1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/src/filters/bilinear.cpp	Sat Mar 03 10:31:27 2012 -0600
     1.3 @@ -0,0 +1,418 @@
     1.4 +/**     Code adapted from Exult source code by Forgotten
     1.5 +**	Scale.cc - Trying to scale with bilinear interpolation.
     1.6 +**
     1.7 +**	Written: 6/14/00 - JSF
     1.8 +**/
     1.9 +
    1.10 +#include "../common/System.h"
    1.11 +
    1.12 +static u8 row_cur[3 * 322];
    1.13 +static u8 row_next[3 * 322];
    1.14 +
    1.15 +static u8 *rgb_row_cur	= row_cur;
    1.16 +static u8 *rgb_row_next = row_next;
    1.17 +
    1.18 +#ifdef RGB
    1.19 +#undef RGB  // wingdi.h has it
    1.20 +#endif
    1.21 +#define RGB(r, g, b) \
    1.22 +    ((r) >> 3) << systemRedShift | \
    1.23 +    ((g) >> 3) << systemGreenShift | \
    1.24 +    ((b) >> 3) << systemBlueShift \
    1.25 +
    1.26 +static void fill_rgb_row_16(u16 *from, int src_width, u8 *row, int width)
    1.27 +{
    1.28 +	u8 *copy_start = row + src_width * 3;
    1.29 +	u8 *all_stop   = row + width * 3;
    1.30 +	while (row < copy_start)
    1.31 +	{
    1.32 +		u16 color = *from++;
    1.33 +		*row++ = ((color >> systemRedShift) & 0x1f) << 3;
    1.34 +		*row++ = ((color >> systemGreenShift) & 0x1f) << 3;
    1.35 +		*row++ = ((color >> systemBlueShift) & 0x1f) << 3;
    1.36 +	}
    1.37 +	// any remaining elements to be written to 'row' are a replica of the
    1.38 +	// preceding pixel
    1.39 +	u8 *p = row - 3;
    1.40 +	while (row < all_stop)
    1.41 +	{
    1.42 +		// we're guaranteed three elements per pixel; could unroll the loop
    1.43 +		// further, especially with a Duff's Device, but the gains would be
    1.44 +		// probably limited (judging by profiler output)
    1.45 +		*row++ = *p++;
    1.46 +		*row++ = *p++;
    1.47 +		*row++ = *p++;
    1.48 +	}
    1.49 +}
    1.50 +
    1.51 +static void fill_rgb_row_32(u32 *from, int src_width, u8 *row, int width)
    1.52 +{
    1.53 +	u8 *copy_start = row + src_width * 3;
    1.54 +	u8 *all_stop   = row + width * 3;
    1.55 +	while (row < copy_start)
    1.56 +	{
    1.57 +		u32 color = *from++;
    1.58 +		*row++ = ((color >> systemRedShift) & 0x1f) << 3;
    1.59 +		*row++ = ((color >> systemGreenShift) & 0x1f) << 3;
    1.60 +		*row++ = ((color >> systemBlueShift) & 0x1f) << 3;
    1.61 +	}
    1.62 +	// any remaining elements to be written to 'row' are a replica of the
    1.63 +	// preceding pixel
    1.64 +	u8 *p = row - 3;
    1.65 +	while (row < all_stop)
    1.66 +	{
    1.67 +		// we're guaranteed three elements per pixel; could unroll the loop
    1.68 +		// further, especially with a Duff's Device, but the gains would be
    1.69 +		// probably limited (judging by profiler output)
    1.70 +		*row++ = *p++;
    1.71 +		*row++ = *p++;
    1.72 +		*row++ = *p++;
    1.73 +	}
    1.74 +}
    1.75 +
    1.76 +void Bilinear(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
    1.77 +              u8 *dstPtr, u32 dstPitch, int width, int height)
    1.78 +{
    1.79 +	u16 *to		= (u16 *)dstPtr;
    1.80 +	u16 *to_odd = (u16 *)(dstPtr + dstPitch);
    1.81 +
    1.82 +	int	 from_width = width;
    1.83 +	u16 *from		= (u16 *)srcPtr;
    1.84 +	fill_rgb_row_16(from, from_width, rgb_row_cur, width + 1);
    1.85 +
    1.86 +	for (int y = 0; y < height; y++)
    1.87 +	{
    1.88 +		u16 *from_orig = from;
    1.89 +		u16 *to_orig   = to;
    1.90 +
    1.91 +		if (y + 1 < height)
    1.92 +			fill_rgb_row_16(from + width + 2, from_width, rgb_row_next,
    1.93 +			                width + 1);
    1.94 +		else
    1.95 +			fill_rgb_row_16(from, from_width, rgb_row_next, width + 1);
    1.96 +
    1.97 +		// every pixel in the src region, is extended to 4 pixels in the
    1.98 +		// destination, arranged in a square 'quad'; if the current src
    1.99 +		// pixel is 'a', then in what follows 'b' is the src pixel to the
   1.100 +		// right, 'c' is the src pixel below, and 'd' is the src pixel to
   1.101 +		// the right and down
   1.102 +		u8 *cur_row	 = rgb_row_cur;
   1.103 +		u8 *next_row = rgb_row_next;
   1.104 +		u8 *ar		 = cur_row++;
   1.105 +		u8 *ag		 = cur_row++;
   1.106 +		u8 *ab		 = cur_row++;
   1.107 +		u8 *cr		 = next_row++;
   1.108 +		u8 *cg		 = next_row++;
   1.109 +		u8 *cb		 = next_row++;
   1.110 +		for (int x = 0; x < width; x++)
   1.111 +		{
   1.112 +			u8 *br = cur_row++;
   1.113 +			u8 *bg = cur_row++;
   1.114 +			u8 *bb = cur_row++;
   1.115 +			u8 *dr = next_row++;
   1.116 +			u8 *dg = next_row++;
   1.117 +			u8 *db = next_row++;
   1.118 +
   1.119 +			// upper left pixel in quad: just copy it in
   1.120 +			*to++ = RGB(*ar, *ag, *ab);
   1.121 +
   1.122 +			// upper right
   1.123 +			*to++ = RGB((*ar + *br) >> 1, (*ag + *bg) >> 1, (*ab + *bb) >> 1);
   1.124 +
   1.125 +			// lower left
   1.126 +			*to_odd++ = RGB((*ar + *cr) >> 1, (*ag + *cg) >> 1, (*ab + *cb) >> 1);
   1.127 +
   1.128 +			// lower right
   1.129 +			*to_odd++ = RGB((*ar + *br + *cr + *dr) >> 2,
   1.130 +			                (*ag + *bg + *cg + *dg) >> 2,
   1.131 +			                (*ab + *bb + *cb + *db) >> 2);
   1.132 +
   1.133 +			// 'b' becomes 'a', 'd' becomes 'c'
   1.134 +			ar = br;
   1.135 +			ag = bg;
   1.136 +			ab = bb;
   1.137 +			cr = dr;
   1.138 +			cg = dg;
   1.139 +			cb = db;
   1.140 +		}
   1.141 +
   1.142 +		// the "next" rgb row becomes the current; the old current rgb row is
   1.143 +		// recycled and serves as the new "next" row
   1.144 +		u8 *temp;
   1.145 +		temp		 = rgb_row_cur;
   1.146 +		rgb_row_cur	 = rgb_row_next;
   1.147 +		rgb_row_next = temp;
   1.148 +
   1.149 +		// update the pointers for start of next pair of lines
   1.150 +		from   = (u16 *)((u8 *)from_orig + srcPitch);
   1.151 +		to	   = (u16 *)((u8 *)to_orig + (dstPitch << 1));
   1.152 +		to_odd = (u16 *)((u8 *)to + dstPitch);
   1.153 +	}
   1.154 +}
   1.155 +
   1.156 +void BilinearPlus(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
   1.157 +                  u8 *dstPtr, u32 dstPitch, int width, int height)
   1.158 +{
   1.159 +	u16 *to		= (u16 *)dstPtr;
   1.160 +	u16 *to_odd = (u16 *)(dstPtr + dstPitch);
   1.161 +
   1.162 +	int	 from_width = width;
   1.163 +	u16 *from		= (u16 *)srcPtr;
   1.164 +	fill_rgb_row_16(from, from_width, rgb_row_cur, width + 1);
   1.165 +
   1.166 +	for (int y = 0; y < height; y++)
   1.167 +	{
   1.168 +		u16 *from_orig = from;
   1.169 +		u16 *to_orig   = to;
   1.170 +
   1.171 +		if (y + 1 < height)
   1.172 +			fill_rgb_row_16(from + width + 2, from_width, rgb_row_next,
   1.173 +			                width + 1);
   1.174 +		else
   1.175 +			fill_rgb_row_16(from, from_width, rgb_row_next, width + 1);
   1.176 +
   1.177 +		// every pixel in the src region, is extended to 4 pixels in the
   1.178 +		// destination, arranged in a square 'quad'; if the current src
   1.179 +		// pixel is 'a', then in what follows 'b' is the src pixel to the
   1.180 +		// right, 'c' is the src pixel below, and 'd' is the src pixel to
   1.181 +		// the right and down
   1.182 +		u8 *cur_row	 = rgb_row_cur;
   1.183 +		u8 *next_row = rgb_row_next;
   1.184 +		u8 *ar		 = cur_row++;
   1.185 +		u8 *ag		 = cur_row++;
   1.186 +		u8 *ab		 = cur_row++;
   1.187 +		u8 *cr		 = next_row++;
   1.188 +		u8 *cg		 = next_row++;
   1.189 +		u8 *cb		 = next_row++;
   1.190 +		for (int x = 0; x < width; x++)
   1.191 +		{
   1.192 +			u8 *br = cur_row++;
   1.193 +			u8 *bg = cur_row++;
   1.194 +			u8 *bb = cur_row++;
   1.195 +			u8 *dr = next_row++;
   1.196 +			u8 *dg = next_row++;
   1.197 +			u8 *db = next_row++;
   1.198 +
   1.199 +			// upper left pixel in quad: just copy it in
   1.200 +			//*to++ = manip.rgb(*ar, *ag, *ab);
   1.201 +#ifdef USE_ORIGINAL_BILINEAR_PLUS
   1.202 +			*to++ = RGB(
   1.203 +			    (((*ar) << 2) + ((*ar)) + (*cr + *br + *br)) >> 3,
   1.204 +			    (((*ag) << 2) + ((*ag)) + (*cg + *bg + *bg)) >> 3,
   1.205 +			    (((*ab) << 2) + ((*ab)) + (*cb + *bb + *bb)) >> 3);
   1.206 +#else
   1.207 +			*to++ = RGB(
   1.208 +			    (((*ar) << 3) + ((*ar) << 1) + (*cr + *br + *br + *cr)) >> 4,
   1.209 +			    (((*ag) << 3) + ((*ag) << 1) + (*cg + *bg + *bg + *cg)) >> 4,
   1.210 +			    (((*ab) << 3) + ((*ab) << 1) + (*cb + *bb + *bb + *cb)) >> 4);
   1.211 +#endif
   1.212 +
   1.213 +			// upper right
   1.214 +			*to++ = RGB((*ar + *br) >> 1, (*ag + *bg) >> 1, (*ab + *bb) >> 1);
   1.215 +
   1.216 +			// lower left
   1.217 +			*to_odd++ = RGB((*ar + *cr) >> 1, (*ag + *cg) >> 1, (*ab + *cb) >> 1);
   1.218 +
   1.219 +			// lower right
   1.220 +			*to_odd++ = RGB((*ar + *br + *cr + *dr) >> 2,
   1.221 +			                (*ag + *bg + *cg + *dg) >> 2,
   1.222 +			                (*ab + *bb + *cb + *db) >> 2);
   1.223 +
   1.224 +			// 'b' becomes 'a', 'd' becomes 'c'
   1.225 +			ar = br;
   1.226 +			ag = bg;
   1.227 +			ab = bb;
   1.228 +			cr = dr;
   1.229 +			cg = dg;
   1.230 +			cb = db;
   1.231 +		}
   1.232 +
   1.233 +		// the "next" rgb row becomes the current; the old current rgb row is
   1.234 +		// recycled and serves as the new "next" row
   1.235 +		u8 *temp;
   1.236 +		temp		 = rgb_row_cur;
   1.237 +		rgb_row_cur	 = rgb_row_next;
   1.238 +		rgb_row_next = temp;
   1.239 +
   1.240 +		// update the pointers for start of next pair of lines
   1.241 +		from   = (u16 *)((u8 *)from_orig + srcPitch);
   1.242 +		to	   = (u16 *)((u8 *)to_orig + (dstPitch << 1));
   1.243 +		to_odd = (u16 *)((u8 *)to + dstPitch);
   1.244 +	}
   1.245 +}
   1.246 +
   1.247 +void Bilinear32(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
   1.248 +                u8 *dstPtr, u32 dstPitch, int width, int height)
   1.249 +{
   1.250 +	u32 *to		= (u32 *)dstPtr;
   1.251 +	u32 *to_odd = (u32 *)(dstPtr + dstPitch);
   1.252 +
   1.253 +	int from_width = width;
   1.254 +	if (width + 1 < from_width)
   1.255 +		from_width = width + 1;
   1.256 +	u32 *from = (u32 *)srcPtr;
   1.257 +	fill_rgb_row_32(from, from_width, rgb_row_cur, width + 1);
   1.258 +
   1.259 +	for (int y = 0; y < height; y++)
   1.260 +	{
   1.261 +		u32 *from_orig = from;
   1.262 +		u32 *to_orig   = to;
   1.263 +
   1.264 +		if (y + 1 < height)
   1.265 +			fill_rgb_row_32(from + width + 1, from_width, rgb_row_next,
   1.266 +			                width + 1);
   1.267 +		else
   1.268 +			fill_rgb_row_32(from, from_width, rgb_row_next, width + 1);
   1.269 +
   1.270 +		// every pixel in the src region, is extended to 4 pixels in the
   1.271 +		// destination, arranged in a square 'quad'; if the current src
   1.272 +		// pixel is 'a', then in what follows 'b' is the src pixel to the
   1.273 +		// right, 'c' is the src pixel below, and 'd' is the src pixel to
   1.274 +		// the right and down
   1.275 +		u8 *cur_row	 = rgb_row_cur;
   1.276 +		u8 *next_row = rgb_row_next;
   1.277 +		u8 *ar		 = cur_row++;
   1.278 +		u8 *ag		 = cur_row++;
   1.279 +		u8 *ab		 = cur_row++;
   1.280 +		u8 *cr		 = next_row++;
   1.281 +		u8 *cg		 = next_row++;
   1.282 +		u8 *cb		 = next_row++;
   1.283 +		for (int x = 0; x < width; x++)
   1.284 +		{
   1.285 +			u8 *br = cur_row++;
   1.286 +			u8 *bg = cur_row++;
   1.287 +			u8 *bb = cur_row++;
   1.288 +			u8 *dr = next_row++;
   1.289 +			u8 *dg = next_row++;
   1.290 +			u8 *db = next_row++;
   1.291 +
   1.292 +			// upper left pixel in quad: just copy it in
   1.293 +			*to++ = RGB(*ar, *ag, *ab);
   1.294 +
   1.295 +			// upper right
   1.296 +			*to++ = RGB((*ar + *br) >> 1, (*ag + *bg) >> 1, (*ab + *bb) >> 1);
   1.297 +
   1.298 +			// lower left
   1.299 +			*to_odd++ = RGB((*ar + *cr) >> 1, (*ag + *cg) >> 1, (*ab + *cb) >> 1);
   1.300 +
   1.301 +			// lower right
   1.302 +			*to_odd++ = RGB((*ar + *br + *cr + *dr) >> 2,
   1.303 +			                (*ag + *bg + *cg + *dg) >> 2,
   1.304 +			                (*ab + *bb + *cb + *db) >> 2);
   1.305 +
   1.306 +			// 'b' becomes 'a', 'd' becomes 'c'
   1.307 +			ar = br;
   1.308 +			ag = bg;
   1.309 +			ab = bb;
   1.310 +			cr = dr;
   1.311 +			cg = dg;
   1.312 +			cb = db;
   1.313 +		}
   1.314 +
   1.315 +		// the "next" rgb row becomes the current; the old current rgb row is
   1.316 +		// recycled and serves as the new "next" row
   1.317 +		u8 *temp;
   1.318 +		temp		 = rgb_row_cur;
   1.319 +		rgb_row_cur	 = rgb_row_next;
   1.320 +		rgb_row_next = temp;
   1.321 +
   1.322 +		// update the pointers for start of next pair of lines
   1.323 +		from   = (u32 *)((u8 *)from_orig + srcPitch);
   1.324 +		to	   = (u32 *)((u8 *)to_orig + (dstPitch << 1));
   1.325 +		to_odd = (u32 *)((u8 *)to + dstPitch);
   1.326 +	}
   1.327 +}
   1.328 +
   1.329 +void BilinearPlus32(u8 *srcPtr, u32 srcPitch, u8 * /* deltaPtr */,
   1.330 +                    u8 *dstPtr, u32 dstPitch, int width, int height)
   1.331 +{
   1.332 +	u32 *to		= (u32 *)dstPtr;
   1.333 +	u32 *to_odd = (u32 *)(dstPtr + dstPitch);
   1.334 +
   1.335 +	int from_width = width;
   1.336 +	if (width + 1 < from_width)
   1.337 +		from_width = width + 1;
   1.338 +	u32 *from = (u32 *)srcPtr;
   1.339 +	fill_rgb_row_32(from, from_width, rgb_row_cur, width + 1);
   1.340 +
   1.341 +	for (int y = 0; y < height; y++)
   1.342 +	{
   1.343 +		u32 *from_orig = from;
   1.344 +		u32 *to_orig   = to;
   1.345 +
   1.346 +		if (y + 1 < height)
   1.347 +			fill_rgb_row_32(from + width + 1, from_width, rgb_row_next,
   1.348 +			                width + 1);
   1.349 +		else
   1.350 +			fill_rgb_row_32(from, from_width, rgb_row_next, width + 1);
   1.351 +
   1.352 +		// every pixel in the src region, is extended to 4 pixels in the
   1.353 +		// destination, arranged in a square 'quad'; if the current src
   1.354 +		// pixel is 'a', then in what follows 'b' is the src pixel to the
   1.355 +		// right, 'c' is the src pixel below, and 'd' is the src pixel to
   1.356 +		// the right and down
   1.357 +		u8 *cur_row	 = rgb_row_cur;
   1.358 +		u8 *next_row = rgb_row_next;
   1.359 +		u8 *ar		 = cur_row++;
   1.360 +		u8 *ag		 = cur_row++;
   1.361 +		u8 *ab		 = cur_row++;
   1.362 +		u8 *cr		 = next_row++;
   1.363 +		u8 *cg		 = next_row++;
   1.364 +		u8 *cb		 = next_row++;
   1.365 +		for (int x = 0; x < width; x++)
   1.366 +		{
   1.367 +			u8 *br = cur_row++;
   1.368 +			u8 *bg = cur_row++;
   1.369 +			u8 *bb = cur_row++;
   1.370 +			u8 *dr = next_row++;
   1.371 +			u8 *dg = next_row++;
   1.372 +			u8 *db = next_row++;
   1.373 +
   1.374 +			// upper left pixel in quad: just copy it in
   1.375 +			//*to++ = manip.rgb(*ar, *ag, *ab);
   1.376 +#ifdef USE_ORIGINAL_BILINEAR_PLUS
   1.377 +			*to++ = RGB(
   1.378 +			    (((*ar) << 2) + ((*ar)) + (*cr + *br + *br)) >> 3,
   1.379 +			    (((*ag) << 2) + ((*ag)) + (*cg + *bg + *bg)) >> 3,
   1.380 +			    (((*ab) << 2) + ((*ab)) + (*cb + *bb + *bb)) >> 3);
   1.381 +#else
   1.382 +			*to++ = RGB(
   1.383 +			    (((*ar) << 3) + ((*ar) << 1) + (*cr + *br + *br + *cr)) >> 4,
   1.384 +			    (((*ag) << 3) + ((*ag) << 1) + (*cg + *bg + *bg + *cg)) >> 4,
   1.385 +			    (((*ab) << 3) + ((*ab) << 1) + (*cb + *bb + *bb + *cb)) >> 4);
   1.386 +#endif
   1.387 +
   1.388 +			// upper right
   1.389 +			*to++ = RGB((*ar + *br) >> 1, (*ag + *bg) >> 1, (*ab + *bb) >> 1);
   1.390 +
   1.391 +			// lower left
   1.392 +			*to_odd++ = RGB((*ar + *cr) >> 1, (*ag + *cg) >> 1, (*ab + *cb) >> 1);
   1.393 +
   1.394 +			// lower right
   1.395 +			*to_odd++ = RGB((*ar + *br + *cr + *dr) >> 2,
   1.396 +			                (*ag + *bg + *cg + *dg) >> 2,
   1.397 +			                (*ab + *bb + *cb + *db) >> 2);
   1.398 +
   1.399 +			// 'b' becomes 'a', 'd' becomes 'c'
   1.400 +			ar = br;
   1.401 +			ag = bg;
   1.402 +			ab = bb;
   1.403 +			cr = dr;
   1.404 +			cg = dg;
   1.405 +			cb = db;
   1.406 +		}
   1.407 +
   1.408 +		// the "next" rgb row becomes the current; the old current rgb row is
   1.409 +		// recycled and serves as the new "next" row
   1.410 +		u8 *temp;
   1.411 +		temp		 = rgb_row_cur;
   1.412 +		rgb_row_cur	 = rgb_row_next;
   1.413 +		rgb_row_next = temp;
   1.414 +
   1.415 +		// update the pointers for start of next pair of lines
   1.416 +		from   = (u32 *)((u8 *)from_orig + srcPitch);
   1.417 +		to	   = (u32 *)((u8 *)to_orig + (dstPitch << 1));
   1.418 +		to_odd = (u32 *)((u8 *)to + dstPitch);
   1.419 +	}
   1.420 +}
   1.421 +