2007-08-27 16:30:52 -07:00
|
|
|
/* -*- Mode: c; tab-width: 8; c-basic-offset: 4; indent-tabs-mode: t; -*- */
|
2005-12-19 22:45:41 +00:00
|
|
|
/* cairo - a vector graphics library with display and print output
|
|
|
|
|
*
|
|
|
|
|
* Copyright © 2002 University of Southern California
|
|
|
|
|
* Copyright © 2005 Red Hat, Inc.
|
|
|
|
|
*
|
|
|
|
|
* This library is free software; you can redistribute it and/or
|
|
|
|
|
* modify it either under the terms of the GNU Lesser General Public
|
|
|
|
|
* License version 2.1 as published by the Free Software Foundation
|
|
|
|
|
* (the "LGPL") or, at your option, under the terms of the Mozilla
|
|
|
|
|
* Public License Version 1.1 (the "MPL"). If you do not alter this
|
|
|
|
|
* notice, a recipient may use your version of this file under either
|
|
|
|
|
* the MPL or the LGPL.
|
|
|
|
|
*
|
|
|
|
|
* You should have received a copy of the LGPL along with this library
|
|
|
|
|
* in the file COPYING-LGPL-2.1; if not, write to the Free Software
|
|
|
|
|
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
|
|
|
* You should have received a copy of the MPL along with this library
|
|
|
|
|
* in the file COPYING-MPL-1.1
|
|
|
|
|
*
|
|
|
|
|
* The contents of this file are subject to the Mozilla 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/MPL/
|
|
|
|
|
*
|
|
|
|
|
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY
|
|
|
|
|
* OF ANY KIND, either express or implied. See the LGPL or the MPL for
|
|
|
|
|
* the specific language governing rights and limitations.
|
|
|
|
|
*
|
|
|
|
|
* The Original Code is the cairo graphics library.
|
|
|
|
|
*
|
|
|
|
|
* The Initial Developer of the Original Code is University of Southern
|
|
|
|
|
* California.
|
|
|
|
|
*
|
|
|
|
|
* Contributor(s):
|
|
|
|
|
* Carl D. Worth <cworth@cworth.org>
|
|
|
|
|
*/
|
|
|
|
|
|
2007-04-03 20:25:30 -04:00
|
|
|
#include "cairoint.h"
|
|
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
#include "cairo-surface-fallback-private.h"
|
|
|
|
|
#include "cairo-clip-private.h"
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
#include "cairo-region-private.h"
|
2005-12-19 22:45:41 +00:00
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
|
cairo_surface_t *dst;
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t extents;
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_image_surface_t *image;
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t image_rect;
|
2005-12-19 22:45:41 +00:00
|
|
|
void *image_extra;
|
|
|
|
|
} fallback_state_t;
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* _fallback_init:
|
2006-06-06 15:35:48 -07:00
|
|
|
*
|
2005-12-19 22:45:41 +00:00
|
|
|
* Acquire destination image surface needed for an image-based
|
|
|
|
|
* fallback.
|
2006-06-06 15:35:48 -07:00
|
|
|
*
|
2008-01-28 20:48:48 -05:00
|
|
|
* Return value: %CAIRO_INT_STATUS_NOTHING_TO_DO if the extents are not
|
|
|
|
|
* visible, %CAIRO_STATUS_SUCCESS if some portion is visible and all
|
2005-12-19 22:45:41 +00:00
|
|
|
* went well, or some error status otherwise.
|
|
|
|
|
**/
|
|
|
|
|
static cairo_int_status_t
|
|
|
|
|
_fallback_init (fallback_state_t *state,
|
|
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int width,
|
|
|
|
|
int height)
|
|
|
|
|
{
|
|
|
|
|
cairo_status_t status;
|
|
|
|
|
|
|
|
|
|
state->extents.x = x;
|
|
|
|
|
state->extents.y = y;
|
|
|
|
|
state->extents.width = width;
|
|
|
|
|
state->extents.height = height;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
state->dst = dst;
|
|
|
|
|
|
|
|
|
|
status = _cairo_surface_acquire_dest_image (dst, &state->extents,
|
|
|
|
|
&state->image, &state->image_rect,
|
|
|
|
|
&state->image_extra);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2005-12-19 22:45:41 +00:00
|
|
|
return status;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
/* XXX: This NULL value tucked away in state->image is a rather
|
|
|
|
|
* ugly interface. Cleaner would be to push the
|
|
|
|
|
* CAIRO_INT_STATUS_NOTHING_TO_DO value down into
|
|
|
|
|
* _cairo_surface_acquire_dest_image and its backend
|
|
|
|
|
* counterparts. */
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
assert (state->image != NULL);
|
2005-12-19 22:45:41 +00:00
|
|
|
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
_fallback_fini (fallback_state_t *state)
|
|
|
|
|
{
|
|
|
|
|
_cairo_surface_release_dest_image (state->dst, &state->extents,
|
|
|
|
|
state->image, &state->image_rect,
|
|
|
|
|
state->image_extra);
|
|
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
typedef cairo_status_t
|
|
|
|
|
(*cairo_draw_func_t) (void *closure,
|
|
|
|
|
cairo_operator_t op,
|
|
|
|
|
const cairo_pattern_t *src,
|
|
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
int dst_x,
|
|
|
|
|
int dst_y,
|
|
|
|
|
const cairo_rectangle_int_t *extents,
|
|
|
|
|
cairo_region_t *clip_region);
|
2005-12-20 09:37:15 +00:00
|
|
|
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_create_composite_mask_pattern (cairo_surface_pattern_t *mask_pattern,
|
|
|
|
|
cairo_clip_t *clip,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_draw_func_t draw_func,
|
2006-05-04 03:43:34 -07:00
|
|
|
void *draw_closure,
|
|
|
|
|
cairo_surface_t *dst,
|
2007-06-18 16:56:24 -07:00
|
|
|
const cairo_rectangle_int_t *extents)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
|
|
|
|
cairo_surface_t *mask;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_region_t *clip_region = NULL;
|
|
|
|
|
cairo_solid_pattern_t solid;
|
2005-12-20 09:37:15 +00:00
|
|
|
cairo_status_t status;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_bool_t clip_surface = FALSE;
|
|
|
|
|
|
|
|
|
|
if (clip != NULL) {
|
|
|
|
|
status = _cairo_clip_get_region (clip, &clip_region);
|
|
|
|
|
assert (! _cairo_status_is_error (status));
|
|
|
|
|
|
|
|
|
|
clip_surface = status == CAIRO_INT_STATUS_UNSUPPORTED;
|
|
|
|
|
|
|
|
|
|
if (clip_region && cairo_region_num_rectangles (clip_region) == 1)
|
|
|
|
|
clip_region = NULL;
|
|
|
|
|
}
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* We need to use solid here, because to use CAIRO_OPERATOR_SOURCE with
|
|
|
|
|
* a mask (as called via _cairo_surface_mask) triggers assertion failures.
|
|
|
|
|
*/
|
|
|
|
|
mask = _cairo_surface_create_similar_solid (dst,
|
|
|
|
|
CAIRO_CONTENT_ALPHA,
|
|
|
|
|
extents->width,
|
|
|
|
|
extents->height,
|
|
|
|
|
CAIRO_COLOR_TRANSPARENT,
|
|
|
|
|
TRUE);
|
|
|
|
|
if (unlikely (mask->status))
|
2007-10-04 13:15:46 +01:00
|
|
|
return mask->status;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_init_solid (&solid, CAIRO_COLOR_WHITE, CAIRO_CONTENT_ALPHA);
|
|
|
|
|
status = draw_func (draw_closure, CAIRO_OPERATOR_ADD,
|
|
|
|
|
&solid.base, mask,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
extents,
|
|
|
|
|
clip_region);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2005-12-20 09:37:15 +00:00
|
|
|
goto CLEANUP_SURFACE;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (clip_surface) {
|
|
|
|
|
cairo_surface_pattern_t pattern;
|
|
|
|
|
cairo_surface_t *surface;
|
|
|
|
|
|
|
|
|
|
surface = _cairo_clip_get_surface (clip, mask);
|
|
|
|
|
_cairo_pattern_init_for_surface (&pattern, surface);
|
|
|
|
|
cairo_surface_destroy (surface);
|
|
|
|
|
|
|
|
|
|
status = _cairo_surface_composite (CAIRO_OPERATOR_IN,
|
|
|
|
|
&pattern.base,
|
|
|
|
|
NULL,
|
|
|
|
|
mask,
|
|
|
|
|
extents->x - clip->path->extents.x,
|
|
|
|
|
extents->y - clip->path->extents.y,
|
|
|
|
|
0, 0,
|
|
|
|
|
0, 0,
|
|
|
|
|
extents->width, extents->height,
|
|
|
|
|
NULL);
|
|
|
|
|
|
|
|
|
|
_cairo_pattern_fini (&pattern.base);
|
|
|
|
|
|
|
|
|
|
if (unlikely (status))
|
|
|
|
|
goto CLEANUP_SURFACE;
|
|
|
|
|
}
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-20 09:37:15 +00:00
|
|
|
_cairo_pattern_init_for_surface (mask_pattern, mask);
|
|
|
|
|
|
|
|
|
|
CLEANUP_SURFACE:
|
|
|
|
|
cairo_surface_destroy (mask);
|
|
|
|
|
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Handles compositing with a clip surface when the operator allows
|
|
|
|
|
* us to combine the clip with the mask
|
|
|
|
|
*/
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_clip_and_composite_with_mask (cairo_clip_t *clip,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *src,
|
2006-05-04 03:43:34 -07:00
|
|
|
cairo_draw_func_t draw_func,
|
|
|
|
|
void *draw_closure,
|
|
|
|
|
cairo_surface_t *dst,
|
2007-06-18 16:56:24 -07:00
|
|
|
const cairo_rectangle_int_t *extents)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
|
|
|
|
cairo_surface_pattern_t mask_pattern;
|
|
|
|
|
cairo_status_t status;
|
|
|
|
|
|
|
|
|
|
status = _create_composite_mask_pattern (&mask_pattern,
|
|
|
|
|
clip,
|
|
|
|
|
draw_func, draw_closure,
|
|
|
|
|
dst, extents);
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (likely (status == CAIRO_STATUS_SUCCESS)) {
|
|
|
|
|
status = _cairo_surface_composite (op,
|
|
|
|
|
src, &mask_pattern.base, dst,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
0, 0,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
extents->width, extents->height,
|
|
|
|
|
NULL);
|
|
|
|
|
|
|
|
|
|
_cairo_pattern_fini (&mask_pattern.base);
|
|
|
|
|
}
|
2005-12-20 09:37:15 +00:00
|
|
|
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Handles compositing with a clip surface when we have to do the operation
|
|
|
|
|
* in two pieces and combine them together.
|
|
|
|
|
*/
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_clip_and_composite_combine (cairo_clip_t *clip,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *src,
|
2006-05-04 03:43:34 -07:00
|
|
|
cairo_draw_func_t draw_func,
|
|
|
|
|
void *draw_closure,
|
|
|
|
|
cairo_surface_t *dst,
|
2007-06-18 16:56:24 -07:00
|
|
|
const cairo_rectangle_int_t *extents)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
|
|
|
|
cairo_surface_t *intermediate;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_surface_pattern_t pattern;
|
|
|
|
|
cairo_surface_pattern_t clip_pattern;
|
|
|
|
|
cairo_surface_t *clip_surface;
|
2005-12-20 09:37:15 +00:00
|
|
|
cairo_status_t status;
|
|
|
|
|
|
|
|
|
|
/* We'd be better off here creating a surface identical in format
|
|
|
|
|
* to dst, but we have no way of getting that information.
|
|
|
|
|
* A CAIRO_CONTENT_CLONE or something might be useful.
|
|
|
|
|
*/
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
intermediate =
|
|
|
|
|
_cairo_surface_create_similar_scratch (dst,
|
|
|
|
|
CAIRO_CONTENT_COLOR_ALPHA,
|
|
|
|
|
extents->width,
|
|
|
|
|
extents->height);
|
|
|
|
|
if (intermediate == NULL) {
|
|
|
|
|
intermediate =
|
|
|
|
|
_cairo_image_surface_create_with_content (CAIRO_CONTENT_COLOR_ALPHA,
|
|
|
|
|
extents->width,
|
|
|
|
|
extents->width);
|
|
|
|
|
}
|
|
|
|
|
if (unlikely (intermediate->status))
|
2007-10-04 13:15:46 +01:00
|
|
|
return intermediate->status;
|
2005-12-20 09:37:15 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* Initialize the intermediate surface from the destination surface */
|
|
|
|
|
_cairo_pattern_init_for_surface (&pattern, dst);
|
2005-12-20 09:37:15 +00:00
|
|
|
status = _cairo_surface_composite (CAIRO_OPERATOR_SOURCE,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
&pattern.base, NULL, intermediate,
|
2005-12-20 09:37:15 +00:00
|
|
|
extents->x, extents->y,
|
|
|
|
|
0, 0,
|
|
|
|
|
0, 0,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
extents->width, extents->height,
|
|
|
|
|
NULL);
|
|
|
|
|
_cairo_pattern_fini (&pattern.base);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2005-12-20 09:37:15 +00:00
|
|
|
goto CLEANUP_SURFACE;
|
|
|
|
|
|
|
|
|
|
status = (*draw_func) (draw_closure, op,
|
|
|
|
|
src, intermediate,
|
|
|
|
|
extents->x, extents->y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
extents,
|
|
|
|
|
NULL);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2005-12-20 09:37:15 +00:00
|
|
|
goto CLEANUP_SURFACE;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
assert (clip->path != NULL);
|
|
|
|
|
clip_surface = _cairo_clip_get_surface (clip, dst);
|
|
|
|
|
if (unlikely (clip_surface->status))
|
2005-12-20 09:37:15 +00:00
|
|
|
goto CLEANUP_SURFACE;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_init_for_surface (&clip_pattern, clip_surface);
|
|
|
|
|
cairo_surface_destroy (clip_surface);
|
|
|
|
|
|
|
|
|
|
/* Combine that with the clip */
|
|
|
|
|
status = _cairo_surface_composite (CAIRO_OPERATOR_DEST_IN,
|
|
|
|
|
&clip_pattern.base, NULL, intermediate,
|
|
|
|
|
extents->x - clip->path->extents.x,
|
|
|
|
|
extents->y - clip->path->extents.y,
|
|
|
|
|
0, 0,
|
|
|
|
|
0, 0,
|
|
|
|
|
extents->width, extents->height,
|
|
|
|
|
NULL);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2005-12-20 09:37:15 +00:00
|
|
|
goto CLEANUP_SURFACE;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* Punch the clip out of the destination */
|
|
|
|
|
status = _cairo_surface_composite (CAIRO_OPERATOR_DEST_OUT,
|
|
|
|
|
&clip_pattern.base, NULL, dst,
|
|
|
|
|
extents->x - clip->path->extents.x,
|
|
|
|
|
extents->y - clip->path->extents.y,
|
|
|
|
|
0, 0,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
extents->width, extents->height,
|
|
|
|
|
NULL);
|
|
|
|
|
if (unlikely (status))
|
|
|
|
|
goto CLEANUP_SURFACE;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* Now add the two results together */
|
|
|
|
|
_cairo_pattern_init_for_surface (&pattern, intermediate);
|
2005-12-20 09:37:15 +00:00
|
|
|
status = _cairo_surface_composite (CAIRO_OPERATOR_ADD,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
&pattern.base, NULL, dst,
|
2005-12-20 09:37:15 +00:00
|
|
|
0, 0,
|
|
|
|
|
0, 0,
|
|
|
|
|
extents->x, extents->y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
extents->width, extents->height,
|
|
|
|
|
NULL);
|
|
|
|
|
_cairo_pattern_fini (&pattern.base);
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-20 09:37:15 +00:00
|
|
|
CLEANUP_SURFACE:
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_fini (&clip_pattern.base);
|
2005-12-20 09:37:15 +00:00
|
|
|
cairo_surface_destroy (intermediate);
|
|
|
|
|
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
2008-01-28 20:48:48 -05:00
|
|
|
/* Handles compositing for %CAIRO_OPERATOR_SOURCE, which is special; it's
|
2005-12-20 09:37:15 +00:00
|
|
|
* defined as (src IN mask IN clip) ADD (dst OUT (mask IN clip))
|
|
|
|
|
*/
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_clip_and_composite_source (cairo_clip_t *clip,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *src,
|
2006-05-04 03:43:34 -07:00
|
|
|
cairo_draw_func_t draw_func,
|
|
|
|
|
void *draw_closure,
|
|
|
|
|
cairo_surface_t *dst,
|
2007-06-18 16:56:24 -07:00
|
|
|
const cairo_rectangle_int_t *extents)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
|
|
|
|
cairo_surface_pattern_t mask_pattern;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_region_t *clip_region = NULL;
|
2005-12-20 09:37:15 +00:00
|
|
|
cairo_status_t status;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* Create a surface that is mask IN clip */
|
2005-12-20 09:37:15 +00:00
|
|
|
status = _create_composite_mask_pattern (&mask_pattern,
|
|
|
|
|
clip,
|
|
|
|
|
draw_func, draw_closure,
|
|
|
|
|
dst, extents);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2005-12-20 09:37:15 +00:00
|
|
|
return status;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (clip != NULL) {
|
|
|
|
|
status = _cairo_clip_get_region (clip, &clip_region);
|
|
|
|
|
assert (! _cairo_status_is_error (status));
|
|
|
|
|
|
|
|
|
|
/* a solitary clip rectangle is already accommodated by extents */
|
|
|
|
|
if (clip_region && cairo_region_num_rectangles (clip_region) == 1)
|
|
|
|
|
clip_region = NULL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Compute dest' = dest OUT (mask IN clip) */
|
2005-12-20 09:37:15 +00:00
|
|
|
status = _cairo_surface_composite (CAIRO_OPERATOR_DEST_OUT,
|
|
|
|
|
&mask_pattern.base, NULL, dst,
|
|
|
|
|
0, 0,
|
|
|
|
|
0, 0,
|
|
|
|
|
extents->x, extents->y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
extents->width, extents->height,
|
|
|
|
|
clip_region);
|
2005-12-20 09:37:15 +00:00
|
|
|
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2005-12-20 09:37:15 +00:00
|
|
|
goto CLEANUP_MASK_PATTERN;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* Now compute (src IN (mask IN clip)) ADD dest' */
|
2005-12-20 09:37:15 +00:00
|
|
|
status = _cairo_surface_composite (CAIRO_OPERATOR_ADD,
|
|
|
|
|
src, &mask_pattern.base, dst,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
0, 0,
|
|
|
|
|
extents->x, extents->y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
extents->width, extents->height,
|
|
|
|
|
clip_region);
|
2005-12-20 09:37:15 +00:00
|
|
|
|
|
|
|
|
CLEANUP_MASK_PATTERN:
|
|
|
|
|
_cairo_pattern_fini (&mask_pattern.base);
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
2007-06-18 16:56:24 -07:00
|
|
|
_cairo_rectangle_empty (const cairo_rectangle_int_t *rect)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
|
|
|
|
return rect->width == 0 || rect->height == 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
|
* _clip_and_composite:
|
|
|
|
|
* @clip: a #cairo_clip_t
|
|
|
|
|
* @op: the operator to draw with
|
|
|
|
|
* @src: source pattern
|
|
|
|
|
* @draw_func: function that can be called to draw with the mask onto a surface.
|
|
|
|
|
* @draw_closure: data to pass to @draw_func.
|
|
|
|
|
* @dst: destination surface
|
|
|
|
|
* @extents: rectangle holding a bounding box for the operation; this
|
|
|
|
|
* rectangle will be used as the size for the temporary
|
|
|
|
|
* surface.
|
|
|
|
|
*
|
|
|
|
|
* When there is a surface clip, we typically need to create an intermediate
|
|
|
|
|
* surface. This function handles the logic of creating a temporary surface
|
|
|
|
|
* drawing to it, then compositing the result onto the target surface.
|
|
|
|
|
*
|
|
|
|
|
* @draw_func is to called to draw the mask; it will be called no more
|
|
|
|
|
* than once.
|
2006-06-06 15:35:48 -07:00
|
|
|
*
|
2005-12-20 09:37:15 +00:00
|
|
|
* Return value: %CAIRO_STATUS_SUCCESS if the drawing succeeded.
|
|
|
|
|
**/
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_clip_and_composite (cairo_clip_t *clip,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *src,
|
2006-05-04 03:43:34 -07:00
|
|
|
cairo_draw_func_t draw_func,
|
|
|
|
|
void *draw_closure,
|
|
|
|
|
cairo_surface_t *dst,
|
2007-06-18 16:56:24 -07:00
|
|
|
const cairo_rectangle_int_t *extents)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
2008-05-28 15:42:09 +01:00
|
|
|
cairo_solid_pattern_t solid_pattern;
|
2005-12-20 09:37:15 +00:00
|
|
|
cairo_status_t status;
|
|
|
|
|
|
|
|
|
|
if (_cairo_rectangle_empty (extents))
|
|
|
|
|
/* Nothing to do */
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
|
|
|
|
|
|
|
|
|
if (op == CAIRO_OPERATOR_CLEAR) {
|
2008-05-28 15:42:09 +01:00
|
|
|
_cairo_pattern_init_solid (&solid_pattern, CAIRO_COLOR_WHITE,
|
2007-04-23 10:24:03 -07:00
|
|
|
CAIRO_CONTENT_COLOR);
|
2005-12-20 09:37:15 +00:00
|
|
|
src = &solid_pattern.base;
|
|
|
|
|
op = CAIRO_OPERATOR_DEST_OUT;
|
|
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (op == CAIRO_OPERATOR_SOURCE) {
|
|
|
|
|
status = _clip_and_composite_source (clip,
|
|
|
|
|
src,
|
|
|
|
|
draw_func, draw_closure,
|
|
|
|
|
dst, extents);
|
|
|
|
|
} else {
|
|
|
|
|
cairo_bool_t clip_surface = FALSE;
|
|
|
|
|
cairo_region_t *clip_region = NULL;
|
|
|
|
|
|
|
|
|
|
if (clip != NULL) {
|
|
|
|
|
status = _cairo_clip_get_region (clip, &clip_region);
|
|
|
|
|
assert (! _cairo_status_is_error (status));
|
|
|
|
|
|
|
|
|
|
clip_surface = status == CAIRO_INT_STATUS_UNSUPPORTED;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (clip_surface) {
|
|
|
|
|
if (_cairo_operator_bounded_by_mask (op)) {
|
|
|
|
|
status = _clip_and_composite_with_mask (clip, op,
|
|
|
|
|
src,
|
|
|
|
|
draw_func, draw_closure,
|
|
|
|
|
dst, extents);
|
|
|
|
|
} else {
|
|
|
|
|
status = _clip_and_composite_combine (clip, op,
|
|
|
|
|
src,
|
|
|
|
|
draw_func, draw_closure,
|
|
|
|
|
dst, extents);
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
/* a solitary clip rectangle is already accommodated by extents */
|
|
|
|
|
if (clip_region && cairo_region_num_rectangles (clip_region) == 1)
|
|
|
|
|
clip_region = NULL;
|
|
|
|
|
|
|
|
|
|
status = draw_func (draw_closure, op,
|
|
|
|
|
src, dst,
|
|
|
|
|
0, 0,
|
|
|
|
|
extents,
|
|
|
|
|
clip_region);
|
|
|
|
|
}
|
2005-12-20 09:37:15 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Composites a region representing a set of trapezoids.
|
|
|
|
|
*/
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_composite_trap_region (cairo_clip_t *clip,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *src,
|
2006-05-04 03:43:34 -07:00
|
|
|
cairo_operator_t op,
|
|
|
|
|
cairo_surface_t *dst,
|
2007-06-18 18:33:29 -07:00
|
|
|
cairo_region_t *trap_region,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
const cairo_rectangle_int_t *extents)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
|
|
|
|
cairo_status_t status;
|
2008-05-28 15:42:09 +01:00
|
|
|
cairo_solid_pattern_t solid_pattern;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_surface_pattern_t mask_pattern;
|
|
|
|
|
cairo_pattern_t *mask = NULL;
|
|
|
|
|
int mask_x = 0, mask_y =0;
|
|
|
|
|
|
|
|
|
|
if (clip != NULL) {
|
|
|
|
|
cairo_surface_t *clip_surface = NULL;
|
|
|
|
|
const cairo_rectangle_int_t *clip_extents;
|
|
|
|
|
|
|
|
|
|
clip_surface = _cairo_clip_get_surface (clip, dst);
|
|
|
|
|
if (unlikely (clip_surface->status))
|
|
|
|
|
return clip_surface->status;
|
|
|
|
|
|
|
|
|
|
if (op == CAIRO_OPERATOR_CLEAR) {
|
|
|
|
|
_cairo_pattern_init_solid (&solid_pattern, CAIRO_COLOR_WHITE,
|
|
|
|
|
CAIRO_CONTENT_COLOR);
|
|
|
|
|
src = &solid_pattern.base;
|
|
|
|
|
op = CAIRO_OPERATOR_DEST_OUT;
|
|
|
|
|
}
|
2005-12-20 09:37:15 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_init_for_surface (&mask_pattern, clip_surface);
|
|
|
|
|
cairo_surface_destroy (clip_surface);
|
2009-05-26 10:41:53 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
clip_extents = _cairo_clip_get_extents (clip);
|
|
|
|
|
mask_x = extents->x - clip_extents->x;
|
|
|
|
|
mask_y = extents->y - clip_extents->y;
|
|
|
|
|
mask = &mask_pattern.base;
|
2005-12-20 09:37:15 +00:00
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* reduce a solitary clipping region to the extents */
|
|
|
|
|
if (cairo_region_num_rectangles (trap_region) == 1)
|
|
|
|
|
trap_region = NULL;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = _cairo_surface_composite (op, src, mask, dst,
|
2005-12-20 09:37:15 +00:00
|
|
|
extents->x, extents->y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
mask_x, mask_y,
|
2005-12-20 09:37:15 +00:00
|
|
|
extents->x, extents->y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
extents->width, extents->height,
|
|
|
|
|
trap_region);
|
2006-04-14 11:01:39 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (mask != NULL)
|
|
|
|
|
_cairo_pattern_fini (mask);
|
2005-12-20 09:37:15 +00:00
|
|
|
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
|
cairo_traps_t *traps;
|
|
|
|
|
cairo_antialias_t antialias;
|
|
|
|
|
} cairo_composite_traps_info_t;
|
|
|
|
|
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_composite_traps_draw_func (void *closure,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *src,
|
2006-05-04 03:43:34 -07:00
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
int dst_x,
|
|
|
|
|
int dst_y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
const cairo_rectangle_int_t *extents,
|
|
|
|
|
cairo_region_t *clip_region)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
|
|
|
|
cairo_composite_traps_info_t *info = closure;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-20 09:37:15 +00:00
|
|
|
if (dst_x != 0 || dst_y != 0)
|
|
|
|
|
_cairo_traps_translate (info->traps, - dst_x, - dst_y);
|
|
|
|
|
|
2009-05-26 10:41:53 +01:00
|
|
|
return _cairo_surface_composite_trapezoids (op,
|
|
|
|
|
src, dst, info->antialias,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
extents->x - dst_x, extents->y - dst_y,
|
|
|
|
|
extents->width, extents->height,
|
|
|
|
|
info->traps->traps,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
info->traps->num_traps,
|
|
|
|
|
clip_region);
|
2005-12-20 09:37:15 +00:00
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
enum {
|
|
|
|
|
HAS_CLEAR_REGION = 0x1,
|
|
|
|
|
};
|
|
|
|
|
|
2005-12-20 09:37:15 +00:00
|
|
|
static cairo_status_t
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_clip_and_composite_region (const cairo_pattern_t *src,
|
|
|
|
|
cairo_operator_t op,
|
|
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
cairo_region_t *trap_region,
|
|
|
|
|
cairo_antialias_t antialias,
|
|
|
|
|
cairo_clip_t *clip,
|
|
|
|
|
cairo_rectangle_int_t *extents)
|
2005-12-20 09:37:15 +00:00
|
|
|
{
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_region_t clear_region;
|
|
|
|
|
cairo_bool_t clip_surface = FALSE;
|
|
|
|
|
unsigned int has_region = 0;
|
2005-12-20 09:37:15 +00:00
|
|
|
cairo_status_t status;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (clip != NULL) {
|
|
|
|
|
cairo_region_t *clip_region;
|
2005-12-20 09:37:15 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = _cairo_clip_get_region (clip, &clip_region);
|
|
|
|
|
assert (! _cairo_status_is_error (status));
|
2007-01-05 15:56:06 -08:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
clip_surface = status == CAIRO_INT_STATUS_UNSUPPORTED;
|
|
|
|
|
}
|
2007-03-14 00:28:49 +01:00
|
|
|
|
|
|
|
|
if (_cairo_operator_bounded_by_mask (op)) {
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t trap_extents;
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_region_get_extents (trap_region, &trap_extents);
|
|
|
|
|
if (! _cairo_rectangle_intersect (extents, &trap_extents))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
|
|
|
|
} else {
|
|
|
|
|
if (! clip_surface) {
|
|
|
|
|
/* If we optimize drawing with an unbounded operator to
|
|
|
|
|
* _cairo_surface_fill_rectangles() or to drawing with a
|
|
|
|
|
* clip region, then we have an additional region to clear.
|
|
|
|
|
*/
|
|
|
|
|
_cairo_region_init_rectangle (&clear_region, extents);
|
|
|
|
|
status = cairo_region_subtract (&clear_region, trap_region);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
return status;
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (! cairo_region_is_empty (&clear_region))
|
|
|
|
|
has_region |= HAS_CLEAR_REGION;
|
2007-03-14 00:28:49 +01:00
|
|
|
}
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
}
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if ((src->type == CAIRO_PATTERN_TYPE_SOLID || op == CAIRO_OPERATOR_CLEAR) &&
|
|
|
|
|
! clip_surface)
|
|
|
|
|
{
|
|
|
|
|
const cairo_color_t *color;
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (op == CAIRO_OPERATOR_CLEAR)
|
|
|
|
|
color = CAIRO_COLOR_TRANSPARENT;
|
|
|
|
|
else
|
|
|
|
|
color = &((cairo_solid_pattern_t *)src)->color;
|
|
|
|
|
|
|
|
|
|
/* Solid rectangles special case */
|
|
|
|
|
status = _cairo_surface_fill_region (dst, op, color, trap_region);
|
2007-03-14 00:28:49 +01:00
|
|
|
} else {
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* For a simple rectangle, we can just use composite(), for more
|
|
|
|
|
* rectangles, we have to set a clip region. The cost of rasterizing
|
|
|
|
|
* trapezoids is pretty high for most backends currently, so it's
|
|
|
|
|
* worthwhile even if a region is needed.
|
|
|
|
|
*
|
|
|
|
|
* If we have a clip surface, we set it as the mask; this only works
|
|
|
|
|
* for bounded operators other than SOURCE; for unbounded operators,
|
|
|
|
|
* clip and mask cannot be interchanged. For SOURCE, the operator
|
|
|
|
|
* as implemented by the backends is different in its handling
|
|
|
|
|
* of the mask then what we want.
|
|
|
|
|
*
|
|
|
|
|
* CAIRO_INT_STATUS_UNSUPPORTED will be returned if the region has
|
|
|
|
|
* more than rectangle and the destination doesn't support clip
|
|
|
|
|
* regions. In that case, we fall through.
|
|
|
|
|
*/
|
|
|
|
|
status = _composite_trap_region (clip_surface ? clip : NULL,
|
|
|
|
|
src, op, dst,
|
|
|
|
|
trap_region, extents);
|
|
|
|
|
}
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (has_region & HAS_CLEAR_REGION) {
|
|
|
|
|
if (status == CAIRO_STATUS_SUCCESS) {
|
|
|
|
|
status = _cairo_surface_fill_region (dst,
|
|
|
|
|
CAIRO_OPERATOR_CLEAR,
|
|
|
|
|
CAIRO_COLOR_TRANSPARENT,
|
|
|
|
|
&clear_region);
|
|
|
|
|
}
|
|
|
|
|
_cairo_region_fini (&clear_region);
|
|
|
|
|
}
|
2008-12-11 15:29:23 -05:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
return status;
|
|
|
|
|
}
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* Warning: This call modifies the coordinates of traps */
|
|
|
|
|
static cairo_status_t
|
|
|
|
|
_clip_and_composite_trapezoids (const cairo_pattern_t *src,
|
|
|
|
|
cairo_operator_t op,
|
|
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
cairo_traps_t *traps,
|
|
|
|
|
cairo_antialias_t antialias,
|
|
|
|
|
cairo_clip_t *clip,
|
|
|
|
|
cairo_rectangle_int_t *extents)
|
|
|
|
|
{
|
|
|
|
|
cairo_composite_traps_info_t traps_info;
|
|
|
|
|
cairo_region_t *clip_region = NULL;
|
|
|
|
|
cairo_bool_t clip_surface = FALSE;
|
|
|
|
|
cairo_status_t status;
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (_cairo_operator_bounded_by_mask (op) && traps->num_traps == 0)
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (clip != NULL) {
|
|
|
|
|
status = _cairo_clip_get_region (clip, &clip_region);
|
|
|
|
|
if (unlikely (_cairo_status_is_error (status)))
|
|
|
|
|
return status;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* The all-clipped state should not have been propagated this far. */
|
|
|
|
|
assert (status != CAIRO_INT_STATUS_NOTHING_TO_DO);
|
|
|
|
|
clip_surface = status == CAIRO_INT_STATUS_UNSUPPORTED;
|
|
|
|
|
}
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* Use a fast path if the trapezoids consist of a simple region,
|
|
|
|
|
* but we can only do this if we do not have a clip surface, or can
|
|
|
|
|
* substitute the mask with the clip.
|
|
|
|
|
*/
|
|
|
|
|
if (! clip_surface ||
|
|
|
|
|
(_cairo_operator_bounded_by_mask (op) && op != CAIRO_OPERATOR_SOURCE))
|
|
|
|
|
{
|
|
|
|
|
cairo_region_t *trap_region = NULL;
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = _cairo_traps_extract_region (traps, &trap_region);
|
|
|
|
|
if (unlikely (_cairo_status_is_error (status)))
|
|
|
|
|
return status;
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (trap_region != NULL) {
|
|
|
|
|
status = cairo_region_intersect_rectangle (trap_region, extents);
|
|
|
|
|
if (unlikely (status)) {
|
|
|
|
|
cairo_region_destroy (trap_region);
|
|
|
|
|
return status;
|
|
|
|
|
}
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (clip_region != NULL) {
|
|
|
|
|
status = cairo_region_intersect (trap_region, clip_region);
|
|
|
|
|
if (unlikely (status)) {
|
|
|
|
|
cairo_region_destroy (trap_region);
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
}
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (cairo_region_is_empty (trap_region) &&
|
|
|
|
|
_cairo_operator_bounded_by_mask (op))
|
|
|
|
|
{
|
|
|
|
|
cairo_region_destroy (trap_region);
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2008-12-11 15:29:23 -05:00
|
|
|
}
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = _clip_and_composite_region (src, op, dst,
|
|
|
|
|
trap_region,
|
|
|
|
|
antialias,
|
|
|
|
|
clip, extents);
|
|
|
|
|
cairo_region_destroy (trap_region);
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (likely (status != CAIRO_INT_STATUS_UNSUPPORTED))
|
|
|
|
|
return status;
|
|
|
|
|
}
|
2005-12-20 09:37:15 +00:00
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* Otherwise we need to render the trapezoids to a mask and composite
|
|
|
|
|
* in the usual fashion.
|
|
|
|
|
*/
|
|
|
|
|
if (_cairo_operator_bounded_by_mask (op)) {
|
|
|
|
|
cairo_rectangle_int_t trap_extents;
|
|
|
|
|
cairo_box_t trap_box;
|
2007-03-14 00:28:49 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_traps_extents (traps, &trap_box);
|
|
|
|
|
_cairo_box_round_to_rectangle (&trap_box, &trap_extents);
|
|
|
|
|
if (! _cairo_rectangle_intersect (extents, &trap_extents))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
|
|
|
|
}
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
traps_info.traps = traps;
|
|
|
|
|
traps_info.antialias = antialias;
|
|
|
|
|
return _clip_and_composite (clip, op, src,
|
|
|
|
|
_composite_traps_draw_func,
|
|
|
|
|
&traps_info, dst, extents);
|
2005-12-20 09:37:15 +00:00
|
|
|
}
|
|
|
|
|
|
2008-07-24 20:47:14 +03:00
|
|
|
typedef struct {
|
|
|
|
|
cairo_path_fixed_t *path;
|
|
|
|
|
cairo_fill_rule_t fill_rule;
|
|
|
|
|
double tolerance;
|
|
|
|
|
cairo_antialias_t antialias;
|
|
|
|
|
} cairo_composite_spans_fill_info_t;
|
|
|
|
|
|
|
|
|
|
static cairo_status_t
|
|
|
|
|
_composite_spans_fill_func (void *closure,
|
|
|
|
|
cairo_operator_t op,
|
|
|
|
|
const cairo_pattern_t *src,
|
|
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
int dst_x,
|
|
|
|
|
int dst_y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
const cairo_rectangle_int_t *extents,
|
|
|
|
|
cairo_region_t *clip_region)
|
2008-07-24 20:47:14 +03:00
|
|
|
{
|
|
|
|
|
cairo_composite_rectangles_t rects;
|
|
|
|
|
cairo_composite_spans_fill_info_t *info = closure;
|
|
|
|
|
|
|
|
|
|
_cairo_composite_rectangles_init (
|
|
|
|
|
&rects, extents->x, extents->y,
|
|
|
|
|
extents->width, extents->height);
|
|
|
|
|
|
|
|
|
|
/* The incoming dst_x/y are where we're pretending the origin of
|
|
|
|
|
* the dst surface is -- *not* the offset of a rectangle where
|
|
|
|
|
* we'd like to place the result. */
|
|
|
|
|
rects.dst.x -= dst_x;
|
|
|
|
|
rects.dst.y -= dst_y;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
return _cairo_path_fixed_fill_using_spans (op, src, info->path, dst,
|
|
|
|
|
info->fill_rule,
|
|
|
|
|
info->tolerance,
|
|
|
|
|
info->antialias,
|
|
|
|
|
&rects,
|
|
|
|
|
clip_region);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static cairo_bool_t
|
|
|
|
|
_rectangle_intersect_clip (cairo_rectangle_int_t *extents, cairo_clip_t *clip)
|
|
|
|
|
{
|
|
|
|
|
if (clip != NULL) {
|
|
|
|
|
const cairo_rectangle_int_t *clip_extents;
|
|
|
|
|
|
|
|
|
|
clip_extents = _cairo_clip_get_extents (clip);
|
|
|
|
|
if (clip_extents != NULL)
|
|
|
|
|
return _cairo_rectangle_intersect (extents, clip_extents);
|
2009-05-26 10:41:53 +01:00
|
|
|
}
|
2008-07-24 20:47:14 +03:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
return ! _cairo_rectangle_empty (extents);
|
2008-07-24 20:47:14 +03:00
|
|
|
}
|
|
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_status_t
|
2008-10-22 19:24:44 +01:00
|
|
|
_cairo_surface_fallback_paint (cairo_surface_t *surface,
|
|
|
|
|
cairo_operator_t op,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
const cairo_pattern_t *source,
|
|
|
|
|
cairo_clip_t *clip)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
cairo_status_t status;
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t extents;
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_box_t box;
|
|
|
|
|
cairo_traps_t traps;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_bool_t is_bounded;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
is_bounded = _cairo_surface_get_extents (surface, &extents);
|
|
|
|
|
assert (is_bounded || clip);
|
2005-12-19 22:45:41 +00:00
|
|
|
|
|
|
|
|
if (_cairo_operator_bounded_by_source (op)) {
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t source_extents;
|
2008-10-22 19:24:44 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_get_extents (source, &source_extents);
|
2008-10-23 14:34:30 +01:00
|
|
|
if (! _cairo_rectangle_intersect (&extents, &source_extents))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (! _rectangle_intersect_clip (&extents, clip))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
2008-09-24 11:12:05 +01:00
|
|
|
_cairo_box_from_rectangle (&box, &extents);
|
2005-12-19 22:45:41 +00:00
|
|
|
|
2008-09-19 10:54:13 +01:00
|
|
|
_cairo_traps_init_box (&traps, &box);
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = _clip_and_composite_trapezoids (source, op, surface,
|
|
|
|
|
&traps, CAIRO_ANTIALIAS_NONE,
|
|
|
|
|
clip, &extents);
|
2005-12-19 22:45:41 +00:00
|
|
|
_cairo_traps_fini (&traps);
|
|
|
|
|
|
2007-05-09 13:39:08 +01:00
|
|
|
return status;
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_cairo_surface_mask_draw_func (void *closure,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *src,
|
2006-05-04 03:43:34 -07:00
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
int dst_x,
|
|
|
|
|
int dst_y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
const cairo_rectangle_int_t *extents,
|
|
|
|
|
cairo_region_t *clip_region)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
cairo_pattern_t *mask = closure;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (src) {
|
2005-12-19 22:45:41 +00:00
|
|
|
return _cairo_surface_composite (op,
|
|
|
|
|
src, mask, dst,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
extents->x - dst_x, extents->y - dst_y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
extents->width, extents->height,
|
|
|
|
|
clip_region);
|
|
|
|
|
} else {
|
2005-12-19 22:45:41 +00:00
|
|
|
return _cairo_surface_composite (op,
|
|
|
|
|
mask, NULL, dst,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
0, 0, /* unused */
|
|
|
|
|
extents->x - dst_x, extents->y - dst_y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
extents->width, extents->height,
|
|
|
|
|
clip_region);
|
|
|
|
|
}
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_status_t
|
|
|
|
|
_cairo_surface_fallback_mask (cairo_surface_t *surface,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *source,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
const cairo_pattern_t *mask,
|
|
|
|
|
cairo_clip_t *clip)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_rectangle_int_t extents;
|
|
|
|
|
cairo_bool_t is_bounded;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
is_bounded = _cairo_surface_get_extents (surface, &extents);
|
|
|
|
|
assert (is_bounded || clip);
|
2005-12-19 22:45:41 +00:00
|
|
|
|
|
|
|
|
if (_cairo_operator_bounded_by_source (op)) {
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_rectangle_int_t source_extents;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_get_extents (source, &source_extents);
|
2008-10-23 14:34:30 +01:00
|
|
|
if (! _cairo_rectangle_intersect (&extents, &source_extents))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
if (_cairo_operator_bounded_by_mask (op)) {
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_rectangle_int_t mask_extents;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_get_extents (mask, &mask_extents);
|
2008-10-23 14:34:30 +01:00
|
|
|
if (! _cairo_rectangle_intersect (&extents, &mask_extents))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (! _rectangle_intersect_clip (&extents, clip))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
return _clip_and_composite (clip, op, source,
|
|
|
|
|
_cairo_surface_mask_draw_func,
|
|
|
|
|
(void *) mask,
|
|
|
|
|
surface, &extents);
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cairo_status_t
|
|
|
|
|
_cairo_surface_fallback_stroke (cairo_surface_t *surface,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *source,
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_path_fixed_t *path,
|
|
|
|
|
cairo_stroke_style_t *stroke_style,
|
|
|
|
|
cairo_matrix_t *ctm,
|
|
|
|
|
cairo_matrix_t *ctm_inverse,
|
|
|
|
|
double tolerance,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_antialias_t antialias,
|
|
|
|
|
cairo_clip_t *clip)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
cairo_status_t status;
|
|
|
|
|
cairo_traps_t traps;
|
2007-04-27 17:23:08 -07:00
|
|
|
cairo_box_t box;
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t extents;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_bool_t is_bounded;
|
2007-04-27 17:23:08 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
is_bounded = _cairo_surface_get_extents (surface, &extents);
|
|
|
|
|
assert (is_bounded || clip);
|
2007-04-27 17:23:08 -07:00
|
|
|
|
|
|
|
|
if (_cairo_operator_bounded_by_source (op)) {
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t source_extents;
|
2007-04-27 17:23:08 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_get_extents (source, &source_extents);
|
2008-10-23 14:34:30 +01:00
|
|
|
if (! _cairo_rectangle_intersect (&extents, &source_extents))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2007-04-27 17:23:08 -07:00
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (! _rectangle_intersect_clip (&extents, clip))
|
2008-09-15 15:41:36 +01:00
|
|
|
return CAIRO_STATUS_SUCCESS;
|
|
|
|
|
|
2008-09-24 11:12:05 +01:00
|
|
|
_cairo_box_from_rectangle (&box, &extents);
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
_cairo_traps_init (&traps);
|
2007-04-27 17:23:08 -07:00
|
|
|
_cairo_traps_limit (&traps, &box);
|
|
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
status = _cairo_path_fixed_stroke_to_traps (path,
|
|
|
|
|
stroke_style,
|
|
|
|
|
ctm, ctm_inverse,
|
|
|
|
|
tolerance,
|
|
|
|
|
&traps);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2008-09-19 10:54:13 +01:00
|
|
|
goto FAIL;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = _clip_and_composite_trapezoids (source, op,
|
|
|
|
|
surface, &traps, antialias,
|
|
|
|
|
clip, &extents);
|
|
|
|
|
FAIL:
|
2005-12-19 22:45:41 +00:00
|
|
|
_cairo_traps_fini (&traps);
|
|
|
|
|
|
2007-05-09 13:39:08 +01:00
|
|
|
return status;
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cairo_status_t
|
|
|
|
|
_cairo_surface_fallback_fill (cairo_surface_t *surface,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *source,
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_path_fixed_t *path,
|
|
|
|
|
cairo_fill_rule_t fill_rule,
|
|
|
|
|
double tolerance,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_antialias_t antialias,
|
|
|
|
|
cairo_clip_t *clip)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
cairo_status_t status;
|
|
|
|
|
cairo_traps_t traps;
|
2007-04-27 17:23:08 -07:00
|
|
|
cairo_box_t box;
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t extents;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_bool_t is_bounded;
|
2007-04-27 17:23:08 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
is_bounded = _cairo_surface_get_extents (surface, &extents);
|
|
|
|
|
assert (is_bounded || clip);
|
2007-04-27 17:23:08 -07:00
|
|
|
|
|
|
|
|
if (_cairo_operator_bounded_by_source (op)) {
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t source_extents;
|
2008-10-23 14:34:30 +01:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
_cairo_pattern_get_extents (source, &source_extents);
|
2008-10-23 14:34:30 +01:00
|
|
|
if (! _cairo_rectangle_intersect (&extents, &source_extents))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2007-04-27 17:23:08 -07:00
|
|
|
}
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (! _rectangle_intersect_clip (&extents, clip))
|
2008-09-15 15:41:36 +01:00
|
|
|
return CAIRO_STATUS_SUCCESS;
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
/* XXX future direction:
|
|
|
|
|
status = _cairo_path_fixed_fill_to_region (path, fill_rule, ®ion);
|
|
|
|
|
if (status != CAIRO_STATUS_INT_UNSUPPORTED) {
|
|
|
|
|
if (unlikely (status))
|
|
|
|
|
return status;
|
|
|
|
|
|
|
|
|
|
status = _clip_and_composite_region ();
|
|
|
|
|
if (status != CAIRO_INT_STATUS_UNSUPPORTED)
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
*/
|
|
|
|
|
|
2008-07-24 20:47:14 +03:00
|
|
|
/* Ask if the surface would like to render this combination of
|
|
|
|
|
* op/source/dst/antialias with spans or not, but don't actually
|
|
|
|
|
* make a renderer yet. We'll try to hit the region optimisations
|
|
|
|
|
* in _clip_and_composite_trapezoids() if it looks like the path
|
|
|
|
|
* is a region. */
|
|
|
|
|
/* TODO: Until we have a mono scan converter we won't even try
|
|
|
|
|
* to use spans for CAIRO_ANTIALIAS_NONE. */
|
|
|
|
|
/* TODO: The region filling code should be lifted from
|
|
|
|
|
* _clip_and_composite_trapezoids() and given first priority
|
|
|
|
|
* explicitly before deciding between spans and trapezoids. */
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (antialias != CAIRO_ANTIALIAS_NONE && ! path->is_rectilinear &&
|
2008-07-24 20:47:14 +03:00
|
|
|
_cairo_surface_check_span_renderer (
|
|
|
|
|
op, source, surface, antialias, NULL))
|
|
|
|
|
{
|
|
|
|
|
cairo_composite_spans_fill_info_t info;
|
|
|
|
|
info.path = path;
|
|
|
|
|
info.fill_rule = fill_rule;
|
|
|
|
|
info.tolerance = tolerance;
|
|
|
|
|
info.antialias = antialias;
|
|
|
|
|
|
|
|
|
|
if (_cairo_operator_bounded_by_mask (op)) {
|
2008-12-18 14:52:03 +00:00
|
|
|
cairo_rectangle_int_t path_extents;
|
2008-12-18 11:50:00 +00:00
|
|
|
|
2009-01-14 18:55:32 +00:00
|
|
|
_cairo_path_fixed_approximate_clip_extents (path,
|
2008-12-18 14:52:03 +00:00
|
|
|
&path_extents);
|
|
|
|
|
if (! _cairo_rectangle_intersect (&extents, &path_extents))
|
2008-07-24 20:47:14 +03:00
|
|
|
return CAIRO_STATUS_SUCCESS;
|
|
|
|
|
}
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
return _clip_and_composite (clip, op, source,
|
|
|
|
|
_composite_spans_fill_func,
|
|
|
|
|
&info,
|
|
|
|
|
surface,
|
|
|
|
|
&extents);
|
2008-07-24 20:47:14 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Fall back to trapezoid fills. */
|
|
|
|
|
_cairo_box_from_rectangle (&box, &extents);
|
2005-12-19 22:45:41 +00:00
|
|
|
_cairo_traps_init (&traps);
|
2007-04-27 17:23:08 -07:00
|
|
|
_cairo_traps_limit (&traps, &box);
|
|
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
status = _cairo_path_fixed_fill_to_traps (path,
|
|
|
|
|
fill_rule,
|
|
|
|
|
tolerance,
|
|
|
|
|
&traps);
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (unlikely (status))
|
|
|
|
|
goto FAIL;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = _clip_and_composite_trapezoids (source, op, surface,
|
|
|
|
|
&traps, antialias,
|
|
|
|
|
clip, &extents);
|
|
|
|
|
FAIL:
|
2005-12-19 22:45:41 +00:00
|
|
|
_cairo_traps_fini (&traps);
|
|
|
|
|
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
|
cairo_scaled_font_t *font;
|
2007-08-31 16:53:21 +01:00
|
|
|
cairo_glyph_t *glyphs;
|
2005-12-19 22:45:41 +00:00
|
|
|
int num_glyphs;
|
|
|
|
|
} cairo_show_glyphs_info_t;
|
|
|
|
|
|
|
|
|
|
static cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_cairo_surface_old_show_glyphs_draw_func (void *closure,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *src,
|
2006-05-04 03:43:34 -07:00
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
int dst_x,
|
|
|
|
|
int dst_y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
const cairo_rectangle_int_t *extents,
|
|
|
|
|
cairo_region_t *clip_region)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
cairo_show_glyphs_info_t *glyph_info = closure;
|
|
|
|
|
cairo_status_t status;
|
|
|
|
|
|
|
|
|
|
/* Modifying the glyph array is fine because we know that this function
|
|
|
|
|
* will be called only once, and we've already made a copy of the
|
|
|
|
|
* glyphs in the wrapper.
|
|
|
|
|
*/
|
|
|
|
|
if (dst_x != 0 || dst_y != 0) {
|
|
|
|
|
int i;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
for (i = 0; i < glyph_info->num_glyphs; ++i) {
|
2005-12-19 22:45:41 +00:00
|
|
|
((cairo_glyph_t *) glyph_info->glyphs)[i].x -= dst_x;
|
|
|
|
|
((cairo_glyph_t *) glyph_info->glyphs)[i].y -= dst_y;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2006-06-06 15:41:31 -07:00
|
|
|
status = _cairo_surface_old_show_glyphs (glyph_info->font, op, src,
|
2005-12-19 22:45:41 +00:00
|
|
|
dst,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
extents->x - dst_x,
|
|
|
|
|
extents->y - dst_y,
|
|
|
|
|
extents->width,
|
|
|
|
|
extents->height,
|
|
|
|
|
glyph_info->glyphs,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
glyph_info->num_glyphs,
|
|
|
|
|
clip_region);
|
2005-12-19 22:45:41 +00:00
|
|
|
if (status != CAIRO_INT_STATUS_UNSUPPORTED)
|
|
|
|
|
return status;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2009-05-26 10:41:53 +01:00
|
|
|
return _cairo_scaled_font_show_glyphs (glyph_info->font,
|
|
|
|
|
op,
|
|
|
|
|
src, dst,
|
|
|
|
|
extents->x, extents->y,
|
|
|
|
|
extents->x - dst_x,
|
|
|
|
|
extents->y - dst_y,
|
|
|
|
|
extents->width, extents->height,
|
|
|
|
|
glyph_info->glyphs,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
glyph_info->num_glyphs,
|
|
|
|
|
clip_region);
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cairo_status_t
|
|
|
|
|
_cairo_surface_fallback_show_glyphs (cairo_surface_t *surface,
|
|
|
|
|
cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *source,
|
2007-08-31 16:53:21 +01:00
|
|
|
cairo_glyph_t *glyphs,
|
2005-12-19 22:45:41 +00:00
|
|
|
int num_glyphs,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_scaled_font_t *scaled_font,
|
|
|
|
|
cairo_clip_t *clip)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
cairo_status_t status;
|
2007-06-18 17:29:04 -07:00
|
|
|
cairo_rectangle_int_t extents;
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_show_glyphs_info_t glyph_info;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_bool_t is_bounded;
|
2005-12-19 22:45:41 +00:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
is_bounded = _cairo_surface_get_extents (surface, &extents);
|
|
|
|
|
assert (is_bounded || clip);
|
2005-12-19 22:45:41 +00:00
|
|
|
|
|
|
|
|
if (_cairo_operator_bounded_by_mask (op)) {
|
2008-01-22 15:32:11 -08:00
|
|
|
cairo_rectangle_int_t glyph_extents;
|
2008-10-21 22:48:17 +01:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
status = _cairo_scaled_font_glyph_device_extents (scaled_font,
|
2006-06-06 15:41:31 -07:00
|
|
|
glyphs,
|
|
|
|
|
num_glyphs,
|
2005-12-19 22:45:41 +00:00
|
|
|
&glyph_extents);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2005-12-19 22:45:41 +00:00
|
|
|
return status;
|
|
|
|
|
|
2008-10-23 14:34:30 +01:00
|
|
|
if (! _cairo_rectangle_intersect (&extents, &glyph_extents))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (! _rectangle_intersect_clip (&extents, clip))
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
glyph_info.font = scaled_font;
|
|
|
|
|
glyph_info.glyphs = glyphs;
|
|
|
|
|
glyph_info.num_glyphs = num_glyphs;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
return _clip_and_composite (clip, op, source,
|
|
|
|
|
_cairo_surface_old_show_glyphs_draw_func,
|
|
|
|
|
&glyph_info,
|
|
|
|
|
surface,
|
|
|
|
|
&extents);
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cairo_surface_t *
|
|
|
|
|
_cairo_surface_fallback_snapshot (cairo_surface_t *surface)
|
|
|
|
|
{
|
|
|
|
|
cairo_surface_t *snapshot;
|
|
|
|
|
cairo_status_t status;
|
2009-05-05 17:05:39 +01:00
|
|
|
cairo_format_t format;
|
2008-05-28 15:42:09 +01:00
|
|
|
cairo_surface_pattern_t pattern;
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_image_surface_t *image;
|
|
|
|
|
void *image_extra;
|
|
|
|
|
|
|
|
|
|
status = _cairo_surface_acquire_source_image (surface,
|
|
|
|
|
&image, &image_extra);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status))
|
2008-01-16 16:23:23 +00:00
|
|
|
return _cairo_surface_create_in_error (status);
|
2005-12-19 22:45:41 +00:00
|
|
|
|
2009-05-05 17:05:39 +01:00
|
|
|
format = image->format;
|
|
|
|
|
if (format == CAIRO_FORMAT_INVALID) {
|
|
|
|
|
/* Non-standard images formats can be generated when retrieving
|
|
|
|
|
* images from unusual xservers, for example.
|
|
|
|
|
*/
|
|
|
|
|
format = _cairo_format_from_content (image->base.content);
|
|
|
|
|
}
|
|
|
|
|
snapshot = cairo_image_surface_create (format,
|
2005-12-19 22:45:41 +00:00
|
|
|
image->width,
|
|
|
|
|
image->height);
|
2007-10-05 09:42:23 +01:00
|
|
|
if (cairo_surface_status (snapshot)) {
|
2009-02-13 12:56:46 +00:00
|
|
|
_cairo_surface_release_source_image (surface, image, image_extra);
|
2005-12-19 22:45:41 +00:00
|
|
|
return snapshot;
|
2007-10-05 09:42:23 +01:00
|
|
|
}
|
2005-12-19 22:45:41 +00:00
|
|
|
|
2008-05-28 15:42:09 +01:00
|
|
|
_cairo_pattern_init_for_surface (&pattern, &image->base);
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = _cairo_surface_paint (snapshot,
|
|
|
|
|
CAIRO_OPERATOR_SOURCE,
|
|
|
|
|
&pattern.base,
|
|
|
|
|
NULL);
|
2005-12-19 22:45:41 +00:00
|
|
|
_cairo_pattern_fini (&pattern.base);
|
2009-02-13 12:56:46 +00:00
|
|
|
_cairo_surface_release_source_image (surface, image, image_extra);
|
|
|
|
|
if (unlikely (status)) {
|
|
|
|
|
cairo_surface_destroy (snapshot);
|
|
|
|
|
return _cairo_surface_create_in_error (status);
|
|
|
|
|
}
|
2007-10-05 09:42:23 +01:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
return snapshot;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cairo_status_t
|
2008-10-22 19:24:44 +01:00
|
|
|
_cairo_surface_fallback_composite (cairo_operator_t op,
|
|
|
|
|
const cairo_pattern_t *src,
|
|
|
|
|
const cairo_pattern_t *mask,
|
|
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
int src_x,
|
|
|
|
|
int src_y,
|
|
|
|
|
int mask_x,
|
|
|
|
|
int mask_y,
|
|
|
|
|
int dst_x,
|
|
|
|
|
int dst_y,
|
|
|
|
|
unsigned int width,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
unsigned int height,
|
|
|
|
|
cairo_region_t *clip_region)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
fallback_state_t state;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_region_t *fallback_region = NULL;
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_status_t status;
|
|
|
|
|
|
|
|
|
|
status = _fallback_init (&state, dst, dst_x, dst_y, width, height);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status)) {
|
2005-12-19 22:45:41 +00:00
|
|
|
if (status == CAIRO_INT_STATUS_NOTHING_TO_DO)
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
2006-05-04 01:45:41 -07:00
|
|
|
/* We know this will never fail with the image backend; but
|
|
|
|
|
* instead of calling into it directly, we call
|
|
|
|
|
* _cairo_surface_composite so that we get the correct device
|
|
|
|
|
* offset handling.
|
|
|
|
|
*/
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
|
|
|
|
|
if (clip_region != NULL && (state.image_rect.x || state.image_rect.y)) {
|
|
|
|
|
fallback_region = cairo_region_copy (clip_region);
|
|
|
|
|
status = fallback_region->status;
|
|
|
|
|
if (unlikely (status))
|
|
|
|
|
goto FAIL;
|
|
|
|
|
|
|
|
|
|
cairo_region_translate (fallback_region,
|
|
|
|
|
-state.image_rect.x,
|
|
|
|
|
-state.image_rect.y);
|
|
|
|
|
clip_region = fallback_region;
|
|
|
|
|
}
|
|
|
|
|
|
2006-05-04 01:45:41 -07:00
|
|
|
status = _cairo_surface_composite (op, src, mask,
|
|
|
|
|
&state.image->base,
|
|
|
|
|
src_x, src_y, mask_x, mask_y,
|
|
|
|
|
dst_x - state.image_rect.x,
|
|
|
|
|
dst_y - state.image_rect.y,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
width, height,
|
|
|
|
|
clip_region);
|
|
|
|
|
FAIL:
|
|
|
|
|
if (fallback_region != NULL)
|
|
|
|
|
cairo_region_destroy (fallback_region);
|
2005-12-19 22:45:41 +00:00
|
|
|
_fallback_fini (&state);
|
|
|
|
|
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cairo_status_t
|
2006-05-04 03:43:34 -07:00
|
|
|
_cairo_surface_fallback_fill_rectangles (cairo_surface_t *surface,
|
|
|
|
|
cairo_operator_t op,
|
|
|
|
|
const cairo_color_t *color,
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t *rects,
|
2006-05-04 03:43:34 -07:00
|
|
|
int num_rects)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
fallback_state_t state;
|
2007-06-18 16:56:24 -07:00
|
|
|
cairo_rectangle_int_t *offset_rects = NULL;
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_status_t status;
|
|
|
|
|
int x1, y1, x2, y2;
|
|
|
|
|
int i;
|
|
|
|
|
|
2009-05-26 21:07:07 +01:00
|
|
|
assert (surface->snapshot_of == NULL);
|
2005-12-19 22:45:41 +00:00
|
|
|
|
|
|
|
|
if (num_rects <= 0)
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
|
|
|
|
|
|
|
|
|
/* Compute the bounds of the rectangles, so that we know what area of the
|
|
|
|
|
* destination surface to fetch
|
|
|
|
|
*/
|
|
|
|
|
x1 = rects[0].x;
|
|
|
|
|
y1 = rects[0].y;
|
|
|
|
|
x2 = rects[0].x + rects[0].width;
|
|
|
|
|
y2 = rects[0].y + rects[0].height;
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
for (i = 1; i < num_rects; i++) {
|
|
|
|
|
if (rects[i].x < x1)
|
|
|
|
|
x1 = rects[i].x;
|
|
|
|
|
if (rects[i].y < y1)
|
|
|
|
|
y1 = rects[i].y;
|
|
|
|
|
|
2008-10-23 14:34:30 +01:00
|
|
|
if ((int) (rects[i].x + rects[i].width) > x2)
|
2005-12-19 22:45:41 +00:00
|
|
|
x2 = rects[i].x + rects[i].width;
|
2008-10-23 14:34:30 +01:00
|
|
|
if ((int) (rects[i].y + rects[i].height) > y2)
|
2005-12-19 22:45:41 +00:00
|
|
|
y2 = rects[i].y + rects[i].height;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
status = _fallback_init (&state, surface, x1, y1, x2 - x1, y2 - y1);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status)) {
|
2005-12-19 22:45:41 +00:00
|
|
|
if (status == CAIRO_INT_STATUS_NOTHING_TO_DO)
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* If the fetched image isn't at 0,0, we need to offset the rectangles */
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
if (state.image_rect.x != 0 || state.image_rect.y != 0) {
|
2007-06-18 16:56:24 -07:00
|
|
|
offset_rects = _cairo_malloc_ab (num_rects, sizeof (cairo_rectangle_int_t));
|
2008-11-18 17:26:55 +00:00
|
|
|
if (unlikely (offset_rects == NULL)) {
|
2007-10-04 13:15:46 +01:00
|
|
|
status = _cairo_error (CAIRO_STATUS_NO_MEMORY);
|
2005-12-19 22:45:41 +00:00
|
|
|
goto DONE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < num_rects; i++) {
|
|
|
|
|
offset_rects[i].x = rects[i].x - state.image_rect.x;
|
|
|
|
|
offset_rects[i].y = rects[i].y - state.image_rect.y;
|
|
|
|
|
offset_rects[i].width = rects[i].width;
|
|
|
|
|
offset_rects[i].height = rects[i].height;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
rects = offset_rects;
|
|
|
|
|
}
|
|
|
|
|
|
2006-05-04 01:45:41 -07:00
|
|
|
status = _cairo_surface_fill_rectangles (&state.image->base,
|
|
|
|
|
op, color,
|
|
|
|
|
rects, num_rects);
|
2005-12-19 22:45:41 +00:00
|
|
|
|
|
|
|
|
free (offset_rects);
|
|
|
|
|
|
|
|
|
|
DONE:
|
|
|
|
|
_fallback_fini (&state);
|
|
|
|
|
|
|
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cairo_status_t
|
|
|
|
|
_cairo_surface_fallback_composite_trapezoids (cairo_operator_t op,
|
2008-10-22 19:24:44 +01:00
|
|
|
const cairo_pattern_t *pattern,
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_surface_t *dst,
|
|
|
|
|
cairo_antialias_t antialias,
|
|
|
|
|
int src_x,
|
|
|
|
|
int src_y,
|
|
|
|
|
int dst_x,
|
|
|
|
|
int dst_y,
|
|
|
|
|
unsigned int width,
|
|
|
|
|
unsigned int height,
|
|
|
|
|
cairo_trapezoid_t *traps,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
int num_traps,
|
|
|
|
|
cairo_region_t *clip_region)
|
2005-12-19 22:45:41 +00:00
|
|
|
{
|
|
|
|
|
fallback_state_t state;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
cairo_region_t *fallback_region = NULL;
|
2005-12-19 22:45:41 +00:00
|
|
|
cairo_trapezoid_t *offset_traps = NULL;
|
|
|
|
|
cairo_status_t status;
|
|
|
|
|
|
|
|
|
|
status = _fallback_init (&state, dst, dst_x, dst_y, width, height);
|
2008-11-18 15:38:37 +00:00
|
|
|
if (unlikely (status)) {
|
2005-12-19 22:45:41 +00:00
|
|
|
if (status == CAIRO_INT_STATUS_NOTHING_TO_DO)
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
status = CAIRO_STATUS_SUCCESS;
|
2005-12-19 22:45:41 +00:00
|
|
|
return status;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* If the destination image isn't at 0,0, we need to offset the trapezoids */
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
if (state.image_rect.x != 0 || state.image_rect.y != 0) {
|
2007-06-19 13:15:21 -07:00
|
|
|
offset_traps = _cairo_malloc_ab (num_traps, sizeof (cairo_trapezoid_t));
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (offset_traps == NULL) {
|
2007-10-04 13:15:46 +01:00
|
|
|
status = _cairo_error (CAIRO_STATUS_NO_MEMORY);
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
goto FAIL;
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
2006-05-04 01:45:41 -07:00
|
|
|
_cairo_trapezoid_array_translate_and_scale (offset_traps, traps, num_traps,
|
|
|
|
|
- state.image_rect.x, - state.image_rect.y,
|
|
|
|
|
1.0, 1.0);
|
2005-12-19 22:45:41 +00:00
|
|
|
traps = offset_traps;
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
|
|
|
|
|
/* similarly we need to adjust the region */
|
|
|
|
|
if (clip_region != NULL) {
|
|
|
|
|
fallback_region = cairo_region_copy (clip_region);
|
|
|
|
|
status = fallback_region->status;
|
|
|
|
|
if (unlikely (status))
|
|
|
|
|
goto FAIL;
|
|
|
|
|
|
|
|
|
|
cairo_region_translate (fallback_region,
|
|
|
|
|
-state.image_rect.x,
|
|
|
|
|
-state.image_rect.y);
|
|
|
|
|
clip_region = fallback_region;
|
|
|
|
|
}
|
2005-12-19 22:45:41 +00:00
|
|
|
}
|
|
|
|
|
|
2007-04-08 21:34:25 +01:00
|
|
|
status = _cairo_surface_composite_trapezoids (op, pattern,
|
|
|
|
|
&state.image->base,
|
|
|
|
|
antialias,
|
|
|
|
|
src_x, src_y,
|
|
|
|
|
dst_x - state.image_rect.x,
|
|
|
|
|
dst_y - state.image_rect.y,
|
|
|
|
|
width, height,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
traps, num_traps,
|
|
|
|
|
clip_region);
|
|
|
|
|
if (offset_traps != NULL)
|
2005-12-19 22:45:41 +00:00
|
|
|
free (offset_traps);
|
|
|
|
|
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
FAIL:
|
|
|
|
|
if (fallback_region != NULL)
|
|
|
|
|
cairo_region_destroy (fallback_region);
|
2005-12-19 22:45:41 +00:00
|
|
|
_fallback_fini (&state);
|
2006-06-06 15:25:49 -07:00
|
|
|
|
2005-12-19 22:45:41 +00:00
|
|
|
return status;
|
|
|
|
|
}
|
2007-08-27 16:30:52 -07:00
|
|
|
|
|
|
|
|
cairo_status_t
|
|
|
|
|
_cairo_surface_fallback_clone_similar (cairo_surface_t *surface,
|
|
|
|
|
cairo_surface_t *src,
|
2009-05-09 10:10:14 +01:00
|
|
|
cairo_content_t content,
|
2007-08-27 16:30:52 -07:00
|
|
|
int src_x,
|
|
|
|
|
int src_y,
|
|
|
|
|
int width,
|
|
|
|
|
int height,
|
2008-09-27 17:24:57 +01:00
|
|
|
int *clone_offset_x,
|
|
|
|
|
int *clone_offset_y,
|
2007-08-27 16:30:52 -07:00
|
|
|
cairo_surface_t **clone_out)
|
|
|
|
|
{
|
2008-12-10 23:06:22 +00:00
|
|
|
cairo_surface_t *new_surface;
|
|
|
|
|
cairo_surface_pattern_t pattern;
|
2007-08-27 16:30:52 -07:00
|
|
|
cairo_status_t status;
|
|
|
|
|
|
|
|
|
|
new_surface = _cairo_surface_create_similar_scratch (surface,
|
2009-05-09 10:10:14 +01:00
|
|
|
src->content & content,
|
2007-08-27 16:30:52 -07:00
|
|
|
width, height);
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
if (new_surface == NULL)
|
|
|
|
|
return CAIRO_INT_STATUS_UNSUPPORTED;
|
|
|
|
|
if (unlikely (new_surface->status))
|
2007-08-27 16:30:52 -07:00
|
|
|
return new_surface->status;
|
|
|
|
|
|
2007-08-28 16:47:24 -07:00
|
|
|
/* We have to copy these here, so that the coordinate spaces are correct */
|
|
|
|
|
new_surface->device_transform = src->device_transform;
|
|
|
|
|
new_surface->device_transform_inverse = src->device_transform_inverse;
|
2007-08-27 16:30:52 -07:00
|
|
|
|
2008-12-10 23:06:22 +00:00
|
|
|
_cairo_pattern_init_for_surface (&pattern, src);
|
|
|
|
|
cairo_matrix_init_translate (&pattern.base.matrix, src_x, src_y);
|
|
|
|
|
pattern.base.filter = CAIRO_FILTER_NEAREST;
|
|
|
|
|
|
|
|
|
|
status = _cairo_surface_paint (new_surface,
|
|
|
|
|
CAIRO_OPERATOR_SOURCE,
|
Remove clip handling from generic surface layer.
Handling clip as part of the surface state, as opposed to being part of
the operation state, is cumbersome and a hindrance to providing true proxy
surface support. For example, the clip must be copied from the surface
onto the fallback image, but this was forgotten causing undue hassle in
each backend. Another example is the contortion the meta surface
endures to ensure the clip is correctly recorded. By contrast passing the
clip along with the operation is quite simple and enables us to write
generic handlers for providing surface wrappers. (And in the future, we
should be able to write more esoteric wrappers, e.g. automatic 2x FSAA,
trivially.)
In brief, instead of the surface automatically applying the clip before
calling the backend, the backend can call into a generic helper to apply
clipping. For raster surfaces, clip regions are handled automatically as
part of the composite interface. For vector surfaces, a clip helper is
introduced to replay and callback into an intersect_clip_path() function
as necessary.
Whilst this is not primarily a performance related change (the change
should just move the computation of the clip from the moment it is applied
by the user to the moment it is required by the backend), it is important
to track any potential regression:
ppc:
Speedups
========
image-rgba evolution-20090607-0 1026085.22 0.18% -> 672972.07 0.77%: 1.52x speedup
▌
image-rgba evolution-20090618-0 680579.98 0.12% -> 573237.66 0.16%: 1.19x speedup
▎
image-rgba swfdec-fill-rate-4xaa-0 460296.92 0.36% -> 407464.63 0.42%: 1.13x speedup
▏
image-rgba swfdec-fill-rate-2xaa-0 128431.95 0.47% -> 115051.86 0.42%: 1.12x speedup
▏
Slowdowns
=========
image-rgba firefox-periodic-table-0 56837.61 0.78% -> 66055.17 3.20%: 1.09x slowdown
▏
2009-07-23 15:32:13 +01:00
|
|
|
&pattern.base,
|
|
|
|
|
NULL);
|
2008-12-10 23:06:22 +00:00
|
|
|
_cairo_pattern_fini (&pattern.base);
|
|
|
|
|
|
|
|
|
|
if (unlikely (status)) {
|
2007-08-27 16:30:52 -07:00
|
|
|
cairo_surface_destroy (new_surface);
|
2008-12-10 23:06:22 +00:00
|
|
|
return status;
|
|
|
|
|
}
|
2007-08-27 16:30:52 -07:00
|
|
|
|
2008-12-10 23:06:22 +00:00
|
|
|
*clone_offset_x = src_x;
|
|
|
|
|
*clone_offset_y = src_y;
|
|
|
|
|
*clone_out = new_surface;
|
|
|
|
|
return CAIRO_STATUS_SUCCESS;
|
2007-08-27 16:30:52 -07:00
|
|
|
}
|