Thomas G. Lane | 36a4ccc | 1994-09-24 00:00:00 +0000 | [diff] [blame^] | 1 | /* |
| 2 | * jccoefct.c |
| 3 | * |
| 4 | * Copyright (C) 1994, Thomas G. Lane. |
| 5 | * This file is part of the Independent JPEG Group's software. |
| 6 | * For conditions of distribution and use, see the accompanying README file. |
| 7 | * |
| 8 | * This file contains the coefficient buffer controller for compression. |
| 9 | * This controller is the top level of the JPEG compressor proper. |
| 10 | * The coefficient buffer lies between forward-DCT and entropy encoding steps. |
| 11 | */ |
| 12 | |
| 13 | #define JPEG_INTERNALS |
| 14 | #include "jinclude.h" |
| 15 | #include "jpeglib.h" |
| 16 | |
| 17 | |
| 18 | /* We use a full-image coefficient buffer when doing Huffman optimization, |
| 19 | * and also for writing multiple-scan JPEG files. In all cases, the DCT |
| 20 | * step is run during the first pass, and subsequent passes need only read |
| 21 | * the buffered coefficients. |
| 22 | */ |
| 23 | #ifdef ENTROPY_OPT_SUPPORTED |
| 24 | #define FULL_COEF_BUFFER_SUPPORTED |
| 25 | #else |
| 26 | #ifdef C_MULTISCAN_FILES_SUPPORTED |
| 27 | #define FULL_COEF_BUFFER_SUPPORTED |
| 28 | #endif |
| 29 | #endif |
| 30 | |
| 31 | |
| 32 | /* Private buffer controller object */ |
| 33 | |
| 34 | typedef struct { |
| 35 | struct jpeg_c_coef_controller pub; /* public fields */ |
| 36 | |
| 37 | JDIMENSION MCU_row_num; /* keep track of MCU row # within image */ |
| 38 | |
| 39 | /* For single-pass compression, it's sufficient to buffer just one MCU |
| 40 | * (although this may prove a bit slow in practice). We allocate a |
| 41 | * workspace of MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each |
| 42 | * MCU constructed and sent. (On 80x86, the workspace is FAR even though |
| 43 | * it's not really very big; this is to keep the module interfaces unchanged |
| 44 | * when a large coefficient buffer is necessary.) |
| 45 | * In multi-pass modes, this array points to the current MCU's blocks |
| 46 | * within the virtual arrays. |
| 47 | */ |
| 48 | JBLOCKROW MCU_buffer[MAX_BLOCKS_IN_MCU]; |
| 49 | |
| 50 | /* In multi-pass modes, we need a virtual block array for each component. */ |
| 51 | jvirt_barray_ptr whole_image[MAX_COMPONENTS]; |
| 52 | } my_coef_controller; |
| 53 | |
| 54 | typedef my_coef_controller * my_coef_ptr; |
| 55 | |
| 56 | |
| 57 | /* Forward declarations */ |
| 58 | METHODDEF void compress_data |
| 59 | JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION *in_mcu_ctr)); |
| 60 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
| 61 | METHODDEF void compress_first_pass |
| 62 | JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION *in_mcu_ctr)); |
| 63 | METHODDEF void compress_output |
| 64 | JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf, JDIMENSION *in_mcu_ctr)); |
| 65 | #endif |
| 66 | |
| 67 | |
| 68 | /* |
| 69 | * Initialize for a processing pass. |
| 70 | */ |
| 71 | |
| 72 | METHODDEF void |
| 73 | start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode) |
| 74 | { |
| 75 | my_coef_ptr coef = (my_coef_ptr) cinfo->coef; |
| 76 | |
| 77 | coef->MCU_row_num = 0; |
| 78 | |
| 79 | switch (pass_mode) { |
| 80 | case JBUF_PASS_THRU: |
| 81 | if (coef->whole_image[0] != NULL) |
| 82 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
| 83 | coef->pub.compress_data = compress_data; |
| 84 | break; |
| 85 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
| 86 | case JBUF_SAVE_AND_PASS: |
| 87 | if (coef->whole_image[0] == NULL) |
| 88 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
| 89 | coef->pub.compress_data = compress_first_pass; |
| 90 | break; |
| 91 | case JBUF_CRANK_DEST: |
| 92 | if (coef->whole_image[0] == NULL) |
| 93 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
| 94 | coef->pub.compress_data = compress_output; |
| 95 | break; |
| 96 | #endif |
| 97 | default: |
| 98 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
| 99 | break; |
| 100 | } |
| 101 | } |
| 102 | |
| 103 | |
| 104 | /* |
| 105 | * Process some data in the single-pass case. |
| 106 | * Up to one MCU row is processed (less if suspension is forced). |
| 107 | * |
| 108 | * NB: input_buf contains a plane for each component in image. |
| 109 | * For single pass, this is the same as the components in the scan. |
| 110 | */ |
| 111 | |
| 112 | METHODDEF void |
| 113 | compress_data (j_compress_ptr cinfo, |
| 114 | JSAMPIMAGE input_buf, JDIMENSION *in_mcu_ctr) |
| 115 | { |
| 116 | my_coef_ptr coef = (my_coef_ptr) cinfo->coef; |
| 117 | JDIMENSION MCU_col_num; /* index of current MCU within row */ |
| 118 | JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1; |
| 119 | JDIMENSION last_MCU_row = cinfo->MCU_rows_in_scan - 1; |
| 120 | int blkn, bi, ci, yindex, blockcnt; |
| 121 | JDIMENSION ypos, xpos; |
| 122 | jpeg_component_info *compptr; |
| 123 | |
| 124 | /* Loop to write as much as one whole MCU row */ |
| 125 | |
| 126 | for (MCU_col_num = *in_mcu_ctr; MCU_col_num <= last_MCU_col; MCU_col_num++) { |
| 127 | /* Determine where data comes from in input_buf and do the DCT thing. |
| 128 | * Each call on forward_DCT processes a horizontal row of DCT blocks |
| 129 | * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks |
| 130 | * sequentially. Dummy blocks at the right or bottom edge are filled in |
| 131 | * specially. The data in them does not matter for image reconstruction, |
| 132 | * so we fill them with values that will encode to the smallest amount of |
| 133 | * data, viz: all zeroes in the AC entries, DC entries equal to previous |
| 134 | * block's DC value. (Thanks to Thomas Kinsman for this idea.) |
| 135 | */ |
| 136 | blkn = 0; |
| 137 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
| 138 | compptr = cinfo->cur_comp_info[ci]; |
| 139 | blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width |
| 140 | : compptr->last_col_width; |
| 141 | xpos = MCU_col_num * compptr->MCU_sample_width; |
| 142 | ypos = 0; |
| 143 | for (yindex = 0; yindex < compptr->MCU_height; yindex++) { |
| 144 | if (coef->MCU_row_num < last_MCU_row || |
| 145 | yindex < compptr->last_row_height) { |
| 146 | (*cinfo->fdct->forward_DCT) (cinfo, compptr, |
| 147 | input_buf[ci], coef->MCU_buffer[blkn], |
| 148 | ypos, xpos, (JDIMENSION) blockcnt); |
| 149 | if (blockcnt < compptr->MCU_width) { |
| 150 | /* Create some dummy blocks at the right edge of the image. */ |
| 151 | jzero_far((void FAR *) coef->MCU_buffer[blkn + blockcnt], |
| 152 | (compptr->MCU_width - blockcnt) * SIZEOF(JBLOCK)); |
| 153 | for (bi = blockcnt; bi < compptr->MCU_width; bi++) { |
| 154 | coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn+bi-1][0][0]; |
| 155 | } |
| 156 | } |
| 157 | } else { |
| 158 | /* Create a whole row of dummy blocks at the bottom of the image. */ |
| 159 | jzero_far((void FAR *) coef->MCU_buffer[blkn], |
| 160 | compptr->MCU_width * SIZEOF(JBLOCK)); |
| 161 | for (bi = 0; bi < compptr->MCU_width; bi++) { |
| 162 | coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn-1][0][0]; |
| 163 | } |
| 164 | } |
| 165 | blkn += compptr->MCU_width; |
| 166 | ypos += DCTSIZE; |
| 167 | } |
| 168 | } |
| 169 | /* Try to write the MCU. In event of a suspension failure, we will |
| 170 | * re-DCT the MCU on restart (a bit inefficient, could be fixed...) |
| 171 | */ |
| 172 | if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) |
| 173 | break; /* suspension forced; exit loop */ |
| 174 | } |
| 175 | if (MCU_col_num > last_MCU_col) |
| 176 | coef->MCU_row_num++; /* advance if we finished the row */ |
| 177 | *in_mcu_ctr = MCU_col_num; |
| 178 | } |
| 179 | |
| 180 | |
| 181 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
| 182 | |
| 183 | /* |
| 184 | * Process some data in the first pass of a multi-pass case. |
| 185 | * We process the equivalent of one fully interleaved MCU row ("iMCU" row) |
| 186 | * per call, ie, v_samp_factor block rows for each component in the image. |
| 187 | * This amount of data is read from the source buffer, DCT'd and quantized, |
| 188 | * and saved into the virtual arrays. We also generate suitable dummy blocks |
| 189 | * as needed at the right and lower edges. (The dummy blocks are constructed |
| 190 | * in the virtual arrays, which have been padded appropriately.) This makes |
| 191 | * it possible for subsequent passes not to worry about real vs. dummy blocks. |
| 192 | * |
| 193 | * We must also emit the data to the entropy encoder. This is conveniently |
| 194 | * done by calling compress_output() after we've loaded the current strip |
| 195 | * of the virtual arrays. |
| 196 | * |
| 197 | * NB: input_buf contains a plane for each component in image. All |
| 198 | * components are DCT'd and loaded into the virtual arrays in this pass. |
| 199 | * However, it may be that only a subset of the components are emitted to |
| 200 | * the entropy encoder during this first pass; be careful about looking |
| 201 | * at the scan-dependent variables (MCU dimensions, etc). |
| 202 | */ |
| 203 | |
| 204 | METHODDEF void |
| 205 | compress_first_pass (j_compress_ptr cinfo, |
| 206 | JSAMPIMAGE input_buf, JDIMENSION *in_mcu_ctr) |
| 207 | { |
| 208 | my_coef_ptr coef = (my_coef_ptr) cinfo->coef; |
| 209 | JDIMENSION last_MCU_row = cinfo->total_iMCU_rows - 1; |
| 210 | JDIMENSION blocks_across, MCUs_across, MCUindex; |
| 211 | int bi, ci, h_samp_factor, block_row, block_rows, ndummy; |
| 212 | JCOEF lastDC; |
| 213 | jpeg_component_info *compptr; |
| 214 | JBLOCKARRAY buffer; |
| 215 | JBLOCKROW thisblockrow, lastblockrow; |
| 216 | |
| 217 | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
| 218 | ci++, compptr++) { |
| 219 | /* Align the virtual buffer for this component. */ |
| 220 | buffer = (*cinfo->mem->access_virt_barray) |
| 221 | ((j_common_ptr) cinfo, coef->whole_image[ci], |
| 222 | coef->MCU_row_num * compptr->v_samp_factor, TRUE); |
| 223 | /* Count non-dummy DCT block rows in this iMCU row. */ |
| 224 | if (coef->MCU_row_num < last_MCU_row) |
| 225 | block_rows = compptr->v_samp_factor; |
| 226 | else { |
| 227 | block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor); |
| 228 | if (block_rows == 0) block_rows = compptr->v_samp_factor; |
| 229 | } |
| 230 | blocks_across = compptr->width_in_blocks; |
| 231 | h_samp_factor = compptr->h_samp_factor; |
| 232 | /* Count number of dummy blocks to be added at the right margin. */ |
| 233 | ndummy = (int) (blocks_across % h_samp_factor); |
| 234 | if (ndummy > 0) |
| 235 | ndummy = h_samp_factor - ndummy; |
| 236 | /* Perform DCT for all non-dummy blocks in this iMCU row. Each call |
| 237 | * on forward_DCT processes a complete horizontal row of DCT blocks. |
| 238 | */ |
| 239 | for (block_row = 0; block_row < block_rows; block_row++) { |
| 240 | thisblockrow = buffer[block_row]; |
| 241 | (*cinfo->fdct->forward_DCT) (cinfo, compptr, |
| 242 | input_buf[ci], thisblockrow, |
| 243 | (JDIMENSION) (block_row * DCTSIZE), |
| 244 | (JDIMENSION) 0, blocks_across); |
| 245 | if (ndummy > 0) { |
| 246 | /* Create dummy blocks at the right edge of the image. */ |
| 247 | thisblockrow += blocks_across; /* => first dummy block */ |
| 248 | jzero_far((void FAR *) thisblockrow, ndummy * SIZEOF(JBLOCK)); |
| 249 | lastDC = thisblockrow[-1][0]; |
| 250 | for (bi = 0; bi < ndummy; bi++) { |
| 251 | thisblockrow[bi][0] = lastDC; |
| 252 | } |
| 253 | } |
| 254 | } |
| 255 | /* If at end of image, create dummy block rows as needed. |
| 256 | * The tricky part here is that within each MCU, we want the DC values |
| 257 | * of the dummy blocks to match the last real block's DC value. |
| 258 | * This squeezes a few more bytes out of the resulting file... |
| 259 | */ |
| 260 | if (coef->MCU_row_num == last_MCU_row) { |
| 261 | blocks_across += ndummy; /* include lower right corner */ |
| 262 | MCUs_across = blocks_across / h_samp_factor; |
| 263 | for (block_row = block_rows; block_row < compptr->v_samp_factor; |
| 264 | block_row++) { |
| 265 | thisblockrow = buffer[block_row]; |
| 266 | lastblockrow = buffer[block_row-1]; |
| 267 | jzero_far((void FAR *) thisblockrow, |
| 268 | (size_t) (blocks_across * SIZEOF(JBLOCK))); |
| 269 | for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) { |
| 270 | lastDC = lastblockrow[h_samp_factor-1][0]; |
| 271 | for (bi = 0; bi < h_samp_factor; bi++) { |
| 272 | thisblockrow[bi][0] = lastDC; |
| 273 | } |
| 274 | thisblockrow += h_samp_factor; /* advance to next MCU in row */ |
| 275 | lastblockrow += h_samp_factor; |
| 276 | } |
| 277 | } |
| 278 | } |
| 279 | } |
| 280 | /* NB: compress_output will increment MCU_row_num */ |
| 281 | |
| 282 | /* Emit data to the entropy encoder, sharing code with subsequent passes */ |
| 283 | compress_output(cinfo, input_buf, in_mcu_ctr); |
| 284 | } |
| 285 | |
| 286 | |
| 287 | /* |
| 288 | * Process some data in subsequent passes of a multi-pass case. |
| 289 | * We process the equivalent of one fully interleaved MCU row ("iMCU" row) |
| 290 | * per call, ie, v_samp_factor block rows for each component in the scan. |
| 291 | * The data is obtained from the virtual arrays and fed to the entropy coder. |
| 292 | * |
| 293 | * Note that output suspension is not supported during multi-pass operation, |
| 294 | * so the complete MCU row will always be emitted to the entropy encoder |
| 295 | * before returning. |
| 296 | * |
| 297 | * NB: input_buf is ignored; it is likely to be a NULL pointer. |
| 298 | */ |
| 299 | |
| 300 | METHODDEF void |
| 301 | compress_output (j_compress_ptr cinfo, |
| 302 | JSAMPIMAGE input_buf, JDIMENSION *in_mcu_ctr) |
| 303 | { |
| 304 | my_coef_ptr coef = (my_coef_ptr) cinfo->coef; |
| 305 | JDIMENSION MCU_col_num; /* index of current MCU within row */ |
| 306 | int blkn, ci, xindex, yindex, yoffset, num_MCU_rows; |
| 307 | JDIMENSION remaining_rows, start_col; |
| 308 | JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN]; |
| 309 | JBLOCKROW buffer_ptr; |
| 310 | jpeg_component_info *compptr; |
| 311 | |
| 312 | /* Align the virtual buffers for the components used in this scan. |
| 313 | * NB: during first pass, this is safe only because the buffers will |
| 314 | * already be aligned properly, so jmemmgr.c won't need to do any I/O. |
| 315 | */ |
| 316 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
| 317 | compptr = cinfo->cur_comp_info[ci]; |
| 318 | buffer[ci] = (*cinfo->mem->access_virt_barray) |
| 319 | ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index], |
| 320 | coef->MCU_row_num * compptr->v_samp_factor, FALSE); |
| 321 | } |
| 322 | |
| 323 | /* In an interleaved scan, we process exactly one MCU row. |
| 324 | * In a noninterleaved scan, we need to process v_samp_factor MCU rows, |
| 325 | * each of which contains a single block row. |
| 326 | */ |
| 327 | if (cinfo->comps_in_scan == 1) { |
| 328 | compptr = cinfo->cur_comp_info[0]; |
| 329 | num_MCU_rows = compptr->v_samp_factor; |
| 330 | /* but watch out for the bottom of the image */ |
| 331 | remaining_rows = cinfo->MCU_rows_in_scan - |
| 332 | coef->MCU_row_num * compptr->v_samp_factor; |
| 333 | if (remaining_rows < (JDIMENSION) num_MCU_rows) |
| 334 | num_MCU_rows = (int) remaining_rows; |
| 335 | } else { |
| 336 | num_MCU_rows = 1; |
| 337 | } |
| 338 | |
| 339 | /* Loop to process one whole iMCU row */ |
| 340 | for (yoffset = 0; yoffset < num_MCU_rows; yoffset++) { |
| 341 | for (MCU_col_num = 0; MCU_col_num < cinfo->MCUs_per_row; MCU_col_num++) { |
| 342 | /* Construct list of pointers to DCT blocks belonging to this MCU */ |
| 343 | blkn = 0; /* index of current DCT block within MCU */ |
| 344 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
| 345 | compptr = cinfo->cur_comp_info[ci]; |
| 346 | start_col = MCU_col_num * compptr->MCU_width; |
| 347 | for (yindex = 0; yindex < compptr->MCU_height; yindex++) { |
| 348 | buffer_ptr = buffer[ci][yindex+yoffset] + start_col; |
| 349 | for (xindex = 0; xindex < compptr->MCU_width; xindex++) { |
| 350 | coef->MCU_buffer[blkn++] = buffer_ptr++; |
| 351 | } |
| 352 | } |
| 353 | } |
| 354 | /* Try to write the MCU. */ |
| 355 | if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { |
| 356 | ERREXIT(cinfo, JERR_CANT_SUSPEND); /* not supported */ |
| 357 | } |
| 358 | } |
| 359 | } |
| 360 | |
| 361 | coef->MCU_row_num++; /* advance to next iMCU row */ |
| 362 | *in_mcu_ctr = cinfo->MCUs_per_row; |
| 363 | } |
| 364 | |
| 365 | #endif /* FULL_COEF_BUFFER_SUPPORTED */ |
| 366 | |
| 367 | |
| 368 | /* |
| 369 | * Initialize coefficient buffer controller. |
| 370 | */ |
| 371 | |
| 372 | GLOBAL void |
| 373 | jinit_c_coef_controller (j_compress_ptr cinfo, boolean need_full_buffer) |
| 374 | { |
| 375 | my_coef_ptr coef; |
| 376 | int ci, i; |
| 377 | jpeg_component_info *compptr; |
| 378 | JBLOCKROW buffer; |
| 379 | |
| 380 | coef = (my_coef_ptr) |
| 381 | (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE, |
| 382 | SIZEOF(my_coef_controller)); |
| 383 | cinfo->coef = (struct jpeg_c_coef_controller *) coef; |
| 384 | coef->pub.start_pass = start_pass_coef; |
| 385 | |
| 386 | /* Create the coefficient buffer. */ |
| 387 | if (need_full_buffer) { |
| 388 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
| 389 | /* Allocate a full-image virtual array for each component, */ |
| 390 | /* padded to a multiple of samp_factor DCT blocks in each direction. */ |
| 391 | /* Note memmgr implicitly pads the vertical direction. */ |
| 392 | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
| 393 | ci++, compptr++) { |
| 394 | coef->whole_image[ci] = (*cinfo->mem->request_virt_barray) |
| 395 | ((j_common_ptr) cinfo, JPOOL_IMAGE, |
| 396 | (JDIMENSION) jround_up((long) compptr->width_in_blocks, |
| 397 | (long) compptr->h_samp_factor), |
| 398 | compptr->height_in_blocks, |
| 399 | (JDIMENSION) compptr->v_samp_factor); |
| 400 | } |
| 401 | #else |
| 402 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
| 403 | #endif |
| 404 | } else { |
| 405 | /* We only need a single-MCU buffer. */ |
| 406 | buffer = (JBLOCKROW) |
| 407 | (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE, |
| 408 | MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK)); |
| 409 | for (i = 0; i < MAX_BLOCKS_IN_MCU; i++) { |
| 410 | coef->MCU_buffer[i] = buffer + i; |
| 411 | } |
| 412 | coef->whole_image[0] = NULL; /* flag for no virtual arrays */ |
| 413 | } |
| 414 | } |