
|
|
#if NAZARA_UTILITY_SAFE
|
|
if (IsCompressed(srcFormat))
|
|
{
|
|
NazaraError("Cannot convert single pixel from compressed format");
|
|
return false;
|
|
}
|
|
|
|
if (IsCompressed(dstFormat))
|
|
{
|
|
NazaraError("Cannot convert single pixel to compressed format");
|
|
return false;
|
|
}
|
|
#endif
|
|
|
|
ConvertFunction func = s_convertFunctions[srcFormat][dstFormat];
|
|
if (!func)
|
|
{
|
|
NazaraError("Pixel format conversion from " + GetName(srcFormat) + " to " + GetName(dstFormat) + " is not supported");
|
|
return false;
|
|
}
|
|
|
|
if (!func(reinterpret_cast<const UInt8*>(src), reinterpret_cast<const UInt8*>(src) + GetBytesPerPixel(srcFormat), reinterpret_cast<UInt8*>(dst)))
|
|
{
|
|
NazaraError("Pixel format conversion from " + GetName(srcFormat) + " to " + GetName(dstFormat) + " failed");
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
inline bool PixelFormat::Convert(PixelFormatType srcFormat, PixelFormatType dstFormat, const void* start, const void* end, void* dst)
|
|
{
|
|
if (srcFormat == dstFormat)
|
|
{
|
|
std::memcpy(dst, start, reinterpret_cast<const UInt8*>(end)-reinterpret_cast<const UInt8*>(start));
|
|
return true;
|
|
}
|
|
|
|
ConvertFunction func = s_convertFunctions[srcFormat][dstFormat];
|
|
if (!func)
|
|
{
|
|
NazaraError("Pixel format conversion from " + GetName(srcFormat) + " to " + GetName(dstFormat) + " is not supported");
|
|
return false;
|
|
}
|
|
|
|
if (!func(reinterpret_cast<const UInt8*>(start), reinterpret_cast<const UInt8*>(end), reinterpret_cast<UInt8*>(dst)))
|
|
{
|
|
NazaraError("Pixel format conversion from " + GetName(srcFormat) + " to " + GetName(dstFormat) + " failed");
|
|
return false;
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
inline UInt8 PixelFormat::GetBitsPerPixel(PixelFormatType format)
|
|
{
|
|
return s_pixelFormatInfos[format].bitsPerPixel;
|
|
}
|
|
|
|
inline UInt8 PixelFormat::GetBytesPerPixel(PixelFormatType format)
|
|
{
|
|
return GetBitsPerPixel(format)/8;
|
|
}
|
|
|
|
inline PixelFormatContent PixelFormat::GetContent(PixelFormatType format)
|
|
{
|
|
return s_pixelFormatInfos[format].content;
|
|
}
|
|
|
|
inline const PixelFormatInfo& PixelFormat::GetInfo(PixelFormatType format)
|
|
{
|
|
return s_pixelFormatInfos[format];
|
|
}
|
|
|
|
inline const String& PixelFormat::GetName(PixelFormatType format)
|
|
{
|
|
return s_pixelFormatInfos[format].name;
|
|
}
|
|
|
|
inline bool PixelFormat::HasAlpha(PixelFormatType format)
|
|
{
|
|
return s_pixelFormatInfos[format].alphaMask.TestAny();
|
|
}
|
|
|
|
inline bool PixelFormat::IsCompressed(PixelFormatType format)
|
|
{
|
|
return s_pixelFormatInfos[format].IsCompressed();
|
|
}
|
|
|
|
inline bool PixelFormat::IsConversionSupported(PixelFormatType srcFormat, PixelFormatType dstFormat)
|
|
{
|
|
if (srcFormat == dstFormat)
|
|
return true;
|
|
|
|
return s_convertFunctions[srcFormat][dstFormat] != nullptr;
|
|
}
|
|
|
|
inline bool PixelFormat::IsValid(PixelFormatType format)
|
|
{
|
|
return format != PixelFormatType_Undefined;
|
|
}
|
|
|
|
inline void PixelFormat::SetConvertFunction(PixelFormatType srcFormat, PixelFormatType dstFormat, ConvertFunction func)
|
|
{
|
|
s_convertFunctions[srcFormat][dstFormat] = func;
|
|
}
|
|
|
|
inline void PixelFormat::SetFlipFunction(PixelFlipping flipping, PixelFormatType format, FlipFunction func)
|
|
{
|
|
s_flipFunctions[flipping][format] = func;
|
|
}
|
|
}
|
|
|
|
#include <Nazara/Utility/DebugOff.hpp>
|