tor%cs.brown.edu e44584673d Bug 68970 - tune libimg throttling constants for better performance.
r=saari, sr=brendan


git-svn-id: svn://10.0.0.236/trunk@87297 18797224-902f-48f8-a5cc-f745e15eee43
2001-02-18 01:05:45 +00:00

1295 lines
44 KiB
C++

/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*-
*
* The contents of this file are subject to the Netscape Public
* License Version 1.1 (the "License"); you may not use this file
* except in compliance with the License. You may obtain a copy of
* the License at http://www.mozilla.org/NPL/
*
* Software distributed under the License is distributed on an "AS
* IS" basis, WITHOUT WARRANTY OF ANY KIND, either express or
* implied. See the License for the specific language governing
* rights and limitations under the License.
*
* The Original Code is mozilla.org code.
*
* The Initial Developer of the Original Code is Netscape
* Communications Corporation. Portions created by Netscape are
* Copyright (C) 1998 Netscape Communications Corporation. All
* Rights Reserved.
*
* Contributor(s):
*/
/* -*- Mode: C; tab-width: 4 -*-
scale.c --- Controls rendering of scan lines to screen, including scaling
and transparency
*/
#include "if.h" /* Image library internal declarations */
#include "il.h" /* Image library external API */
#include "nsCRT.h"
#include "xpcompat.h" //temporary, for timers
#include "nsVoidArray.h"
#include "nsITimer.h"
#include "nsITimerCallback.h"
/* Approximate size of pixel data chunks sent to the FE for display */
#define OUTPUT_CHUNK_SIZE 150000
/* Delay from decode to display of first scanline, in milliseconds. */
#define ROW_OUTPUT_INITIAL_DELAY 32
/* Delays between subsequent sets of scanlines */
#define ROW_OUTPUT_DELAY 300
/* for png */
typedef struct _IL_IRGBGA {
PRUint8 index;
PRUint8 red, green, blue, gray, alpha;
} IL_IRGBGA;
static void
il_timeout_callback(void *closure)
{
int delay;
il_container *ic = (il_container *)closure;
NI_PixmapHeader *img_header = &ic->image->header;
ic->row_output_timeout = NULL;
if (ic->state == IC_ABORT_PENDING)
return;
/*
* Don't schedule any more timeouts once we start decoding the
* second image in a multipart image sequence. Instead display
* will take place when the entire image is decoded.
*/
if (ic->multi &&
((PRUint32)img_header->height * img_header->width < 100000)) {
return;
}
il_flush_image_data(ic);
delay = (ic->pass > 1) ? 2 * ROW_OUTPUT_DELAY : ROW_OUTPUT_DELAY;
ic->row_output_timeout = IL_SetTimeout(il_timeout_callback, ic, delay);
}
/*-----------------------------------------------------------------------------
* Display a specified number of rows of pixels for the purpose of
* progressive image display. The data is accumulated without forwarding
* it to the front-end for display until either the row-output timeout
* fires or the image is fully decoded.
*---------------------------------------------------------------------------*/
void
il_partial(
il_container *ic, /* The image container */
int row, /* Starting row; zero is top row of image */
int row_count, /* Number of rows to output, including starting row */
int pass) /* Zero, unless interlaced GIF,
in which case ranges 1-4, or progressive JPEG,
in which case ranges from 1-n. */
{
NI_PixmapHeader *img_header = &ic->image->header;
if (!ic->new_data_for_fe) {
ic->update_start_row = row;
ic->update_end_row = row + row_count - 1;
ic->new_data_for_fe = PR_TRUE;
} else {
if (row < ic->update_start_row)
ic->update_start_row = row;
if ((row + row_count - 1) > ic->update_end_row)
ic->update_end_row = row + row_count - 1;
}
ic->pass = pass;
if (ic->img_cx->progressive_display)
{
#ifdef XP_WIN
/* The last pass of an image is displayed with less delay. */
if (!ic->multi && (pass == IL_FINAL_PASS))
#else
/* The first and last pass of an image are displayed with less
delay. */
if (!ic->multi && ((pass <= 1) || (pass == IL_FINAL_PASS)))
#endif
{
int num_rows = ic->update_end_row - ic->update_start_row + 1;
if (num_rows * img_header->widthBytes > OUTPUT_CHUNK_SIZE)
il_flush_image_data(ic);
}
/*
* Don't schedule any more timeouts once we start decoding the
* second image in a multipart image sequence. Instead display
* will take place when the entire image is decoded.
*/
if (ic->multi &&
((PRUint32)img_header->height * img_header->width < 100000)) {
return;
}
if (!ic->row_output_timeout){
/* Set a timer that will actually display the image data. */
ic->row_output_timeout = IL_SetTimeout(il_timeout_callback, ic,
ROW_OUTPUT_INITIAL_DELAY);
}
}
}
/*-----------------------------------------------------------------------------
* Force the front-end to to display any lines in the image bitmap
* that have been decoded, but haven't yet been sent to the screen.
* (Progressively displayed images are normally displayed several
* lines at a time for efficiency. This routine flushes out the last
* few undisplayed lines in the image.)
*---------------------------------------------------------------------------*/
void
il_flush_image_data(il_container *ic)
{
IL_GroupContext *img_cx = ic->img_cx;
IL_Pixmap *image = ic->image;
IL_Pixmap *mask = ic->mask;
NI_PixmapHeader *img_header = &image->header;
NI_PixmapHeader *mask_header = mask ? &mask->header : NULL;
int row, start_row, end_row, row_interval;
/* If we never called il_size(), we have no data for the FE. There
may also be no new data if a previous flush has occurred. */
if (!image->haveBits || !ic->new_data_for_fe)
return;
start_row = ic->update_start_row;
end_row = ic->update_end_row;
row_interval = (2 * OUTPUT_CHUNK_SIZE) / img_header->widthBytes;
row = start_row;
#ifdef XP_UNIX
/* If the amount of image data becomes really large, break it
* up into chunks to BLT out to the screen. Otherwise, there
* can be a noticeable delay as the FE processes a large image.
* (In the case of the XFE, it can take a long time to send
* it to the server.)
*/
for (;row < (end_row - row_interval); row += row_interval) {
img_cx->img_cb->UpdatePixmap(img_cx->dpy_cx, image, 0, row,
img_header->width, row_interval);
if (mask) {
img_cx->img_cb->UpdatePixmap(img_cx->dpy_cx, mask, 0, row,
mask_header->width, row_interval);
}
}
#endif /* XP_UNIX */
/* Draw whatever is leftover after sending the chunks */
img_cx->img_cb->UpdatePixmap(img_cx->dpy_cx, image,
0, row, img_header->width, end_row - row + 1);
if (mask) {
img_cx->img_cb->UpdatePixmap(img_cx->dpy_cx, mask, 0, row,
mask_header->width, end_row - row + 1);
}
/* Update the displayable area of the pixmap. */
ic->displayable_rect.x_origin = 0;
ic->displayable_rect.y_origin = 0;
ic->displayable_rect.width = (PRUint16)img_header->width;
ic->displayable_rect.height = MAX(ic->displayable_rect.height,
end_row + 1);
img_cx->img_cb->SetDecodedRect(image, 0, 0,
ic->displayable_rect.width, ic->displayable_rect.height);
/* Notify observers that the image pixmap has been updated. */
il_pixmap_update_notify(ic);
/* Notify observers of image progress. */
il_progress_notify(ic);
ic->new_data_for_fe = PR_FALSE;
ic->update_end_row = ic->update_start_row = 0;
}
/* Copy a packed RGB triple */
#define COPY_RGB(src, dest) \
{dest[0] = src[0]; dest[1] = src[1]; dest[2] = src[2];}
/*-----------------------------------------------------------------------------
* Scale a row of packed RGB pixels using the Bresenham algorithm.
* Output is also packed RGB pixels.
*---------------------------------------------------------------------------*/
static void
il_scale_RGB_row(
PRUint8 XP_HUGE *src, /* Source row of packed RGB pixels */
int src_len, /* Number of pixels in source row */
PRUint8 *dest, /* Destination, packed RGB pixels */
int dest_len) /* Length of target row, in pixels */
{
PRUint8 *dest_end = dest + (3 * dest_len);
int n = 0;
PR_ASSERT(dest);
PR_ASSERT(src_len != dest_len);
/* Two cases */
/* Scaling down ? ... */
if (src_len > dest_len)
{
while (dest < dest_end) {
COPY_RGB(src, dest);
dest += 3;
n += src_len;
while (n >= dest_len) {
src += 3;
n -= dest_len;
}
}
}
else
/* Scaling up. */
{
while (dest < dest_end) {
n += dest_len;
while (n >= src_len) {
COPY_RGB(src, dest);
dest += 3;
n -= src_len;
}
src += 3;
}
}
}
#ifdef M12N
/*-----------------------------------------------------------------------------
* Scale a row of single-byte pixels using the Bresenham algorithm.
* Output is also single-byte pixels.
*---------------------------------------------------------------------------*/
static void
il_scale_CI_row(
PRUint8 XP_HUGE *src, /* Source row of packed RGB pixels */
int src_len, /* Number of pixels in source row */
PRUint8 *dest, /* Destination, packed RGB pixels */
int dest_len, /* Length of target row, in pixels */
PRUint8* indirect_map,/* image-to-FE color mapping */
int transparent_pixel_color)
{
int src_pixel, mapped_src_pixel;
PRUint8 *dest_end = dest + dest_len;
int n = 0;
PR_ASSERT(dest);
PR_ASSERT(src_len != dest_len);
/* Two cases */
/* Scaling down ? ... */
if (src_len > dest_len)
{
while (dest < dest_end) {
if (*src != transparent_pixel_color)
*dest = indirect_map[*src];
dest ++;
n += src_len;
while (n >= dest_len) {
src ++;
n -= dest_len;
}
}
}
else
/* Scaling up. */
{
while (dest < dest_end) {
n += dest_len;
src_pixel = *src;
mapped_src_pixel = indirect_map[src_pixel];
while (n >= src_len) {
if (src_pixel != transparent_pixel_color)
*dest = mapped_src_pixel;
dest ++;
n -= src_len;
}
src++;
}
}
}
#endif /* M12N */
/* Convert row coordinate from image space to display space. */
#define SCALE_YCOORD(ih, sh, y) \
((int)((PRUint32)(y) * (ih)->height / (sh)->height))
#define SCALE_XCOORD(ih, sh, x) \
((int)((PRUint32)(x) * (ih)->width / (sh)->width))
/* Add a bit to the row of mask bits. Flush accumulator to memory if full. */
#define SHIFT_IMAGE_MASK(not_transparent_flag) \
{ \
fgmask32 |= ((PRUint32)not_transparent_flag ) << M32(mask_bit); \
bgmask32 |= ((PRUint32)not_transparent_flag ^ 1) << M32(mask_bit); \
\
/* Filled up 32-bit mask word. Write it to memory. */ \
if (mask_bit-- == 0) { \
PRUint32 mtmp = *m; \
mtmp |= fgmask32; \
if (draw_mode == ilErase) \
mtmp &= ~bgmask32; \
*m++ = mtmp; \
mask_bit = 31; \
bgmask32 = 0; \
fgmask32 = 0; \
} \
output_bits_remaining--; \
}
/*-----------------------------------------------------------------------------
*
* Create a 1 bit mask bitmap from alpha channel.
* Accumulate the mask in 32-bit chunks for efficiency.
*******
A Special thanks to Glenn Randers-Pehrson <randeg@alum.rpi.edu> for
contributing a nice, fast dithering function, the
"GlennRP_random_dither_alpha" function, to convert
8 bit alpha masks to 1bit dithered masks for use where 8 bit
masks are not supported.
04-11-2000. pnunn.
********
*---------------------------------------------------------------------------*/
static void
il_alpha_mask(
PRUint8 *src, /* RGBa, input data */
int src_len, /* Number of pixels in source row */
int x_offset, /* Destination offset from left edge */
PRUint8 XP_HUGE *maskp, /* Output pointer, left-justified bitmask */
int mask_len, /* Number of pixels in output row */
il_draw_mode draw_mode) /* ilOverlay or ilErase */
{
int not_transparent,n =0;
int output_bits_remaining = mask_len;
PRUint32 bgmask32 = 0; /* 32-bit temporary mask accumulators */
PRUint32 fgmask32 = 0;
int mask_bit; /* next bit to write in setmask32 */
PRUint32 *m = ((PRUint32*)maskp) + (x_offset >> 5);
mask_bit = ~x_offset & 0x1f;
PR_ASSERT(mask_len);
// mask_len = (int)mask_len;
/* Handle case in which we have a mask for a non-transparent
image. This can happen when we have a LOSRC that is a transparent
GIF and a SRC that is a JPEG. For now, we avoid crashing. Later
we should fix that case so it does the right thing and gets rid
of the mask. */
if (!src)
return;
/* Scaling down (or source and dest same size) ... */
if (src_len >= mask_len)
{
while (output_bits_remaining ) {
int tmprand=0;
tmprand = 1+(int)(200.0*rand()/(RAND_MAX+1.0));
not_transparent = (*(src+3) > tmprand);
if(!not_transparent) { *src = 0; src[1] = 0; src[2] = 0; }
SHIFT_IMAGE_MASK(not_transparent);
n += src_len;
while ( n >= mask_len){
src += 4;
n -= mask_len;
}
}
}
else
/* Scaling up */
{
while (output_bits_remaining) {
int tmprand=0;
n += mask_len;
while (n >= src_len) {
tmprand = 1+(int)(200.0*rand()/(RAND_MAX+1.0));
not_transparent = (*(src+3) > tmprand);
if(!not_transparent) { *src = 0; src[1] = 0; src[2] = 0; }
SHIFT_IMAGE_MASK(not_transparent);
n -= src_len;
}
src +=4;
}
}
/* End of scan line. Write out any remaining mask bits. */
if (mask_bit < 31) {
PRUint32 mtmp = *m;
mtmp |= fgmask32;
if (draw_mode == ilErase)
mtmp &= ~bgmask32;
*m = mtmp;
}
}
/*-----------------------------------------------------------------------------
* Create a transparency mask bitmap. Perform horizontal scaling if
* requested using a Bresenham algorithm. Accumulate the mask in
* 32-bit chunks for efficiency.
*---------------------------------------------------------------------------*/
static void
il_generate_scaled_transparency_mask(
IL_IRGB *transparent_pixel, /* The transparent pixel */
PRUint8 *src, /* Row of pixels, 8-bit pseudocolor data */
int src_len, /* Number of pixels in source row */
int x_offset, /* Destination offset from left edge */
PRUint8 XP_HUGE *maskp, /* Output pointer, left-justified bitmask */
int mask_len, /* Number of pixels in output row */
il_draw_mode draw_mode) /* ilOverlay or ilErase */
{
int not_transparent, n = 0;
int src_trans_pixel_index =
transparent_pixel ? transparent_pixel->index : -1;
int output_bits_remaining = mask_len;
PRUint32 bgmask32 = 0; /* 32-bit temporary mask accumulators */
PRUint32 fgmask32 = 0;
int mask_bit; /* next bit to write in setmask32 */
PRUint32 *m = ((PRUint32*)maskp) + (x_offset >> 5);
mask_bit = ~x_offset & 0x1f;
PR_ASSERT(maskp);
/* Handle case in which we have a mask for a non-transparent
image. This can happen when we have a LOSRC that is a transparent
GIF and a SRC that is a JPEG. For now, we avoid crashing. Later
we should fix that case so it does the right thing and gets rid
of the mask. */
if (!src)
return;
/* Two cases */
/* Scaling down ? (or source and dest same size) ... */
if (src_len >= mask_len)
{
while (output_bits_remaining) {
not_transparent = (*src != src_trans_pixel_index);
SHIFT_IMAGE_MASK(not_transparent);
n += src_len;
while (n >= mask_len) {
src++;
n -= mask_len;
}
}
}
else
/* Scaling up */
{
while (output_bits_remaining) {
n += mask_len;
not_transparent = (*src != src_trans_pixel_index);
while (n >= src_len) {
SHIFT_IMAGE_MASK(not_transparent);
n -= src_len;
}
src++;
}
}
/* End of scan line. Write out any remaining mask bits. */
if (mask_bit < 31) {
PRUint32 mtmp = *m;
mtmp |= fgmask32;
if (draw_mode == ilErase)
mtmp &= ~bgmask32;
*m = mtmp;
}
}
/*-----------------------------------------------------------------------------
* Scale 1-bit transparency mask:
* Perform horizontal scaling if requested using a Bresenham algorithm.
* Accumulate the mask in 32-bit chunks for efficiency.
*---------------------------------------------------------------------------*/
static void
il_scale_mask(
PRUint8 *src, /* input mask data */
int src_len, /* Number of pixels in source row */
int x_offset, /* Destination offset from left edge */
PRUint8 XP_HUGE *maskp, /* Output pointer, left-justified bitmask */
int mask_len, /* Number of pixels in output row */
il_draw_mode draw_mode) /* ilOverlay or ilErase */
{
int not_transparent,n = 0;
int output_bits_remaining = mask_len;
PRUint32 bgmask32 = 0; /* 32-bit temporary mask accumulators */
PRUint32 fgmask32 = 0;
PRUint8 src_bit; /* next bit to read in setmask32 */
int mask_bit; /* next bit to write in setmask32 */
PRUint32 *m = ((PRUint32*)maskp) + (x_offset >> 5);
mask_bit = ~x_offset & 0x1f;
src_bit = 0x07;
PR_ASSERT(mask_len);
/* Handle case in which we have a mask for a non-transparent
image. This can happen when we have a LOSRC that is a transparent
GIF and a SRC that is a JPEG. For now, we avoid crashing. Later
we should fix that case so it does the right thing and gets rid
of the mask. */
if (!src)
return;
if (src_len >= mask_len) /* Scaling down */
{
while (output_bits_remaining ) {
not_transparent = ((*src & ((PRUint8)1 << src_bit)) != 0);
SHIFT_IMAGE_MASK(not_transparent);
n += src_len;
while ( n >= mask_len){
if (src_bit-- == 0){
src ++;
src_bit = 7;
}
n -= mask_len;
}
}
} else /* Scaling up */
{
while (output_bits_remaining){
n += mask_len;
not_transparent = ((*src & ((PRUint8)1 << src_bit)) != 0);
while (n >= src_len){
SHIFT_IMAGE_MASK(not_transparent);
n -= src_len;
}
if (src_bit-- == 0){
src ++;
src_bit = 7;
}
}
}
/* End of scan line. Write out any remaining mask bits. */
if (mask_bit < 31){
PRUint32 mtmp = *m;
mtmp |= fgmask32;
if (draw_mode == ilErase)
mtmp &= ~bgmask32;
*m = mtmp;
}
}
#undef SHIFT_IMAGE_MASK
/*-----------------------------------------------------------------------------
* When color quantization (possibly accompanied by dithering) takes
* place, the background pixels in a transparent image that overlays a
* solid-color background, e.g. <BODY BGCOLOR=#c5c5c5>, will get
* mapped to a color in the color-cube. The real background color,
* however, may not be one of these colors reserved for images. This
* routine serves to return transparent pixels to their background
* color. This routine must performing scaling because the source
* pixels are in the image space and the target pixels are in the
* display space.
*---------------------------------------------------------------------------*/
static void
il_reset_background_pixels(
il_container *ic, /* The image container */
PRUint8 *src, /* Row of pixels, 8-bit pseudocolor data */
int src_len, /* Number of pixels in row */
PRUint8 XP_HUGE *dest, /* Output pointer, 8-bit pseudocolor data */
int dest_len) /* Width of output pixel row */
{
int is_transparent, n = 0;
PRUint8 XP_HUGE *dest_end = dest + dest_len;
NI_PixmapHeader *img_header = &ic->image->header;
int src_trans_pixel_index = ic->src_header->transparent_pixel->index;
int img_trans_pixel_index = img_header->transparent_pixel->index;
int dpy_trans_pixel_index = img_header->color_space->cmap.index ?
img_header->color_space->cmap.index[img_trans_pixel_index] :
il_identity_index_map[img_trans_pixel_index];
/* Two cases */
/* Scaling down ? (or not scaling ?) ... */
if (src_len >= dest_len) {
while (dest < dest_end) {
is_transparent = (*src == src_trans_pixel_index);
if (is_transparent)
*dest = dpy_trans_pixel_index;
dest++;
n += src_len;
while (n >= dest_len) {
src++;
n -= dest_len;
}
}
} else {
/* Scaling up */
while (dest < dest_end) {
n += dest_len;
is_transparent = (*src++ == src_trans_pixel_index);
if (is_transparent)
while (n >= src_len) {
*dest++ = dpy_trans_pixel_index;
n -= src_len;
}
else
while (n >= src_len) {
dest++;
n -= src_len;
}
}
}
}
static void
il_generate_byte_mask(
il_container *ic, /* The image container */
PRUint8 *src, /* Row of pixels, 8-bit pseudocolor data */
int src_len, /* Number of pixels in row */
PRUint8 *dest, /* Output pointer, 8-bit pseudocolor data */
int dest_len) /* Width of output pixel row */
{
int is_transparent, n = 0;
PRUint8 XP_HUGE *dest_end = dest + dest_len;
int src_trans_pixel_index = ic->src_header->transparent_pixel->index;
/* Two cases */
/* Scaling down ? (or not scaling ?) ... */
if (src_len >= dest_len) {
while (dest < dest_end) {
is_transparent = (*src == src_trans_pixel_index);
*dest = is_transparent - 1;
dest++;
n += src_len;
while (n >= dest_len) {
src++;
n -= dest_len;
}
}
} else {
/* Scaling up */
while (dest < dest_end) {
n += dest_len;
is_transparent = (*src++ == src_trans_pixel_index);
if (is_transparent)
while (n >= src_len) {
*dest++ = 0;
n -= src_len;
}
else
while (n >= src_len) {
*dest++ = (PRUint8)-1;
n -= src_len;
}
}
}
}
static void
il_overlay(PRUint8 *src, PRUint8 *dest, PRUint8 *byte_mask, int num_cols,
int bytes_per_pixel)
{
int i, col;
#if 0
PRUint8 *s = src;
PRUint8 *s_end = src + (num_cols * bytes_per_pixel);
#endif
for (col = num_cols; col > 0; col--) {
if (*byte_mask++) {
for (i = bytes_per_pixel-1; i >= 0; i--) {
dest[i] = src[i];
}
}
dest += bytes_per_pixel;
src += bytes_per_pixel;
}
}
/*-----------------------------------------------------------------------------
* Scale a row of packed RGB pixels using the Bresenham algorithm.
* Output is also packed RGB pixels.
*---------------------------------------------------------------------------*/
static void
il_scale_alpha8(
PRUint8 XP_HUGE *src, /* Source row of packed RGB pixels */
int src_len, /* Number of pixels in source row */
PRUint8 *dest, /* Destination, packed RGB pixels */
int dest_len) /* Length of target row, in pixels */
{
PRUint8 *dest_end = dest + dest_len;
int n = 0;
PR_ASSERT(dest);
PR_ASSERT(src_len != dest_len);
/* Two cases */
/* Scaling down ? ... */
if (src_len > dest_len)
{
while (dest < dest_end) {
*dest= *src;
dest ++;
n += src_len;
while (n >= dest_len) {
src++;
n -= dest_len;
}
}
}
else
/* Scaling up. */
{
while (dest < dest_end) {
n += dest_len;
while (n >= src_len) {
*dest = *src;
dest ++;
n -= src_len;
}
src ++;
}
}
}
static PRUint8 il_tmpbuf[MAX_IMAGE_WIDTH];
/*-----------------------------------------------------------------------------
* Emit a complete row of pixel data into the image. This routine
* provides any necessary conversion to the display depth, optional dithering
* for pseudocolor displays, scaling and transparency, including mask
* generation, if necessary. If sufficient data is accumulated, the screen
* image is updated, as well.
*
* PN note: Function too long. Needs to be split. Return true on success.
*---------------------------------------------------------------------------*/
PRBool
il_emit_row(
il_container *ic, /* The image container */
PRUint8 *cbuf, /* Color index data source, or NULL if source
is RGB data */
PRUint8 *rgbbuf, /* Packed RGBa data or RGBa workspace if <cbuf> != NULL */
int x_offset, /* First column to write data into */
int len, /* Width of source image, in pixels */
int row, /* Starting row of image */
int dup_row_count, /* Number of times to duplicate row */
il_draw_mode draw_mode, /* ilOverlay or ilErase */
int pass) /* Zero, unless interlaced GIF,
in which case ranges 1-4, or progressive JPEG,
in which case ranges from 1-n. */
{
IL_GroupContext *img_cx = ic->img_cx;
IL_Pixmap *image = ic->image;
IL_Pixmap *mask = ic->mask;
NI_PixmapHeader *src_header = ic->src_header;
NI_PixmapHeader *img_header = &image->header;
NI_PixmapHeader *mask_header = NULL;
NI_ColorSpace *src_color_space = src_header->color_space;
NI_ColorSpace *img_color_space = img_header->color_space;
PRUint8 XP_HUGE *out;
PRUint8 XP_HUGE *dp;
PRUint8 XP_HUGE *mp;
PRUint8 XP_HUGE *maskp = NULL;
PRUint8 XP_HUGE *alphabitstart = NULL;
PRUint8 *byte_mask = NULL;
PRUint8 XP_HUGE *srcbuf = rgbbuf;
PRUint8 *p = cbuf;
PRUint8 *pl = cbuf+len;
int drow_start, drow_end, row_count, color_index, dup, do_dither;
int dcolumn_start, dcolumn_end, column_count, offset, src_len, dest_len;
PR_ASSERT(row >= 0);
PR_ASSERT(image->haveBits); /* we'd better have some bits... */
if(row >= (int) src_header->height) {
ILTRACE(2,("il: ignoring extra row (%d)", row));
return PR_TRUE;
}
/* Set first and last destination row in the image. Assume no scaling. */
drow_start = row;
drow_end = row + dup_row_count - 1;
dcolumn_start = x_offset;
dcolumn_end = x_offset + len - 1;
/* If scaling, convert vertical image coordinates to display space. */
if (img_header->height != src_header->height) {
int d = drow_start;
int next_drow_start = SCALE_YCOORD(img_header, src_header, drow_end+1);
drow_start = SCALE_YCOORD(img_header, src_header, drow_start);
/*
* Don't emit a row of pixels that will be overwritten later.
* (as may happen during when images are being reduced vertically).
*/
if (drow_start == next_drow_start) {
/*
* Except that the bottom line of pixels can never be
* overwritten by a subsequent line.
*/
if (d != (int) (src_header->height - 1))
return PR_TRUE;
else
drow_end = drow_start;
} else {
drow_end = next_drow_start - 1;
if (drow_end >= (int)img_header->height)
drow_end = img_header->height - 1;
}
}
/* If scaling, convert horizontal image coordinates to display space. */
if (img_header->width != src_header->width) {
int d = dcolumn_start;
int next_dcolumn_start = SCALE_XCOORD(img_header, src_header,
dcolumn_end+1);
dcolumn_start = SCALE_XCOORD(img_header, src_header, dcolumn_start);
/*
* Don't emit a column of pixels that will be overwritten later.
* (as may happen during when images are being reduced vertically).
*/
if (dcolumn_start == next_dcolumn_start) {
/*
* Except that the right column of pixels can never be
* overwritten by a subsequent column.
*/
if (d != (int)(src_header->width - 1))
return PR_TRUE;
else
dcolumn_end = dcolumn_start;
} else {
dcolumn_end = next_dcolumn_start - 1;
if (dcolumn_end >= (int)img_header->width)
dcolumn_end = img_header->width - 1;
}
}
/* Before we do anything else, lock down the pixels of the image, and
mask (if we have one). We need to be carefull not to return
between here and the end
*/
nsresult rv;
rv = img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, image, IL_LOCK_BITS);
if (NS_FAILED(rv)) return PR_FALSE;
if (mask)
{
rv = img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, mask, IL_LOCK_BITS);
if (NS_FAILED(rv)) return PR_FALSE;
}
/* Number of pixel rows and columns to emit into framebuffer */
row_count = drow_end - drow_start + 1;
column_count = dcolumn_end - dcolumn_start + 1;
/* If an alphamask (8 or 1 bit) exists......*/
if(img_header->alpha_bits) {
mask_header = &mask->header;
// img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, mask,
// IL_LOCK_BITS);
#ifdef _USD
alphabitstart = maskp = (PRUint8 XP_HUGE *)mask->bits +
(mask_header->height - drow_start - 1) * mask_header->widthBytes;
#else
alphabitstart = maskp = (PRUint8 XP_HUGE *)mask->bits +
drow_start * mask_header->widthBytes;
#endif
if(img_header->alpha_bits == 1)
{
// picks off alphachannel, packs it for 1bit mask and scales if necessary
il_alpha_mask(rgbbuf, (int)len, dcolumn_start,
alphabitstart, column_count, draw_mode);
PRUint8 *tmpbuf, *rgbbuf_p;
int i;
rgbbuf_p = tmpbuf = rgbbuf;
for(i=0; i < len; i++){
*rgbbuf_p++ = *tmpbuf++;
*rgbbuf_p++ = *tmpbuf++;
*rgbbuf_p++ = *tmpbuf++;
tmpbuf++; /* strip off alpha channel */
}
}
else{ /* alpha_bits=8*/
PRUint8 *tmpbuf, *rgbbuf_p;
int i;
unsigned char *scalemask;
unsigned char *alphabits_1= (unsigned char *)PR_MALLOC(len);
unsigned char *alphabits;
alphabits = alphabits_1;
rgbbuf_p = tmpbuf = rgbbuf;
for(i=0; i < len; i++){
*rgbbuf_p++ = *tmpbuf++;
*rgbbuf_p++ = *tmpbuf++;
*rgbbuf_p++ = *tmpbuf++;
*alphabits++ = *tmpbuf++; /* put alpha channel in separate buffer */
}
alphabits = alphabits_1;
if(len != column_count ){
if ((scalemask= (unsigned char *)PR_MALLOC(mask_header->widthBytes)) != NULL)
{
il_scale_alpha8( alphabits, len, scalemask, mask_header->widthBytes);
nsCRT::memcpy(alphabitstart, scalemask, mask_header->widthBytes);
PR_DELETE(scalemask);
}
}
else
nsCRT::memcpy(alphabitstart, alphabits_1, mask_header->widthBytes);
PR_DELETE(alphabits_1);
} /* else */
// img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, mask,
// IL_UNLOCK_BITS);
} else if(mask){/* If a mask not attributed to an alpha channel exists..ie:
a transparent image appears over a background image ... */
PR_ASSERT(mask->haveBits); /* we'd better have some bits... */
mask_header = &mask->header;
/* Bug, we retain the mask from a transparent
LOSRC GIF when the SRC is a JPEG. */
/* PR_ASSERT(cbuf); */
// img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, mask,
// IL_LOCK_BITS);
#ifdef _USD
maskp = (PRUint8 XP_HUGE *)mask->bits +
(mask_header->height - drow_start - 1) * mask_header->widthBytes;
#else
maskp = (PRUint8 XP_HUGE *)mask->bits +
drow_start * mask_header->widthBytes;
#endif
PR_ASSERT(maskp);
/* We know this mask is prescaled: */
if (nsCRT::strncasecmp(ic->type, "image/x-jg", 9)==0){
/* No scaling needed*/
if (len == column_count)
nsCRT::memcpy(maskp, cbuf, mask_header->widthBytes);
else /* Scale */
il_scale_mask(cbuf, (int)len,
dcolumn_start,
maskp, column_count,
draw_mode);
/* Set to NULL, don't let anyone use it by mistake. */
cbuf = NULL;
}else{
il_generate_scaled_transparency_mask(src_header->transparent_pixel,
cbuf, (int)len,
dcolumn_start,
maskp, column_count,
draw_mode);
}
// img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, mask,
// IL_UNLOCK_BITS);
}
if (!ic->converter) {
#ifdef M12N
int i;
int src_trans_pixel_index;
PRUint8 XP_HUGE * dest;
PRUint8 *indirect_map = src_color_space->cmap.index;
PR_ASSERT(image->haveBits);
if (indirect_map == NULL) {
indirect_map = il_identity_index_map;
}
if ((draw_mode == ilErase) || !src_header->transparent_pixel)
src_trans_pixel_index = -1; /* no transparency */
else
src_trans_pixel_index = src_header->transparent_pixel->index;
// img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, image,
// IL_LOCK_BITS);
/* No converter, image is already rendered in pseudocolor. */
#ifdef _USD
out = (PRUint8 XP_HUGE *)image->bits +
(img_header->height - drow_start - 1) * img_header->widthBytes;
#else
out = (PRUint8 XP_HUGE *)image->bits +
drow_start * img_header->widthBytes;
#endif
dest = out + dcolumn_start;
/* If horizontal scaling ... */
if (len != column_count) {
il_scale_CI_row(cbuf, len, dest, column_count,
indirect_map, src_trans_pixel_index);
} else {
/* Convert to FE's palette indices */
for (i = 0; i < len; i++)
if (cbuf[i] != src_trans_pixel_index)
dest[i] = indirect_map[cbuf[i]];
}
// img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, image,
// IL_UNLOCK_BITS);
#endif /* M12N */
} else {
/* Generate the output row in RGB space, regardless of screen depth. */
if (cbuf) {
PRUint8 *r = rgbbuf;
IL_RGB *map = src_color_space->cmap.map, *entry;
if (!src_header->transparent_pixel) {
/* Simple case: no transparency */
while (p < pl) {
color_index = *p++;
entry = map + color_index;
r[0] = entry->red;
r[1] = entry->green;
r[2] = entry->blue;
r += 3 ;
}
} else {
/*
* There are two kinds of transparency, depending on whether
* the image is overlaying:
* 1. a solid color background, or
* 2. another image
*
* The first case is easy. We just substitute the background
* color for all the transparent pixels in the image. No mask
* is necessary. The second case requires that we generate a
* bit mask (see the code above). It also seems to require that
* all the transparent pixels in the image be set to black.
* XXX - Why ? Is this some platform-specific thing ? - fur
*/
int background_r, background_g, background_b;
IL_IRGB *src_trans_pixel = src_header->transparent_pixel;
int src_trans_pixel_index = src_trans_pixel->index;
background_r = background_g = background_b = 0;
if (!ic->mask) {
/* Solid background color */
background_r = src_trans_pixel->red;
background_g = src_trans_pixel->green;
background_b = src_trans_pixel->blue;
}
/* Remap transparent pixels */
while (p < pl) {
color_index = *p++;
if (color_index == src_trans_pixel_index) {
r[0] = background_r;
r[1] = background_g;
r[2] = background_b;
r += 3;
} else {
entry = map + color_index;
r[0] = entry->red;
r[1] = entry->green;
r[2] = entry->blue;
r += 3;
}
}
}
}
/* Now we are in RGB space. */
/* Simple anamorphic scaling (in RGB space for now) */
src_len = len;
dest_len = column_count;
if (src_len != dest_len) {
PRUint8 XP_HUGE *src = rgbbuf;
PRUint8 *dest = ic->scalerow;
srcbuf = dest;
/* Scale the pixel data (mask data already scaled) */
il_scale_RGB_row(src, src_len, dest, dest_len);
}
img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, image,
IL_LOCK_BITS);
#ifdef _USD
out = (PRUint8 XP_HUGE *)image->bits +
(img_header->height-drow_start-1) * (PRUint32)img_header->widthBytes;
#else
out = (PRUint8 XP_HUGE *)image->bits +
drow_start * img_header->widthBytes;
#endif
if (src_header->transparent_pixel && (draw_mode == ilOverlay))
{
if( cbuf ){
il_generate_byte_mask(ic, cbuf, len, il_tmpbuf, column_count);
byte_mask = il_tmpbuf;
}
}
/*
* Convert RGB to display depth. If display is pseudocolor, this may
* also color-quantize and dither.
*/
(*ic->converter)(ic, byte_mask, srcbuf, dcolumn_start,
column_count, out);
// img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, image,
// IL_UNLOCK_BITS);
/*
* Have to reset transparent pixels to background color
* because color quantization may have mutated them.
*/
if (src_header->transparent_pixel &&
(img_color_space->type == NI_PseudoColor) &&
!mask && (draw_mode == ilErase))
il_reset_background_pixels(ic, cbuf, len,
out + dcolumn_start, column_count);
}
/*
* We now have one row of pixels and, if required for transparency, a row
* of mask data. If the pixels in this row of the image cover span more
* than one pixel vertically when displayed, the row needs to be
* replicated in the framebuffer. (This replication is necessary when
* displaying interlaced GIFs and/or vertical scaling of any image type.)
* Actually, pixel rows are not simply copied: Dithering may need to be
* applied on a line-by-line basis.
*/
dp = out;
mp = maskp;
dup = row_count - 1;
offset = dcolumn_start * (img_color_space->pixmap_depth / 8);
#ifndef M12N /* Clean this up */
if (ic->image->pixmap_depth == 1)
do_dither = PR_TRUE;
else
do_dither = ic->converter && (row_count <= 4) &&
((ic->dither_mode == IL_Dither) ||
(nsCRT::strncasecmp(ic->type, "image/jpeg",10)==0);
#else
do_dither = (ic->dither_mode == IL_Dither);
if ((nsCRT::strncasecmp(ic->type, "image/gif",9)==0)||
(nsCRT::strncasecmp(ic->type, "image/png",9)==0) &&
(!ic->converter || (row_count > 4)))
do_dither = PR_FALSE;
#endif /* M12N */
while (dup--) {
#ifdef _USD
dp -= img_header->widthBytes;
if (mask)
mp -= mask_header->widthBytes;
#else
dp += img_header->widthBytes;
if (mask)
mp += mask_header->widthBytes;
#endif
/* Is dithering being done (either mono or pseudocolor) ... ? */
if (do_dither) {
/* Dither / color-quantize */
(*ic->converter)(ic, byte_mask, srcbuf, dcolumn_start,
column_count, dp);
/*
* Have to reset transparent pixels to background color
* because color quantization may have mutated them.
*/
if (img_header->transparent_pixel &&
(img_color_space->type == NI_PseudoColor) &&
!mask && (draw_mode == ilErase))
il_reset_background_pixels(ic, cbuf, len, dp + dcolumn_start,
column_count);
} else {
/* If no dithering, each row of pixels is exactly the same. */
if (byte_mask)
il_overlay(out + offset, dp + offset, byte_mask, column_count,
(img_color_space->pixmap_depth/8));
else
nsCRT::memcpy(dp + offset, out + offset,
(img_color_space->pixmap_depth/8) * column_count);
}
/* Duplicate the mask also. */
if (maskp) {
nsCRT::memcpy(mp, maskp, mask_header->widthBytes);
}
}
/* Now we can unlock the bits */
img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, image, IL_UNLOCK_BITS);
if (mask)
img_cx->img_cb->ControlPixmapBits(img_cx->dpy_cx, mask, IL_UNLOCK_BITS);
/* If enough rows accumulated, send to the front-end for display. */
il_partial(ic, drow_start, row_count, pass);
return PR_TRUE;
}