michael@0: /* michael@0: * jccoefct.c michael@0: * michael@0: * Copyright (C) 1994-1997, Thomas G. Lane. michael@0: * This file is part of the Independent JPEG Group's software. michael@0: * For conditions of distribution and use, see the accompanying README file. michael@0: * michael@0: * This file contains the coefficient buffer controller for compression. michael@0: * This controller is the top level of the JPEG compressor proper. michael@0: * The coefficient buffer lies between forward-DCT and entropy encoding steps. michael@0: */ michael@0: michael@0: #define JPEG_INTERNALS michael@0: #include "jinclude.h" michael@0: #include "jpeglib.h" michael@0: michael@0: michael@0: /* We use a full-image coefficient buffer when doing Huffman optimization, michael@0: * and also for writing multiple-scan JPEG files. In all cases, the DCT michael@0: * step is run during the first pass, and subsequent passes need only read michael@0: * the buffered coefficients. michael@0: */ michael@0: #ifdef ENTROPY_OPT_SUPPORTED michael@0: #define FULL_COEF_BUFFER_SUPPORTED michael@0: #else michael@0: #ifdef C_MULTISCAN_FILES_SUPPORTED michael@0: #define FULL_COEF_BUFFER_SUPPORTED michael@0: #endif michael@0: #endif michael@0: michael@0: michael@0: /* Private buffer controller object */ michael@0: michael@0: typedef struct { michael@0: struct jpeg_c_coef_controller pub; /* public fields */ michael@0: michael@0: JDIMENSION iMCU_row_num; /* iMCU row # within image */ michael@0: JDIMENSION mcu_ctr; /* counts MCUs processed in current row */ michael@0: int MCU_vert_offset; /* counts MCU rows within iMCU row */ michael@0: int MCU_rows_per_iMCU_row; /* number of such rows needed */ michael@0: michael@0: /* For single-pass compression, it's sufficient to buffer just one MCU michael@0: * (although this may prove a bit slow in practice). We allocate a michael@0: * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each michael@0: * MCU constructed and sent. (On 80x86, the workspace is FAR even though michael@0: * it's not really very big; this is to keep the module interfaces unchanged michael@0: * when a large coefficient buffer is necessary.) michael@0: * In multi-pass modes, this array points to the current MCU's blocks michael@0: * within the virtual arrays. michael@0: */ michael@0: JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU]; michael@0: michael@0: /* In multi-pass modes, we need a virtual block array for each component. */ michael@0: jvirt_barray_ptr whole_image[MAX_COMPONENTS]; michael@0: } my_coef_controller; michael@0: michael@0: typedef my_coef_controller * my_coef_ptr; michael@0: michael@0: michael@0: /* Forward declarations */ michael@0: METHODDEF(boolean) compress_data michael@0: JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf)); michael@0: #ifdef FULL_COEF_BUFFER_SUPPORTED michael@0: METHODDEF(boolean) compress_first_pass michael@0: JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf)); michael@0: METHODDEF(boolean) compress_output michael@0: JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf)); michael@0: #endif michael@0: michael@0: michael@0: LOCAL(void) michael@0: start_iMCU_row (j_compress_ptr cinfo) michael@0: /* Reset within-iMCU-row counters for a new row */ michael@0: { michael@0: my_coef_ptr coef = (my_coef_ptr) cinfo->coef; michael@0: michael@0: /* In an interleaved scan, an MCU row is the same as an iMCU row. michael@0: * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows. michael@0: * But at the bottom of the image, process only what's left. michael@0: */ michael@0: if (cinfo->comps_in_scan > 1) { michael@0: coef->MCU_rows_per_iMCU_row = 1; michael@0: } else { michael@0: if (coef->iMCU_row_num < (cinfo->total_iMCU_rows-1)) michael@0: coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor; michael@0: else michael@0: coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height; michael@0: } michael@0: michael@0: coef->mcu_ctr = 0; michael@0: coef->MCU_vert_offset = 0; michael@0: } michael@0: michael@0: michael@0: /* michael@0: * Initialize for a processing pass. michael@0: */ michael@0: michael@0: METHODDEF(void) michael@0: start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode) michael@0: { michael@0: my_coef_ptr coef = (my_coef_ptr) cinfo->coef; michael@0: michael@0: coef->iMCU_row_num = 0; michael@0: start_iMCU_row(cinfo); michael@0: michael@0: switch (pass_mode) { michael@0: case JBUF_PASS_THRU: michael@0: if (coef->whole_image[0] != NULL) michael@0: ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); michael@0: coef->pub.compress_data = compress_data; michael@0: break; michael@0: #ifdef FULL_COEF_BUFFER_SUPPORTED michael@0: case JBUF_SAVE_AND_PASS: michael@0: if (coef->whole_image[0] == NULL) michael@0: ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); michael@0: coef->pub.compress_data = compress_first_pass; michael@0: break; michael@0: case JBUF_CRANK_DEST: michael@0: if (coef->whole_image[0] == NULL) michael@0: ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); michael@0: coef->pub.compress_data = compress_output; michael@0: break; michael@0: #endif michael@0: default: michael@0: ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); michael@0: break; michael@0: } michael@0: } michael@0: michael@0: michael@0: /* michael@0: * Process some data in the single-pass case. michael@0: * We process the equivalent of one fully interleaved MCU row ("iMCU" row) michael@0: * per call, ie, v_samp_factor block rows for each component in the image. michael@0: * Returns TRUE if the iMCU row is completed, FALSE if suspended. michael@0: * michael@0: * NB: input_buf contains a plane for each component in image, michael@0: * which we index according to the component's SOF position. michael@0: */ michael@0: michael@0: METHODDEF(boolean) michael@0: compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_buf) michael@0: { michael@0: my_coef_ptr coef = (my_coef_ptr) cinfo->coef; michael@0: JDIMENSION MCU_col_num; /* index of current MCU within row */ michael@0: JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1; michael@0: JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; michael@0: int blkn, bi, ci, yindex, yoffset, blockcnt; michael@0: JDIMENSION ypos, xpos; michael@0: jpeg_component_info *compptr; michael@0: michael@0: /* Loop to write as much as one whole iMCU row */ michael@0: for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; michael@0: yoffset++) { michael@0: for (MCU_col_num = coef->mcu_ctr; MCU_col_num <= last_MCU_col; michael@0: MCU_col_num++) { michael@0: /* Determine where data comes from in input_buf and do the DCT thing. michael@0: * Each call on forward_DCT processes a horizontal row of DCT blocks michael@0: * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks michael@0: * sequentially. Dummy blocks at the right or bottom edge are filled in michael@0: * specially. The data in them does not matter for image reconstruction, michael@0: * so we fill them with values that will encode to the smallest amount of michael@0: * data, viz: all zeroes in the AC entries, DC entries equal to previous michael@0: * block's DC value. (Thanks to Thomas Kinsman for this idea.) michael@0: */ michael@0: blkn = 0; michael@0: for (ci = 0; ci < cinfo->comps_in_scan; ci++) { michael@0: compptr = cinfo->cur_comp_info[ci]; michael@0: blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width michael@0: : compptr->last_col_width; michael@0: xpos = MCU_col_num * compptr->MCU_sample_width; michael@0: ypos = yoffset * DCTSIZE; /* ypos == (yoffset+yindex) * DCTSIZE */ michael@0: for (yindex = 0; yindex < compptr->MCU_height; yindex++) { michael@0: if (coef->iMCU_row_num < last_iMCU_row || michael@0: yoffset+yindex < compptr->last_row_height) { michael@0: (*cinfo->fdct->forward_DCT) (cinfo, compptr, michael@0: input_buf[compptr->component_index], michael@0: coef->MCU_buffer[blkn], michael@0: ypos, xpos, (JDIMENSION) blockcnt); michael@0: if (blockcnt < compptr->MCU_width) { michael@0: /* Create some dummy blocks at the right edge of the image. */ michael@0: jzero_far((void FAR *) coef->MCU_buffer[blkn + blockcnt], michael@0: (compptr->MCU_width - blockcnt) * SIZEOF(JBLOCK)); michael@0: for (bi = blockcnt; bi < compptr->MCU_width; bi++) { michael@0: coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn+bi-1][0][0]; michael@0: } michael@0: } michael@0: } else { michael@0: /* Create a row of dummy blocks at the bottom of the image. */ michael@0: jzero_far((void FAR *) coef->MCU_buffer[blkn], michael@0: compptr->MCU_width * SIZEOF(JBLOCK)); michael@0: for (bi = 0; bi < compptr->MCU_width; bi++) { michael@0: coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn-1][0][0]; michael@0: } michael@0: } michael@0: blkn += compptr->MCU_width; michael@0: ypos += DCTSIZE; michael@0: } michael@0: } michael@0: /* Try to write the MCU. In event of a suspension failure, we will michael@0: * re-DCT the MCU on restart (a bit inefficient, could be fixed...) michael@0: */ michael@0: if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { michael@0: /* Suspension forced; update state counters and exit */ michael@0: coef->MCU_vert_offset = yoffset; michael@0: coef->mcu_ctr = MCU_col_num; michael@0: return FALSE; michael@0: } michael@0: } michael@0: /* Completed an MCU row, but perhaps not an iMCU row */ michael@0: coef->mcu_ctr = 0; michael@0: } michael@0: /* Completed the iMCU row, advance counters for next one */ michael@0: coef->iMCU_row_num++; michael@0: start_iMCU_row(cinfo); michael@0: return TRUE; michael@0: } michael@0: michael@0: michael@0: #ifdef FULL_COEF_BUFFER_SUPPORTED michael@0: michael@0: /* michael@0: * Process some data in the first pass of a multi-pass case. michael@0: * We process the equivalent of one fully interleaved MCU row ("iMCU" row) michael@0: * per call, ie, v_samp_factor block rows for each component in the image. michael@0: * This amount of data is read from the source buffer, DCT'd and quantized, michael@0: * and saved into the virtual arrays. We also generate suitable dummy blocks michael@0: * as needed at the right and lower edges. (The dummy blocks are constructed michael@0: * in the virtual arrays, which have been padded appropriately.) This makes michael@0: * it possible for subsequent passes not to worry about real vs. dummy blocks. michael@0: * michael@0: * We must also emit the data to the entropy encoder. This is conveniently michael@0: * done by calling compress_output() after we've loaded the current strip michael@0: * of the virtual arrays. michael@0: * michael@0: * NB: input_buf contains a plane for each component in image. All michael@0: * components are DCT'd and loaded into the virtual arrays in this pass. michael@0: * However, it may be that only a subset of the components are emitted to michael@0: * the entropy encoder during this first pass; be careful about looking michael@0: * at the scan-dependent variables (MCU dimensions, etc). michael@0: */ michael@0: michael@0: METHODDEF(boolean) michael@0: compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf) michael@0: { michael@0: my_coef_ptr coef = (my_coef_ptr) cinfo->coef; michael@0: JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; michael@0: JDIMENSION blocks_across, MCUs_across, MCUindex; michael@0: int bi, ci, h_samp_factor, block_row, block_rows, ndummy; michael@0: JCOEF lastDC; michael@0: jpeg_component_info *compptr; michael@0: JBLOCKARRAY buffer; michael@0: JBLOCKROW thisblockrow, lastblockrow; michael@0: michael@0: for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; michael@0: ci++, compptr++) { michael@0: /* Align the virtual buffer for this component. */ michael@0: buffer = (*cinfo->mem->access_virt_barray) michael@0: ((j_common_ptr) cinfo, coef->whole_image[ci], michael@0: coef->iMCU_row_num * compptr->v_samp_factor, michael@0: (JDIMENSION) compptr->v_samp_factor, TRUE); michael@0: /* Count non-dummy DCT block rows in this iMCU row. */ michael@0: if (coef->iMCU_row_num < last_iMCU_row) michael@0: block_rows = compptr->v_samp_factor; michael@0: else { michael@0: /* NB: can't use last_row_height here, since may not be set! */ michael@0: block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor); michael@0: if (block_rows == 0) block_rows = compptr->v_samp_factor; michael@0: } michael@0: blocks_across = compptr->width_in_blocks; michael@0: h_samp_factor = compptr->h_samp_factor; michael@0: /* Count number of dummy blocks to be added at the right margin. */ michael@0: ndummy = (int) (blocks_across % h_samp_factor); michael@0: if (ndummy > 0) michael@0: ndummy = h_samp_factor - ndummy; michael@0: /* Perform DCT for all non-dummy blocks in this iMCU row. Each call michael@0: * on forward_DCT processes a complete horizontal row of DCT blocks. michael@0: */ michael@0: for (block_row = 0; block_row < block_rows; block_row++) { michael@0: thisblockrow = buffer[block_row]; michael@0: (*cinfo->fdct->forward_DCT) (cinfo, compptr, michael@0: input_buf[ci], thisblockrow, michael@0: (JDIMENSION) (block_row * DCTSIZE), michael@0: (JDIMENSION) 0, blocks_across); michael@0: if (ndummy > 0) { michael@0: /* Create dummy blocks at the right edge of the image. */ michael@0: thisblockrow += blocks_across; /* => first dummy block */ michael@0: jzero_far((void FAR *) thisblockrow, ndummy * SIZEOF(JBLOCK)); michael@0: lastDC = thisblockrow[-1][0]; michael@0: for (bi = 0; bi < ndummy; bi++) { michael@0: thisblockrow[bi][0] = lastDC; michael@0: } michael@0: } michael@0: } michael@0: /* If at end of image, create dummy block rows as needed. michael@0: * The tricky part here is that within each MCU, we want the DC values michael@0: * of the dummy blocks to match the last real block's DC value. michael@0: * This squeezes a few more bytes out of the resulting file... michael@0: */ michael@0: if (coef->iMCU_row_num == last_iMCU_row) { michael@0: blocks_across += ndummy; /* include lower right corner */ michael@0: MCUs_across = blocks_across / h_samp_factor; michael@0: for (block_row = block_rows; block_row < compptr->v_samp_factor; michael@0: block_row++) { michael@0: thisblockrow = buffer[block_row]; michael@0: lastblockrow = buffer[block_row-1]; michael@0: jzero_far((void FAR *) thisblockrow, michael@0: (size_t) (blocks_across * SIZEOF(JBLOCK))); michael@0: for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) { michael@0: lastDC = lastblockrow[h_samp_factor-1][0]; michael@0: for (bi = 0; bi < h_samp_factor; bi++) { michael@0: thisblockrow[bi][0] = lastDC; michael@0: } michael@0: thisblockrow += h_samp_factor; /* advance to next MCU in row */ michael@0: lastblockrow += h_samp_factor; michael@0: } michael@0: } michael@0: } michael@0: } michael@0: /* NB: compress_output will increment iMCU_row_num if successful. michael@0: * A suspension return will result in redoing all the work above next time. michael@0: */ michael@0: michael@0: /* Emit data to the entropy encoder, sharing code with subsequent passes */ michael@0: return compress_output(cinfo, input_buf); michael@0: } michael@0: michael@0: michael@0: /* michael@0: * Process some data in subsequent passes of a multi-pass case. michael@0: * We process the equivalent of one fully interleaved MCU row ("iMCU" row) michael@0: * per call, ie, v_samp_factor block rows for each component in the scan. michael@0: * The data is obtained from the virtual arrays and fed to the entropy coder. michael@0: * Returns TRUE if the iMCU row is completed, FALSE if suspended. michael@0: * michael@0: * NB: input_buf is ignored; it is likely to be a NULL pointer. michael@0: */ michael@0: michael@0: METHODDEF(boolean) michael@0: compress_output (j_compress_ptr cinfo, JSAMPIMAGE input_buf) michael@0: { michael@0: my_coef_ptr coef = (my_coef_ptr) cinfo->coef; michael@0: JDIMENSION MCU_col_num; /* index of current MCU within row */ michael@0: int blkn, ci, xindex, yindex, yoffset; michael@0: JDIMENSION start_col; michael@0: JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN]; michael@0: JBLOCKROW buffer_ptr; michael@0: jpeg_component_info *compptr; michael@0: michael@0: /* Align the virtual buffers for the components used in this scan. michael@0: * NB: during first pass, this is safe only because the buffers will michael@0: * already be aligned properly, so jmemmgr.c won't need to do any I/O. michael@0: */ michael@0: for (ci = 0; ci < cinfo->comps_in_scan; ci++) { michael@0: compptr = cinfo->cur_comp_info[ci]; michael@0: buffer[ci] = (*cinfo->mem->access_virt_barray) michael@0: ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index], michael@0: coef->iMCU_row_num * compptr->v_samp_factor, michael@0: (JDIMENSION) compptr->v_samp_factor, FALSE); michael@0: } michael@0: michael@0: /* Loop to process one whole iMCU row */ michael@0: for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; michael@0: yoffset++) { michael@0: for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row; michael@0: MCU_col_num++) { michael@0: /* Construct list of pointers to DCT blocks belonging to this MCU */ michael@0: blkn = 0; /* index of current DCT block within MCU */ michael@0: for (ci = 0; ci < cinfo->comps_in_scan; ci++) { michael@0: compptr = cinfo->cur_comp_info[ci]; michael@0: start_col = MCU_col_num * compptr->MCU_width; michael@0: for (yindex = 0; yindex < compptr->MCU_height; yindex++) { michael@0: buffer_ptr = buffer[ci][yindex+yoffset] + start_col; michael@0: for (xindex = 0; xindex < compptr->MCU_width; xindex++) { michael@0: coef->MCU_buffer[blkn++] = buffer_ptr++; michael@0: } michael@0: } michael@0: } michael@0: /* Try to write the MCU. */ michael@0: if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { michael@0: /* Suspension forced; update state counters and exit */ michael@0: coef->MCU_vert_offset = yoffset; michael@0: coef->mcu_ctr = MCU_col_num; michael@0: return FALSE; michael@0: } michael@0: } michael@0: /* Completed an MCU row, but perhaps not an iMCU row */ michael@0: coef->mcu_ctr = 0; michael@0: } michael@0: /* Completed the iMCU row, advance counters for next one */ michael@0: coef->iMCU_row_num++; michael@0: start_iMCU_row(cinfo); michael@0: return TRUE; michael@0: } michael@0: michael@0: #endif /* FULL_COEF_BUFFER_SUPPORTED */ michael@0: michael@0: michael@0: /* michael@0: * Initialize coefficient buffer controller. michael@0: */ michael@0: michael@0: GLOBAL(void) michael@0: jinit_c_coef_controller (j_compress_ptr cinfo, boolean need_full_buffer) michael@0: { michael@0: my_coef_ptr coef; michael@0: michael@0: coef = (my_coef_ptr) michael@0: (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE, michael@0: SIZEOF(my_coef_controller)); michael@0: cinfo->coef = (struct jpeg_c_coef_controller *) coef; michael@0: coef->pub.start_pass = start_pass_coef; michael@0: michael@0: /* Create the coefficient buffer. */ michael@0: if (need_full_buffer) { michael@0: #ifdef FULL_COEF_BUFFER_SUPPORTED michael@0: /* Allocate a full-image virtual array for each component, */ michael@0: /* padded to a multiple of samp_factor DCT blocks in each direction. */ michael@0: int ci; michael@0: jpeg_component_info *compptr; michael@0: michael@0: for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; michael@0: ci++, compptr++) { michael@0: coef->whole_image[ci] = (*cinfo->mem->request_virt_barray) michael@0: ((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE, michael@0: (JDIMENSION) jround_up((long) compptr->width_in_blocks, michael@0: (long) compptr->h_samp_factor), michael@0: (JDIMENSION) jround_up((long) compptr->height_in_blocks, michael@0: (long) compptr->v_samp_factor), michael@0: (JDIMENSION) compptr->v_samp_factor); michael@0: } michael@0: #else michael@0: ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); michael@0: #endif michael@0: } else { michael@0: /* We only need a single-MCU buffer. */ michael@0: JBLOCKROW buffer; michael@0: int i; michael@0: michael@0: buffer = (JBLOCKROW) michael@0: (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE, michael@0: C_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK)); michael@0: for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) { michael@0: coef->MCU_buffer[i] = buffer + i; michael@0: } michael@0: coef->whole_image[0] = NULL; /* flag for no virtual arrays */ michael@0: } michael@0: }