1 /* 2 * jdmainct.c 3 * 4 * This file was part of the Independent JPEG Group's software: 5 * Copyright (C) 1994-1996, Thomas G. Lane. 6 * libjpeg-turbo Modifications: 7 * Copyright (C) 2010, D. R. Commander. 8 * For conditions of distribution and use, see the accompanying README file. 9 * 10 * This file contains the main buffer controller for decompression. 11 * The main buffer lies between the JPEG decompressor proper and the 12 * post-processor; it holds downsampled data in the JPEG colorspace. 13 * 14 * Note that this code is bypassed in raw-data mode, since the application 15 * supplies the equivalent of the main buffer in that case. 16 */ 17 18 #include "jdmainct.h" 19 20 21 /* 22 * In the current system design, the main buffer need never be a full-image 23 * buffer; any full-height buffers will be found inside the coefficient or 24 * postprocessing controllers. Nonetheless, the main controller is not 25 * trivial. Its responsibility is to provide context rows for upsampling/ 26 * rescaling, and doing this in an efficient fashion is a bit tricky. 27 * 28 * Postprocessor input data is counted in "row groups". A row group 29 * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size) 30 * sample rows of each component. (We require DCT_scaled_size values to be 31 * chosen such that these numbers are integers. In practice DCT_scaled_size 32 * values will likely be powers of two, so we actually have the stronger 33 * condition that DCT_scaled_size / min_DCT_scaled_size is an integer.) 34 * Upsampling will typically produce max_v_samp_factor pixel rows from each 35 * row group (times any additional scale factor that the upsampler is 36 * applying). 37 * 38 * The coefficient controller will deliver data to us one iMCU row at a time; 39 * each iMCU row contains v_samp_factor * DCT_scaled_size sample rows, or 40 * exactly min_DCT_scaled_size row groups. (This amount of data corresponds 41 * to one row of MCUs when the image is fully interleaved.) Note that the 42 * number of sample rows varies across components, but the number of row 43 * groups does not. Some garbage sample rows may be included in the last iMCU 44 * row at the bottom of the image. 45 * 46 * Depending on the vertical scaling algorithm used, the upsampler may need 47 * access to the sample row(s) above and below its current input row group. 48 * The upsampler is required to set need_context_rows TRUE at global selection 49 * time if so. When need_context_rows is FALSE, this controller can simply 50 * obtain one iMCU row at a time from the coefficient controller and dole it 51 * out as row groups to the postprocessor. 52 * 53 * When need_context_rows is TRUE, this controller guarantees that the buffer 54 * passed to postprocessing contains at least one row group's worth of samples 55 * above and below the row group(s) being processed. Note that the context 56 * rows "above" the first passed row group appear at negative row offsets in 57 * the passed buffer. At the top and bottom of the image, the required 58 * context rows are manufactured by duplicating the first or last real sample 59 * row; this avoids having special cases in the upsampling inner loops. 60 * 61 * The amount of context is fixed at one row group just because that's a 62 * convenient number for this controller to work with. The existing 63 * upsamplers really only need one sample row of context. An upsampler 64 * supporting arbitrary output rescaling might wish for more than one row 65 * group of context when shrinking the image; tough, we don't handle that. 66 * (This is justified by the assumption that downsizing will be handled mostly 67 * by adjusting the DCT_scaled_size values, so that the actual scale factor at 68 * the upsample step needn't be much less than one.) 69 * 70 * To provide the desired context, we have to retain the last two row groups 71 * of one iMCU row while reading in the next iMCU row. (The last row group 72 * can't be processed until we have another row group for its below-context, 73 * and so we have to save the next-to-last group too for its above-context.) 74 * We could do this most simply by copying data around in our buffer, but 75 * that'd be very slow. We can avoid copying any data by creating a rather 76 * strange pointer structure. Here's how it works. We allocate a workspace 77 * consisting of M+2 row groups (where M = min_DCT_scaled_size is the number 78 * of row groups per iMCU row). We create two sets of redundant pointers to 79 * the workspace. Labeling the physical row groups 0 to M+1, the synthesized 80 * pointer lists look like this: 81 * M+1 M-1 82 * master pointer --> 0 master pointer --> 0 83 * 1 1 84 * ... ... 85 * M-3 M-3 86 * M-2 M 87 * M-1 M+1 88 * M M-2 89 * M+1 M-1 90 * 0 0 91 * We read alternate iMCU rows using each master pointer; thus the last two 92 * row groups of the previous iMCU row remain un-overwritten in the workspace. 93 * The pointer lists are set up so that the required context rows appear to 94 * be adjacent to the proper places when we pass the pointer lists to the 95 * upsampler. 96 * 97 * The above pictures describe the normal state of the pointer lists. 98 * At top and bottom of the image, we diddle the pointer lists to duplicate 99 * the first or last sample row as necessary (this is cheaper than copying 100 * sample rows around). 101 * 102 * This scheme breaks down if M < 2, ie, min_DCT_scaled_size is 1. In that 103 * situation each iMCU row provides only one row group so the buffering logic 104 * must be different (eg, we must read two iMCU rows before we can emit the 105 * first row group). For now, we simply do not support providing context 106 * rows when min_DCT_scaled_size is 1. That combination seems unlikely to 107 * be worth providing --- if someone wants a 1/8th-size preview, they probably 108 * want it quick and dirty, so a context-free upsampler is sufficient. 109 */ 110 111 112 /* Forward declarations */ 113 METHODDEF(void) process_data_simple_main 114 (j_decompress_ptr cinfo, JSAMPARRAY output_buf, 115 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); 116 METHODDEF(void) process_data_context_main 117 (j_decompress_ptr cinfo, JSAMPARRAY output_buf, 118 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); 119 #ifdef QUANT_2PASS_SUPPORTED 120 METHODDEF(void) process_data_crank_post 121 (j_decompress_ptr cinfo, JSAMPARRAY output_buf, 122 JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); 123 #endif 124 125 126 LOCAL(void) 127 alloc_funny_pointers (j_decompress_ptr cinfo) 128 /* Allocate space for the funny pointer lists. 129 * This is done only once, not once per pass. 130 */ 131 { 132 my_main_ptr main_ptr = (my_main_ptr) cinfo->main; 133 int ci, rgroup; 134 int M = cinfo->_min_DCT_scaled_size; 135 jpeg_component_info *compptr; 136 JSAMPARRAY xbuf; 137 138 /* Get top-level space for component array pointers. 139 * We alloc both arrays with one call to save a few cycles. 140 */ 141 main_ptr->xbuffer[0] = (JSAMPIMAGE) 142 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE, 143 cinfo->num_components * 2 * sizeof(JSAMPARRAY)); 144 main_ptr->xbuffer[1] = main_ptr->xbuffer[0] + cinfo->num_components; 145 146 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; 147 ci++, compptr++) { 148 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) / 149 cinfo->_min_DCT_scaled_size; /* height of a row group of component */ 150 /* Get space for pointer lists --- M+4 row groups in each list. 151 * We alloc both pointer lists with one call to save a few cycles. 152 */ 153 xbuf = (JSAMPARRAY) 154 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE, 155 2 * (rgroup * (M + 4)) * sizeof(JSAMPROW)); 156 xbuf += rgroup; /* want one row group at negative offsets */ 157 main_ptr->xbuffer[0][ci] = xbuf; 158 xbuf += rgroup * (M + 4); 159 main_ptr->xbuffer[1][ci] = xbuf; 160 } 161 } 162 163 164 LOCAL(void) 165 make_funny_pointers (j_decompress_ptr cinfo) 166 /* Create the funny pointer lists discussed in the comments above. 167 * The actual workspace is already allocated (in main_ptr->buffer), 168 * and the space for the pointer lists is allocated too. 169 * This routine just fills in the curiously ordered lists. 170 * This will be repeated at the beginning of each pass. 171 */ 172 { 173 my_main_ptr main_ptr = (my_main_ptr) cinfo->main; 174 int ci, i, rgroup; 175 int M = cinfo->_min_DCT_scaled_size; 176 jpeg_component_info *compptr; 177 JSAMPARRAY buf, xbuf0, xbuf1; 178 179 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; 180 ci++, compptr++) { 181 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) / 182 cinfo->_min_DCT_scaled_size; /* height of a row group of component */ 183 xbuf0 = main_ptr->xbuffer[0][ci]; 184 xbuf1 = main_ptr->xbuffer[1][ci]; 185 /* First copy the workspace pointers as-is */ 186 buf = main_ptr->buffer[ci]; 187 for (i = 0; i < rgroup * (M + 2); i++) { 188 xbuf0[i] = xbuf1[i] = buf[i]; 189 } 190 /* In the second list, put the last four row groups in swapped order */ 191 for (i = 0; i < rgroup * 2; i++) { 192 xbuf1[rgroup*(M-2) + i] = buf[rgroup*M + i]; 193 xbuf1[rgroup*M + i] = buf[rgroup*(M-2) + i]; 194 } 195 /* The wraparound pointers at top and bottom will be filled later 196 * (see set_wraparound_pointers, below). Initially we want the "above" 197 * pointers to duplicate the first actual data line. This only needs 198 * to happen in xbuffer[0]. 199 */ 200 for (i = 0; i < rgroup; i++) { 201 xbuf0[i - rgroup] = xbuf0[0]; 202 } 203 } 204 } 205 206 207 LOCAL(void) 208 set_bottom_pointers (j_decompress_ptr cinfo) 209 /* Change the pointer lists to duplicate the last sample row at the bottom 210 * of the image. whichptr indicates which xbuffer holds the final iMCU row. 211 * Also sets rowgroups_avail to indicate number of nondummy row groups in row. 212 */ 213 { 214 my_main_ptr main_ptr = (my_main_ptr) cinfo->main; 215 int ci, i, rgroup, iMCUheight, rows_left; 216 jpeg_component_info *compptr; 217 JSAMPARRAY xbuf; 218 219 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; 220 ci++, compptr++) { 221 /* Count sample rows in one iMCU row and in one row group */ 222 iMCUheight = compptr->v_samp_factor * compptr->_DCT_scaled_size; 223 rgroup = iMCUheight / cinfo->_min_DCT_scaled_size; 224 /* Count nondummy sample rows remaining for this component */ 225 rows_left = (int) (compptr->downsampled_height % (JDIMENSION) iMCUheight); 226 if (rows_left == 0) rows_left = iMCUheight; 227 /* Count nondummy row groups. Should get same answer for each component, 228 * so we need only do it once. 229 */ 230 if (ci == 0) { 231 main_ptr->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1); 232 } 233 /* Duplicate the last real sample row rgroup*2 times; this pads out the 234 * last partial rowgroup and ensures at least one full rowgroup of context. 235 */ 236 xbuf = main_ptr->xbuffer[main_ptr->whichptr][ci]; 237 for (i = 0; i < rgroup * 2; i++) { 238 xbuf[rows_left + i] = xbuf[rows_left-1]; 239 } 240 } 241 } 242 243 244 /* 245 * Initialize for a processing pass. 246 */ 247 248 METHODDEF(void) 249 start_pass_main (j_decompress_ptr cinfo, J_BUF_MODE pass_mode) 250 { 251 my_main_ptr main_ptr = (my_main_ptr) cinfo->main; 252 253 switch (pass_mode) { 254 case JBUF_PASS_THRU: 255 if (cinfo->upsample->need_context_rows) { 256 main_ptr->pub.process_data = process_data_context_main; 257 make_funny_pointers(cinfo); /* Create the xbuffer[] lists */ 258 main_ptr->whichptr = 0; /* Read first iMCU row into xbuffer[0] */ 259 main_ptr->context_state = CTX_PREPARE_FOR_IMCU; 260 main_ptr->iMCU_row_ctr = 0; 261 } else { 262 /* Simple case with no context needed */ 263 main_ptr->pub.process_data = process_data_simple_main; 264 } 265 main_ptr->buffer_full = FALSE; /* Mark buffer empty */ 266 main_ptr->rowgroup_ctr = 0; 267 break; 268 #ifdef QUANT_2PASS_SUPPORTED 269 case JBUF_CRANK_DEST: 270 /* For last pass of 2-pass quantization, just crank the postprocessor */ 271 main_ptr->pub.process_data = process_data_crank_post; 272 break; 273 #endif 274 default: 275 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); 276 break; 277 } 278 } 279 280 281 /* 282 * Process some data. 283 * This handles the simple case where no context is required. 284 */ 285 286 METHODDEF(void) 287 process_data_simple_main (j_decompress_ptr cinfo, 288 JSAMPARRAY output_buf, JDIMENSION *out_row_ctr, 289 JDIMENSION out_rows_avail) 290 { 291 my_main_ptr main_ptr = (my_main_ptr) cinfo->main; 292 JDIMENSION rowgroups_avail; 293 294 /* Read input data if we haven't filled the main buffer yet */ 295 if (! main_ptr->buffer_full) { 296 if (! (*cinfo->coef->decompress_data) (cinfo, main_ptr->buffer)) 297 return; /* suspension forced, can do nothing more */ 298 main_ptr->buffer_full = TRUE; /* OK, we have an iMCU row to work with */ 299 } 300 301 /* There are always min_DCT_scaled_size row groups in an iMCU row. */ 302 rowgroups_avail = (JDIMENSION) cinfo->_min_DCT_scaled_size; 303 /* Note: at the bottom of the image, we may pass extra garbage row groups 304 * to the postprocessor. The postprocessor has to check for bottom 305 * of image anyway (at row resolution), so no point in us doing it too. 306 */ 307 308 /* Feed the postprocessor */ 309 (*cinfo->post->post_process_data) (cinfo, main_ptr->buffer, 310 &main_ptr->rowgroup_ctr, rowgroups_avail, 311 output_buf, out_row_ctr, out_rows_avail); 312 313 /* Has postprocessor consumed all the data yet? If so, mark buffer empty */ 314 if (main_ptr->rowgroup_ctr >= rowgroups_avail) { 315 main_ptr->buffer_full = FALSE; 316 main_ptr->rowgroup_ctr = 0; 317 } 318 } 319 320 321 /* 322 * Process some data. 323 * This handles the case where context rows must be provided. 324 */ 325 326 METHODDEF(void) 327 process_data_context_main (j_decompress_ptr cinfo, 328 JSAMPARRAY output_buf, JDIMENSION *out_row_ctr, 329 JDIMENSION out_rows_avail) 330 { 331 my_main_ptr main_ptr = (my_main_ptr) cinfo->main; 332 333 /* Read input data if we haven't filled the main buffer yet */ 334 if (! main_ptr->buffer_full) { 335 if (! (*cinfo->coef->decompress_data) (cinfo, 336 main_ptr->xbuffer[main_ptr->whichptr])) 337 return; /* suspension forced, can do nothing more */ 338 main_ptr->buffer_full = TRUE; /* OK, we have an iMCU row to work with */ 339 main_ptr->iMCU_row_ctr++; /* count rows received */ 340 } 341 342 /* Postprocessor typically will not swallow all the input data it is handed 343 * in one call (due to filling the output buffer first). Must be prepared 344 * to exit and restart. This switch lets us keep track of how far we got. 345 * Note that each case falls through to the next on successful completion. 346 */ 347 switch (main_ptr->context_state) { 348 case CTX_POSTPONED_ROW: 349 /* Call postprocessor using previously set pointers for postponed row */ 350 (*cinfo->post->post_process_data) (cinfo, main_ptr->xbuffer[main_ptr->whichptr], 351 &main_ptr->rowgroup_ctr, main_ptr->rowgroups_avail, 352 output_buf, out_row_ctr, out_rows_avail); 353 if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail) 354 return; /* Need to suspend */ 355 main_ptr->context_state = CTX_PREPARE_FOR_IMCU; 356 if (*out_row_ctr >= out_rows_avail) 357 return; /* Postprocessor exactly filled output buf */ 358 /*FALLTHROUGH*/ 359 case CTX_PREPARE_FOR_IMCU: 360 /* Prepare to process first M-1 row groups of this iMCU row */ 361 main_ptr->rowgroup_ctr = 0; 362 main_ptr->rowgroups_avail = (JDIMENSION) (cinfo->_min_DCT_scaled_size - 1); 363 /* Check for bottom of image: if so, tweak pointers to "duplicate" 364 * the last sample row, and adjust rowgroups_avail to ignore padding rows. 365 */ 366 if (main_ptr->iMCU_row_ctr == cinfo->total_iMCU_rows) 367 set_bottom_pointers(cinfo); 368 main_ptr->context_state = CTX_PROCESS_IMCU; 369 /*FALLTHROUGH*/ 370 case CTX_PROCESS_IMCU: 371 /* Call postprocessor using previously set pointers */ 372 (*cinfo->post->post_process_data) (cinfo, main_ptr->xbuffer[main_ptr->whichptr], 373 &main_ptr->rowgroup_ctr, main_ptr->rowgroups_avail, 374 output_buf, out_row_ctr, out_rows_avail); 375 if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail) 376 return; /* Need to suspend */ 377 /* After the first iMCU, change wraparound pointers to normal state */ 378 if (main_ptr->iMCU_row_ctr == 1) 379 set_wraparound_pointers(cinfo); 380 /* Prepare to load new iMCU row using other xbuffer list */ 381 main_ptr->whichptr ^= 1; /* 0=>1 or 1=>0 */ 382 main_ptr->buffer_full = FALSE; 383 /* Still need to process last row group of this iMCU row, */ 384 /* which is saved at index M+1 of the other xbuffer */ 385 main_ptr->rowgroup_ctr = (JDIMENSION) (cinfo->_min_DCT_scaled_size + 1); 386 main_ptr->rowgroups_avail = (JDIMENSION) (cinfo->_min_DCT_scaled_size + 2); 387 main_ptr->context_state = CTX_POSTPONED_ROW; 388 } 389 } 390 391 392 /* 393 * Process some data. 394 * Final pass of two-pass quantization: just call the postprocessor. 395 * Source data will be the postprocessor controller's internal buffer. 396 */ 397 398 #ifdef QUANT_2PASS_SUPPORTED 399 400 METHODDEF(void) 401 process_data_crank_post (j_decompress_ptr cinfo, 402 JSAMPARRAY output_buf, JDIMENSION *out_row_ctr, 403 JDIMENSION out_rows_avail) 404 { 405 (*cinfo->post->post_process_data) (cinfo, (JSAMPIMAGE) NULL, 406 (JDIMENSION *) NULL, (JDIMENSION) 0, 407 output_buf, out_row_ctr, out_rows_avail); 408 } 409 410 #endif /* QUANT_2PASS_SUPPORTED */ 411 412 413 /* 414 * Initialize main buffer controller. 415 */ 416 417 GLOBAL(void) 418 jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer) 419 { 420 my_main_ptr main_ptr; 421 int ci, rgroup, ngroups; 422 jpeg_component_info *compptr; 423 424 main_ptr = (my_main_ptr) 425 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE, 426 sizeof(my_main_controller)); 427 cinfo->main = (struct jpeg_d_main_controller *) main_ptr; 428 main_ptr->pub.start_pass = start_pass_main; 429 430 if (need_full_buffer) /* shouldn't happen */ 431 ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); 432 433 /* Allocate the workspace. 434 * ngroups is the number of row groups we need. 435 */ 436 if (cinfo->upsample->need_context_rows) { 437 if (cinfo->_min_DCT_scaled_size < 2) /* unsupported, see comments above */ 438 ERREXIT(cinfo, JERR_NOTIMPL); 439 alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */ 440 ngroups = cinfo->_min_DCT_scaled_size + 2; 441 } else { 442 ngroups = cinfo->_min_DCT_scaled_size; 443 } 444 445 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; 446 ci++, compptr++) { 447 rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) / 448 cinfo->_min_DCT_scaled_size; /* height of a row group of component */ 449 main_ptr->buffer[ci] = (*cinfo->mem->alloc_sarray) 450 ((j_common_ptr) cinfo, JPOOL_IMAGE, 451 compptr->width_in_blocks * compptr->_DCT_scaled_size, 452 (JDIMENSION) (rgroup * ngroups)); 453 } 454 } 455
