pix_convolve.cpp 23.7 KB
Newer Older
1 2 3 4 5 6 7 8 9
////////////////////////////////////////////////////////
//
// GEM - Graphics Environment for Multimedia
//
// zmoelnig@iem.kug.ac.at
//
// Implementation file
//
//    Copyright (c) 1997-1998 Mark Danks.
zmoelnig's avatar
zmoelnig committed
10
//    Copyright (c) Günther Geiger.
zmoelnig's avatar
zmoelnig committed
11
//    Copyright (c) 2001-2011 IOhannes m zmölnig. forum::für::umläute. IEM. zmoelnig@iem.at
12
//    Copyright (c) 2002 James Tittle & Chris Clepper
13 14 15 16 17 18
//    For information on usage and redistribution, and for a DISCLAIMER OF ALL
//    WARRANTIES, see the file, "GEM.LICENSE.TERMS" in this distribution.
//
/////////////////////////////////////////////////////////

#include "pix_convolve.h"
19
#include "Gem/Exception.h"
20
#include "Utils/Functions.h"
21

22 23
CPPEXTERN_NEW_WITH_TWO_ARGS(pix_convolve, t_floatarg, A_DEFFLOAT,
                            t_floatarg, A_DEFFLOAT);
24 25 26 27 28 29 30 31 32

/////////////////////////////////////////////////////////
//
// pix_convolve
//
/////////////////////////////////////////////////////////
// Constructor
//
/////////////////////////////////////////////////////////
33 34 35 36 37
pix_convolve :: pix_convolve(t_floatarg fRow, t_floatarg fCol) :
  m_imatrix(NULL),
  m_irange(255),
  m_rows(0), m_cols(0),
  m_chroma(0)
38
{
zmoelnig's avatar
zmoelnig committed
39 40
  int row = static_cast<int>(fRow);
  int col = static_cast<int>(fCol);
41

42 43 44
  if (!row || !col ) {
    throw(GemException("matrix must have some dimension"));
  }
45

46 47 48
  if (!(row % 2) || !(col % 2) ) {
    throw(GemException("matrix must have odd dimensions"));
  }
49

50 51 52
  m_rows = row;
  m_cols = col;
  m_imatrix = new signed short[m_rows * m_cols];
53

54 55 56 57 58 59 60
  // zero out the matrix
  int i;
  for (i = 0; i < m_cols * m_rows; i++) {
    m_imatrix[i] = 0;
  }
  // insert a one for the default center value (identity matrix)
  m_imatrix[ ((m_cols / 2 + 1) * m_rows) + (m_rows / 2 + 1) ] = 255;
61

62 63 64 65
  inlet_new(this->x_obj, &this->x_obj->ob_pd, gensym("float"),
            gensym("ft1"));
  inlet_new(this->x_obj, &this->x_obj->ob_pd, gensym("list"),
            gensym("matrix"));
66 67 68 69 70 71 72 73
}

/////////////////////////////////////////////////////////
// Destructor
//
/////////////////////////////////////////////////////////
pix_convolve :: ~pix_convolve()
{
74 75 76
  if (m_imatrix) {
    delete [] m_imatrix;
  }
77 78 79 80 81 82
}

/////////////////////////////////////////////////////////
// processImage
//
/////////////////////////////////////////////////////////
83 84
void pix_convolve :: calculateRGBA3x3(imageStruct &image,
                                      imageStruct &tempImg)
85 86 87
{
  int i;
  int j;
88
//  int k;
89 90 91
  int xsize =  tempImg.xsize;
  int ysize =  tempImg.ysize;
  int size = xsize*ysize - xsize-1;
92
  int csize = tempImg.csize;
93 94 95

  int* src = (int*) tempImg.data;
  int* dest = (int*)image.data;
96 97


cclepper's avatar
cclepper committed
98
//unroll this to do R G B in one pass?? (too many registers?)
99 100 101 102 103 104 105 106 107 108 109
  i = xsize;
  int* val1 = 0;
  int* val2 = src+i-xsize;
  int* val3 = src+i-xsize+1;
  int* val4 = src+i-1;
  int* val5 = src+i;
  int* val6 = src+i+1;
  int* val7 = src+i+xsize-1;
  int* val8 = src+i+xsize;
  int* val9 = src+i+xsize+1;
  int res;
110
  for (i=xsize+1; i<size; i++) {
111 112 113 114 115 116 117 118 119
    val1 = val2;
    val2 = val3;
    val3 = src+i-xsize+1;
    val4 = val5;
    val5 = val6;
    val6 = src+i+1;
    val7 = val8;
    val8 = val9;
    val9 = src+i+xsize+1;
120 121 122 123 124 125 126 127
    if (i%xsize == 0 || i%xsize == xsize-1) {
      continue;
    }
#ifndef __APPLE__
    for (j=0; j<3; j++)
#else
    for (j=1; j<4; j++)
#endif
cclepper's avatar
cclepper committed
128
    {
zmoelnig's avatar
zmoelnig committed
129
      //      res =  m_imatrix[0]*(int)((unsigned char*)val1)[j];
130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147
      res =  m_imatrix[0]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val1)[j]);
      res += m_imatrix[1]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val2)[j]);
      res += m_imatrix[2]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val3)[j]);
      res += m_imatrix[3]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val4)[j]);
      res += m_imatrix[4]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val5)[j]);
      res += m_imatrix[5]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val6)[j]);
      res += m_imatrix[6]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val7)[j]);
      res += m_imatrix[7]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val8)[j]);
      res += m_imatrix[8]*static_cast<int>(reinterpret_cast<unsigned char*>
                                           (val9)[j]);
148 149
      res*=m_irange;
      res>>=16;
150
      ((unsigned char*)dest)[i*csize+j] = CLAMP(res);
151 152 153 154
    }

  }

155

156 157
}

zmoelnig's avatar
zmoelnig committed
158
void pix_convolve :: processRGBAImage(imageStruct &image)
159
{
160 161 162 163 164 165 166 167 168 169 170 171 172
  image.copy2Image(&tempImg);
  int initX = m_rows / 2;
  int initY = m_cols / 2;
  int maxX = tempImg.xsize - initX;
  int maxY = tempImg.ysize - initY;
  int xTimesc = tempImg.xsize * tempImg.csize;
  int initOffset = initY * xTimesc + initX * tempImg.csize;
  const int csize = tempImg.csize;

  if (m_rows == 3 && m_cols == 3) {
    calculateRGBA3x3(image,tempImg);
    return;
  }
173

174 175 176
  for (int y = initY; y < maxY; y++) {
    int realY = y * xTimesc;
    int offsetY = realY - initOffset;
zmoelnig's avatar
zmoelnig committed
177

178 179 180
    for (int x = initX; x < maxX; x++) {
      int realPos = x * csize + realY;
      int offsetXY = x * csize + offsetY;
181

182 183
      // skip the alpha value
      for (int c = 1; c < csize; c++) {
184
        int new_val = 0;
185 186
        int offsetXYC = offsetXY + c;
        for (int matY = 0; matY < m_cols; matY++) {
187 188
          int offsetXYCMat = matY * xTimesc + offsetXYC;
          int realMatY = matY * m_rows;
189 190
          for (int matX = 0; matX < m_rows; matX++) {
            new_val += (tempImg.data[offsetXYCMat + matX * csize] *
191 192 193
                        m_imatrix[realMatY + matX])>>8;
          }
        }
194
        image.data[realPos + c] = CLAMP(new_val);
195 196
        //removes insult from injury ??
        // we do not use the m_irange anymore ...  remove it ??
197

zmoelnig's avatar
zmoelnig committed
198 199
      }
    }
200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236
  }

}

void pix_convolve :: processGrayImage(imageStruct &image)
{
  const int csize=image.csize;
  image.copy2Image(&tempImg);
  int initX = m_rows / 2;
  int initY = m_cols / 2;
  int maxX = tempImg.xsize - initX;
  int maxY = tempImg.ysize - initY;
  int xTimesc = tempImg.xsize * csize;
  int initOffset = initY * xTimesc + initX * csize;

  for (int y = initY; y < maxY; y++)    {
    int realY = y * xTimesc;
    int offsetY = realY - initOffset;

    for (int x = initX; x < maxX; x++)        {
      int offsetXY = x + offsetY;

      int new_val = 0;
      int offsetXYC = offsetXY;
      for (int matY = 0; matY < m_cols; matY++)   {
        int offsetXYCMat = matY * xTimesc + offsetXYC;
        int realMatY = matY * m_rows;
        for (int matX = 0; matX < m_rows; matX++)     {
          new_val += (tempImg.data[offsetXYCMat + matX] *
                      m_imatrix[realMatY + matX])>>8;
        }
      }
      image.data[x+realY] = CLAMP(new_val);
      //removes insult from injury ??
      // we do not use the m_irange anymore ...  remove it ??
    }
  }
zmoelnig's avatar
zmoelnig committed
237
}
238

239 240
void pix_convolve :: processYUVImage(imageStruct &image)
{
zmoelnig's avatar
zmoelnig committed
241 242
  image.copy2Image(&tempImg);
  //float range = 1;
243 244 245 246 247 248
  int initX = m_rows / 2;
  int initY = m_cols / 2;
  int maxX = tempImg.xsize - initX;
  int maxY = tempImg.ysize - initY;
  int xTimesc = tempImg.xsize * tempImg.csize;
  int initOffset = initY * xTimesc + initX * tempImg.csize;
249

250
//   calculate3x3YUV(image,tempImg);
251

252
//quick fix for Intel 3x3YUV problems
253
#ifdef __BIG_ENDIAN__
254 255 256 257
  if (m_rows == 3 && m_cols == 3) {
    calculate3x3YUV(image,tempImg);
    return;
  }
258
#endif
259
  if (m_chroma) {
zmoelnig's avatar
zmoelnig committed
260
    for (int y = initY; y < maxY; y++)   {
261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277
      int realY = y * xTimesc;
      int offsetY = realY - initOffset;

      for (int x = initX; x < maxX; x++) {
        int realPos = x * tempImg.csize + realY;
        int offsetXY = x * tempImg.csize + offsetY;

        // skip the UV
        for (int c = 1; c < 3; c+=2) {
          int new_val = 0;
          int offsetXYC = offsetXY + c;
          for (int matY = 0; matY < m_cols; matY++) {
            int offsetXYCMat = matY * xTimesc + offsetXYC;
            int realMatY = matY * m_rows;
            for (int matX = 0; matX < m_rows; matX++) {
              new_val += (tempImg.data[offsetXYCMat + matX * tempImg.csize] *
                          m_imatrix[realMatY + matX])>>8;
278
            }
279 280 281
          }
          image.data[realPos + c] = CLAMP(new_val);
          // image.data[realPos + c-1] = 128;  //remove the U+V
282
        }
283
      }
cclepper's avatar
cclepper committed
284
    }
285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
  } else {
    for (int y = initY; y < maxY; y++) {
      int realY = y * xTimesc;
      int offsetY = realY - initOffset;

      for (int x = initX; x < maxX; x++) {
        int realPos = x * tempImg.csize + realY;
        int offsetXY = x * tempImg.csize + offsetY;

        // skip the UV
        for (int c = 1; c < 3; c+=2) {
          int new_val = 0;
          int offsetXYC = offsetXY + c;
          for (int matY = 0; matY < m_cols; matY++) {
            int offsetXYCMat = matY * xTimesc + offsetXYC;
            int realMatY = matY * m_rows;
            for (int matX = 0; matX < m_rows; matX++) {
              new_val += (tempImg.data[offsetXYCMat + matX * tempImg.csize] *
                          m_imatrix[realMatY + matX])>>8;
304
            }
305 306 307
          }
          image.data[realPos + c] = CLAMP(new_val);
          image.data[realPos + c-1] = 128;  //remove the U+V
308
        }
309
      }
cclepper's avatar
cclepper committed
310
    }
311
  }
312

313
}
cclepper's avatar
cclepper committed
314

315
//make two functions - one for chroma one without
316 317
void pix_convolve :: calculate3x3YUV(imageStruct &image,
                                     imageStruct &tempImg)
cclepper's avatar
cclepper committed
318
{
319

cclepper's avatar
cclepper committed
320
#ifdef __VEC__
321 322
  calculate3x3YUVAltivec(image,tempImg);
  return;
323 324
#else

cclepper's avatar
cclepper committed
325 326 327
  int i;
  int j;
  int k;
328 329
  int xsize =  tempImg.xsize -1;
  int ysize =  tempImg.ysize -1;
cclepper's avatar
cclepper committed
330
  int size = xsize*ysize - xsize-1;
331
  int length;
cclepper's avatar
cclepper committed
332 333 334

  short* src = (short*) tempImg.data;
  short* dest = (short*)image.data;
335 336 337
  register int mat1,mat2,mat3,mat4,mat5,mat6,mat7,mat8,mat9;
  register int res1,res2,res3,res4,res5,res6,res7,res8,res9;
  register int range;
338

339 340 341 342 343 344 345 346
  mat1 = m_imatrix[0];
  mat2 = m_imatrix[1];
  mat3 = m_imatrix[2];
  mat4 = m_imatrix[3];
  mat5 = m_imatrix[4];
  mat6 = m_imatrix[5];
  mat7 = m_imatrix[6];
  mat8 = m_imatrix[7];
347
  mat9 = m_imatrix[8];
348
  range =m_irange;
349

350 351
  if (m_chroma) {
    i = xsize;
352

353
#ifdef i386
354 355 356 357 358 359 360 361 362
    register unsigned char val1 = 0;
    register unsigned char val2 = src[i-xsize+1];
    register unsigned char val3 = src[i-xsize+3];
    register unsigned char val4 = src[i-1];
    register unsigned char val5 = src[i+1];
    register unsigned char val6 = src[i+3];
    register unsigned char val7 = src[i+xsize-1];
    register unsigned char val8 = src[i+xsize+1];
    register unsigned char val9 = src[i+xsize+3];
363
#else
364 365 366 367 368 369 370 371 372
    register unsigned char val1 = 0;
    register unsigned char val2 = src[i-xsize+1];
    register unsigned char val3 = src[i-xsize+3];
    register unsigned char val4 = src[i-1];
    register unsigned char val5 = src[i+1];
    register unsigned char val6 = src[i+3];
    register unsigned char val7 = src[i+xsize-1];
    register unsigned char val8 = src[i+xsize+1];
    register unsigned char val9 = src[i+xsize+3];
373
#endif
374

375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413
    //unroll this 2x to fill the registers? (matrix*y1*y2= 9*9*9 =27)
    //messed up looking on x86
    i=xsize+2;

    for (k=1; k<ysize; k++) {
      for (j=1; j<xsize; j++) {
        //load furthest value first...the rest should be in cache

        val7 = val8;
        val8 = val9;
        val9 = src[i+xsize+3]; //this will come from main mem
        val1 = val2;
        val2 = val3;
        val3 = src[i-xsize+3]; //should be in cache from previous pass
        val4 = val5;
        val5 = val6;
        val6 = src[i+3];

        //unroll??
        res1 = mat1*static_cast<int>(val1);
        res2 = mat2*static_cast<int>(val2);
        res3 = mat3*static_cast<int>(val3);
        res4 = mat4*static_cast<int>(val4);
        res5 = mat5*static_cast<int>(val5);
        res6 = mat6*static_cast<int>(val6);
        res7 = mat7*static_cast<int>(val7);
        res8 = mat8*static_cast<int>(val8);
        res9 = mat9*static_cast<int>(val9);


        res1 += res2 + res3;
        res4 += res5 + res6;
        res7 += res8 + res9;
        res1 += res4 + res7;

        res1*=range;
        res1>>=16;
        ((unsigned char*)dest)[i*2+1] = CLAMP(res1);
        i++;
414

cclepper's avatar
cclepper committed
415
      }
416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456
      i=k*tempImg.xsize;
    }
  } else {

    i = xsize;
    //make these temp register vars rather than pointers?

    short* val1 = 0;
    short* val2 = src+i-xsize; //val2 = src[i-xsize];
    short* val3 = src+i-xsize+1; //val3 = src[i-xsize+1];
    short* val4 = src+i-1; //val4 = src[i-1];
    short* val5 = src+i; //val5 = src[i];
    short* val6 = src+i+1; //val6 = src[i+1];
    short* val7 = src+i+xsize-1; //val7 = src[i+xsize-1];
    short* val8 = src+i+xsize; //val8 = src[i+xsize];
    short* val9 = src+i+xsize+1; //val9 = src[i+xsize+1];
    /*
    register short* val1 = 0;
    register short* val2 = src+i-xsize; //val2 = src[i-xsize];
    register short* val3 = src+i-xsize+1; //val3 = src[i-xsize+1];
    register short* val4 = src+i-1; //val4 = src[i-1];
    register short* val5 = src+i; //val5 = src[i];
    register short* val6 = src+i+1; //val6 = src[i+1];
    register short* val7 = src+i+xsize-1; //val7 = src[i+xsize-1];
    register short* val8 = src+i+xsize; //val8 = src[i+xsize];
    register short* val9 = src+i+xsize+1; //val9 = src[i+xsize+1];*/
    //int res;
// for (i=xsize+1;i<size;i++) {
    for (k=1; k<ysize; k++) {
      for (j=1; j<xsize; j++) {
        val1 = val2;
        val2 = val3;
        val3 = src+i-xsize+1;
        val4 = val5;
        val5 = val6;
        val6 = src+i+1;
        val7 = val8;
        val8 = val9;
        val9 = src+i+xsize+1;

        /* if (i%xsize == 0 || i%xsize == xsize-1) continue;
457
#ifndef __APPLE__
458
         for (j=0;j<3;j++)
459
#else
460
         for (j=1;j<3;j+=2)
461
#endif
462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488
         { */

        res1 = mat1*static_cast<int>(reinterpret_cast<unsigned char*>(val1)[j]);
        res2 = mat2*static_cast<int>(reinterpret_cast<unsigned char*>(val2)[j]);
        res3 = mat3*static_cast<int>(reinterpret_cast<unsigned char*>(val3)[j]);
        res4 = mat4*static_cast<int>(reinterpret_cast<unsigned char*>(val4)[j]);
        res5 = mat5*static_cast<int>(reinterpret_cast<unsigned char*>(val5)[j]);
        res6 = mat6*static_cast<int>(reinterpret_cast<unsigned char*>(val6)[j]);
        res7 = mat7*static_cast<int>(reinterpret_cast<unsigned char*>(val7)[j]);
        res8 = mat8*static_cast<int>(reinterpret_cast<unsigned char*>(val8)[j]);
        res9 = mat9*static_cast<int>(reinterpret_cast<unsigned char*>(val9)[j]);
        res1 += res2 + res3;
        res4 += res5 + res6;
        res7 += res8 + res9;
        res1 += res4 + res7;
        res1*=range;
        res1>>=16;
        // ((unsigned char*)dest)[i*2] = 128;
        // ((unsigned char*)dest)[i*2+2] = 128;
        ((unsigned char*)dest)[i*2+1] = CLAMP(res1);
        // }
        ((unsigned char*)dest)[i*2] = 128;
        // ((unsigned char*)dest)[i*2+2] = 128;
        i++;
      }
      i=k*tempImg.xsize;
    }
cclepper's avatar
cclepper committed
489
  }
490
#endif
cclepper's avatar
cclepper committed
491 492
}

cclepper's avatar
cclepper committed
493
//too many temps for all the registers - reuse some
494 495
void pix_convolve :: calculate3x3YUVAltivec(imageStruct &image,
    imageStruct &tempImg)
496
{
497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
#ifdef __VEC__
  int h,w,width,i;
  int xsize =  (tempImg.xsize)*2;

  width = (tempImg.xsize)/8;
  //format is U Y V Y

  union {
    short   elements[8];
    vector  signed short v;
  } shortBuffer;

  union {
    unsigned int    elements[4];
    vector  unsigned int v;
  } intBuffer;

  vector unsigned char one;
  vector signed short mat1,mat2,mat3,mat4,mat5,mat6,mat7,mat8,mat9;
  vector unsigned char  val1,val2,val3,val4,val5,val6,val7,val8,val9;
  register vector signed int  res1,res2,res3,res4,res5,res6,res7,res8,res9;
  vector signed int  yhi,ylo;
519 520
  register vector signed int  res1a,res2a,res3a,res4a,res5a,res6a,res7a,
           res8a,res9a;
521
  vector unsigned int bitshift;
522 523
  register vector signed short y1,y2,y3,y4,y5,y6,y7,y8,y9,yres,uvres,hiImage,
           loImage;
524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584
  vector signed short range,uvnone,uv128;
  unsigned char *dst =  (unsigned char*) image.data;
  unsigned char *src =  (unsigned char*) tempImg.data;


  one =  vec_splat_u8( 1 );

  intBuffer.elements[0] = 8;
  //Load it into the vector unit
  bitshift = intBuffer.v;
  bitshift = (vector unsigned int)vec_splat((vector unsigned int)bitshift,0);

  shortBuffer.elements[0] = m_irange;
  range = shortBuffer.v;
  range = (vector signed short)vec_splat((vector signed short)range, 0);

  shortBuffer.elements[0] = 128;
  uvnone = shortBuffer.v;
  uvnone = (vector signed short)vec_splat((vector signed short)uvnone, 0);

  //load the matrix values into vectors
  shortBuffer.elements[0] = m_imatrix[0];
  mat1 = shortBuffer.v;
  mat1 = (vector signed short)vec_splat((vector signed short)mat1,0);

  shortBuffer.elements[0] = m_imatrix[1];
  mat2 = shortBuffer.v;
  mat2 = (vector signed short)vec_splat((vector signed short)mat2,0);

  shortBuffer.elements[0] = m_imatrix[2];
  mat3 = shortBuffer.v;
  mat3 = (vector signed short)vec_splat((vector signed short)mat3,0);

  shortBuffer.elements[0] = m_imatrix[3];
  mat4 = shortBuffer.v;
  mat4 = (vector signed short)vec_splat((vector signed short)mat4,0);

  shortBuffer.elements[0] = m_imatrix[4];
  mat5 = shortBuffer.v;
  mat5 = (vector signed short)vec_splat((vector signed short)mat5,0);

  shortBuffer.elements[0] = m_imatrix[5];
  mat6 = shortBuffer.v;
  mat6 = (vector signed short)vec_splat((vector signed short)mat6,0);

  shortBuffer.elements[0] = m_imatrix[6];
  mat7 = shortBuffer.v;
  mat7 = (vector signed short)vec_splat((vector signed short)mat7,0);

  shortBuffer.elements[0] = m_imatrix[7];
  mat8 = shortBuffer.v;
  mat8 = (vector signed short)vec_splat((vector signed short)mat8,0);

  shortBuffer.elements[0] = m_imatrix[8];
  mat9 = shortBuffer.v;
  mat9 = (vector signed short)vec_splat((vector signed short)mat9,0);

  shortBuffer.elements[0] = 128;
  uv128 = shortBuffer.v;
  uv128 = (vector signed short)vec_splat((vector signed short)uv128,0);
#ifndef PPC970
585 586
  UInt32                      prefetchSize = GetPrefetchConstant( 16, 1,
      256 );
587 588 589
  vec_dst( src, prefetchSize, 0 );
  vec_dst( dst, prefetchSize, 0 );
#endif
590

591
  i = xsize*2;
cclepper's avatar
cclepper committed
592 593 594

//need to treat the first rows as a special case for accuracy and keep it from crashing
//or just skip the first 2 rows ;)
595

596 597 598 599 600 601 602 603
  for ( h=2; h<image.ysize-1; h++) {
    // i+=2; //this gets rid of the echoes but kills the vertical edge-detects???
    i+=8;
    for (w=0; w<width-1; w++) {
#ifndef PPC970
      vec_dst( src, prefetchSize, 0 );
      vec_dst( dst, prefetchSize, 1 );
#endif
604

605 606 607 608 609 610 611 612 613 614 615
      val1 = vec_ld(0,src+(i-xsize-2));//this might crash?
      val2 = vec_ld(0,src+(i-xsize));
      val3 = vec_ld(0,src+(i-xsize+2));
      val4 = vec_ld(0,src+(i-2));
      val5 = vec_ld(0,src+i);
      val6 = vec_ld(0,src+(i+2));
      val7 = vec_ld(0,src+(i+xsize-2));
      val8 = vec_ld(0,src+(i+xsize));
      val9 = vec_ld(0,src+(i+xsize+2));

      //extract the Y for processing
616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636
      y1 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val1);
      y2 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val2);
      y3 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val3);
      y4 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val4);
      y5 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val5);
      y6 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val6);
      y7 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val7);
      y8 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val8);
      y9 = (vector signed short)vec_mulo((vector unsigned char)one,
                                         (vector unsigned char)val9);

      uvres = (vector signed short)vec_mule((vector unsigned char)one,
                                            (vector unsigned char)val5);
637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706

      //mult the Y by the matrix coefficient
      res1 = vec_mulo(mat1,y1);
      res2 = vec_mulo(mat2,y2);
      res3 = vec_mulo(mat3,y3);
      res4 = vec_mulo(mat4,y4);
      res5 = vec_mulo(mat5,y5);
      res6 = vec_mulo(mat6,y6);
      res7 = vec_mulo(mat7,y7);
      res8 = vec_mulo(mat8,y8);
      res9 = vec_mulo(mat9,y9);

      res1a = vec_mule(mat1,y1);
      res2a = vec_mule(mat2,y2);
      res3a = vec_mule(mat3,y3);
      res4a = vec_mule(mat4,y4);
      res5a = vec_mule(mat5,y5);
      res6a = vec_mule(mat6,y6);
      res7a = vec_mule(mat7,y7);
      res8a = vec_mule(mat8,y8);
      res9a = vec_mule(mat9,y9);

      //sum the results - these are only 1 cycle ops so no dependency issues
      res1 = vec_adds(res1,res2); //1+2
      res3 = vec_adds(res3,res4);//3+4
      res5 = vec_adds(res5,res6);//5+6
      res7 = vec_adds(res7,res8);//7+8
      res1 = vec_adds(res1,res3);//(1+2)+(3+4)
      res7 = vec_adds(res7,res9);//7+8+9
      res1 = vec_adds(res1,res5);//(1+2)+(3+4)+(5+6)
      res1 = vec_adds(res1,res7);//(1+2)+(3+4)+(5+6)+(7+8+9)

      res1a = vec_adds(res1a,res2a); //1+2
      res3a = vec_adds(res3a,res4a);//3+4
      res5a = vec_adds(res5a,res6a);//5+6
      res7a = vec_adds(res7a,res8a);//7+8
      res1a = vec_adds(res1a,res3a);//(1+2)+(3+4)
      res7a = vec_adds(res7a,res9a);//7+8+9
      res1a = vec_adds(res1a,res5a);//(1+2)+(3+4)+(5+6)
      res1a = vec_adds(res1a,res7a);//(1+2)+(3+4)+(5+6)+(7+8+9)


      //do the bitshift on the results here??
      res1 = vec_sra(res1,bitshift);
      res1a = vec_sra(res1a,bitshift);

      //pack back to one short vector??
      yhi = vec_mergeh(res1a,res1);
      ylo = vec_mergel(res1a,res1);


      yres = vec_packs(yhi,ylo);


      //combine with the UV
      //vec_mergel + vec_mergeh Y and UV
      hiImage =  vec_mergeh(uvres,yres);
      loImage =  vec_mergel(uvres,yres);

      val1 = vec_packsu(hiImage,loImage);
      vec_st(val1,0,dst+i);
      i+=16;

    }
    i = h * xsize;
#ifndef PPC970
    vec_dss( 0 );
    vec_dss( 1 );
#endif
  }  /*end of working altivec function */
707

cclepper's avatar
cclepper committed
708

709 710
#endif
}
cclepper's avatar
cclepper committed
711

712 713 714 715 716 717
/////////////////////////////////////////////////////////
// rangeMess
//
/////////////////////////////////////////////////////////
void pix_convolve :: rangeMess(float range)
{
718 719
  m_irange = (int)(range*255.f);
  setPixModified();
720 721 722 723 724 725 726 727
}

/////////////////////////////////////////////////////////
// matrixMess
//
/////////////////////////////////////////////////////////
void pix_convolve :: matrixMess(int argc, t_atom *argv)
{
728 729 730 731
  if (argc != m_cols * m_rows) {
    error("matrix size not correct");
    return;
  }
cclepper's avatar
cclepper committed
732

733 734 735 736
  int i;
  for (i = 0; i < argc; i++) {
    m_imatrix[i] = (int)(atom_getfloat(&argv[i])*255.);
  }
737

cclepper's avatar
cclepper committed
738

739
  setPixModified();
740 741 742 743 744 745 746 747
}

/////////////////////////////////////////////////////////
// static member function
//
/////////////////////////////////////////////////////////
void pix_convolve :: obj_setupCallback(t_class *classPtr)
{
748 749
  class_addmethod(classPtr,
                  reinterpret_cast<t_method>(&pix_convolve::matrixMessCallback),
750
                  gensym("matrix"), A_GIMME, A_NULL);
751 752
  class_addmethod(classPtr,
                  reinterpret_cast<t_method>(&pix_convolve::rangeMessCallback),
753
                  gensym("ft1"), A_FLOAT, A_NULL);
754 755
  class_addmethod(classPtr,
                  reinterpret_cast<t_method>(&pix_convolve::chromaMessCallback),
756
                  gensym("chroma"), A_FLOAT, A_NULL);
757
}
758 759
void pix_convolve :: matrixMessCallback(void *data, t_symbol *, int argc,
                                        t_atom *argv)
760
{
761
  GetMyClass(data)->matrixMess(argc, argv);
762
}
763
void pix_convolve :: rangeMessCallback(void *data, t_float range)
764
{
765
  GetMyClass(data)->rangeMess(range);
766
}
cclepper's avatar
cclepper committed
767

768
void pix_convolve :: chromaMessCallback(void *data, t_float value)
cclepper's avatar
cclepper committed
769
{
770
  GetMyClass(data)->m_chroma=static_cast<int>(value);
cclepper's avatar
cclepper committed
771
}