#include "traster.h"
#include "tpixelutils.h"
#include "trop.h"
//***********************************************************************************************
// Local namespace stuff
//***********************************************************************************************
namespace {
template <typename T>
struct RaylitFuncTraits {
typedef void (*function_type)(T *, T *, int, int, int, int, const TRect &,
const TRect &, const TRop::RaylitParams &);
};
//--------------------------------------------------------------------------------------------
template <typename T>
void performStandardRaylit(T *bufIn, T *bufOut, int dxIn, int dyIn, int dxOut,
int dyOut, const TRect &srcRect,
const TRect &dstRect,
const TRop::RaylitParams ¶ms) {
/* NOTATION: Diagram assuming octant 1
/ |
/ |
/ - ray_final_y | octLy
/ 1 |
+---- |
_____ octLx
So, octLx and octLy are the octant's lx and ly; ray_final_y is the final height
of the ray we're tracing
*/
// Build colors-related variables
int max = T::maxChannelValue;
/*-- 透明部分の色 --*/
int transp_val = (params.m_invert) ? max : 0, opaque_val = max - transp_val;
int value, val_r, val_g, val_b, val_m;
double lightness, r_fac, g_fac, b_fac, m_fac;
/*-- 8bit/16bitの違いを吸収する係数 --*/
double factor = max / 255.0;
double scale =
params.m_scale; // NOTE: These variable initializations are, well,
double decay = log(params.m_decay / 100.0 + 1.0) +
1.0; // heuristic at least. They were probably tested
double intensity =
1e8 * log(params.m_intensity / 100.0 + 1.0) /
scale; // to be good, but didn't quite make any REAL sense.
double smoothness = log(params.m_smoothness * 5.0 / 100.0 + 1.0); //
// They could be done MUCH better, but changing them
/*-- 1ステップ進んだ時、次のピクセルで光源が無かったときの光の弱まる割合 --*/
double neg_delta_p =
smoothness *
intensity; // would alter the way raylit has been applied until now.
/*-- 1ステップ進んだ時、次のピクセルで光源が有ったときの光の強まる割合 --*/
double quot_delta_p = intensity / max; //
// Should be changed at some point, though...
/*--
* m_colorはRaylitFxのColor値。r_fac、g_fac、b_facは各チャンネルをPremultiplyした値
* --*/
m_fac = (params.m_color.m / 255.0);
r_fac = m_fac * (params.m_color.r / 255.0);
g_fac = m_fac * (params.m_color.g / 255.0);
b_fac = m_fac * (params.m_color.b / 255.0);
// Geometry-related variables
int x, y, ray_final_y;
int octLx = dstRect.x1 - dstRect.x0;
double rayPosIncrementX = 1.0 / scale;
double sq_z = sq(params.m_lightOriginSrc.z); // We'll be making square
// distances from p, so square
// it once now
// Perform raylit
T *pixIn, *pixOut;
for (ray_final_y = 0; ray_final_y < octLx; ++ray_final_y) {
// Initialize increment variables
lightness = 0.0;
double rayPosIncrementY = rayPosIncrementX * (ray_final_y / (double)octLx);
// Use an integer counter to know when y must increase. Will add ray_final_y
// as long as
// a multiple of octLx-1 is reached, then increase
int yIncrementCounter = 0, yIncrementThreshold = octLx - 1;
// Trace a single ray of light
TPointD rayPos(rayPosIncrementX, rayPosIncrementY);
for (x = dstRect.x0, y = dstRect.y0, pixIn = bufIn, pixOut = bufOut;
(x < dstRect.x1) && (y < dstRect.y1); ++x) {
bool insideSrc = (x >= srcRect.x0) && (x < srcRect.x1) &&
(y >= srcRect.y0) && (y < srcRect.y1);
if (insideSrc) {
// Add a light component depending on source's matte
if (pixIn->m == opaque_val)
lightness = std::max(
0.0, lightness - neg_delta_p); // No light source - ray fading
else {
if (pixIn->m == transp_val)
lightness += intensity; // Full light source - ray enforcing
else
lightness = std::max(
0.0, lightness + // Half light source
(params.m_invert ? pixIn->m : (max - pixIn->m)) *
quot_delta_p); // matte-linear enforcing
}
if (params.m_includeInput) {
val_r = pixIn->r;
val_g = pixIn->g;
val_b = pixIn->b;
val_m = pixIn->m;
} else
val_r = val_g = val_b = val_m = 0;
} else {
if (!params.m_invert)
lightness += intensity;
else
lightness = std::max(0.0, lightness - neg_delta_p);
val_r = val_g = val_b = val_m = 0;
}
bool insideDst = (x >= 0) && (y >= 0);
if (insideDst) {
// Write the corresponding destination pixel
if (lightness > 0.0)
value = (int)(factor * lightness /
(rayPos.x *
pow((double)(sq(rayPos.x) + sq(rayPos.y) + sq_z),
decay)) +
0.5); // * ^-d... 0.5 rounds
else
value = 0;
// NOTE: pow() could be slow. If that is the case, it could be cached
// for the whole octant along the longest ray at integer positions,
// and then linearly interpolated between those... Have to profile this
// before resorting to that...
val_r += value * r_fac;
val_g += value * g_fac;
val_b += value * b_fac;
val_m += value * m_fac;
pixOut->r = (val_r > max) ? max : val_r;
pixOut->g = (val_g > max) ? max : val_g;
pixOut->b = (val_b > max) ? max : val_b;
pixOut->m = (val_m > max) ? max : val_m;
}
// Increment variables along the x-axis
pixIn += dxIn, pixOut += dxOut;
rayPos.x += rayPosIncrementX, rayPos.y += rayPosIncrementY;
// Increment variables along the y-axis
if ((yIncrementCounter += ray_final_y) >= yIncrementThreshold) {
++y, pixIn += dyIn, pixOut += dyOut;
yIncrementCounter -= yIncrementThreshold;
}
}
}
}
//--------------------------------------------------------------------------------------------
template <typename T>
void performColorRaylit(T *bufIn, T *bufOut, int dxIn, int dyIn, int dxOut,
int dyOut, const TRect &srcRect, const TRect &dstRect,
const TRop::RaylitParams ¶ms) {
// Build colors-related variables
int max = T::maxChannelValue;
int val_r, val_g, val_b, val_m;
double lightness_r, lightness_g, lightness_b;
double factor = max / 255.0;
double scale =
params.m_scale; // NOTE: These variable initializations are, well,
double decay = log(params.m_decay / 100.0 + 1.0) +
1.0; // heuristic at least. They were probably tested
double intensity =
1e8 * log(params.m_intensity / 100.0 + 1.0) /
scale; // to be good, but didn't quite make any REAL sense.
double smoothness = log(params.m_smoothness * 5.0 / 100.0 + 1.0); //
// They could be done MUCH better, but changing them
double neg_delta_p =
smoothness *
intensity; // would alter the way raylit has been applied until now.
double quot_delta_p = intensity / max; //
// Should be changed at some point, though...
// Geometry-related variables
int x, y, ray_final_y;
int octLx = dstRect.x1 - dstRect.x0;
double rayPosIncrementX = 1.0 / scale;
double fac, sq_z = sq(params.m_lightOriginSrc.z); // We'll be making square
// distances from p, so
// square it once now
// Perform raylit
T *pixIn, *pixOut;
for (ray_final_y = 0; ray_final_y < octLx; ++ray_final_y) {
// Initialize increment variables
lightness_r = lightness_g = lightness_b = 0.0;
int l, l_max;
double rayPosIncrementY = rayPosIncrementX * (ray_final_y / (double)octLx);
// Use an integer counter to know when y must increase. Will add ray_final_y
// as long as
// a multiple of octLx-1 is reached, then increase
int yIncrementCounter = 0, yIncrementThreshold = octLx - 1;
// Trace a single ray of light
TPointD rayPos(rayPosIncrementX, rayPosIncrementY);
for (x = dstRect.x0, y = dstRect.y0, pixIn = bufIn, pixOut = bufOut;
(x < dstRect.x1) && (y < dstRect.y1); ++x) {
bool insideSrc = (x >= srcRect.x0) && (x < srcRect.x1) &&
(y >= srcRect.y0) && (y < srcRect.y1);
if (insideSrc) {
val_r = pixIn->r;
val_g = pixIn->g;
val_b = pixIn->b;
val_m = pixIn->m;
lightness_r = std::max(0.0, val_r ? lightness_r + val_r * quot_delta_p
: lightness_r - neg_delta_p);
lightness_g = std::max(0.0, val_g ? lightness_g + val_g * quot_delta_p
: lightness_g - neg_delta_p);
lightness_b = std::max(0.0, val_b ? lightness_b + val_b * quot_delta_p
: lightness_b - neg_delta_p);
if (!params.m_includeInput) val_r = val_g = val_b = val_m = 0;
} else {
lightness_r = std::max(0.0, lightness_r - neg_delta_p);
lightness_g = std::max(0.0, lightness_g - neg_delta_p);
lightness_b = std::max(0.0, lightness_b - neg_delta_p);
val_r = val_g = val_b = val_m = 0;
}
bool insideDst = (x >= 0) && (y >= 0);
if (insideDst) {
// Write the corresponding destination pixel
fac =
factor / (rayPos.x *
pow((double)(sq(rayPos.x) + sq(rayPos.y) + sq_z), decay));
// NOTE: pow() could be slow. If that is the case, it could be cached
// for the whole octant along the longest ray at integer positions,
// and then linearly interpolated between those... Have to profile this
// before resorting to that...
val_r += l = (int)(fac * lightness_r + 0.5);
l_max = l;
val_g += l = (int)(fac * lightness_g + 0.5);
l_max = std::max(l, l_max);
val_b += l = (int)(fac * lightness_b + 0.5);
l_max = std::max(l, l_max);
val_m += l_max;
pixOut->r = (val_r > max) ? max : val_r;
pixOut->g = (val_g > max) ? max : val_g;
pixOut->b = (val_b > max) ? max : val_b;
pixOut->m = (val_m > max) ? max : val_m;
}
// Increment variables along the x-axis
pixIn += dxIn, pixOut += dxOut;
rayPos.x += rayPosIncrementX, rayPos.y += rayPosIncrementY;
// Increment variables along the y-axis
if ((yIncrementCounter += ray_final_y) >= yIncrementThreshold) {
++y, pixIn += dyIn, pixOut += dyOut;
yIncrementCounter -= yIncrementThreshold;
}
}
}
}
//--------------------------------------------------------------------------------------------
/*-- ピザ状に8分割された領域の1つを計算する --*/
template <typename T>
void computeOctant(const TRasterPT<T> &src, const TRasterPT<T> &dst, int octant,
const TRop::RaylitParams ¶ms,
typename RaylitFuncTraits<T>::function_type raylitFunc) {
// Build octant geometry variables
int x0, x1, lxIn, lxOut, dxIn, dxOut;
int y0, y1, lyIn, lyOut, dyIn, dyOut;
const T3DPoint &pIn = params.m_lightOriginSrc,
&pOut = params.m_lightOriginDst;
int srcWrap = src->getWrap(), dstWrap = dst->getWrap();
T *bufIn = (T *)src->getRawData() + tfloor(pIn.y) * srcWrap + tfloor(pIn.x);
T *bufOut =
(T *)dst->getRawData() + tfloor(pOut.y) * dstWrap + tfloor(pOut.x);
TRect srcRect(src->getBounds() +
TPoint(tround(pOut.x - pIn.x), tround(pOut.y - pIn.y)));
lxIn = src->getLx(), lxOut = dst->getLx();
lyIn = src->getLy(), lyOut = dst->getLy();
/*-- 1ピクセルずつ進むときの移動値 --*/
// Vertical octant pairs
if (octant == 1 || octant == 8)
dxIn = 1, dxOut = 1, x0 = tfloor(pOut.x), x1 = lxOut;
if (octant == 2 || octant == 7)
dyIn = 1, dyOut = 1, y0 = tfloor(pOut.x), y1 = lxOut;
if (octant == 3 || octant == 6)
dyIn = -1, dyOut = -1, y0 = lxOut - tfloor(pOut.x) - 1, y1 = lxOut,
tswap(srcRect.x0, srcRect.x1), srcRect.x0 = lxOut - srcRect.x0,
srcRect.x1 = lxOut - srcRect.x1;
if (octant == 4 || octant == 5)
dxIn = -1, dxOut = -1, x0 = lxOut - tfloor(pOut.x) - 1, x1 = lxOut,
tswap(srcRect.x0, srcRect.x1), srcRect.x0 = lxOut - srcRect.x0,
srcRect.x1 = lxOut - srcRect.x1;
// Horizontal octant pairs
if (octant == 2 || octant == 3)
dxIn = srcWrap, dxOut = dstWrap, x0 = tfloor(pOut.y), x1 = lyOut;
if (octant == 1 || octant == 4)
dyIn = srcWrap, dyOut = dstWrap, y0 = tfloor(pOut.y), y1 = lyOut;
if (octant == 5 || octant == 8)
dyIn = -srcWrap, dyOut = -dstWrap, y0 = lyOut - tfloor(pOut.y) - 1,
y1 = lyOut, tswap(srcRect.y0, srcRect.y1), srcRect.y0 = lyOut - srcRect.y0,
srcRect.y1 = lyOut - srcRect.y1;
if (octant == 6 || octant == 7)
dxIn = -srcWrap, dxOut = -dstWrap, x0 = lyOut - tfloor(pOut.y) - 1,
x1 = lyOut, tswap(srcRect.y0, srcRect.y1), srcRect.y0 = lyOut - srcRect.y0,
srcRect.y1 = lyOut - srcRect.y1;
/*-- 縦向きのピザ領域を計算する場合は、90度回転してから --*/
// Swap x and y axis where necessary
if (octant == 2 || octant == 3 || octant == 6 || octant == 7) {
tswap(lxIn, lyIn), tswap(lxOut, lyOut);
tswap(srcRect.x0, srcRect.y0), tswap(srcRect.x1, srcRect.y1);
}
int octLx = (x1 - x0), octLy = (y1 - y0);
assert(octLx > 0 && octLy > 0);
if (octLx <= 0 && octLy <= 0) return;
raylitFunc(bufIn, bufOut, dxIn, dyIn, dxOut, dyOut, srcRect,
TRect(x0, y0, x1, y1), params);
}
//--------------------------------------------------------------------------------------------
/*
OCTANTS:
\ 3 | 2 /
\ | /
\ | /
4 \|/ 1
----+----
5 /|\ 8
/ | \
/ | \
/ 6 | 7 \
*/
template <typename T>
void doRaylit(const TRasterPT<T> &src, const TRasterPT<T> &dst,
const TRop::RaylitParams ¶ms,
typename RaylitFuncTraits<T>::function_type raylitFunc) {
int lxOut = dst->getLx(), lyOut = dst->getLy();
const T3DPoint &p = params.m_lightOriginDst;
src->lock();
dst->lock();
// Depending on the position of p, only some of the quadrants need to be built
if (p.y < lyOut) {
if (p.x < lxOut) {
// Compute the raylit fx on each octant independently
computeOctant(src, dst, 1, params, raylitFunc);
computeOctant(src, dst, 2, params, raylitFunc);
}
if (p.x >= 0) {
computeOctant(src, dst, 3, params, raylitFunc);
computeOctant(src, dst, 4, params, raylitFunc);
}
}
if (p.y >= 0) {
if (p.x >= 0) {
computeOctant(src, dst, 5, params, raylitFunc);
computeOctant(src, dst, 6, params, raylitFunc);
}
if (p.x < lxOut) {
computeOctant(src, dst, 7, params, raylitFunc);
computeOctant(src, dst, 8, params, raylitFunc);
}
}
dst->unlock();
src->unlock();
}
} // namespace
//***********************************************************************************************
// TRop::raylit implementation
//***********************************************************************************************
void TRop::raylit(const TRasterP &dstRas, const TRasterP &srcRas,
const RaylitParams ¶ms) {
if ((TRaster32P)dstRas && (TRaster32P)srcRas)
doRaylit<TPixel32>(srcRas, dstRas, params,
&performStandardRaylit<TPixel32>);
else if ((TRaster64P)dstRas && (TRaster64P)srcRas)
doRaylit<TPixel64>(srcRas, dstRas, params,
&performStandardRaylit<TPixel64>);
else
throw TException("TRop::raylit unsupported pixel type");
}
//--------------------------------------------------------------------------------------------
void TRop::glassRaylit(const TRasterP &dstRas, const TRasterP &srcRas,
const RaylitParams ¶ms) {
if ((TRaster32P)dstRas && (TRaster32P)srcRas)
doRaylit<TPixel32>(srcRas, dstRas, params, &performColorRaylit<TPixel32>);
else if ((TRaster64P)dstRas && (TRaster64P)srcRas)
doRaylit<TPixel64>(srcRas, dstRas, params, &performColorRaylit<TPixel64>);
else
throw TException("TRop::raylit unsupported pixel type");
}