rlm@1: /**     Code adapted from Exult source code by Forgotten
rlm@1: **	Scale.cc - Trying to scale with bilinear interpolation.
rlm@1: **
rlm@1: **	Written: 6/14/00 - JSF
rlm@1: **/
rlm@1: 
rlm@1: #include "../common/System.h"
rlm@1: 
rlm@1: static u8 row_cur[3 * 322];
rlm@1: static u8 row_next[3 * 322];
rlm@1: 
rlm@1: static u8 *rgb_row_cur	= row_cur;
rlm@1: static u8 *rgb_row_next = row_next;
rlm@1: 
rlm@1: #ifdef RGB
rlm@1: #undef RGB  // wingdi.h has it
rlm@1: #endif
rlm@1: #define RGB(r, g, b) \
rlm@1:     ((r) >> 3) << systemRedShift | \
rlm@1:     ((g) >> 3) << systemGreenShift | \
rlm@1:     ((b) >> 3) << systemBlueShift \
rlm@1: 
rlm@1: static void fill_rgb_row_16(u16 *from, int src_width, u8 *row, int width)
rlm@1: {
rlm@1: 	u8 *copy_start = row + src_width * 3;
rlm@1: 	u8 *all_stop   = row + width * 3;
rlm@1: 	while (row < copy_start)
rlm@1: 	{
rlm@1: 		u16 color = *from++;
rlm@1: 		*row++ = ((color >> systemRedShift) & 0x1f) << 3;
rlm@1: 		*row++ = ((color >> systemGreenShift) & 0x1f) << 3;
rlm@1: 		*row++ = ((color >> systemBlueShift) & 0x1f) << 3;
rlm@1: 	}
rlm@1: 	// any remaining elements to be written to 'row' are a replica of the
rlm@1: 	// preceding pixel
rlm@1: 	u8 *p = row - 3;
rlm@1: 	while (row < all_stop)
rlm@1: 	{
rlm@1: 		// we're guaranteed three elements per pixel; could unroll the loop
rlm@1: 		// further, especially with a Duff's Device, but the gains would be
rlm@1: 		// probably limited (judging by profiler output)
rlm@1: 		*row++ = *p++;
rlm@1: 		*row++ = *p++;
rlm@1: 		*row++ = *p++;
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: static void fill_rgb_row_32(u32 *from, int src_width, u8 *row, int width)
rlm@1: {
rlm@1: 	u8 *copy_start = row + src_width * 3;
rlm@1: 	u8 *all_stop   = row + width * 3;
rlm@1: 	while (row < copy_start)
rlm@1: 	{
rlm@1: 		u32 color = *from++;
rlm@1: 		*row++ = ((color >> systemRedShift) & 0x1f) << 3;
rlm@1: 		*row++ = ((color >> systemGreenShift) & 0x1f) << 3;
rlm@1: 		*row++ = ((color >> systemBlueShift) & 0x1f) << 3;
rlm@1: 	}
rlm@1: 	// any remaining elements to be written to 'row' are a replica of the
rlm@1: 	// preceding pixel
rlm@1: 	u8 *p = row - 3;
rlm@1: 	while (row < all_stop)
rlm@1: 	{
rlm@1: 		// we're guaranteed three elements per pixel; could unroll the loop
rlm@1: 		// further, especially with a Duff's Device, but the gains would be
rlm@1: 		// probably limited (judging by profiler output)
rlm@1: 		*row++ = *p++;
rlm@1: 		*row++ = *p++;
rlm@1: 		*row++ = *p++;
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: void Bilinear(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:               u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u16 *to		= (u16 *)dstPtr;
rlm@1: 	u16 *to_odd = (u16 *)(dstPtr + dstPitch);
rlm@1: 
rlm@1: 	int	 from_width = width;
rlm@1: 	u16 *from		= (u16 *)srcPtr;
rlm@1: 	fill_rgb_row_16(from, from_width, rgb_row_cur, width + 1);
rlm@1: 
rlm@1: 	for (int y = 0; y < height; y++)
rlm@1: 	{
rlm@1: 		u16 *from_orig = from;
rlm@1: 		u16 *to_orig   = to;
rlm@1: 
rlm@1: 		if (y + 1 < height)
rlm@1: 			fill_rgb_row_16(from + width + 2, from_width, rgb_row_next,
rlm@1: 			                width + 1);
rlm@1: 		else
rlm@1: 			fill_rgb_row_16(from, from_width, rgb_row_next, width + 1);
rlm@1: 
rlm@1: 		// every pixel in the src region, is extended to 4 pixels in the
rlm@1: 		// destination, arranged in a square 'quad'; if the current src
rlm@1: 		// pixel is 'a', then in what follows 'b' is the src pixel to the
rlm@1: 		// right, 'c' is the src pixel below, and 'd' is the src pixel to
rlm@1: 		// the right and down
rlm@1: 		u8 *cur_row	 = rgb_row_cur;
rlm@1: 		u8 *next_row = rgb_row_next;
rlm@1: 		u8 *ar		 = cur_row++;
rlm@1: 		u8 *ag		 = cur_row++;
rlm@1: 		u8 *ab		 = cur_row++;
rlm@1: 		u8 *cr		 = next_row++;
rlm@1: 		u8 *cg		 = next_row++;
rlm@1: 		u8 *cb		 = next_row++;
rlm@1: 		for (int x = 0; x < width; x++)
rlm@1: 		{
rlm@1: 			u8 *br = cur_row++;
rlm@1: 			u8 *bg = cur_row++;
rlm@1: 			u8 *bb = cur_row++;
rlm@1: 			u8 *dr = next_row++;
rlm@1: 			u8 *dg = next_row++;
rlm@1: 			u8 *db = next_row++;
rlm@1: 
rlm@1: 			// upper left pixel in quad: just copy it in
rlm@1: 			*to++ = RGB(*ar, *ag, *ab);
rlm@1: 
rlm@1: 			// upper right
rlm@1: 			*to++ = RGB((*ar + *br) >> 1, (*ag + *bg) >> 1, (*ab + *bb) >> 1);
rlm@1: 
rlm@1: 			// lower left
rlm@1: 			*to_odd++ = RGB((*ar + *cr) >> 1, (*ag + *cg) >> 1, (*ab + *cb) >> 1);
rlm@1: 
rlm@1: 			// lower right
rlm@1: 			*to_odd++ = RGB((*ar + *br + *cr + *dr) >> 2,
rlm@1: 			                (*ag + *bg + *cg + *dg) >> 2,
rlm@1: 			                (*ab + *bb + *cb + *db) >> 2);
rlm@1: 
rlm@1: 			// 'b' becomes 'a', 'd' becomes 'c'
rlm@1: 			ar = br;
rlm@1: 			ag = bg;
rlm@1: 			ab = bb;
rlm@1: 			cr = dr;
rlm@1: 			cg = dg;
rlm@1: 			cb = db;
rlm@1: 		}
rlm@1: 
rlm@1: 		// the "next" rgb row becomes the current; the old current rgb row is
rlm@1: 		// recycled and serves as the new "next" row
rlm@1: 		u8 *temp;
rlm@1: 		temp		 = rgb_row_cur;
rlm@1: 		rgb_row_cur	 = rgb_row_next;
rlm@1: 		rgb_row_next = temp;
rlm@1: 
rlm@1: 		// update the pointers for start of next pair of lines
rlm@1: 		from   = (u16 *)((u8 *)from_orig + srcPitch);
rlm@1: 		to	   = (u16 *)((u8 *)to_orig + (dstPitch << 1));
rlm@1: 		to_odd = (u16 *)((u8 *)to + dstPitch);
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: void BilinearPlus(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:                   u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u16 *to		= (u16 *)dstPtr;
rlm@1: 	u16 *to_odd = (u16 *)(dstPtr + dstPitch);
rlm@1: 
rlm@1: 	int	 from_width = width;
rlm@1: 	u16 *from		= (u16 *)srcPtr;
rlm@1: 	fill_rgb_row_16(from, from_width, rgb_row_cur, width + 1);
rlm@1: 
rlm@1: 	for (int y = 0; y < height; y++)
rlm@1: 	{
rlm@1: 		u16 *from_orig = from;
rlm@1: 		u16 *to_orig   = to;
rlm@1: 
rlm@1: 		if (y + 1 < height)
rlm@1: 			fill_rgb_row_16(from + width + 2, from_width, rgb_row_next,
rlm@1: 			                width + 1);
rlm@1: 		else
rlm@1: 			fill_rgb_row_16(from, from_width, rgb_row_next, width + 1);
rlm@1: 
rlm@1: 		// every pixel in the src region, is extended to 4 pixels in the
rlm@1: 		// destination, arranged in a square 'quad'; if the current src
rlm@1: 		// pixel is 'a', then in what follows 'b' is the src pixel to the
rlm@1: 		// right, 'c' is the src pixel below, and 'd' is the src pixel to
rlm@1: 		// the right and down
rlm@1: 		u8 *cur_row	 = rgb_row_cur;
rlm@1: 		u8 *next_row = rgb_row_next;
rlm@1: 		u8 *ar		 = cur_row++;
rlm@1: 		u8 *ag		 = cur_row++;
rlm@1: 		u8 *ab		 = cur_row++;
rlm@1: 		u8 *cr		 = next_row++;
rlm@1: 		u8 *cg		 = next_row++;
rlm@1: 		u8 *cb		 = next_row++;
rlm@1: 		for (int x = 0; x < width; x++)
rlm@1: 		{
rlm@1: 			u8 *br = cur_row++;
rlm@1: 			u8 *bg = cur_row++;
rlm@1: 			u8 *bb = cur_row++;
rlm@1: 			u8 *dr = next_row++;
rlm@1: 			u8 *dg = next_row++;
rlm@1: 			u8 *db = next_row++;
rlm@1: 
rlm@1: 			// upper left pixel in quad: just copy it in
rlm@1: 			//*to++ = manip.rgb(*ar, *ag, *ab);
rlm@1: #ifdef USE_ORIGINAL_BILINEAR_PLUS
rlm@1: 			*to++ = RGB(
rlm@1: 			    (((*ar) << 2) + ((*ar)) + (*cr + *br + *br)) >> 3,
rlm@1: 			    (((*ag) << 2) + ((*ag)) + (*cg + *bg + *bg)) >> 3,
rlm@1: 			    (((*ab) << 2) + ((*ab)) + (*cb + *bb + *bb)) >> 3);
rlm@1: #else
rlm@1: 			*to++ = RGB(
rlm@1: 			    (((*ar) << 3) + ((*ar) << 1) + (*cr + *br + *br + *cr)) >> 4,
rlm@1: 			    (((*ag) << 3) + ((*ag) << 1) + (*cg + *bg + *bg + *cg)) >> 4,
rlm@1: 			    (((*ab) << 3) + ((*ab) << 1) + (*cb + *bb + *bb + *cb)) >> 4);
rlm@1: #endif
rlm@1: 
rlm@1: 			// upper right
rlm@1: 			*to++ = RGB((*ar + *br) >> 1, (*ag + *bg) >> 1, (*ab + *bb) >> 1);
rlm@1: 
rlm@1: 			// lower left
rlm@1: 			*to_odd++ = RGB((*ar + *cr) >> 1, (*ag + *cg) >> 1, (*ab + *cb) >> 1);
rlm@1: 
rlm@1: 			// lower right
rlm@1: 			*to_odd++ = RGB((*ar + *br + *cr + *dr) >> 2,
rlm@1: 			                (*ag + *bg + *cg + *dg) >> 2,
rlm@1: 			                (*ab + *bb + *cb + *db) >> 2);
rlm@1: 
rlm@1: 			// 'b' becomes 'a', 'd' becomes 'c'
rlm@1: 			ar = br;
rlm@1: 			ag = bg;
rlm@1: 			ab = bb;
rlm@1: 			cr = dr;
rlm@1: 			cg = dg;
rlm@1: 			cb = db;
rlm@1: 		}
rlm@1: 
rlm@1: 		// the "next" rgb row becomes the current; the old current rgb row is
rlm@1: 		// recycled and serves as the new "next" row
rlm@1: 		u8 *temp;
rlm@1: 		temp		 = rgb_row_cur;
rlm@1: 		rgb_row_cur	 = rgb_row_next;
rlm@1: 		rgb_row_next = temp;
rlm@1: 
rlm@1: 		// update the pointers for start of next pair of lines
rlm@1: 		from   = (u16 *)((u8 *)from_orig + srcPitch);
rlm@1: 		to	   = (u16 *)((u8 *)to_orig + (dstPitch << 1));
rlm@1: 		to_odd = (u16 *)((u8 *)to + dstPitch);
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: void Bilinear32(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:                 u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u32 *to		= (u32 *)dstPtr;
rlm@1: 	u32 *to_odd = (u32 *)(dstPtr + dstPitch);
rlm@1: 
rlm@1: 	int from_width = width;
rlm@1: 	if (width + 1 < from_width)
rlm@1: 		from_width = width + 1;
rlm@1: 	u32 *from = (u32 *)srcPtr;
rlm@1: 	fill_rgb_row_32(from, from_width, rgb_row_cur, width + 1);
rlm@1: 
rlm@1: 	for (int y = 0; y < height; y++)
rlm@1: 	{
rlm@1: 		u32 *from_orig = from;
rlm@1: 		u32 *to_orig   = to;
rlm@1: 
rlm@1: 		if (y + 1 < height)
rlm@1: 			fill_rgb_row_32(from + width + 1, from_width, rgb_row_next,
rlm@1: 			                width + 1);
rlm@1: 		else
rlm@1: 			fill_rgb_row_32(from, from_width, rgb_row_next, width + 1);
rlm@1: 
rlm@1: 		// every pixel in the src region, is extended to 4 pixels in the
rlm@1: 		// destination, arranged in a square 'quad'; if the current src
rlm@1: 		// pixel is 'a', then in what follows 'b' is the src pixel to the
rlm@1: 		// right, 'c' is the src pixel below, and 'd' is the src pixel to
rlm@1: 		// the right and down
rlm@1: 		u8 *cur_row	 = rgb_row_cur;
rlm@1: 		u8 *next_row = rgb_row_next;
rlm@1: 		u8 *ar		 = cur_row++;
rlm@1: 		u8 *ag		 = cur_row++;
rlm@1: 		u8 *ab		 = cur_row++;
rlm@1: 		u8 *cr		 = next_row++;
rlm@1: 		u8 *cg		 = next_row++;
rlm@1: 		u8 *cb		 = next_row++;
rlm@1: 		for (int x = 0; x < width; x++)
rlm@1: 		{
rlm@1: 			u8 *br = cur_row++;
rlm@1: 			u8 *bg = cur_row++;
rlm@1: 			u8 *bb = cur_row++;
rlm@1: 			u8 *dr = next_row++;
rlm@1: 			u8 *dg = next_row++;
rlm@1: 			u8 *db = next_row++;
rlm@1: 
rlm@1: 			// upper left pixel in quad: just copy it in
rlm@1: 			*to++ = RGB(*ar, *ag, *ab);
rlm@1: 
rlm@1: 			// upper right
rlm@1: 			*to++ = RGB((*ar + *br) >> 1, (*ag + *bg) >> 1, (*ab + *bb) >> 1);
rlm@1: 
rlm@1: 			// lower left
rlm@1: 			*to_odd++ = RGB((*ar + *cr) >> 1, (*ag + *cg) >> 1, (*ab + *cb) >> 1);
rlm@1: 
rlm@1: 			// lower right
rlm@1: 			*to_odd++ = RGB((*ar + *br + *cr + *dr) >> 2,
rlm@1: 			                (*ag + *bg + *cg + *dg) >> 2,
rlm@1: 			                (*ab + *bb + *cb + *db) >> 2);
rlm@1: 
rlm@1: 			// 'b' becomes 'a', 'd' becomes 'c'
rlm@1: 			ar = br;
rlm@1: 			ag = bg;
rlm@1: 			ab = bb;
rlm@1: 			cr = dr;
rlm@1: 			cg = dg;
rlm@1: 			cb = db;
rlm@1: 		}
rlm@1: 
rlm@1: 		// the "next" rgb row becomes the current; the old current rgb row is
rlm@1: 		// recycled and serves as the new "next" row
rlm@1: 		u8 *temp;
rlm@1: 		temp		 = rgb_row_cur;
rlm@1: 		rgb_row_cur	 = rgb_row_next;
rlm@1: 		rgb_row_next = temp;
rlm@1: 
rlm@1: 		// update the pointers for start of next pair of lines
rlm@1: 		from   = (u32 *)((u8 *)from_orig + srcPitch);
rlm@1: 		to	   = (u32 *)((u8 *)to_orig + (dstPitch << 1));
rlm@1: 		to_odd = (u32 *)((u8 *)to + dstPitch);
rlm@1: 	}
rlm@1: }
rlm@1: 
rlm@1: void BilinearPlus32(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
rlm@1:                     u8 *dstPtr, u32 dstPitch, int width, int height)
rlm@1: {
rlm@1: 	u32 *to		= (u32 *)dstPtr;
rlm@1: 	u32 *to_odd = (u32 *)(dstPtr + dstPitch);
rlm@1: 
rlm@1: 	int from_width = width;
rlm@1: 	if (width + 1 < from_width)
rlm@1: 		from_width = width + 1;
rlm@1: 	u32 *from = (u32 *)srcPtr;
rlm@1: 	fill_rgb_row_32(from, from_width, rgb_row_cur, width + 1);
rlm@1: 
rlm@1: 	for (int y = 0; y < height; y++)
rlm@1: 	{
rlm@1: 		u32 *from_orig = from;
rlm@1: 		u32 *to_orig   = to;
rlm@1: 
rlm@1: 		if (y + 1 < height)
rlm@1: 			fill_rgb_row_32(from + width + 1, from_width, rgb_row_next,
rlm@1: 			                width + 1);
rlm@1: 		else
rlm@1: 			fill_rgb_row_32(from, from_width, rgb_row_next, width + 1);
rlm@1: 
rlm@1: 		// every pixel in the src region, is extended to 4 pixels in the
rlm@1: 		// destination, arranged in a square 'quad'; if the current src
rlm@1: 		// pixel is 'a', then in what follows 'b' is the src pixel to the
rlm@1: 		// right, 'c' is the src pixel below, and 'd' is the src pixel to
rlm@1: 		// the right and down
rlm@1: 		u8 *cur_row	 = rgb_row_cur;
rlm@1: 		u8 *next_row = rgb_row_next;
rlm@1: 		u8 *ar		 = cur_row++;
rlm@1: 		u8 *ag		 = cur_row++;
rlm@1: 		u8 *ab		 = cur_row++;
rlm@1: 		u8 *cr		 = next_row++;
rlm@1: 		u8 *cg		 = next_row++;
rlm@1: 		u8 *cb		 = next_row++;
rlm@1: 		for (int x = 0; x < width; x++)
rlm@1: 		{
rlm@1: 			u8 *br = cur_row++;
rlm@1: 			u8 *bg = cur_row++;
rlm@1: 			u8 *bb = cur_row++;
rlm@1: 			u8 *dr = next_row++;
rlm@1: 			u8 *dg = next_row++;
rlm@1: 			u8 *db = next_row++;
rlm@1: 
rlm@1: 			// upper left pixel in quad: just copy it in
rlm@1: 			//*to++ = manip.rgb(*ar, *ag, *ab);
rlm@1: #ifdef USE_ORIGINAL_BILINEAR_PLUS
rlm@1: 			*to++ = RGB(
rlm@1: 			    (((*ar) << 2) + ((*ar)) + (*cr + *br + *br)) >> 3,
rlm@1: 			    (((*ag) << 2) + ((*ag)) + (*cg + *bg + *bg)) >> 3,
rlm@1: 			    (((*ab) << 2) + ((*ab)) + (*cb + *bb + *bb)) >> 3);
rlm@1: #else
rlm@1: 			*to++ = RGB(
rlm@1: 			    (((*ar) << 3) + ((*ar) << 1) + (*cr + *br + *br + *cr)) >> 4,
rlm@1: 			    (((*ag) << 3) + ((*ag) << 1) + (*cg + *bg + *bg + *cg)) >> 4,
rlm@1: 			    (((*ab) << 3) + ((*ab) << 1) + (*cb + *bb + *bb + *cb)) >> 4);
rlm@1: #endif
rlm@1: 
rlm@1: 			// upper right
rlm@1: 			*to++ = RGB((*ar + *br) >> 1, (*ag + *bg) >> 1, (*ab + *bb) >> 1);
rlm@1: 
rlm@1: 			// lower left
rlm@1: 			*to_odd++ = RGB((*ar + *cr) >> 1, (*ag + *cg) >> 1, (*ab + *cb) >> 1);
rlm@1: 
rlm@1: 			// lower right
rlm@1: 			*to_odd++ = RGB((*ar + *br + *cr + *dr) >> 2,
rlm@1: 			                (*ag + *bg + *cg + *dg) >> 2,
rlm@1: 			                (*ab + *bb + *cb + *db) >> 2);
rlm@1: 
rlm@1: 			// 'b' becomes 'a', 'd' becomes 'c'
rlm@1: 			ar = br;
rlm@1: 			ag = bg;
rlm@1: 			ab = bb;
rlm@1: 			cr = dr;
rlm@1: 			cg = dg;
rlm@1: 			cb = db;
rlm@1: 		}
rlm@1: 
rlm@1: 		// the "next" rgb row becomes the current; the old current rgb row is
rlm@1: 		// recycled and serves as the new "next" row
rlm@1: 		u8 *temp;
rlm@1: 		temp		 = rgb_row_cur;
rlm@1: 		rgb_row_cur	 = rgb_row_next;
rlm@1: 		rgb_row_next = temp;
rlm@1: 
rlm@1: 		// update the pointers for start of next pair of lines
rlm@1: 		from   = (u32 *)((u8 *)from_orig + srcPitch);
rlm@1: 		to	   = (u32 *)((u8 *)to_orig + (dstPitch << 1));
rlm@1: 		to_odd = (u32 *)((u8 *)to + dstPitch);
rlm@1: 	}
rlm@1: }
rlm@1: