reactos/dll/3rdparty/libjpeg/jccoefct.c

   1 /*
   2  * jccoefct.c
   3  *
   4  * Copyright (C) 1994-1997, Thomas G. Lane.
   5  * Modified 2003-2011 by Guido Vollbeding.
   6  * This file is part of the Independent JPEG Group's software.
   7  * For conditions of distribution and use, see the accompanying README file.
   8  *
   9  * This file contains the coefficient buffer controller for compression.
  10  * This controller is the top level of the JPEG compressor proper.
  11  * The coefficient buffer lies between forward-DCT and entropy encoding steps.
  12  */
  13
  14 #define JPEG_INTERNALS
  15 #include "jinclude.h"
  16 #include "jpeglib.h"
  17
  18
  19 /* We use a full-image coefficient buffer when doing Huffman optimization,
  20  * and also for writing multiple-scan JPEG files.  In all cases, the DCT
  21  * step is run during the first pass, and subsequent passes need only read
  22  * the buffered coefficients.
  23  */
  24 #ifdef ENTROPY_OPT_SUPPORTED
  25 #define FULL_COEF_BUFFER_SUPPORTED
  26 #else
  27 #ifdef C_MULTISCAN_FILES_SUPPORTED
  28 #define FULL_COEF_BUFFER_SUPPORTED
  29 #endif
  30 #endif
  31
  32
  33 /* Private buffer controller object */
  34
  35 typedef struct {
  36   struct jpeg_c_coef_controller pub; /* public fields */
  37
  38   JDIMENSION iMCU_row_num;      /* iMCU row # within image */
  39   JDIMENSION mcu_ctr;           /* counts MCUs processed in current row */
  40   int MCU_vert_offset;          /* counts MCU rows within iMCU row */
  41   int MCU_rows_per_iMCU_row;    /* number of such rows needed */
  42
  43   /* For single-pass compression, it's sufficient to buffer just one MCU
  44    * (although this may prove a bit slow in practice).  We allocate a
  45    * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each
  46    * MCU constructed and sent.  (On 80x86, the workspace is FAR even though
  47    * it's not really very big; this is to keep the module interfaces unchanged
  48    * when a large coefficient buffer is necessary.)
  49    * In multi-pass modes, this array points to the current MCU's blocks
  50    * within the virtual arrays.
  51    */
  52   JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU];
  53
  54   /* In multi-pass modes, we need a virtual block array for each component. */
  55   jvirt_barray_ptr whole_image[MAX_COMPONENTS];
  56 } my_coef_controller;
  57
  58 typedef my_coef_controller * my_coef_ptr;
  59
  60
  61 /* Forward declarations */
  62 METHODDEF(boolean) compress_data
  63     JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
  64 #ifdef FULL_COEF_BUFFER_SUPPORTED
  65 METHODDEF(boolean) compress_first_pass
  66     JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
  67 METHODDEF(boolean) compress_output
  68     JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf));
  69 #endif
  70
  71
  72 LOCAL(void)
  73 start_iMCU_row (j_compress_ptr cinfo)
  74 /* Reset within-iMCU-row counters for a new row */
  75 {
  76   my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
  77
  78   /* In an interleaved scan, an MCU row is the same as an iMCU row.
  79    * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
  80    * But at the bottom of the image, process only what's left.
  81    */
  82   if (cinfo->comps_in_scan > 1) {
  83     coef->MCU_rows_per_iMCU_row = 1;
  84   } else {
  85     if (coef->iMCU_row_num < (cinfo->total_iMCU_rows-1))
  86       coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
  87     else
  88       coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
  89   }
  90
  91   coef->mcu_ctr = 0;
  92   coef->MCU_vert_offset = 0;
  93 }
  94
  95
  96 /*
  97  * Initialize for a processing pass.
  98  */
  99
 100 METHODDEF(void)
 101 start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode)
 102 {
 103   my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
 104
 105   coef->iMCU_row_num = 0;
 106   start_iMCU_row(cinfo);
 107
 108   switch (pass_mode) {
 109   case JBUF_PASS_THRU:
 110     if (coef->whole_image[0] != NULL)
 111       ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 112     coef->pub.compress_data = compress_data;
 113     break;
 114 #ifdef FULL_COEF_BUFFER_SUPPORTED
 115   case JBUF_SAVE_AND_PASS:
 116     if (coef->whole_image[0] == NULL)
 117       ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 118     coef->pub.compress_data = compress_first_pass;
 119     break;
 120   case JBUF_CRANK_DEST:
 121     if (coef->whole_image[0] == NULL)
 122       ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 123     coef->pub.compress_data = compress_output;
 124     break;
 125 #endif
 126   default:
 127     ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 128     break;
 129   }
 130 }
 131
 132
 133 /*
 134  * Process some data in the single-pass case.
 135  * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
 136  * per call, ie, v_samp_factor block rows for each component in the image.
 137  * Returns TRUE if the iMCU row is completed, FALSE if suspended.
 138  *
 139  * NB: input_buf contains a plane for each component in image,
 140  * which we index according to the component's SOF position.
 141  */
 142
 143 METHODDEF(boolean)
 144 compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
 145 {
 146   my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
 147   JDIMENSION MCU_col_num;       /* index of current MCU within row */
 148   JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
 149   JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
 150   int blkn, bi, ci, yindex, yoffset, blockcnt;
 151   JDIMENSION ypos, xpos;
 152   jpeg_component_info *compptr;
 153   forward_DCT_ptr forward_DCT;
 154
 155   /* Loop to write as much as one whole iMCU row */
 156   for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
 157        yoffset++) {
 158     for (MCU_col_num = coef->mcu_ctr; MCU_col_num <= last_MCU_col;
 159          MCU_col_num++) {
 160       /* Determine where data comes from in input_buf and do the DCT thing.
 161        * Each call on forward_DCT processes a horizontal row of DCT blocks
 162        * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks
 163        * sequentially.  Dummy blocks at the right or bottom edge are filled in
 164        * specially.  The data in them does not matter for image reconstruction,
 165        * so we fill them with values that will encode to the smallest amount of
 166        * data, viz: all zeroes in the AC entries, DC entries equal to previous
 167        * block's DC value.  (Thanks to Thomas Kinsman for this idea.)
 168        */
 169       blkn = 0;
 170       for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
 171         compptr = cinfo->cur_comp_info[ci];
 172         forward_DCT = cinfo->fdct->forward_DCT[compptr->component_index];
 173         blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
 174                                                 : compptr->last_col_width;
 175         xpos = MCU_col_num * compptr->MCU_sample_width;
 176         ypos = yoffset * compptr->DCT_v_scaled_size;
 177         /* ypos == (yoffset+yindex) * DCTSIZE */
 178         for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
 179           if (coef->iMCU_row_num < last_iMCU_row ||
 180               yoffset+yindex < compptr->last_row_height) {
 181             (*forward_DCT) (cinfo, compptr,
 182                             input_buf[compptr->component_index],
 183                             coef->MCU_buffer[blkn],
 184                             ypos, xpos, (JDIMENSION) blockcnt);
 185             if (blockcnt < compptr->MCU_width) {
 186               /* Create some dummy blocks at the right edge of the image. */
 187               FMEMZERO((void FAR *) coef->MCU_buffer[blkn + blockcnt],
 188                        (compptr->MCU_width - blockcnt) * SIZEOF(JBLOCK));
 189               for (bi = blockcnt; bi < compptr->MCU_width; bi++) {
 190                 coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn+bi-1][0][0];
 191               }
 192             }
 193           } else {
 194             /* Create a row of dummy blocks at the bottom of the image. */
 195             FMEMZERO((void FAR *) coef->MCU_buffer[blkn],
 196                      compptr->MCU_width * SIZEOF(JBLOCK));
 197             for (bi = 0; bi < compptr->MCU_width; bi++) {
 198               coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn-1][0][0];
 199             }
 200           }
 201           blkn += compptr->MCU_width;
 202           ypos += compptr->DCT_v_scaled_size;
 203         }
 204       }
 205       /* Try to write the MCU.  In event of a suspension failure, we will
 206        * re-DCT the MCU on restart (a bit inefficient, could be fixed...)
 207        */
 208       if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
 209         /* Suspension forced; update state counters and exit */
 210         coef->MCU_vert_offset = yoffset;
 211         coef->mcu_ctr = MCU_col_num;
 212         return FALSE;
 213       }
 214     }
 215     /* Completed an MCU row, but perhaps not an iMCU row */
 216     coef->mcu_ctr = 0;
 217   }
 218   /* Completed the iMCU row, advance counters for next one */
 219   coef->iMCU_row_num++;
 220   start_iMCU_row(cinfo);
 221   return TRUE;
 222 }
 223
 224
 225 #ifdef FULL_COEF_BUFFER_SUPPORTED
 226
 227 /*
 228  * Process some data in the first pass of a multi-pass case.
 229  * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
 230  * per call, ie, v_samp_factor block rows for each component in the image.
 231  * This amount of data is read from the source buffer, DCT'd and quantized,
 232  * and saved into the virtual arrays.  We also generate suitable dummy blocks
 233  * as needed at the right and lower edges.  (The dummy blocks are constructed
 234  * in the virtual arrays, which have been padded appropriately.)  This makes
 235  * it possible for subsequent passes not to worry about real vs. dummy blocks.
 236  *
 237  * We must also emit the data to the entropy encoder.  This is conveniently
 238  * done by calling compress_output() after we've loaded the current strip
 239  * of the virtual arrays.
 240  *
 241  * NB: input_buf contains a plane for each component in image.  All
 242  * components are DCT'd and loaded into the virtual arrays in this pass.
 243  * However, it may be that only a subset of the components are emitted to
 244  * the entropy encoder during this first pass; be careful about looking
 245  * at the scan-dependent variables (MCU dimensions, etc).
 246  */
 247
 248 METHODDEF(boolean)
 249 compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
 250 {
 251   my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
 252   JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
 253   JDIMENSION blocks_across, MCUs_across, MCUindex;
 254   int bi, ci, h_samp_factor, block_row, block_rows, ndummy;
 255   JCOEF lastDC;
 256   jpeg_component_info *compptr;
 257   JBLOCKARRAY buffer;
 258   JBLOCKROW thisblockrow, lastblockrow;
 259   forward_DCT_ptr forward_DCT;
 260
 261   for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
 262        ci++, compptr++) {
 263     /* Align the virtual buffer for this component. */
 264     buffer = (*cinfo->mem->access_virt_barray)
 265       ((j_common_ptr) cinfo, coef->whole_image[ci],
 266        coef->iMCU_row_num * compptr->v_samp_factor,
 267        (JDIMENSION) compptr->v_samp_factor, TRUE);
 268     /* Count non-dummy DCT block rows in this iMCU row. */
 269     if (coef->iMCU_row_num < last_iMCU_row)
 270       block_rows = compptr->v_samp_factor;
 271     else {
 272       /* NB: can't use last_row_height here, since may not be set! */
 273       block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
 274       if (block_rows == 0) block_rows = compptr->v_samp_factor;
 275     }
 276     blocks_across = compptr->width_in_blocks;
 277     h_samp_factor = compptr->h_samp_factor;
 278     /* Count number of dummy blocks to be added at the right margin. */
 279     ndummy = (int) (blocks_across % h_samp_factor);
 280     if (ndummy > 0)
 281       ndummy = h_samp_factor - ndummy;
 282     forward_DCT = cinfo->fdct->forward_DCT[ci];
 283     /* Perform DCT for all non-dummy blocks in this iMCU row.  Each call
 284      * on forward_DCT processes a complete horizontal row of DCT blocks.
 285      */
 286     for (block_row = 0; block_row < block_rows; block_row++) {
 287       thisblockrow = buffer[block_row];
 288       (*forward_DCT) (cinfo, compptr, input_buf[ci], thisblockrow,
 289                       (JDIMENSION) (block_row * compptr->DCT_v_scaled_size),
 290                       (JDIMENSION) 0, blocks_across);
 291       if (ndummy > 0) {
 292         /* Create dummy blocks at the right edge of the image. */
 293         thisblockrow += blocks_across; /* => first dummy block */
 294         FMEMZERO((void FAR *) thisblockrow, ndummy * SIZEOF(JBLOCK));
 295         lastDC = thisblockrow[-1][0];
 296         for (bi = 0; bi < ndummy; bi++) {
 297           thisblockrow[bi][0] = lastDC;
 298         }
 299       }
 300     }
 301     /* If at end of image, create dummy block rows as needed.
 302      * The tricky part here is that within each MCU, we want the DC values
 303      * of the dummy blocks to match the last real block's DC value.
 304      * This squeezes a few more bytes out of the resulting file...
 305      */
 306     if (coef->iMCU_row_num == last_iMCU_row) {
 307       blocks_across += ndummy;  /* include lower right corner */
 308       MCUs_across = blocks_across / h_samp_factor;
 309       for (block_row = block_rows; block_row < compptr->v_samp_factor;
 310            block_row++) {
 311         thisblockrow = buffer[block_row];
 312         lastblockrow = buffer[block_row-1];
 313         FMEMZERO((void FAR *) thisblockrow,
 314                  (size_t) (blocks_across * SIZEOF(JBLOCK)));
 315         for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) {
 316           lastDC = lastblockrow[h_samp_factor-1][0];
 317           for (bi = 0; bi < h_samp_factor; bi++) {
 318             thisblockrow[bi][0] = lastDC;
 319           }
 320           thisblockrow += h_samp_factor; /* advance to next MCU in row */
 321           lastblockrow += h_samp_factor;
 322         }
 323       }
 324     }
 325   }
 326   /* NB: compress_output will increment iMCU_row_num if successful.
 327    * A suspension return will result in redoing all the work above next time.
 328    */
 329
 330   /* Emit data to the entropy encoder, sharing code with subsequent passes */
 331   return compress_output(cinfo, input_buf);
 332 }
 333
 334
 335 /*
 336  * Process some data in subsequent passes of a multi-pass case.
 337  * We process the equivalent of one fully interleaved MCU row ("iMCU" row)
 338  * per call, ie, v_samp_factor block rows for each component in the scan.
 339  * The data is obtained from the virtual arrays and fed to the entropy coder.
 340  * Returns TRUE if the iMCU row is completed, FALSE if suspended.
 341  *
 342  * NB: input_buf is ignored; it is likely to be a NULL pointer.
 343  */
 344
 345 METHODDEF(boolean)
 346 compress_output (j_compress_ptr cinfo, JSAMPIMAGE input_buf)
 347 {
 348   my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
 349   JDIMENSION MCU_col_num;       /* index of current MCU within row */
 350   int blkn, ci, xindex, yindex, yoffset;
 351   JDIMENSION start_col;
 352   JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
 353   JBLOCKROW buffer_ptr;
 354   jpeg_component_info *compptr;
 355
 356   /* Align the virtual buffers for the components used in this scan.
 357    * NB: during first pass, this is safe only because the buffers will
 358    * already be aligned properly, so jmemmgr.c won't need to do any I/O.
 359    */
 360   for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
 361     compptr = cinfo->cur_comp_info[ci];
 362     buffer[ci] = (*cinfo->mem->access_virt_barray)
 363       ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
 364        coef->iMCU_row_num * compptr->v_samp_factor,
 365        (JDIMENSION) compptr->v_samp_factor, FALSE);
 366   }
 367
 368   /* Loop to process one whole iMCU row */
 369   for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
 370        yoffset++) {
 371     for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row;
 372          MCU_col_num++) {
 373       /* Construct list of pointers to DCT blocks belonging to this MCU */
 374       blkn = 0;                 /* index of current DCT block within MCU */
 375       for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
 376         compptr = cinfo->cur_comp_info[ci];
 377         start_col = MCU_col_num * compptr->MCU_width;
 378         for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
 379           buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
 380           for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
 381             coef->MCU_buffer[blkn++] = buffer_ptr++;
 382           }
 383         }
 384       }
 385       /* Try to write the MCU. */
 386       if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) {
 387         /* Suspension forced; update state counters and exit */
 388         coef->MCU_vert_offset = yoffset;
 389         coef->mcu_ctr = MCU_col_num;
 390         return FALSE;
 391       }
 392     }
 393     /* Completed an MCU row, but perhaps not an iMCU row */
 394     coef->mcu_ctr = 0;
 395   }
 396   /* Completed the iMCU row, advance counters for next one */
 397   coef->iMCU_row_num++;
 398   start_iMCU_row(cinfo);
 399   return TRUE;
 400 }
 401
 402 #endif /* FULL_COEF_BUFFER_SUPPORTED */
 403
 404
 405 /*
 406  * Initialize coefficient buffer controller.
 407  */
 408
 409 GLOBAL(void)
 410 jinit_c_coef_controller (j_compress_ptr cinfo, boolean need_full_buffer)
 411 {
 412   my_coef_ptr coef;
 413
 414   coef = (my_coef_ptr)
 415     (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 416                                 SIZEOF(my_coef_controller));
 417   cinfo->coef = (struct jpeg_c_coef_controller *) coef;
 418   coef->pub.start_pass = start_pass_coef;
 419
 420   /* Create the coefficient buffer. */
 421   if (need_full_buffer) {
 422 #ifdef FULL_COEF_BUFFER_SUPPORTED
 423     /* Allocate a full-image virtual array for each component, */
 424     /* padded to a multiple of samp_factor DCT blocks in each direction. */
 425     int ci;
 426     jpeg_component_info *compptr;
 427
 428     for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
 429          ci++, compptr++) {
 430       coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
 431         ((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
 432          (JDIMENSION) jround_up((long) compptr->width_in_blocks,
 433                                 (long) compptr->h_samp_factor),
 434          (JDIMENSION) jround_up((long) compptr->height_in_blocks,
 435                                 (long) compptr->v_samp_factor),
 436          (JDIMENSION) compptr->v_samp_factor);
 437     }
 438 #else
 439     ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
 440 #endif
 441   } else {
 442     /* We only need a single-MCU buffer. */
 443     JBLOCKROW buffer;
 444     int i;
 445
 446     buffer = (JBLOCKROW)
 447       (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
 448                                   C_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
 449     for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) {
 450       coef->MCU_buffer[i] = buffer + i;
 451     }
 452     coef->whole_image[0] = NULL; /* flag for no virtual arrays */
 453   }
 454 }