irender_proc_t gs_image_class_1_simple(gx_image_enum * penum) { irender_proc_t rproc; fixed ox = dda_current(penum->dda.pixel0.x); fixed oy = dda_current(penum->dda.pixel0.y); if (penum->use_rop || penum->spp != 1 || penum->bps != 1) return 0; switch (penum->posture) { case image_portrait: { /* Use fast portrait algorithm. */ long dev_width = fixed2long_pixround(ox + penum->x_extent.x) - fixed2long_pixround(ox); if (dev_width != penum->rect.w) { /* * Add an extra align_bitmap_mod of padding so that * we can align scaled rows with the device. */ long line_size = bitmap_raster(any_abs(dev_width)) + align_bitmap_mod; if (penum->adjust != 0 || line_size > max_uint) return 0; /* Must buffer a scan line. */ penum->line_width = any_abs(dev_width); penum->line_size = (uint) line_size; penum->line = gs_alloc_bytes(penum->memory, penum->line_size, "image line"); if (penum->line == 0) { gx_default_end_image(penum->dev, (gx_image_enum_common_t *)penum, false); return 0; } } if_debug2('b', "[b]render=simple, unpack=copy; rect.w=%d, dev_width=%ld\n", penum->rect.w, dev_width); rproc = image_render_simple; break; } case image_landscape: { /* Use fast landscape algorithm. */ long dev_width = fixed2long_pixround(oy + penum->x_extent.y) - fixed2long_pixround(oy); long line_size = (dev_width = any_abs(dev_width), bitmap_raster(dev_width) * 8 + ROUND_UP(dev_width, 8) * align_bitmap_mod); if ((dev_width != penum->rect.w && penum->adjust != 0) || line_size > max_uint ) return 0; /* Must buffer a group of 8N scan lines. */ penum->line_width = dev_width; penum->line_size = (uint) line_size; penum->line = gs_alloc_bytes(penum->memory, penum->line_size, "image line"); if (penum->line == 0) { gx_default_end_image(penum->dev, (gx_image_enum_common_t *) penum, false); return 0; } penum->xi_next = penum->line_xy = fixed2int_var_rounded(ox); if_debug3('b', "[b]render=landscape, unpack=copy; rect.w=%d, dev_width=%ld, line_size=%ld\n", penum->rect.w, dev_width, line_size); rproc = image_render_landscape; /* Precompute values needed for rasterizing. */ penum->dxy = float2fixed(penum->matrix.xy + fixed2float(fixed_epsilon) / 2); break; } default: return 0; } /* Precompute values needed for rasterizing. */ penum->dxx = float2fixed(penum->matrix.xx + fixed2float(fixed_epsilon) / 2); /* * We don't want to spread the samples, but we have to reset unpack_bps * to prevent the buffer pointer from being incremented by 8 bytes per * input byte. */ penum->unpack = sample_unpack_copy; penum->unpack_bps = 8; if (penum->use_mask_color) { /* * Set the masked color as 'no_color' to make it transparent * according to the mask color range and the decoding. */ penum->masked = true; if (penum->mask_color.values[0] == 1) { /* if v0 == 1, 1 is transparent since v1 must be == 1 to be a valid range */ set_nonclient_dev_color(penum->map[0].inverted ? penum->icolor0 : penum->icolor1, gx_no_color_index); } else if (penum->mask_color.values[1] == 0) { /* if v1 == 0, 0 is transparent since v0 must be == 0 to be a valid range */ set_nonclient_dev_color(penum->map[0].inverted ? penum->icolor1 : penum->icolor0, gx_no_color_index); } else { /* * The only other possible in-range value is v0 = 0, v1 = 1. * The image is completely transparent! */ rproc = image_render_skip; } penum->map[0].decoding = sd_none; } return rproc; }
int gxht_thresh_image_init(gx_image_enum *penum) { int code = 0; fixed ox, oy; int temp; int dev_width, max_height; int spp_out; int k; gx_ht_order *d_order; if (gx_device_must_halftone(penum->dev)) { if (penum->pis != NULL && penum->pis->dev_ht != NULL) { for (k = 0; k < penum->pis->dev_ht->num_comp; k++) { d_order = &(penum->pis->dev_ht->components[k].corder); code = gx_ht_construct_threshold(d_order, penum->dev, penum->pis, k); if (code < 0 ) { return gs_rethrow(code, "threshold creation failed"); } } } else { return -1; } } spp_out = penum->dev->color_info.num_components; /* If the image is landscaped then we want to maintain a buffer that is sufficiently large so that we can hold a byte of halftoned data along the column. This way we avoid doing multiple writes into the same position over and over. The size of the buffer we need depends upon the bitdepth of the output device, the number of device coloranants and the number of colorants in the source space. Note we will need to eventually consider multi-level halftone case here too. For now, to make use of the SSE2 stuff, we would like to have 16 bytes of data to process at a time. So we will collect the columns of data in a buffer that is 16 wide. We will also keep track of the widths of each column. When the total width count reaches 16, we will create our threshold array and apply it. We may have one column that is buffered between calls in this case. Also if a call is made with h=0 we will flush the buffer as we are at the end of the data. */ if (penum->posture == image_landscape) { int col_length = fixed2int_var_rounded(any_abs(penum->x_extent.y)) * spp_out; ox = dda_current(penum->dda.pixel0.x); oy = dda_current(penum->dda.pixel0.y); temp = (int) ceil((float) col_length/16.0); penum->line_size = temp * 16; /* The stride */ /* Now we need at most 16 of these */ penum->line = gs_alloc_bytes(penum->memory, 16 * penum->line_size + 16, "gxht_thresh"); /* Same with this */ penum->thresh_buffer = gs_alloc_bytes(penum->memory, penum->line_size * 16 + 16, "gxht_thresh"); /* That maps into 2 bytes of Halftone data */ penum->ht_buffer = gs_alloc_bytes(penum->memory, penum->line_size * 2, "gxht_thresh"); penum->ht_stride = penum->line_size; if (penum->line == NULL || penum->thresh_buffer == NULL || penum->ht_buffer == NULL) return -1; penum->ht_landscape.count = 0; penum->ht_landscape.num_contones = 0; if (penum->y_extent.x < 0) { /* Going right to left */ penum->ht_landscape.curr_pos = 15; penum->ht_landscape.index = -1; } else { /* Going left to right */ penum->ht_landscape.curr_pos = 0; penum->ht_landscape.index = 1; } if (penum->x_extent.y < 0) { penum->ht_landscape.flipy = true; penum->ht_landscape.y_pos = fixed2int_pixround_perfect(dda_current(penum->dda.pixel0.y) + penum->x_extent.y); } else { penum->ht_landscape.flipy = false; penum->ht_landscape.y_pos = fixed2int_pixround_perfect(dda_current(penum->dda.pixel0.y)); } memset(&(penum->ht_landscape.widths[0]), 0, sizeof(int)*16); penum->ht_landscape.offset_set = false; penum->ht_offset_bits = 0; /* Will get set in call to render */ if (code >= 0) { #if defined(DEBUG) || defined(PACIFY_VALGRIND) memset(penum->line, 0, 16 * penum->line_size + 16); memset(penum->ht_buffer, 0, penum->line_size * 2); memset(penum->thresh_buffer, 0, 16 * penum->line_size + 16); #endif } } else { /* In the portrait case we allocate a single line buffer in device width, a threshold buffer of the same size and possibly wider and the buffer for the halftoned bits. We have to do a bit of work to enable 16 byte boundary after an offset to ensure that we can make use of the SSE2 operations for thresholding. We do the allocations now to avoid doing them with every line */ /* Initialize the ht_landscape stuff to zero */ memset(&(penum->ht_landscape), 0, sizeof(ht_landscape_info_t)); ox = dda_current(penum->dda.pixel0.x); oy = dda_current(penum->dda.pixel0.y); dev_width = (int) fabs((long) fixed2long_pixround(ox + penum->x_extent.x) - fixed2long_pixround(ox)); /* Get the bit position so that we can do a copy_mono for the left remainder and then 16 bit aligned copies for the rest. The right remainder will be OK as it will land in the MSBit positions. Note the #define chunk bits16 in gdevm1.c. Allow also for a 15 sample over run. */ penum->ht_offset_bits = (-fixed2int_var_pixround(ox)) & 15; if (penum->ht_offset_bits > 0) { penum->ht_stride = ((7 + (dev_width + 4)) / 8) + ARCH_SIZEOF_LONG; } else { penum->ht_stride = ((7 + (dev_width + 2)) / 8) + ARCH_SIZEOF_LONG; } /* We want to figure out the maximum height that we may have in taking a single source row and going to device space */ max_height = (int) ceil(fixed2float(any_abs(penum->dst_height)) / (float) penum->Height); penum->ht_buffer = gs_alloc_bytes(penum->memory, penum->ht_stride * max_height * spp_out, "gxht_thresh"); /* We want to have 128 bit alignement for our contone and threshold strips so that we can use SSE operations in the threshold operation. Add in a minor buffer and offset to ensure this. If gs_alloc_bytes provides at least 16 bit alignment so we may need to move 14 bytes. However, the HT process is split in two operations. One that involves the HT of a left remainder and the rest which ensures that we pack in the HT data in the bits with no skew for a fast copy into the gdevm1 device (16 bit copies). So, we need to account for those pixels which occur first and which are NOT aligned for the contone buffer. After we offset by this remainder portion we should be 128 bit aligned. Also allow a 15 sample over run during the execution. */ temp = (int) ceil((float) ((dev_width + 15.0) + 15.0)/16.0); penum->line_size = temp * 16; /* The stride */ penum->line = gs_alloc_bytes(penum->memory, penum->line_size * spp_out, "gxht_thresh"); penum->thresh_buffer = gs_alloc_bytes(penum->memory, penum->line_size * max_height * spp_out, "gxht_thresh"); if (penum->line == NULL || penum->thresh_buffer == NULL || penum->ht_buffer == NULL) { return -1; } else { #if defined(DEBUG) || defined(PACIFY_VALGRIND) memset(penum->line, 0, penum->line_size * spp_out); memset(penum->ht_buffer, 0, penum->ht_stride * max_height * spp_out); memset(penum->thresh_buffer, 0, penum->line_size * max_height * spp_out); #endif } } /* Precompute values needed for rasterizing. */ penum->dxx = float2fixed(penum->matrix.xx + fixed2float(fixed_epsilon) / 2); return code; }