| 1 | /* |
| 2 | * transformfixedpoint.c |
| 3 | * |
| 4 | * Fixed point implementation of image transformations (see also transformfloat.c/h) |
| 5 | * |
| 6 | * Copyright (C) Georg Martius - June 2011 |
| 7 | * georg dot martius at web dot de |
| 8 | * |
| 9 | * This file is part of vid.stab video stabilization library |
| 10 | * |
| 11 | * vid.stab is free software; you can redistribute it and/or modify |
| 12 | * it under the terms of the GNU General Public License, |
| 13 | * as published by the Free Software Foundation; either version 2, or |
| 14 | * (at your option) any later version. |
| 15 | * |
| 16 | * vid.stab is distributed in the hope that it will be useful, |
| 17 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 18 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 19 | * GNU General Public License for more details. |
| 20 | * |
| 21 | * You should have received a copy of the GNU General Public License |
| 22 | * along with GNU Make; see the file COPYING. If not, write to |
| 23 | * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. |
| 24 | * |
| 25 | * |
| 26 | */ |
| 27 | #include "transformfixedpoint.h" |
| 28 | #include "transform.h" |
| 29 | #include "transformtype_operations.h" |
| 30 | |
| 31 | // the orc code does not work at the moment (BUG in ORC?) |
| 32 | // #include "orc/transformorc.h" |
| 33 | |
| 34 | //#include <math.h> |
| 35 | //#include <libgen.h> |
| 36 | |
| 37 | #define iToFp8(v) ((v)<<8) |
| 38 | #define fToFp8(v) ((int32_t)((v)*((float)0xFF))) |
| 39 | #define iToFp16(v) ((v)<<16) |
| 40 | #define fToFp16(v) ((int32_t)((v)*((double)0xFFFF))) |
| 41 | #define fp16To8(v) ((v)>>8) |
| 42 | //#define fp16To8(v) ( (v) && 0x80 == 1 ? ((v)>>8 + 1) : ((v)>>8) ) |
| 43 | #define fp24To8(v) ((v)>>16) |
| 44 | |
| 45 | #define fp8ToI(v) ((v)>>8) |
| 46 | #define fp16ToI(v) ((v)>>16) |
| 47 | #define fp8ToF(v) ((v)/((double)(1<<8))) |
| 48 | #define fp16ToF(v) ((v)/((double)(1<<16))) |
| 49 | |
| 50 | // #define fp8ToIRound(v) ( (((v)>>7) & 0x1) == 0 ? ((v)>>8) : ((v)>>8)+1 ) |
| 51 | #define fp8_0_5 (1<<7) |
| 52 | #define fp8ToIRound(v) (((v) + fp8_0_5) >> 7) |
| 53 | //#define fp16ToIRound(v) ( (((v)>>15) & 0x1) == 0 ? ((v)>>16) : ((v)>>16)+1 ) |
| 54 | #define fp16_0_5 (1<<15) |
| 55 | #define fp16ToIRound(v) (((v) + fp16_0_5) >> 16) |
| 56 | |
| 57 | /** interpolateBiLinBorder: bi-linear interpolation function that also works at the border. |
| 58 | This is used by many other interpolation methods at and outsize the border, see interpolate */ |
| 59 | inline void interpolateBiLinBorder(uint8_t *rv, fp16 x, fp16 y, |
| 60 | const uint8_t *img, int img_linesize, |
| 61 | int32_t width, int32_t height, uint8_t def) |
| 62 | { |
| 63 | int32_t ix_f = fp16ToI(x); |
| 64 | int32_t iy_f = fp16ToI(y); |
| 65 | int32_t ix_c = ix_f + 1; |
| 66 | int32_t iy_c = iy_f + 1; |
| 67 | if (ix_f < 0 || ix_c >= width || iy_f < 0 || iy_c >= height) { |
| 68 | int32_t w = 10; // number of pixels to blur out the border pixel outwards |
| 69 | int32_t xl = - w - ix_f; |
| 70 | int32_t yl = - w - iy_f; |
| 71 | int32_t xh = ix_c - w - width; |
| 72 | int32_t yh = iy_c - w - height; |
| 73 | int32_t c = VS_MAX(VS_MIN(VS_MAX(xl, VS_MAX(yl, VS_MAX(xh, yh))),w),0); |
| 74 | // pixel at border of source image |
| 75 | short val_border = PIX(img, img_linesize, VS_MAX(VS_MIN(ix_f, width-1),0), |
| 76 | VS_MAX(VS_MIN(iy_f, height-1),0)); |
| 77 | *rv = (def * c + val_border * (w - c)) / w; |
| 78 | }else{ |
| 79 | short v1 = PIXEL(img, img_linesize, ix_c, iy_c, width, height, def); |
| 80 | short v2 = PIXEL(img, img_linesize, ix_c, iy_f, width, height, def); |
| 81 | short v3 = PIXEL(img, img_linesize, ix_f, iy_c, width, height, def); |
| 82 | short v4 = PIXEL(img, img_linesize, ix_f, iy_f, width, height, def); |
| 83 | fp16 x_f = iToFp16(ix_f); |
| 84 | fp16 x_c = iToFp16(ix_c); |
| 85 | fp16 y_f = iToFp16(iy_f); |
| 86 | fp16 y_c = iToFp16(iy_c); |
| 87 | fp16 s = fp16To8(v1*(x - x_f)+v3*(x_c - x))*fp16To8(y - y_f) + |
| 88 | fp16To8(v2*(x - x_f) + v4*(x_c - x))*fp16To8(y_c - y) + 1; |
| 89 | *rv = fp16ToIRound(s); |
| 90 | } |
| 91 | } |
| 92 | |
| 93 | /** taken from http://en.wikipedia.org/wiki/Bicubic_interpolation for alpha=-0.5 |
| 94 | in matrix notation: |
| 95 | a0-a3 are the neigthboring points where the target point is between a1 and a2 |
| 96 | t is the point of interpolation (position between a1 and a2) value between 0 and 1 |
| 97 | | 0, 2, 0, 0 | |a0| |
| 98 | |-1, 0, 1, 0 | |a1| |
| 99 | (1,t,t^2,t^3) | 2,-5, 4,-1 | |a2| |
| 100 | |-1, 3,-3, 1 | |a3| |
| 101 | */ |
| 102 | /* inline static short bicub_kernel(fp16 t, short a0, short a1, short a2, short a3){ */ |
| 103 | /* // (2*a1 + t*((-a0+a2) + t*((2*a0-5*a1+4*a2-a3) + t*(-a0+3*a1-3*a2+a3) )) ) / 2; */ |
| 104 | /* return ((iToFp16(2*a1) + t*(-a0+a2 */ |
| 105 | /* + fp16ToI(t*((2*a0-5*a1+4*a2-a3) */ |
| 106 | /* + fp16ToI(t*(-a0+3*a1-3*a2+a3)) )) ) */ |
| 107 | /* ) ) >> 17; */ |
| 108 | /* } */ |
| 109 | |
| 110 | inline static short bicub_kernel(fp16 t, short a0, short a1, short a2, short a3){ |
| 111 | // (2*a1 + t*((-a0+a2) + t*((2*a0-5*a1+4*a2-a3) + t*(-a0+3*a1-3*a2+a3) )) ) / 2; |
| 112 | // we add 1/2 because of truncation errors |
| 113 | return fp16ToIRound((iToFp16(2*a1) + t*(-a0+a2 |
| 114 | + fp16ToIRound(t*((2*a0-5*a1+4*a2-a3) |
| 115 | + fp16ToIRound(t*(-a0+3*a1-3*a2+a3)) )) ) |
| 116 | ) >> 1); |
| 117 | } |
| 118 | |
| 119 | /** interpolateBiCub: bi-cubic interpolation function using 4x4 pixel, see interpolate */ |
| 120 | inline void interpolateBiCub(uint8_t *rv, fp16 x, fp16 y, |
| 121 | const uint8_t *img, int img_linesize, |
| 122 | int width, int height, uint8_t def) |
| 123 | { |
| 124 | // do a simple linear interpolation at the border |
| 125 | int32_t ix_f = fp16ToI(x); |
| 126 | int32_t iy_f = fp16ToI(y); |
| 127 | if (unlikely(ix_f < 1 || ix_f > width - 3 || iy_f < 1 || iy_f > height - 3)) { |
| 128 | interpolateBiLinBorder(rv, x, y, img, img_linesize, width, height, def); |
| 129 | } else { |
| 130 | fp16 x_f = iToFp16(ix_f); |
| 131 | fp16 y_f = iToFp16(iy_f); |
| 132 | fp16 tx = x-x_f; |
| 133 | short v1 = bicub_kernel(tx, |
| 134 | PIX(img, img_linesize, ix_f-1, iy_f-1), |
| 135 | PIX(img, img_linesize, ix_f, iy_f-1), |
| 136 | PIX(img, img_linesize, ix_f+1, iy_f-1), |
| 137 | PIX(img, img_linesize, ix_f+2, iy_f-1)); |
| 138 | short v2 = bicub_kernel(tx, |
| 139 | PIX(img, img_linesize, ix_f-1, iy_f), |
| 140 | PIX(img, img_linesize, ix_f, iy_f), |
| 141 | PIX(img, img_linesize, ix_f+1, iy_f), |
| 142 | PIX(img, img_linesize, ix_f+2, iy_f)); |
| 143 | short v3 = bicub_kernel(tx, |
| 144 | PIX(img, img_linesize, ix_f-1, iy_f+1), |
| 145 | PIX(img, img_linesize, ix_f, iy_f+1), |
| 146 | PIX(img, img_linesize, ix_f+1, iy_f+1), |
| 147 | PIX(img, img_linesize, ix_f+2, iy_f+1)); |
| 148 | short v4 = bicub_kernel(tx, |
| 149 | PIX(img, img_linesize, ix_f-1, iy_f+2), |
| 150 | PIX(img, img_linesize, ix_f, iy_f+2), |
| 151 | PIX(img, img_linesize, ix_f+1, iy_f+2), |
| 152 | PIX(img, img_linesize, ix_f+2, iy_f+2)); |
| 153 | short res = bicub_kernel(y-y_f, v1, v2, v3, v4); |
| 154 | *rv = res < 255 ? res : 255; |
| 155 | } |
| 156 | } |
| 157 | |
| 158 | |
| 159 | /** interpolateBiLin: bi-linear interpolation function, see interpolate */ |
| 160 | inline void interpolateBiLin(uint8_t *rv, fp16 x, fp16 y, |
| 161 | const uint8_t *img, int img_linesize, |
| 162 | int32_t width, int32_t height, uint8_t def) |
| 163 | { |
| 164 | int32_t ix_f = fp16ToI(x); |
| 165 | int32_t iy_f = fp16ToI(y); |
| 166 | if (unlikely(ix_f < 0 || ix_f > width - 2 || iy_f < 0 || iy_f > height - 2)) { |
| 167 | interpolateBiLinBorder(rv, x, y, img, img_linesize, width, height, def); |
| 168 | } else { |
| 169 | int32_t ix_c = ix_f + 1; |
| 170 | int32_t iy_c = iy_f + 1; |
| 171 | short v1 = PIX(img, img_linesize, ix_c, iy_c); |
| 172 | short v2 = PIX(img, img_linesize, ix_c, iy_f); |
| 173 | short v3 = PIX(img, img_linesize, ix_f, iy_c); |
| 174 | short v4 = PIX(img, img_linesize, ix_f, iy_f); |
| 175 | fp16 x_f = iToFp16(ix_f); |
| 176 | fp16 x_c = iToFp16(ix_c); |
| 177 | fp16 y_f = iToFp16(iy_f); |
| 178 | fp16 y_c = iToFp16(iy_c); |
| 179 | fp16 s = fp16To8(v1*(x - x_f) + v3*(x_c - x))*fp16To8(y - y_f) + |
| 180 | fp16To8(v2*(x - x_f) + v4*(x_c - x))*fp16To8(y_c - y); |
| 181 | // it is underestimated due to truncation, so we add one |
| 182 | short res = fp16ToI(s); |
| 183 | *rv = res < 255 ? res+1 : 255; |
| 184 | } |
| 185 | } |
| 186 | |
| 187 | /** interpolateLin: linear (only x) interpolation function, see interpolate */ |
| 188 | inline void interpolateLin(uint8_t *rv, fp16 x, fp16 y, |
| 189 | const uint8_t *img, int img_linesize, |
| 190 | int width, int height, uint8_t def) |
| 191 | { |
| 192 | int32_t ix_f = fp16ToI(x); |
| 193 | int32_t ix_c = ix_f + 1; |
| 194 | fp16 x_c = iToFp16(ix_c); |
| 195 | fp16 x_f = iToFp16(ix_f); |
| 196 | int y_n = fp16ToIRound(y); |
| 197 | |
| 198 | short v1 = PIXEL(img, img_linesize, ix_c, y_n, width, height, def); |
| 199 | short v2 = PIXEL(img, img_linesize, ix_f, y_n, width, height, def); |
| 200 | fp16 s = v1*(x - x_f) + v2*(x_c - x); |
| 201 | short res = fp16ToI(s); |
| 202 | *rv = res < 255 ? res : 255; |
| 203 | } |
| 204 | |
| 205 | /** interpolateZero: nearest neighbor interpolation function, see interpolate */ |
| 206 | inline void interpolateZero(uint8_t *rv, fp16 x, fp16 y, |
| 207 | const uint8_t *img, int img_linesize, |
| 208 | int width, int height, uint8_t def) |
| 209 | { |
| 210 | int32_t ix_n = fp16ToIRound(x); |
| 211 | int32_t iy_n = fp16ToIRound(y); |
| 212 | *rv = (uint8_t) PIXEL(img, img_linesize, ix_n, iy_n, width, height, def); |
| 213 | } |
| 214 | |
| 215 | |
| 216 | /** |
| 217 | * interpolateN: Bi-linear interpolation function for N channel image. |
| 218 | * |
| 219 | * Parameters: |
| 220 | * rv: destination pixel (call by reference) |
| 221 | * x,y: the source coordinates in the image img. Note this |
| 222 | * are real-value coordinates, that's why we interpolate |
| 223 | * img: source image |
| 224 | * width,height: dimension of image |
| 225 | * N: number of channels |
| 226 | * channel: channel number (0..N-1) |
| 227 | * def: default value if coordinates are out of range |
| 228 | * Return value: None |
| 229 | */ |
| 230 | inline void interpolateN(uint8_t *rv, fp16 x, fp16 y, |
| 231 | const uint8_t *img, int img_linesize, |
| 232 | int width, int height, |
| 233 | uint8_t N, uint8_t channel, |
| 234 | uint8_t def) |
| 235 | { |
| 236 | int32_t ix_f = fp16ToI(x); |
| 237 | int32_t iy_f = fp16ToI(y); |
| 238 | if (ix_f < 0 || ix_f > width-1 || iy_f < 0 || iy_f > height - 1) { |
| 239 | *rv = def; |
| 240 | } else { |
| 241 | int32_t ix_c = ix_f + 1; |
| 242 | int32_t iy_c = iy_f + 1; |
| 243 | short v1 = PIXN(img, img_linesize, ix_c, iy_c, N, channel); |
| 244 | short v2 = PIXN(img, img_linesize, ix_c, iy_f, N, channel); |
| 245 | short v3 = PIXN(img, img_linesize, ix_f, iy_c, N, channel); |
| 246 | short v4 = PIXN(img, img_linesize, ix_f, iy_f, N, channel); |
| 247 | fp16 x_f = iToFp16(ix_f); |
| 248 | fp16 x_c = iToFp16(ix_c); |
| 249 | fp16 y_f = iToFp16(iy_f); |
| 250 | fp16 y_c = iToFp16(iy_c); |
| 251 | fp16 s = fp16To8(v1*(x - x_f)+v3*(x_c - x))*fp16To8(y - y_f) + |
| 252 | fp16To8(v2*(x - x_f) + v4*(x_c - x))*fp16To8(y_c - y); |
| 253 | *rv = fp16ToIRound(s); |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | |
| 258 | /** |
| 259 | * transformPacked: applies current transformation to frame |
| 260 | * Parameters: |
| 261 | * td: private data structure of this filter |
| 262 | * Return value: |
| 263 | * 0 for failture, 1 for success |
| 264 | * Preconditions: |
| 265 | * The frame must be in Packed format |
| 266 | */ |
| 267 | int transformPacked(VSTransformData* td, VSTransform t) |
| 268 | { |
| 269 | int x = 0, y = 0, k = 0; |
| 270 | uint8_t *D_1, *D_2; |
| 271 | |
| 272 | D_1 = td->src.data[0]; |
| 273 | D_2 = td->destbuf.data[0]; |
| 274 | fp16 c_s_x = iToFp16(td->fiSrc.width/2); |
| 275 | fp16 c_s_y = iToFp16(td->fiSrc.height/2); |
| 276 | int32_t c_d_x = td->fiDest.width/2; |
| 277 | int32_t c_d_y = td->fiDest.height/2; |
| 278 | |
| 279 | /* for each pixel in the destination image we calc the source |
| 280 | * coordinate and make an interpolation: |
| 281 | * p_d = c_d + M(p_s - c_s) + t |
| 282 | * where p are the points, c the center coordinate, |
| 283 | * _s source and _d destination, |
| 284 | * t the translation, and M the rotation matrix |
| 285 | * p_s = M^{-1}(p_d - c_d - t) + c_s |
| 286 | */ |
| 287 | float z = 1.0-t.zoom/100.0; |
| 288 | fp16 zcos_a = fToFp16(z*cos(-t.alpha)); // scaled cos |
| 289 | fp16 zsin_a = fToFp16(z*sin(-t.alpha)); // scaled sin |
| 290 | fp16 c_tx = c_s_x - fToFp16(t.x); |
| 291 | fp16 c_ty = c_s_y - fToFp16(t.y); |
| 292 | int channels = td->fiSrc.bytesPerPixel; |
| 293 | /* All channels */ |
| 294 | for (y = 0; y < td->fiDest.height; y++) { |
| 295 | int32_t y_d1 = (y - c_d_y); |
| 296 | for (x = 0; x < td->fiDest.width; x++) { |
| 297 | int32_t x_d1 = (x - c_d_x); |
| 298 | fp16 x_s = zcos_a * x_d1 + zsin_a * y_d1 + c_tx; |
| 299 | fp16 y_s = -zsin_a * x_d1 + zcos_a * y_d1 + c_ty; |
| 300 | |
| 301 | for (k = 0; k < channels; k++) { // iterate over colors |
| 302 | uint8_t *dest = &D_2[x + y * td->destbuf.linesize[0]+k]; |
| 303 | interpolateN(dest, x_s, y_s, D_1, td->src.linesize[0], |
| 304 | td->fiSrc.width, td->fiSrc.height, |
| 305 | channels, k, td->conf.crop ? 16 : *dest); |
| 306 | } |
| 307 | } |
| 308 | } |
| 309 | return VS_OK; |
| 310 | } |
| 311 | |
| 312 | /** |
| 313 | * transformPlanar: applies current transformation to frame |
| 314 | * |
| 315 | * Parameters: |
| 316 | * td: private data structure of this filter |
| 317 | * Return value: |
| 318 | * 0 for failture, 1 for success |
| 319 | * Preconditions: |
| 320 | * The frame must be in Planar format |
| 321 | * |
| 322 | * Fixed-point format 32 bit integer: |
| 323 | * for image coords we use val<<8 |
| 324 | * for angle and zoom we use val<<16 |
| 325 | * |
| 326 | */ |
| 327 | int transformPlanar(VSTransformData* td, VSTransform t) |
| 328 | { |
| 329 | int32_t x = 0, y = 0; |
| 330 | uint8_t *dat_1, *dat_2; |
| 331 | |
| 332 | if (t.alpha==0 && t.x==0 && t.y==0 && t.zoom == 0){ |
| 333 | if(vsFramesEqual(&td->src,&td->destbuf)) |
| 334 | return VS_OK; // noop |
| 335 | else { |
| 336 | vsFrameCopy(&td->destbuf, &td->src, &td->fiSrc); |
| 337 | return VS_OK; |
| 338 | } |
| 339 | } |
| 340 | |
| 341 | int plane; |
| 342 | for(plane=0; plane< td->fiSrc.planes; plane++){ |
| 343 | dat_1 = td->src.data[plane]; |
| 344 | dat_2 = td->destbuf.data[plane]; |
| 345 | int wsub = vsGetPlaneWidthSubS(&td->fiSrc,plane); |
| 346 | int hsub = vsGetPlaneHeightSubS(&td->fiSrc,plane); |
| 347 | int dw = CHROMA_SIZE(td->fiDest.width , wsub); |
| 348 | int dh = CHROMA_SIZE(td->fiDest.height, hsub); |
| 349 | int sw = CHROMA_SIZE(td->fiSrc.width , wsub); |
| 350 | int sh = CHROMA_SIZE(td->fiSrc.height , hsub); |
| 351 | uint8_t black = plane==0 ? 0 : 0x80; |
| 352 | |
| 353 | fp16 c_s_x = iToFp16(sw / 2); |
| 354 | fp16 c_s_y = iToFp16(sh / 2); |
| 355 | int32_t c_d_x = dw / 2; |
| 356 | int32_t c_d_y = dh / 2; |
| 357 | |
| 358 | float z = 1.0-t.zoom/100.0; |
| 359 | fp16 zcos_a = fToFp16(z*cos(-t.alpha)); // scaled cos |
| 360 | fp16 zsin_a = fToFp16(z*sin(-t.alpha)); // scaled sin |
| 361 | fp16 c_tx = c_s_x - (fToFp16(t.x) >> wsub); |
| 362 | fp16 c_ty = c_s_y - (fToFp16(t.y) >> hsub); |
| 363 | |
| 364 | /* for each pixel in the destination image we calc the source |
| 365 | * coordinate and make an interpolation: |
| 366 | * p_d = c_d + M(p_s - c_s) + t |
| 367 | * where p are the points, c the center coordinate, |
| 368 | * _s source and _d destination, |
| 369 | * t the translation, and M the rotation and scaling matrix |
| 370 | * p_s = M^{-1}(p_d - c_d - t) + c_s |
| 371 | */ |
| 372 | for (y = 0; y < dh; y++) { |
| 373 | // swapping of the loops brought 15% performace gain |
| 374 | int32_t y_d1 = (y - c_d_y); |
| 375 | for (x = 0; x < dw; x++) { |
| 376 | int32_t x_d1 = (x - c_d_x); |
| 377 | fp16 x_s = zcos_a * x_d1 + zsin_a * y_d1 + c_tx; |
| 378 | fp16 y_s = -zsin_a * x_d1 + zcos_a * y_d1 + c_ty; |
| 379 | uint8_t *dest = &dat_2[x + y * td->destbuf.linesize[plane]]; |
| 380 | // inlining the interpolation function would bring 10% |
| 381 | // (but then we cannot use the function pointer anymore...) |
| 382 | td->interpolate(dest, x_s, y_s, dat_1, |
| 383 | td->src.linesize[plane], sw, sh, |
| 384 | td->conf.crop ? black : *dest); |
| 385 | } |
| 386 | } |
| 387 | } |
| 388 | |
| 389 | return VS_OK; |
| 390 | } |
| 391 | |
| 392 | |
| 393 | |
| 394 | /* /\** TESTING */ |
| 395 | /* * transformPlanar_orc: applies current transformation to frame */ |
| 396 | /* * */ |
| 397 | /* * Parameters: */ |
| 398 | /* * td: private data structure of this filter */ |
| 399 | /* * Return value: */ |
| 400 | /* * 0 for failture, 1 for success */ |
| 401 | /* * Preconditions: */ |
| 402 | /* * The frame must be in Planar format */ |
| 403 | /* * */ |
| 404 | /* * Fixed-point format 32 bit integer: */ |
| 405 | /* * for image coords we use val<<8 */ |
| 406 | /* * for angle and zoom we use val<<16 */ |
| 407 | /* * */ |
| 408 | /* *\/ */ |
| 409 | /* int transformPlanar_orc(VSTransformData* td, VSTransform t) */ |
| 410 | /* { */ |
| 411 | /* int32_t x = 0, y = 0; */ |
| 412 | /* uint8_t *Y_1, *Y_2, *Cb_1, *Cb_2, *Cr_1, *Cr_2; */ |
| 413 | |
| 414 | /* if (t.alpha==0 && t.x==0 && t.y==0 && t.zoom == 0) return VS_OK; // noop */ |
| 415 | |
| 416 | /* Y_1 = td->src; */ |
| 417 | /* Y_2 = td->destbuf; */ |
| 418 | /* Cb_1 = td->src + td->fiSrc.width * td->fiSrc.height; */ |
| 419 | /* Cb_2 = td->destbuf + td->fiDest.width * td->fiDest.height; */ |
| 420 | /* Cr_1 = td->src + 5*td->fiSrc.width * td->fiSrc.height/4; */ |
| 421 | /* Cr_2 = td->destbuf + 5*td->fiDest.width * td->fiDest.height/4; */ |
| 422 | /* fp16 c_s_x = iToFp16(td->fiSrc.width / 2); */ |
| 423 | /* fp16 c_s_y = iToFp16(td->fiSrc.height / 2); */ |
| 424 | /* int32_t c_d_x = td->fiDest.width / 2; */ |
| 425 | /* int32_t c_d_y = td->fiDest.height / 2; */ |
| 426 | |
| 427 | /* float z = 1.0-t.zoom/100.0; */ |
| 428 | /* fp16 zcos_a = fToFp16(z*cos(-t.alpha)); // scaled cos */ |
| 429 | /* fp16 zsin_a = fToFp16(z*sin(-t.alpha)); // scaled sin */ |
| 430 | /* fp16 c_tx = c_s_x - fToFp16(t.x); */ |
| 431 | /* fp16 c_ty = c_s_y - fToFp16(t.y); */ |
| 432 | |
| 433 | /* /\* for each pixel in the destination image we calc the source */ |
| 434 | /* * coordinate and make an interpolation: */ |
| 435 | /* * p_d = c_d + M(p_s - c_s) + t */ |
| 436 | /* * where p are the points, c the center coordinate, */ |
| 437 | /* * _s source and _d destination, */ |
| 438 | /* * t the translation, and M the rotation and scaling matrix */ |
| 439 | /* * p_s = M^{-1}(p_d - c_d - t) + c_s */ |
| 440 | /* *\/ */ |
| 441 | /* /\* Luminance channel *\/ */ |
| 442 | /* fp16* x_ss = (fp16*)malloc(sizeof(fp16)*td->fiDest.width); */ |
| 443 | /* fp16* y_ss = (fp16*)malloc(sizeof(fp16)*td->fiDest.width); */ |
| 444 | /* int32_t* xs = (int32_t*)malloc(sizeof(int32_t)*td->fiDest.width); */ |
| 445 | /* for (x = 0; x < td->fiDest.width; x++) { // this can go to td */ |
| 446 | /* xs[x]=x; */ |
| 447 | /* } */ |
| 448 | |
| 449 | /* for (y = 0; y < td->fiDest.height; y++) { */ |
| 450 | /* int32_t y_d1 = (y - c_d_y); */ |
| 451 | /* fp16 sin_y = zsin_a * y_d1; */ |
| 452 | /* fp16 cos_y = zcos_a * y_d1; */ |
| 453 | /* for (x = 0; x < td->fiDest.width; x++) { */ |
| 454 | /* int32_t x_d1 = (xs[x] - c_d_x); */ |
| 455 | /* //x_ss[x] = zcos_a * x_d1 + zsin_a * y_d1 + c_tx; */ |
| 456 | /* y_ss[x] = -zsin_a * x_d1 + zcos_a * y_d1 + c_ty; */ |
| 457 | /* } */ |
| 458 | /* transform_one_line_optimized1 (x_ss, y_ss, xs, y_d1, c_d_x, */ |
| 459 | /* c_tx, c_ty, zcos_a, zsin_a, sin_y, cos_y, */ |
| 460 | /* td->fiDest.width); */ |
| 461 | /* // transform_one_line_optimized (x_ss, y_ss, xs, y_d1, c_d_x, */ |
| 462 | /* // c_tx, c_ty, zcos_a, zsin_a, td->fiDest.width); */ |
| 463 | |
| 464 | /* for (x = 0; x < td->fiDest.width; x++) { */ |
| 465 | /* uint8_t *dest = &Y_2[x + y * td->fiDest.width]; */ |
| 466 | /* td->interpolate(dest, x_ss[x], y_ss[x], Y_1, */ |
| 467 | /* td->fiSrc.width, td->fiSrc.height, */ |
| 468 | /* td->crop ? 16 : *dest); */ |
| 469 | /* } */ |
| 470 | /* } */ |
| 471 | |
| 472 | /* /\* Color channels *\/ */ |
| 473 | /* int32_t ws2 = td->fiSrc.width/2; */ |
| 474 | /* int32_t wd2 = td->fiDest.width/2; */ |
| 475 | /* int32_t hs2 = td->fiSrc.height/2; */ |
| 476 | /* int32_t hd2 = td->fiDest.height/2; */ |
| 477 | /* fp16 c_tx2 = c_tx/2; */ |
| 478 | /* fp16 c_ty2 = c_ty/2; */ |
| 479 | |
| 480 | /* for (y = 0; y < hd2; y++) { */ |
| 481 | /* int32_t y_d1 = y - (c_d_y)/2; */ |
| 482 | /* for (x = 0; x < wd2; x++) { */ |
| 483 | /* int32_t x_d1 = x - (c_d_x)/2; */ |
| 484 | /* fp16 x_s = zcos_a * x_d1 + zsin_a * y_d1 + c_tx2; */ |
| 485 | /* fp16 y_s = -zsin_a * x_d1 + zcos_a * y_d1 + c_ty2; */ |
| 486 | /* uint8_t *dest = &Cr_2[x + y * wd2]; */ |
| 487 | /* td->interpolate(dest, x_s, y_s, Cr_1, ws2, hs2, */ |
| 488 | /* td->crop ? 128 : *dest); */ |
| 489 | /* dest = &Cb_2[x + y * wd2]; */ |
| 490 | /* td->interpolate(dest, x_s, y_s, Cb_1, ws2, hs2, */ |
| 491 | /* td->crop ? 128 : *dest); */ |
| 492 | /* } */ |
| 493 | /* } */ |
| 494 | |
| 495 | /* return VS_OK; */ |
| 496 | /* } */ |
| 497 | |
| 498 | /* |
| 499 | some debugging stuff |
| 500 | FILE* f1 = fopen("transFP.pos","w"); |
| 501 | fprintf(f1,"%i,%i:\t %f,%f\n", x, y, x_s / (float)(1<<16), y_s / (float)(1<<16)); |
| 502 | fclose(f1); |
| 503 | |
| 504 | */ |
| 505 | |
| 506 | |
| 507 | |
| 508 | /* |
| 509 | * Local variables: |
| 510 | * c-file-style: "stroustrup" |
| 511 | * c-file-offsets: ((case-label . *) (statement-case-intro . *)) |
| 512 | * indent-tabs-mode: nil |
| 513 | * c-basic-offset: 2 t |
| 514 | * |
| 515 | * End: |
| 516 | * |
| 517 | * vim: expandtab shiftwidth=2: |
| 518 | */ |