scummvm/engines/cine/pal.cpp
Kari Salminen af0efd8938 Add saturatedAddColor-function in which you can specify the added color's format (This replaces saturatedAddNormalizedColor).
Make colorFormat()-accessor function to return a const reference for speed.

svn-id: r39851
2009-04-04 17:57:00 +00:00

373 lines
13 KiB
C++

/* ScummVM - Graphic Adventure Engine
*
* ScummVM is the legal property of its developers, whose names
* are too numerous to list here. Please refer to the COPYRIGHT
* file distributed with this source distribution.
*
* 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; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*
* $URL$
* $Id$
*
*/
#include "cine/cine.h"
#include "cine/various.h"
#include "cine/pal.h"
#include "common/system.h" // For g_system->setPalette
namespace Cine {
Common::Array<PalEntry> palArray;
static byte paletteBuffer1[16];
static byte paletteBuffer2[16];
void loadPal(const char *fileName) {
char buffer[20];
removeExtention(buffer, fileName);
strcat(buffer, ".PAL");
palArray.clear();
Common::File palFileHandle;
if (!palFileHandle.open(buffer))
error("loadPal(): Cannot open file %s", fileName);
uint16 palEntriesCount = palFileHandle.readUint16LE();
palFileHandle.readUint16LE(); // entry size
palArray.resize(palEntriesCount);
for (uint i = 0; i < palArray.size(); ++i) {
palFileHandle.read(palArray[i].name, 10);
palFileHandle.read(palArray[i].pal1, 16);
palFileHandle.read(palArray[i].pal2, 16);
}
palFileHandle.close();
}
int16 findPaletteFromName(const char *fileName) {
char buffer[10];
uint16 position = 0;
uint16 i;
strcpy(buffer, fileName);
while (position < strlen(fileName)) {
if (buffer[position] > 'a' && buffer[position] < 'z') {
buffer[position] += 'A' - 'a';
}
position++;
}
for (i = 0; i < palArray.size(); i++) {
if (!strcmp(buffer, palArray[i].name)) {
return i;
}
}
return -1;
}
void loadRelatedPalette(const char *fileName) {
char localName[16];
byte i;
int16 paletteIndex;
removeExtention(localName, fileName);
paletteIndex = findPaletteFromName(localName);
if (paletteIndex == -1) {
for (i = 0; i < 16; i++) { // generate default palette
paletteBuffer1[i] = paletteBuffer2[i] = (i << 4) + i;
}
} else {
assert(paletteIndex < (int32)palArray.size());
memcpy(paletteBuffer1, palArray[paletteIndex].pal1, 16);
memcpy(paletteBuffer2, palArray[paletteIndex].pal2, 16);
}
}
void palRotate(uint16 *pal, byte a, byte b, byte c) {
assert(pal);
if (c == 1) {
uint16 currentColor = pal[b];
for (int i = b; i > a; i--) {
pal[i] = pal[i - 1];
}
pal[a] = currentColor;
}
}
void palRotate(byte *pal, byte a, byte b, byte c) {
assert(pal);
if (c == 1) {
byte currentR = pal[3 * b + 0];
byte currentG = pal[3 * b + 1];
byte currentB = pal[3 * b + 2];
for (int i = b; i > a; i--) {
pal[3 * i + 0] = pal[3 * (i - 1) + 0];
pal[3 * i + 1] = pal[3 * (i - 1) + 1];
pal[3 * i + 2] = pal[3 * (i - 1) + 2];
}
pal[3 * a + 0] = currentR;
pal[3 * a + 1] = currentG;
pal[3 * a + 2] = currentB;
}
}
uint16 transformColor(uint16 baseColor, int r, int g, int b) {
int8 oriR = CLIP( (baseColor & 0x007) + b, 0, 7);
int8 oriG = CLIP(((baseColor & 0x070) >> 4) + g, 0, 7);
int8 oriB = CLIP(((baseColor & 0x700) >> 8) + r, 0, 7);
return oriR | (oriG << 4) | (oriB << 8);
}
void transformPaletteRange(uint16 *dstPal, uint16 *srcPal, int startColor, int stopColor, int r, int g, int b) {
assert(srcPal && dstPal);
for (int i = startColor; i <= stopColor; i++)
dstPal[i] = transformColor(srcPal[i], r, g, b);
}
void transformPaletteRange(byte *dstPal, byte *srcPal, int startColor, int stopColor, int r, int g, int b) {
assert(srcPal && dstPal);
for (int i = startColor; i <= stopColor; i++) {
dstPal[3 * i + 0] = CLIP(srcPal[3 * i + 0] + r * 36, 0, 252);
dstPal[3 * i + 1] = CLIP(srcPal[3 * i + 1] + g * 36, 0, 252);
dstPal[3 * i + 2] = CLIP(srcPal[3 * i + 2] + b * 36, 0, 252);
}
}
/*! \brief Shift byte to the left by given amount (Handles negative shifting amounts too, otherwise this would be trivial). */
byte shiftByteLeft(const byte value, const signed shiftLeft) {
if (shiftLeft >= 0)
return value << shiftLeft;
else // right shift with negative shiftLeft values
return value >> abs(shiftLeft);
}
/*! \brief Is given endian type big endian? (Handles native endian type too, otherwise this would be trivial). */
bool isBigEndian(const EndianType endian) {
assert(endian == CINE_NATIVE_ENDIAN || endian == CINE_LITTLE_ENDIAN || endian == CINE_BIG_ENDIAN);
// Handle explicit little and big endian types here
if (endian != CINE_NATIVE_ENDIAN) {
return (endian == CINE_BIG_ENDIAN);
}
// Handle native endian type here
#if defined(SCUMM_BIG_ENDIAN)
return true;
#elif defined(SCUMM_LITTLE_ENDIAN)
return false;
#else
#error No endianness defined
#endif
}
/*! \brief Calculate byte position of given bit position in a multibyte variable using defined endianness. */
int bytePos(const int bitPos, const int numBytes, const bool bigEndian) {
if (bigEndian)
return (numBytes - 1) - (bitPos / 8);
else // little endian
return bitPos / 8;
}
// a.k.a. palRotate
Palette &Palette::rotateRight(byte firstIndex, byte lastIndex) {
const Color lastColor = _colors[lastIndex];
for (int i = lastIndex; i > firstIndex; i--)
_colors[i] = _colors[i - 1];
_colors[firstIndex] = lastColor;
return *this;
}
bool Palette::empty() const {
return _colors.empty();
}
uint Palette::colorCount() const {
return _colors.size();
}
Palette &Palette::fillWithBlack() {
for (uint i = 0; i < _colors.size(); i++) {
_colors[i].r = 0;
_colors[i].g = 0;
_colors[i].b = 0;
}
return *this;
}
// TODO: Add better heuristic for checking whether the color format is valid
bool Palette::isValid() const {
// Check that the color format has been actually set and not just default constructed.
// Also check that the alpha channel is discarded.
return _format != Graphics::PixelFormat() && _format.aLoss == 8;
}
const Graphics::PixelFormat &Palette::colorFormat() const {
return _format;
}
void Palette::setGlobalOSystemPalette() const {
byte buf[256 * 4]; // Allocate space for the largest possible palette
save(buf, sizeof(buf), Cine::kSystemPalFormat, CINE_LITTLE_ENDIAN);
g_system->setPalette(buf, 0, colorCount());
}
void Palette::setColorFormat(const Graphics::PixelFormat format) {
_format = format;
}
// a.k.a. transformPaletteRange
Palette &Palette::saturatedAddColor(Palette& output, byte firstIndex, byte lastIndex, signed r, signed g, signed b) {
assert(firstIndex < colorCount() && lastIndex < colorCount());
assert(firstIndex < output.colorCount() && lastIndex < output.colorCount());
assert(output.colorFormat() == colorFormat());
for (uint i = firstIndex; i <= lastIndex; i++)
output._colors[i] = saturatedAddColor(_colors[i], r, g, b);
return output;
}
Palette &Palette::saturatedAddColor(Palette& output, byte firstIndex, byte lastIndex, signed rSource, signed gSource, signed bSource, const Graphics::PixelFormat &sourceFormat) {
// Convert the source color to the internal color format ensuring that no divide by zero will happen
const signed r = _format.rMax() * rSource / MAX<int>(sourceFormat.rMax(), 1);
const signed g = _format.gMax() * gSource / MAX<int>(sourceFormat.gMax(), 1);
const signed b = _format.bMax() * bSource / MAX<int>(sourceFormat.bMax(), 1);
return saturatedAddColor(output, firstIndex, lastIndex, r, g, b);
}
Palette &Palette::saturatedAddNormalizedGray(Palette& output, byte firstIndex, byte lastIndex, int grayDividend, int grayDenominator) {
assert(grayDenominator != 0);
const signed r = _format.rMax() * grayDividend / grayDenominator;
const signed g = _format.gMax() * grayDividend / grayDenominator;
const signed b = _format.bMax() * grayDividend / grayDenominator;
return saturatedAddColor(output, firstIndex, lastIndex, r, g, b);
}
// a.k.a. transformColor
// Parameter color components (i.e. r, g and b) are in range [-7, 7]
// e.g. r = 7 sets the resulting color's red component to maximum
// e.g. r = -7 sets the resulting color's red component to minimum (i.e. zero)
Cine::Palette::Color Palette::saturatedAddColor(Cine::Palette::Color baseColor, signed r, signed g, signed b) const {
Cine::Palette::Color result;
result.r = CLIP<int>(baseColor.r + r, 0, _format.rMax());
result.g = CLIP<int>(baseColor.g + g, 0, _format.gMax());
result.b = CLIP<int>(baseColor.b + b, 0, _format.bMax());
return result;
}
Palette::Palette(const Graphics::PixelFormat format, const uint numColors) : _format(format), _colors() {
_colors.resize(numColors);
fillWithBlack();
}
Palette &Palette::clear() {
_format = Graphics::PixelFormat();
_colors.clear();
return *this;
}
Palette &Palette::load(const byte *buf, const uint size, const Graphics::PixelFormat format, const uint numColors, const EndianType endian) {
assert(format.bytesPerPixel * numColors <= size); // Make sure there's enough input space
assert(format.aLoss == 8); // No alpha
assert(format.rShift / 8 == (format.rShift + MAX<int>(0, format.rBits() - 1)) / 8); // R must be inside one byte
assert(format.gShift / 8 == (format.gShift + MAX<int>(0, format.gBits() - 1)) / 8); // G must be inside one byte
assert(format.bShift / 8 == (format.bShift + MAX<int>(0, format.bBits() - 1)) / 8); // B must be inside one byte
setColorFormat(format);
_colors.clear();
_colors.resize(numColors);
const int rBytePos = bytePos(format.rShift, format.bytesPerPixel, isBigEndian(endian));
const int gBytePos = bytePos(format.gShift, format.bytesPerPixel, isBigEndian(endian));
const int bBytePos = bytePos(format.bShift, format.bytesPerPixel, isBigEndian(endian));
for (uint i = 0; i < numColors; i++) {
// format.rMax(), format.gMax(), format.bMax() are also used as masks here
_colors[i].r = (buf[i * format.bytesPerPixel + rBytePos] >> (format.rShift % 8)) & format.rMax();
_colors[i].g = (buf[i * format.bytesPerPixel + gBytePos] >> (format.gShift % 8)) & format.gMax();
_colors[i].b = (buf[i * format.bytesPerPixel + bBytePos] >> (format.bShift % 8)) & format.bMax();
}
return *this;
}
byte *Palette::save(byte *buf, const uint size, const EndianType endian) const {
return save(buf, size, colorFormat(), colorCount(), endian);
}
byte *Palette::save(byte *buf, const uint size, const Graphics::PixelFormat format, const EndianType endian) const {
return save(buf, size, format, colorCount(), endian);
}
byte *Palette::save(byte *buf, const uint size, const Graphics::PixelFormat format, const uint numColors, const EndianType endian, const byte firstIndex) const {
assert(format.bytesPerPixel * numColors <= size); // Make sure there's enough output space
assert(format.aLoss == 8); // No alpha
assert(format.rShift / 8 == (format.rShift + MAX<int>(0, format.rBits() - 1)) / 8); // R must be inside one byte
assert(format.gShift / 8 == (format.gShift + MAX<int>(0, format.gBits() - 1)) / 8); // G must be inside one byte
assert(format.bShift / 8 == (format.bShift + MAX<int>(0, format.bBits() - 1)) / 8); // B must be inside one byte
// Clear the part of the output palette we're going to be writing to with all black
memset(buf, 0, format.bytesPerPixel * numColors);
// Calculate how much bit shifting the color components need (for positioning them correctly)
const signed rShiftLeft = (colorFormat().rLoss - (signed) format.rLoss) + (format.rShift % 8);
const signed gShiftLeft = (colorFormat().gLoss - (signed) format.gLoss) + (format.gShift % 8);
const signed bShiftLeft = (colorFormat().bLoss - (signed) format.bLoss) + (format.bShift % 8);
// Calculate the byte masks for each color component (for masking away excess bits)
const byte rMask = format.rMax() << (format.rShift % 8);
const byte gMask = format.gMax() << (format.gShift % 8);
const byte bMask = format.bMax() << (format.bShift % 8);
const int rBytePos = bytePos(format.rShift, format.bytesPerPixel, isBigEndian(endian));
const int gBytePos = bytePos(format.gShift, format.bytesPerPixel, isBigEndian(endian));
const int bBytePos = bytePos(format.bShift, format.bytesPerPixel, isBigEndian(endian));
// Save the palette to the output in the specified format
for (uint i = firstIndex; i < firstIndex + numColors; i++) {
buf[i * format.bytesPerPixel + rBytePos] |= (shiftByteLeft(_colors[i].r, rShiftLeft) & rMask);
buf[i * format.bytesPerPixel + gBytePos] |= (shiftByteLeft(_colors[i].g, gShiftLeft) & gMask);
buf[i * format.bytesPerPixel + bBytePos] |= (shiftByteLeft(_colors[i].b, bShiftLeft) & bMask);
}
// Return the pointer to the output palette
return buf;
}
} // End of namespace Cine