METHODDEF void
dump_scan_MCUs (compress_info_ptr cinfo, MCU_output_method_ptr output_method)
/* Dump the MCUs saved in whole_scan_MCUs to the output method. */
/* The method may be either the entropy encoder or some routine supplied */
/* by the entropy optimizer. */
{
/* On an 80x86 machine, the entropy encoder expects the passed data block
* to be in NEAR memory (for performance reasons), so we have to copy it
* back from the big array to a local array. On less brain-damaged CPUs
* we needn't do that.
*/
#ifdef NEED_FAR_POINTERS
JBLOCK MCU_data[MAX_BLOCKS_IN_MCU];
#endif
long mcurow, mcuindex, next_row;
int next_index;
JBLOCKARRAY rowptr = NULL; /* init only to suppress compiler complaint */
next_row = 0;
next_index = MCUs_in_big_row;
for (mcurow = 0; mcurow < cinfo->MCU_rows_in_scan; mcurow++) {
(*cinfo->methods->progress_monitor) (cinfo, mcurow,
cinfo->MCU_rows_in_scan);
for (mcuindex = 0; mcuindex < cinfo->MCUs_per_row; mcuindex++) {
if (next_index >= MCUs_in_big_row) {
rowptr = (*cinfo->emethods->access_big_barray) (whole_scan_MCUs,
next_row, FALSE);
next_row++;
next_index = 0;
}
#ifdef NEED_FAR_POINTERS
jcopy_block_row(rowptr[0] + next_index * cinfo->blocks_in_MCU,
(JBLOCKROW) MCU_data, /* casts near to far pointer! */
(long) cinfo->blocks_in_MCU);
(*output_method) (cinfo, MCU_data);
#else
(*output_method) (cinfo, rowptr[0] + next_index * cinfo->blocks_in_MCU);
#endif
next_index++;
}
}
cinfo->completed_passes++;
}
METHODDEF void
MCU_output_catcher (compress_info_ptr cinfo, JBLOCK *MCU_data)
/* Output method for siphoning off extract_MCUs output into a big array */
{
static JBLOCKARRAY rowptr;
if (next_MCU_index >= MCUs_in_big_row) {
rowptr = (*cinfo->emethods->access_big_barray) (whole_scan_MCUs,
next_whole_row, TRUE);
next_whole_row++;
next_MCU_index = 0;
}
/*
* note that on 80x86, the cast applied to MCU_data implies
* near to far pointer conversion.
*/
jcopy_block_row((JBLOCKROW) MCU_data,
rowptr[0] + next_MCU_index * cinfo->blocks_in_MCU,
(long) cinfo->blocks_in_MCU);
next_MCU_index++;
}
do_flip_v (j_decompress_ptr srcinfo, j_compress_ptr dstinfo,
jvirt_barray_ptr *src_coef_arrays,
jvirt_barray_ptr *dst_coef_arrays)
/* Vertical flip */
{
JDIMENSION MCU_rows, comp_height, dst_blk_x, dst_blk_y;
int ci, i, j, offset_y;
JBLOCKARRAY src_buffer, dst_buffer;
JBLOCKROW src_row_ptr, dst_row_ptr;
JCOEFPTR src_ptr, dst_ptr;
jpeg_component_info *compptr;
/* We output into a separate array because we can't touch different
* rows of the source virtual array simultaneously. Otherwise, this
* is a pretty straightforward analog of horizontal flip.
* Within a DCT block, vertical mirroring is done by changing the signs
* of odd-numbered rows.
* Partial iMCUs at the bottom edge are copied verbatim.
*/
MCU_rows = dstinfo->image_height / (dstinfo->max_v_samp_factor * DCTSIZE);
for (ci = 0; ci < dstinfo->num_components; ci++) {
compptr = dstinfo->comp_info + ci;
comp_height = MCU_rows * compptr->v_samp_factor;
for (dst_blk_y = 0; dst_blk_y < compptr->height_in_blocks;
dst_blk_y += compptr->v_samp_factor) {
dst_buffer = (*srcinfo->mem->access_virt_barray)
((j_common_ptr) srcinfo, dst_coef_arrays[ci], dst_blk_y,
(JDIMENSION) compptr->v_samp_factor, TRUE);
if (dst_blk_y < comp_height) {
/* Row is within the mirrorable area. */
src_buffer = (*srcinfo->mem->access_virt_barray)
((j_common_ptr) srcinfo, src_coef_arrays[ci],
comp_height - dst_blk_y - (JDIMENSION) compptr->v_samp_factor,
(JDIMENSION) compptr->v_samp_factor, FALSE);
} else {
/* Bottom-edge blocks will be copied verbatim. */
src_buffer = (*srcinfo->mem->access_virt_barray)
((j_common_ptr) srcinfo, src_coef_arrays[ci], dst_blk_y,
(JDIMENSION) compptr->v_samp_factor, FALSE);
}
for (offset_y = 0; offset_y < compptr->v_samp_factor; offset_y++) {
if (dst_blk_y < comp_height) {
/* Row is within the mirrorable area. */
dst_row_ptr = dst_buffer[offset_y];
src_row_ptr = src_buffer[compptr->v_samp_factor - offset_y - 1];
for (dst_blk_x = 0; dst_blk_x < compptr->width_in_blocks;
dst_blk_x++) {
dst_ptr = dst_row_ptr[dst_blk_x];
src_ptr = src_row_ptr[dst_blk_x];
for (i = 0; i < DCTSIZE; i += 2) {
/* copy even row */
for (j = 0; j < DCTSIZE; j++)
*dst_ptr++ = *src_ptr++;
/* copy odd row with sign change */
for (j = 0; j < DCTSIZE; j++)
*dst_ptr++ = - *src_ptr++;
}
}
} else {
/* Just copy row verbatim. */
jcopy_block_row(src_buffer[offset_y], dst_buffer[offset_y],
compptr->width_in_blocks);
}
}
}
}
}
decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf) {
my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
JDIMENSION block_num, last_block_column;
int ci, block_row, block_rows, access_rows;
JBLOCKARRAY buffer;
JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
JSAMPARRAY output_ptr;
JDIMENSION output_col;
jpeg_component_info *compptr;
inverse_DCT_method_ptr inverse_DCT;
boolean first_row, last_row;
JBLOCK workspace;
int *coef_bits;
JQUANT_TBL *quanttbl;
INT32 Q00, Q01, Q02, Q10, Q11, Q20, num;
int DC1, DC2, DC3, DC4, DC5, DC6, DC7, DC8, DC9;
int Al, pred;
/* Force some input to be done if we are getting ahead of the input. */
while (cinfo->input_scan_number <= cinfo->output_scan_number &&
!cinfo->inputctl->eoi_reached) {
if (cinfo->input_scan_number == cinfo->output_scan_number) {
/* If input is working on current scan, we ordinarily want it to
* have completed the current row. But if input scan is DC,
* we want it to keep one row ahead so that next block row's DC
* values are up to date.
*/
JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
if (cinfo->input_iMCU_row > cinfo->output_iMCU_row + delta)
break;
}
if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
return JPEG_SUSPENDED;
}
/* OK, output from the virtual arrays. */
for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
ci++, compptr++) {
/* Don't bother to IDCT an uninteresting component. */
if (!compptr->component_needed)
continue;
/* Count non-dummy DCT block rows in this iMCU row. */
if (cinfo->output_iMCU_row < last_iMCU_row) {
block_rows = compptr->v_samp_factor;
access_rows = block_rows * 2; /* this and next iMCU row */
last_row = FALSE;
} else {
/* NB: can't use last_row_height here; it is input-side-dependent! */
block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
if (block_rows == 0) block_rows = compptr->v_samp_factor;
access_rows = block_rows; /* this iMCU row only */
last_row = TRUE;
}
/* Align the virtual buffer for this component. */
if (cinfo->output_iMCU_row > 0) {
access_rows += compptr->v_samp_factor; /* prior iMCU row too */
buffer = (*cinfo->mem->access_virt_barray)
((j_common_ptr) cinfo, coef->whole_image[ci],
(cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
(JDIMENSION) access_rows, FALSE);
buffer += compptr->v_samp_factor; /* point to current iMCU row */
first_row = FALSE;
} else {
buffer = (*cinfo->mem->access_virt_barray)
((j_common_ptr) cinfo, coef->whole_image[ci],
(JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
first_row = TRUE;
}
/* Fetch component-dependent info */
coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
quanttbl = compptr->quant_table;
Q00 = quanttbl->quantval[0];
Q01 = quanttbl->quantval[Q01_POS];
Q10 = quanttbl->quantval[Q10_POS];
Q20 = quanttbl->quantval[Q20_POS];
Q11 = quanttbl->quantval[Q11_POS];
Q02 = quanttbl->quantval[Q02_POS];
inverse_DCT = cinfo->idct->inverse_DCT[ci];
output_ptr = output_buf[ci];
/* Loop over all DCT blocks to be processed. */
for (block_row = 0; block_row < block_rows; block_row++) {
buffer_ptr = buffer[block_row];
if (first_row && block_row == 0)
prev_block_row = buffer_ptr;
else
prev_block_row = buffer[block_row - 1];
if (last_row && block_row == block_rows - 1)
next_block_row = buffer_ptr;
else
next_block_row = buffer[block_row + 1];
/* We fetch the surrounding DC values using a sliding-register approach.
* Initialize all nine here so as to do the right thing on narrow pics.
*/
DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
DC7 = DC8 = DC9 = (int) next_block_row[0][0];
output_col = 0;
last_block_column = compptr->width_in_blocks - 1;
for (block_num = 0; block_num <= last_block_column; block_num++) {
/* Fetch current DCT block into workspace so we can modify it. */
jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
/* Update DC values */
if (block_num < last_block_column) {
DC3 = (int) prev_block_row[1][0];
DC6 = (int) buffer_ptr[1][0];
DC9 = (int) next_block_row[1][0];
}
/* Compute coefficient estimates per K.8.
* An estimate is applied only if coefficient is still zero,
* and is not known to be fully accurate.
*/
/* AC01 */
if ((Al = coef_bits[1]) != 0 && workspace[1] == 0) {
num = 36 * Q00 * (DC4 - DC6);
if (num >= 0) {
pred = (int) (((Q01 << 7) + num) / (Q01 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
} else {
pred = (int) (((Q01 << 7) - num) / (Q01 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
pred = -pred;
}
workspace[1] = (JCOEF) pred;
}
/* AC10 */
if ((Al = coef_bits[2]) != 0 && workspace[8] == 0) {
num = 36 * Q00 * (DC2 - DC8);
if (num >= 0) {
pred = (int) (((Q10 << 7) + num) / (Q10 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
} else {
pred = (int) (((Q10 << 7) - num) / (Q10 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
pred = -pred;
}
workspace[8] = (JCOEF) pred;
}
/* AC20 */
if ((Al = coef_bits[3]) != 0 && workspace[16] == 0) {
num = 9 * Q00 * (DC2 + DC8 - 2 * DC5);
if (num >= 0) {
pred = (int) (((Q20 << 7) + num) / (Q20 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
} else {
pred = (int) (((Q20 << 7) - num) / (Q20 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
pred = -pred;
}
workspace[16] = (JCOEF) pred;
}
/* AC11 */
if ((Al = coef_bits[4]) != 0 && workspace[9] == 0) {
num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
if (num >= 0) {
pred = (int) (((Q11 << 7) + num) / (Q11 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
} else {
pred = (int) (((Q11 << 7) - num) / (Q11 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
pred = -pred;
}
workspace[9] = (JCOEF) pred;
}
/* AC02 */
if ((Al = coef_bits[5]) != 0 && workspace[2] == 0) {
num = 9 * Q00 * (DC4 + DC6 - 2 * DC5);
if (num >= 0) {
pred = (int) (((Q02 << 7) + num) / (Q02 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
} else {
pred = (int) (((Q02 << 7) - num) / (Q02 << 8));
if (Al > 0 && pred >= (1 << Al))
pred = (1 << Al) - 1;
pred = -pred;
}
workspace[2] = (JCOEF) pred;
}
/* OK, do the IDCT */
(*inverse_DCT)(cinfo, compptr, (JCOEFPTR) workspace,
output_ptr, output_col);
/* Advance for next column */
DC1 = DC2;
DC2 = DC3;
DC4 = DC5;
DC5 = DC6;
DC7 = DC8;
DC8 = DC9;
buffer_ptr++, prev_block_row++, next_block_row++;
output_col += compptr->DCT_scaled_size;
}
output_ptr += compptr->DCT_scaled_size;
}
}
if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
return JPEG_ROW_COMPLETED;
return JPEG_SCAN_COMPLETED;
}
METHODDEF void
smooth_coefficients (decompress_info_ptr cinfo,
jpeg_component_info *compptr,
JBLOCKROW above,
JBLOCKROW currow,
JBLOCKROW below,
JBLOCKROW output)
{
QUANT_TBL_PTR Qptr = cinfo->quant_tbl_ptrs[compptr->quant_tbl_no];
long blocks_in_row = compptr->downsampled_width / DCTSIZE;
long col;
/* First, copy the block row as-is.
* This takes care of the first & last blocks in the row, the top/bottom
* special cases, and the higher-order coefficients in each block.
*/
jcopy_block_row(currow, output, blocks_in_row);
/* Now apply the smoothing calculation, but not to any blocks on the
* edges of the image.
*/
if (above != NULL && below != NULL) {
for (col = 1; col < blocks_in_row-1; col++) {
/* See section K.8 of the JPEG standard.
*
* As I understand it, this produces approximations
* for the low frequency AC components, based on the
* DC values of the block and its eight neighboring blocks.
* (Thus it can't be used for blocks on the image edges.)
*/
/* The layout of these variables corresponds to text and figure in K.8 */
JCOEF DC1, DC2, DC3;
JCOEF DC4, DC5, DC6;
JCOEF DC7, DC8, DC9;
long AC01, AC02;
long AC10, AC11;
long AC20;
DC1 = above [col-1][0];
DC2 = above [col ][0];
DC3 = above [col+1][0];
DC4 = currow[col-1][0];
DC5 = currow[col ][0];
DC6 = currow[col+1][0];
DC7 = below [col-1][0];
DC8 = below [col ][0];
DC9 = below [col+1][0];
#define DIVIDE_256(x) x = ( (x) < 0 ? -((128-(x))/256) : ((x)+128)/256 )
AC01 = (36 * (DC4 - DC6));
DIVIDE_256(AC01);
AC10 = (36 * (DC2 - DC8));
DIVIDE_256(AC10);
AC20 = (9 * (DC2 + DC8 - 2*DC5));
DIVIDE_256(AC20);
AC11 = (5 * ((DC1 - DC3) - (DC7 - DC9)));
DIVIDE_256(AC11);
AC02 = (9 * (DC4 + DC6 - 2*DC5));
DIVIDE_256(AC02);
/* I think that this checks to see if the quantisation
* on the transmitting side would have produced this
* answer. If so, then we use our (hopefully better)
* estimate.
*/
#define ABS(x) ((x) < 0 ? -(x) : (x))
#define COND_ASSIGN(_ac,_n,_z) if ((ABS(output[col][_n] - (_ac))<<1) <= Qptr[_z]) output[col][_n] = (JCOEF) (_ac)
COND_ASSIGN(AC01, 1, 1);
COND_ASSIGN(AC02, 2, 5);
COND_ASSIGN(AC10, 8, 2);
COND_ASSIGN(AC11, 9, 4);
COND_ASSIGN(AC20, 16, 3);
}
}
}
