0ct0pu5/wesnoth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update xBRZ to v1.8 (#9208)

Browse source 
* Update xBRZ to v1.8 (resolves #8307)

This replaces our implementation with stock xBRZ sans any of our custom code. Made these changes from plain source:
- Renaming .h -> .hpp
- Conditionally use [[likely]]
- Added a trailing newline
- Comment out unused parameters
- Disable `-Wunused-function` warning on macOS
This commit is contained in:
Charles Dang 2024-08-14 01:13:37 -04:00 • committed by GitHub
parent dac8f6f5e0
commit b874b8c6d4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
7 changed files with 1194 additions and 881 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

19
src/global.hpp
View file

@ -53,10 +53,29 @@
* attributes in the gnu:: namespace, so has to have the #if, and the #if means we need the #ifdef.
*/
#ifdef __has_cpp_attribute
#if __has_cpp_attribute(gnu::no_dangling)
#define NOT_DANGLING [[gnu::no_dangling]]
#endif
#if __has_cpp_attribute(likely)
#define LIKELY [[likely]]
#endif
#if __has_cpp_attribute(unlikely)
#define UNLIKELY [[unlikely]]
#endif
#endif // __has_cpp_attribute
#ifndef NOT_DANGLING
#define NOT_DANGLING
#endif
#ifndef LIKELY
#define LIKELY
#endif
#ifndef UNLIKELY
#define UNLIKELY
#endif

4
src/sdl/utils.cpp
View file

@ -85,7 +85,7 @@ surface scale_surface_xbrz(const surface & surf, std::size_t z)
const_surface_lock src_lock(surf);
surface_lock dst_lock(dst);
xbrz::scale(z, src_lock.pixels(), dst_lock.pixels(), surf->w, surf->h);
xbrz::scale(z, src_lock.pixels(), dst_lock.pixels(), surf->w, surf->h, xbrz::ColorFormat::ARGB);
}
return dst;
@ -1838,7 +1838,7 @@ namespace
constexpr bool not_alpha(uint32_t pixel)
{
return (pixel >> 24) != 0x00;
}
}
}
rect get_non_transparent_portion(const surface &surf)

62
src/xBRZ/config.hpp
View file

@ -1,62 +0,0 @@
/*
Copyright (C) 2014 - 2018 by Chris Beck <render787@gmail.com>
Part of the Battle for Wesnoth Project https://www.wesnoth.org/
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY.
See the COPYING file for more details.
This is a derivative work of the xBRZ component of the HqMAME project
by Zenju. The original Licensing statement follows, indented with //
The primary changes are, syntactic to make it compile with C99+Boost,
and to make it handle an alpha channel in the image in a manner proper
for SDL.
It is not possible to extend the MAME 'special exception' to all of
the Battle for Wesnoth project, however, the special exception is
granted for my derivative forms of this work.
*/
// ****************************************************************************
// * This file is part of the HqMAME project. It is distributed under *
// * GNU General Public License: http://www.gnu.org/licenses/gpl.html *
// * Copyright (C) Zenju (zenju AT gmx DOT de) - All Rights Reserved *
// * *
// * Additionally and as a special exception, the author gives permission *
// * to link the code of this program with the MAME library (or with modified *
// * versions of MAME that use the same license as MAME), and distribute *
// * linked combinations including the two. You must obey the GNU General *
// * Public License in all respects for all of the code used other than MAME. *
// * If you modify this file, you may extend this exception to your version *
// * of the file, but you are not obligated to do so. If you do not wish to *
// * do so, delete this exception statement from your version. *
// ****************************************************************************
#ifndef XBRZ_CONFIG_HEADER_284578425345
#define XBRZ_CONFIG_HEADER_284578425345
//do NOT include any headers here! used by xBRZ_dll!!!
namespace xbrz
{
struct ScalerCfg
{
ScalerCfg() :
luminanceWeight_(1),
equalColorTolerance_(30),
dominantDirectionThreshold(3.6),
steepDirectionThreshold(2.2),
newTestAttribute_(0) {}
double luminanceWeight_;
double equalColorTolerance_;
double dominantDirectionThreshold;
double steepDirectionThreshold;
double newTestAttribute_; //unused; test new parameters
};
}
#endif

1591
src/xBRZ/xbrz.cpp
View file

File diff suppressed because it is too large Load diff

98
src/xBRZ/xbrz.hpp
View file

@ -1,35 +1,14 @@
/*
Copyright (C) 2014 - 2018 by Chris Beck <render787@gmail.com>
Part of the Battle for Wesnoth Project https://www.wesnoth.org/
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY.
See the COPYING file for more details.
This is a derivative work of the xBRZ component of the HqMAME project
by Zenju. The original Licensing statement follows, indented with //
The primary changes are, syntactic to make it compile with C99+Boost,
and to make it handle an alpha channel in the image in a manner proper
for SDL.
It is not possible to extend the MAME 'special exception' to all of
the Battle for Wesnoth project, however, the special exception is
granted for my derivative forms of this work.
*/
// ****************************************************************************
// * This file is part of the HqMAME project. It is distributed under *
// * GNU General Public License: http://www.gnu.org/licenses/gpl.html *
// * This file is part of the xBRZ project. It is distributed under *
// * GNU General Public License: https://www.gnu.org/licenses/gpl-3.0 *
// * Copyright (C) Zenju (zenju AT gmx DOT de) - All Rights Reserved *
// * *
// * Additionally and as a special exception, the author gives permission *
// * to link the code of this program with the MAME library (or with modified *
// * versions of MAME that use the same license as MAME), and distribute *
// * linked combinations including the two. You must obey the GNU General *
// * Public License in all respects for all of the code used other than MAME. *
// * to link the code of this program with the following libraries *
// * (or with modified versions that use the same licenses), and distribute *
// * linked combinations including the two: MAME, FreeFileSync, Snes9x, ePSXe *
// * You must obey the GNU General Public License in all respects for all of *
// * the code used other than MAME, FreeFileSync, Snes9x, ePSXe. *
// * If you modify this file, you may extend this exception to your version *
// * of the file, but you are not obligated to do so. If you do not wish to *
// * do so, delete this exception statement from your version. *
@ -41,7 +20,8 @@
#include <cstddef> //size_t
#include <cstdint> //uint32_t
#include <limits>
#include "config.hpp"
#include "xbrz_config.hpp"
namespace xbrz
{
@ -52,58 +32,48 @@ namespace xbrz
using a modified approach of xBR:
http://board.byuu.org/viewtopic.php?f=10&t=2248
- new rule set preserving small image features
- highly optimized for performance
- support alpha channel
- support multithreading
- support 64 bit architectures
- support 64-bit architectures
- support processing image slices
- support scaling up to 6xBRZ
*/
enum class ColorFormat //from high bits -> low bits, 8 bit per channel
{
RGB, //8 bit for each red, green, blue, upper 8 bits unused
ARGB, //including alpha channel, BGRA byte order on little-endian machines
ARGB_UNBUFFERED, //like ARGB, but without the one-time buffer creation overhead (ca. 100 - 300 ms) at the expense of a slightly slower scaling time
};
const int SCALE_FACTOR_MAX = 6;
/*
-> map source (srcWidth * srcHeight) to target (scale * width x scale * height) image, optionally processing a half-open slice of rows [yFirst, yLast) only
-> color format: ARGB (BGRA byte order), alpha channel unused
-> support for source/target pitch in bytes!
-> if your emulator changes only a few image slices during each cycle (e.g. Dosbox) then there's no need to run xBRZ on the complete image:
-> if your emulator changes only a few image slices during each cycle (e.g. DOSBox) then there's no need to run xBRZ on the complete image:
Just make sure you enlarge the source image slice by 2 rows on top and 2 on bottom (this is the additional range the xBRZ algorithm is using during analysis)
Caveat: If there are multiple changed slices, make sure they do not overlap after adding these additional rows in order to avoid a memory race condition
if you are using multiple threads for processing each enlarged slice!
CAVEAT: If there are multiple changed slices, make sure they do not overlap after adding these additional rows in order to avoid a memory race condition
in the target image data if you are using multiple threads for processing each enlarged slice!
THREAD-SAFETY: - parts of the same image may be scaled by multiple threads as long as the [yFirst, yLast) ranges do not overlap!
- there is a minor inefficiency for the first row of a slice, so avoid processing single rows only
- there is a minor inefficiency for the first row of a slice, so avoid processing single rows only; suggestion: process at least 8-16 rows
*/
void scale(size_t factor, //valid range: 2 - 5
void scale(size_t factor, //valid range: 2 - SCALE_FACTOR_MAX
const uint32_t* src, uint32_t* trg, int srcWidth, int srcHeight,
ColorFormat colFmt,
const ScalerCfg& cfg = ScalerCfg(),
int yFirst = 0, int yLast = std::numeric_limits<int>::max()); //slice of source image
void nearestNeighborScale(const uint32_t* src, int srcWidth, int srcHeight,
uint32_t* trg, int trgWidth, int trgHeight);
void bilinearScale(const uint32_t* src, int srcWidth, int srcHeight,
/**/ uint32_t* trg, int trgWidth, int trgHeight);
void nearestNeighborScale(const uint32_t* src, int srcWidth, int srcHeight,
/**/ uint32_t* trg, int trgWidth, int trgHeight);
enum SliceType
{
NN_SCALE_SLICE_SOURCE,
NN_SCALE_SLICE_TARGET,
};
void nearestNeighborScale(const uint32_t* src, int srcWidth, int srcHeight, int srcPitch, //pitch in bytes!
uint32_t* trg, int trgWidth, int trgHeight, int trgPitch,
SliceType st, int yFirst, int yLast);
//parameter tuning
bool equalColor(uint32_t col1, uint32_t col2, double luminanceWeight, double equalColorTolerance);
//########################### implementation ###########################
inline
void nearestNeighborScale(const uint32_t* src, int srcWidth, int srcHeight,
uint32_t* trg, int trgWidth, int trgHeight)
{
nearestNeighborScale(src, srcWidth, srcHeight, srcWidth * sizeof(uint32_t),
trg, trgWidth, trgHeight, trgWidth * sizeof(uint32_t),
NN_SCALE_SLICE_TARGET, 0, trgHeight);
}
bool equalColorTest(uint32_t col1, uint32_t col2, ColorFormat colFmt, double luminanceWeight, double equalColorTolerance);
}
#endif

35
src/xBRZ/xbrz_config.hpp Normal file
View file

@ -0,0 +1,35 @@
// ****************************************************************************
// * This file is part of the xBRZ project. It is distributed under *
// * GNU General Public License: https://www.gnu.org/licenses/gpl-3.0 *
// * Copyright (C) Zenju (zenju AT gmx DOT de) - All Rights Reserved *
// * *
// * Additionally and as a special exception, the author gives permission *
// * to link the code of this program with the following libraries *
// * (or with modified versions that use the same licenses), and distribute *
// * linked combinations including the two: MAME, FreeFileSync, Snes9x, ePSXe *
// * You must obey the GNU General Public License in all respects for all of *
// * the code used other than MAME, FreeFileSync, Snes9x, ePSXe. *
// * If you modify this file, you may extend this exception to your version *
// * of the file, but you are not obligated to do so. If you do not wish to *
// * do so, delete this exception statement from your version. *
// ****************************************************************************
#ifndef XBRZ_CONFIG_HEADER_284578425345
#define XBRZ_CONFIG_HEADER_284578425345
//do NOT include any headers here! used by xBRZ_dll!!!
namespace xbrz
{
struct ScalerCfg
{
double luminanceWeight = 1;
double equalColorTolerance = 30;
double centerDirectionBias = 4;
double dominantDirectionThreshold = 3.6;
double steepDirectionThreshold = 2.2;
double newTestAttribute = 0; //unused; test new parameters
};
}
#endif

266
src/xBRZ/xbrz_tools.hpp Normal file
View file

@ -0,0 +1,266 @@
// ****************************************************************************
// * This file is part of the xBRZ project. It is distributed under *
// * GNU General Public License: https://www.gnu.org/licenses/gpl-3.0 *
// * Copyright (C) Zenju (zenju AT gmx DOT de) - All Rights Reserved *
// * *
// * Additionally and as a special exception, the author gives permission *
// * to link the code of this program with the following libraries *
// * (or with modified versions that use the same licenses), and distribute *
// * linked combinations including the two: MAME, FreeFileSync, Snes9x, ePSXe *
// * You must obey the GNU General Public License in all respects for all of *
// * the code used other than MAME, FreeFileSync, Snes9x, ePSXe. *
// * If you modify this file, you may extend this exception to your version *
// * of the file, but you are not obligated to do so. If you do not wish to *
// * do so, delete this exception statement from your version. *
// ****************************************************************************
#ifndef XBRZ_TOOLS_H_825480175091875
#define XBRZ_TOOLS_H_825480175091875
#include <cassert>
#include <algorithm>
#include <type_traits>
namespace xbrz
{
template <uint32_t N> inline
unsigned char getByte(uint32_t val) { return static_cast<unsigned char>((val >> (8 * N)) & 0xff); }
inline unsigned char getAlpha(uint32_t pix) { return getByte<3>(pix); }
inline unsigned char getRed (uint32_t pix) { return getByte<2>(pix); }
inline unsigned char getGreen(uint32_t pix) { return getByte<1>(pix); }
inline unsigned char getBlue (uint32_t pix) { return getByte<0>(pix); }
inline uint32_t makePixel(unsigned char a, unsigned char r, unsigned char g, unsigned char b) { return (a << 24) | (r << 16) | (g << 8) | b; }
inline uint32_t makePixel( unsigned char r, unsigned char g, unsigned char b) { return (r << 16) | (g << 8) | b; }
inline uint32_t rgb555to888(uint16_t pix) { return ((pix & 0x7C00) << 9) | ((pix & 0x03E0) << 6) | ((pix & 0x001F) << 3); }
inline uint32_t rgb565to888(uint16_t pix) { return ((pix & 0xF800) << 8) | ((pix & 0x07E0) << 5) | ((pix & 0x001F) << 3); }
inline uint16_t rgb888to555(uint32_t pix) { return static_cast<uint16_t>(((pix & 0xF80000) >> 9) | ((pix & 0x00F800) >> 6) | ((pix & 0x0000F8) >> 3)); }
inline uint16_t rgb888to565(uint32_t pix) { return static_cast<uint16_t>(((pix & 0xF80000) >> 8) | ((pix & 0x00FC00) >> 5) | ((pix & 0x0000F8) >> 3)); }
template <class Pix> inline
Pix* byteAdvance(Pix* ptr, int bytes)
{
using PixNonConst = typename std::remove_cv<Pix>::type;
using PixByte = typename std::conditional<std::is_same<Pix, PixNonConst>::value, char, const char>::type;
static_assert(std::is_integral<PixNonConst>::value, "Pix* is expected to be cast-able to char*");
return reinterpret_cast<Pix*>(reinterpret_cast<PixByte*>(ptr) + bytes);
}
//fill block with the given color
template <class Pix> inline
void fillBlock(Pix* trg, int pitch /*[bytes]*/, Pix col, int blockWidth, int blockHeight)
{
//for (int y = 0; y < blockHeight; ++y, trg = byteAdvance(trg, pitch))
// std::fill(trg, trg + blockWidth, col);
for (int y = 0; y < blockHeight; ++y, trg = byteAdvance(trg, pitch))
for (int x = 0; x < blockWidth; ++x)
trg[x] = col;
}
//nearest-neighbor (going over target image - slow for upscaling, since source is read multiple times missing out on cache! Fast for similar image sizes!)
template <class PixSrc, class PixTrg, class PixConverter>
void nearestNeighborScale(const PixSrc* src, int srcWidth, int srcHeight, int srcPitch /*[bytes]*/,
/**/ PixTrg* trg, int trgWidth, int trgHeight, int trgPitch /*[bytes]*/,
int yFirst, int yLast, PixConverter pixCvrt /*convert PixSrc to PixTrg*/)
{
static_assert(std::is_integral<PixSrc>::value, "PixSrc* is expected to be cast-able to char*");
static_assert(std::is_integral<PixTrg>::value, "PixTrg* is expected to be cast-able to char*");
static_assert(std::is_same<decltype(pixCvrt(PixSrc())), PixTrg>::value, "PixConverter returning wrong pixel format");
if (srcPitch < srcWidth * static_cast<int>(sizeof(PixSrc)) ||
trgPitch < trgWidth * static_cast<int>(sizeof(PixTrg)))
{
assert(false);
return;
}
yFirst = std::max(yFirst, 0);
yLast = std::min(yLast, trgHeight);
if (yFirst >= yLast || srcHeight <= 0 || srcWidth <= 0) return;
for (int y = yFirst; y < yLast; ++y)
{
const int ySrc = srcHeight * y / trgHeight;
const PixSrc* const srcLine = byteAdvance(src, ySrc * srcPitch);
PixTrg* const trgLine = byteAdvance(trg, y * trgPitch);
for (int x = 0; x < trgWidth; ++x)
{
const int xSrc = srcWidth * x / trgWidth;
trgLine[x] = pixCvrt(srcLine[xSrc]);
}
}
}
//nearest-neighbor (going over source image - fast for upscaling, since source is read only once
template <class PixSrc, class PixTrg, class PixConverter>
void nearestNeighborScaleOverSource(const PixSrc* src, int srcWidth, int srcHeight, int srcPitch /*[bytes]*/,
/**/ PixTrg* trg, int trgWidth, int trgHeight, int trgPitch /*[bytes]*/,
int yFirst, int yLast, PixConverter pixCvrt /*convert PixSrc to PixTrg*/)
{
static_assert(std::is_integral<PixSrc>::value, "PixSrc* is expected to be cast-able to char*");
static_assert(std::is_integral<PixTrg>::value, "PixTrg* is expected to be cast-able to char*");
static_assert(std::is_same<decltype(pixCvrt(PixSrc())), PixTrg>::value, "PixConverter returning wrong pixel format");
if (srcPitch < srcWidth * static_cast<int>(sizeof(PixSrc)) ||
trgPitch < trgWidth * static_cast<int>(sizeof(PixTrg)))
{
assert(false);
return;
}
yFirst = std::max(yFirst, 0);
yLast = std::min(yLast, srcHeight);
if (yFirst >= yLast || trgWidth <= 0 || trgHeight <= 0) return;
for (int y = yFirst; y < yLast; ++y)
{
//mathematically: ySrc = floor(srcHeight * yTrg / trgHeight)
// => search for integers in: [ySrc, ySrc + 1) * trgHeight / srcHeight
//keep within for loop to support MT input slices!
const int yTrgFirst = ( y * trgHeight + srcHeight - 1) / srcHeight; //=ceil(y * trgHeight / srcHeight)
const int yTrgLast = ((y + 1) * trgHeight + srcHeight - 1) / srcHeight; //=ceil(((y + 1) * trgHeight) / srcHeight)
const int blockHeight = yTrgLast - yTrgFirst;
if (blockHeight > 0)
{
const PixSrc* srcLine = byteAdvance(src, y * srcPitch);
/**/ PixTrg* trgLine = byteAdvance(trg, yTrgFirst * trgPitch);
int xTrgFirst = 0;
for (int x = 0; x < srcWidth; ++x)
{
const int xTrgLast = ((x + 1) * trgWidth + srcWidth - 1) / srcWidth;
const int blockWidth = xTrgLast - xTrgFirst;
if (blockWidth > 0)
{
xTrgFirst = xTrgLast;
const auto trgPix = pixCvrt(srcLine[x]);
fillBlock(trgLine, trgPitch, trgPix, blockWidth, blockHeight);
trgLine += blockWidth;
}
}
}
}
}
template <class PixTrg, class PixConverter>
void bilinearScale(const uint32_t* src, int srcWidth, int srcHeight, int srcPitch,
/**/ PixTrg* trg, int trgWidth, int trgHeight, int trgPitch,
int yFirst, int yLast, PixConverter pixCvrt /*convert uint32_t to PixTrg*/)
{
static_assert(std::is_integral<PixTrg>::value, "PixTrg* is expected to be cast-able to char*");
static_assert(std::is_same<decltype(pixCvrt(uint32_t())), PixTrg>::value, "PixConverter returning wrong pixel format");
if (srcPitch < srcWidth * static_cast<int>(sizeof(uint32_t)) ||
trgPitch < trgWidth * static_cast<int>(sizeof(PixTrg)))
{
assert(false);
return;
}
yFirst = std::max(yFirst, 0);
yLast = std::min(yLast, trgHeight);
if (yFirst >= yLast || srcHeight <= 0 || srcWidth <= 0) return;
const double scaleX = static_cast<double>(trgWidth ) / srcWidth;
const double scaleY = static_cast<double>(trgHeight) / srcHeight;
//perf notes:
// -> double-based calculation is (slightly) faster than float
// -> pre-calculation gives significant boost; std::vector<> memory allocation is negligible!
struct CoeffsX
{
int x1 = 0;
int x2 = 0;
double xx1 = 0;
double x2x = 0;
};
std::vector<CoeffsX> buf(trgWidth);
for (int x = 0; x < trgWidth; ++x)
{
const int x1 = srcWidth * x / trgWidth;
int x2 = x1 + 1;
if (x2 == srcWidth) --x2;
const double xx1 = x / scaleX - x1;
const double x2x = 1 - xx1;
buf[x] = { x1, x2, xx1, x2x };
}
for (int y = yFirst; y < yLast; ++y)
{
const int y1 = srcHeight * y / trgHeight;
int y2 = y1 + 1;
if (y2 == srcHeight) --y2;
const double yy1 = y / scaleY - y1;
const double y2y = 1 - yy1;
const uint32_t* const srcLine = byteAdvance(src, y1 * srcPitch);
const uint32_t* const srcLineNext = byteAdvance(src, y2 * srcPitch);
PixTrg* const trgLine = byteAdvance(trg, y * trgPitch);
for (int x = 0; x < trgWidth; ++x)
{
//perf: do NOT "simplify" the variable layout without measurement!
const int x1 = buf[x].x1;
const int x2 = buf[x].x2;
const double xx1 = buf[x].xx1;
const double x2x = buf[x].x2x;
const double x2xy2y = x2x * y2y;
const double xx1y2y = xx1 * y2y;
const double x2xyy1 = x2x * yy1;
const double xx1yy1 = xx1 * yy1;
auto interpolate = [=](int offset)
{
/* https://en.wikipedia.org/wiki/Bilinear_interpolation
(c11(x2 - x) + c21(x - x1)) * (y2 - y ) +
(c12(x2 - x) + c22(x - x1)) * (y - y1) */
const auto c11 = (srcLine [x1] >> (8 * offset)) & 0xff;
const auto c21 = (srcLine [x2] >> (8 * offset)) & 0xff;
const auto c12 = (srcLineNext[x1] >> (8 * offset)) & 0xff;
const auto c22 = (srcLineNext[x2] >> (8 * offset)) & 0xff;
return c11 * x2xy2y + c21 * xx1y2y +
c12 * x2xyy1 + c22 * xx1yy1;
};
const double bi = interpolate(0);
const double gi = interpolate(1);
const double ri = interpolate(2);
const double ai = interpolate(3);
const auto b = static_cast<uint32_t>(bi + 0.5);
const auto g = static_cast<uint32_t>(gi + 0.5);
const auto r = static_cast<uint32_t>(ri + 0.5);
const auto a = static_cast<uint32_t>(ai + 0.5);
const uint32_t trgPix = (a << 24) | (r << 16) | (g << 8) | b;
trgLine[x] = pixCvrt(trgPix);
}
}
}
}
#endif //XBRZ_TOOLS_H_825480175091875