|
|
|
|
/*
|
|
|
|
|
* tight.c
|
|
|
|
|
*
|
|
|
|
|
* Routines to implement Tight Encoding
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
*
|
|
|
|
|
* Our Tight encoder is based roughly on the TurboVNC v0.6 encoder with some
|
|
|
|
|
* additional enhancements from TurboVNC 1.1. For lower compression levels,
|
|
|
|
|
* this encoder provides a tremendous reduction in CPU usage (and subsequently,
|
|
|
|
|
* an increase in throughput for CPU-limited environments) relative to the
|
|
|
|
|
* TightVNC encoder, whereas Compression Level 9 provides a low-bandwidth mode
|
|
|
|
|
* that behaves similarly to Compression Levels 5-9 in the old TightVNC
|
|
|
|
|
* encoder.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
/*
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
* Copyright (C) 2010-2012 D. R. Commander. All Rights Reserved.
|
|
|
|
|
* Copyright (C) 2005-2008 Sun Microsystems, Inc. All Rights Reserved.
|
|
|
|
|
* Copyright (C) 2004 Landmark Graphics Corporation. All Rights Reserved.
|
|
|
|
|
* Copyright (C) 2000, 2001 Const Kaplinsky. All Rights Reserved.
|
|
|
|
|
* Copyright (C) 1999 AT&T Laboratories Cambridge. All Rights Reserved.
|
|
|
|
|
*
|
|
|
|
|
* This is free software; you can redistribute it and/or modify
|
|
|
|
|
* it under the terms of the GNU General Public License as published by
|
|
|
|
|
* the Free Software Foundation; either version 2 of the License, or
|
|
|
|
|
* (at your option) any later version.
|
|
|
|
|
*
|
|
|
|
|
* This software is distributed in the hope that it will be useful,
|
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
|
* GNU General Public License for more details.
|
|
|
|
|
*
|
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
|
* along with this software; if not, write to the Free Software
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
|
|
|
|
|
* USA.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include <rfb/rfb.h>
|
|
|
|
|
#include "private.h"
|
|
|
|
|
|
|
|
|
|
#ifdef LIBVNCSERVER_HAVE_LIBPNG
|
|
|
|
|
#include <png.h>
|
|
|
|
|
#endif
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
#include "turbojpeg.h"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/* Note: The following constant should not be changed. */
|
|
|
|
|
#define TIGHT_MIN_TO_COMPRESS 12
|
|
|
|
|
|
|
|
|
|
/* The parameters below may be adjusted. */
|
|
|
|
|
#define MIN_SPLIT_RECT_SIZE 4096
|
|
|
|
|
#define MIN_SOLID_SUBRECT_SIZE 2048
|
|
|
|
|
#define MAX_SPLIT_TILE_SIZE 16
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* There is so much access of the Tight encoding static data buffers
|
|
|
|
|
* that we resort to using thread local storage instead of having
|
|
|
|
|
* per-client data.
|
|
|
|
|
*/
|
|
|
|
|
#if LIBVNCSERVER_HAVE_LIBPTHREAD && LIBVNCSERVER_HAVE_TLS && !defined(TLS) && defined(__linux__)
|
|
|
|
|
#define TLS __thread
|
|
|
|
|
#endif
|
|
|
|
|
#ifndef TLS
|
|
|
|
|
#define TLS
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
/* This variable is set on every rfbSendRectEncodingTight() call. */
|
|
|
|
|
static TLS rfbBool usePixelFormat24 = FALSE;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
/* Compression level stuff. The following array contains various
|
|
|
|
|
encoder parameters for each of 10 compression levels (0..9).
|
|
|
|
|
Last three parameters correspond to JPEG quality levels (0..9). */
|
|
|
|
|
|
|
|
|
|
typedef struct TIGHT_CONF_s {
|
|
|
|
|
int maxRectSize, maxRectWidth;
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
int monoMinRectSize;
|
|
|
|
|
int idxZlibLevel, monoZlibLevel, rawZlibLevel;
|
|
|
|
|
int idxMaxColorsDivisor;
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
int palMaxColorsWithJPEG;
|
|
|
|
|
} TIGHT_CONF;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static TIGHT_CONF tightConf[4] = {
|
|
|
|
|
{ 65536, 2048, 6, 0, 0, 0, 4, 24 }, /* 0 (used only without JPEG) */
|
|
|
|
|
{ 65536, 2048, 32, 1, 1, 1, 96, 24 }, /* 1 */
|
|
|
|
|
{ 65536, 2048, 32, 3, 3, 2, 96, 96 }, /* 2 (used only with JPEG) */
|
|
|
|
|
{ 65536, 2048, 32, 7, 7, 5, 96, 256 } /* 9 */
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
#ifdef LIBVNCSERVER_HAVE_LIBPNG
|
|
|
|
|
typedef struct TIGHT_PNG_CONF_s {
|
|
|
|
|
int png_zlib_level, png_filters;
|
|
|
|
|
} TIGHT_PNG_CONF;
|
|
|
|
|
|
|
|
|
|
static TIGHT_PNG_CONF tightPngConf[10] = {
|
|
|
|
|
{ 0, PNG_NO_FILTERS },
|
|
|
|
|
{ 1, PNG_NO_FILTERS },
|
|
|
|
|
{ 2, PNG_NO_FILTERS },
|
|
|
|
|
{ 3, PNG_NO_FILTERS },
|
|
|
|
|
{ 4, PNG_NO_FILTERS },
|
|
|
|
|
{ 5, PNG_ALL_FILTERS },
|
|
|
|
|
{ 6, PNG_ALL_FILTERS },
|
|
|
|
|
{ 7, PNG_ALL_FILTERS },
|
|
|
|
|
{ 8, PNG_ALL_FILTERS },
|
|
|
|
|
{ 9, PNG_ALL_FILTERS },
|
|
|
|
|
};
|
|
|
|
|
#endif
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static TLS int compressLevel = 1;
|
|
|
|
|
static TLS int qualityLevel = 95;
|
|
|
|
|
static TLS int subsampLevel = TJ_444;
|
|
|
|
|
|
|
|
|
|
static const int subsampLevel2tjsubsamp[4] = {
|
|
|
|
|
TJ_444, TJ_420, TJ_422, TJ_GRAYSCALE
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/* Stuff dealing with palettes. */
|
|
|
|
|
|
|
|
|
|
typedef struct COLOR_LIST_s {
|
|
|
|
|
struct COLOR_LIST_s *next;
|
|
|
|
|
int idx;
|
|
|
|
|
uint32_t rgb;
|
|
|
|
|
} COLOR_LIST;
|
|
|
|
|
|
|
|
|
|
typedef struct PALETTE_ENTRY_s {
|
|
|
|
|
COLOR_LIST *listNode;
|
|
|
|
|
int numPixels;
|
|
|
|
|
} PALETTE_ENTRY;
|
|
|
|
|
|
|
|
|
|
typedef struct PALETTE_s {
|
|
|
|
|
PALETTE_ENTRY entry[256];
|
|
|
|
|
COLOR_LIST *hash[256];
|
|
|
|
|
COLOR_LIST list[256];
|
|
|
|
|
} PALETTE;
|
|
|
|
|
|
|
|
|
|
/* TODO: move into rfbScreen struct */
|
|
|
|
|
static TLS int paletteNumColors = 0;
|
|
|
|
|
static TLS int paletteMaxColors = 0;
|
|
|
|
|
static TLS uint32_t monoBackground = 0;
|
|
|
|
|
static TLS uint32_t monoForeground = 0;
|
|
|
|
|
static TLS PALETTE palette;
|
|
|
|
|
|
|
|
|
|
/* Pointers to dynamically-allocated buffers. */
|
|
|
|
|
|
|
|
|
|
static TLS int tightBeforeBufSize = 0;
|
|
|
|
|
static TLS char *tightBeforeBuf = NULL;
|
|
|
|
|
|
|
|
|
|
static TLS int tightAfterBufSize = 0;
|
|
|
|
|
static TLS char *tightAfterBuf = NULL;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static TLS tjhandle j = NULL;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
void rfbTightCleanup (rfbScreenInfoPtr screen)
|
|
|
|
|
{
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (tightBeforeBufSize) {
|
|
|
|
|
free (tightBeforeBuf);
|
|
|
|
|
tightBeforeBufSize = 0;
|
|
|
|
|
tightBeforeBuf = NULL;
|
|
|
|
|
}
|
|
|
|
|
if (tightAfterBufSize) {
|
|
|
|
|
free (tightAfterBuf);
|
|
|
|
|
tightAfterBufSize = 0;
|
|
|
|
|
tightAfterBuf = NULL;
|
|
|
|
|
}
|
|
|
|
|
if (j) tjDestroy(j);
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
/* Prototypes for static functions. */
|
|
|
|
|
|
|
|
|
|
static rfbBool SendRectEncodingTight(rfbClientPtr cl, int x, int y,
|
|
|
|
|
int w, int h);
|
|
|
|
|
static void FindBestSolidArea (rfbClientPtr cl, int x, int y, int w, int h,
|
|
|
|
|
uint32_t colorValue, int *w_ptr, int *h_ptr);
|
|
|
|
|
static void ExtendSolidArea (rfbClientPtr cl, int x, int y, int w, int h,
|
|
|
|
|
uint32_t colorValue,
|
|
|
|
|
int *x_ptr, int *y_ptr, int *w_ptr, int *h_ptr);
|
|
|
|
|
static rfbBool CheckSolidTile (rfbClientPtr cl, int x, int y, int w, int h,
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
uint32_t *colorPtr, rfbBool needSameColor);
|
|
|
|
|
static rfbBool CheckSolidTile8 (rfbClientPtr cl, int x, int y, int w, int h,
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
uint32_t *colorPtr, rfbBool needSameColor);
|
|
|
|
|
static rfbBool CheckSolidTile16 (rfbClientPtr cl, int x, int y, int w, int h,
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
uint32_t *colorPtr, rfbBool needSameColor);
|
|
|
|
|
static rfbBool CheckSolidTile32 (rfbClientPtr cl, int x, int y, int w, int h,
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
uint32_t *colorPtr, rfbBool needSameColor);
|
|
|
|
|
|
|
|
|
|
static rfbBool SendRectSimple (rfbClientPtr cl, int x, int y, int w, int h);
|
|
|
|
|
static rfbBool SendSubrect (rfbClientPtr cl, int x, int y, int w, int h);
|
|
|
|
|
static rfbBool SendTightHeader (rfbClientPtr cl, int x, int y, int w, int h);
|
|
|
|
|
|
|
|
|
|
static rfbBool SendSolidRect (rfbClientPtr cl);
|
|
|
|
|
static rfbBool SendMonoRect (rfbClientPtr cl, int x, int y, int w, int h);
|
|
|
|
|
static rfbBool SendIndexedRect (rfbClientPtr cl, int x, int y, int w, int h);
|
|
|
|
|
static rfbBool SendFullColorRect (rfbClientPtr cl, int x, int y, int w, int h);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static rfbBool CompressData (rfbClientPtr cl, int streamId, int dataLen,
|
|
|
|
|
int zlibLevel, int zlibStrategy);
|
|
|
|
|
static rfbBool SendCompressedData (rfbClientPtr cl, char *buf,
|
|
|
|
|
int compressedLen);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static void FillPalette8 (int count);
|
|
|
|
|
static void FillPalette16 (int count);
|
|
|
|
|
static void FillPalette32 (int count);
|
|
|
|
|
static void FastFillPalette16 (rfbClientPtr cl, uint16_t *data, int w,
|
|
|
|
|
int pitch, int h);
|
|
|
|
|
static void FastFillPalette32 (rfbClientPtr cl, uint32_t *data, int w,
|
|
|
|
|
int pitch, int h);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static void PaletteReset (void);
|
|
|
|
|
static int PaletteInsert (uint32_t rgb, int numPixels, int bpp);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static void Pack24 (rfbClientPtr cl, char *buf, rfbPixelFormat *fmt,
|
|
|
|
|
int count);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static void EncodeIndexedRect16 (uint8_t *buf, int count);
|
|
|
|
|
static void EncodeIndexedRect32 (uint8_t *buf, int count);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static void EncodeMonoRect8 (uint8_t *buf, int w, int h);
|
|
|
|
|
static void EncodeMonoRect16 (uint8_t *buf, int w, int h);
|
|
|
|
|
static void EncodeMonoRect32 (uint8_t *buf, int w, int h);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static rfbBool SendJpegRect (rfbClientPtr cl, int x, int y, int w, int h,
|
|
|
|
|
int quality);
|
|
|
|
|
static void PrepareRowForImg(rfbClientPtr cl, uint8_t *dst, int x, int y, int count);
|
|
|
|
|
static void PrepareRowForImg24(rfbClientPtr cl, uint8_t *dst, int x, int y, int count);
|
|
|
|
|
static void PrepareRowForImg16(rfbClientPtr cl, uint8_t *dst, int x, int y, int count);
|
|
|
|
|
static void PrepareRowForImg32(rfbClientPtr cl, uint8_t *dst, int x, int y, int count);
|
|
|
|
|
|
|
|
|
|
#ifdef LIBVNCSERVER_HAVE_LIBPNG
|
|
|
|
|
static rfbBool SendPngRect(rfbClientPtr cl, int x, int y, int w, int h);
|
|
|
|
|
static rfbBool CanSendPngRect(rfbClientPtr cl, int w, int h);
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Tight encoding implementation.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
|
rfbNumCodedRectsTight(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
int maxRectSize, maxRectWidth;
|
|
|
|
|
int subrectMaxWidth, subrectMaxHeight;
|
|
|
|
|
|
|
|
|
|
/* No matter how many rectangles we will send if LastRect markers
|
|
|
|
|
are used to terminate rectangle stream. */
|
|
|
|
|
if (cl->enableLastRectEncoding && w * h >= MIN_SPLIT_RECT_SIZE)
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
return 0;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
maxRectSize = tightConf[compressLevel].maxRectSize;
|
|
|
|
|
maxRectWidth = tightConf[compressLevel].maxRectWidth;
|
|
|
|
|
|
|
|
|
|
if (w > maxRectWidth || w * h > maxRectSize) {
|
|
|
|
|
subrectMaxWidth = (w > maxRectWidth) ? maxRectWidth : w;
|
|
|
|
|
subrectMaxHeight = maxRectSize / subrectMaxWidth;
|
|
|
|
|
return (((w - 1) / maxRectWidth + 1) *
|
|
|
|
|
((h - 1) / subrectMaxHeight + 1));
|
|
|
|
|
} else {
|
|
|
|
|
return 1;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
rfbBool
|
|
|
|
|
rfbSendRectEncodingTight(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
cl->tightEncoding = rfbEncodingTight;
|
|
|
|
|
return SendRectEncodingTight(cl, x, y, w, h);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
rfbBool
|
|
|
|
|
rfbSendRectEncodingTightPng(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
cl->tightEncoding = rfbEncodingTightPng;
|
|
|
|
|
return SendRectEncodingTight(cl, x, y, w, h);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
rfbBool
|
|
|
|
|
SendRectEncodingTight(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
int nMaxRows;
|
|
|
|
|
uint32_t colorValue;
|
|
|
|
|
int dx, dy, dw, dh;
|
|
|
|
|
int x_best, y_best, w_best, h_best;
|
|
|
|
|
char *fbptr;
|
|
|
|
|
|
|
|
|
|
rfbSendUpdateBuf(cl);
|
|
|
|
|
|
|
|
|
|
compressLevel = cl->tightCompressLevel;
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
qualityLevel = cl->turboQualityLevel;
|
|
|
|
|
subsampLevel = cl->turboSubsampLevel;
|
|
|
|
|
|
|
|
|
|
/* We only allow compression levels that have a demonstrable performance
|
|
|
|
|
benefit. CL 0 with JPEG reduces CPU usage for workloads that have low
|
|
|
|
|
numbers of unique colors, but the same thing can be accomplished by
|
|
|
|
|
using CL 0 without JPEG (AKA "Lossless Tight.") For those same
|
|
|
|
|
low-color workloads, CL 2 can provide typically 20-40% better
|
|
|
|
|
compression than CL 1 (with a commensurate increase in CPU usage.) For
|
|
|
|
|
high-color workloads, CL 1 should always be used, as higher compression
|
|
|
|
|
levels increase CPU usage for these workloads without providing any
|
|
|
|
|
significant reduction in bandwidth. */
|
|
|
|
|
if (qualityLevel != -1) {
|
|
|
|
|
if (compressLevel < 1) compressLevel = 1;
|
|
|
|
|
if (compressLevel > 2) compressLevel = 2;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* With JPEG disabled, CL 2 offers no significant bandwidth savings over
|
|
|
|
|
CL 1, so we don't include it. */
|
|
|
|
|
else if (compressLevel > 1) compressLevel = 1;
|
|
|
|
|
|
|
|
|
|
/* CL 9 (which maps internally to CL 3) is included mainly for backward
|
|
|
|
|
compatibility with TightVNC Compression Levels 5-9. It should be used
|
|
|
|
|
only in extremely low-bandwidth cases in which it can be shown to have a
|
|
|
|
|
benefit. For low-color workloads, it provides typically only 10-20%
|
|
|
|
|
better compression than CL 2 with JPEG and CL 1 without JPEG, and it
|
|
|
|
|
uses, on average, twice as much CPU time. */
|
|
|
|
|
if (cl->tightCompressLevel == 9) compressLevel = 3;
|
|
|
|
|
|
|
|
|
|
if ( cl->format.depth == 24 && cl->format.redMax == 0xFF &&
|
|
|
|
|
cl->format.greenMax == 0xFF && cl->format.blueMax == 0xFF ) {
|
|
|
|
|
usePixelFormat24 = TRUE;
|
|
|
|
|
} else {
|
|
|
|
|
usePixelFormat24 = FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (!cl->enableLastRectEncoding || w * h < MIN_SPLIT_RECT_SIZE)
|
|
|
|
|
return SendRectSimple(cl, x, y, w, h);
|
|
|
|
|
|
|
|
|
|
/* Make sure we can write at least one pixel into tightBeforeBuf. */
|
|
|
|
|
|
|
|
|
|
if (tightBeforeBufSize < 4) {
|
|
|
|
|
tightBeforeBufSize = 4;
|
|
|
|
|
if (tightBeforeBuf == NULL)
|
|
|
|
|
tightBeforeBuf = (char *)malloc(tightBeforeBufSize);
|
|
|
|
|
else
|
|
|
|
|
tightBeforeBuf = (char *)realloc(tightBeforeBuf,
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
tightBeforeBufSize);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Calculate maximum number of rows in one non-solid rectangle. */
|
|
|
|
|
|
|
|
|
|
{
|
|
|
|
|
int maxRectSize, maxRectWidth, nMaxWidth;
|
|
|
|
|
|
|
|
|
|
maxRectSize = tightConf[compressLevel].maxRectSize;
|
|
|
|
|
maxRectWidth = tightConf[compressLevel].maxRectWidth;
|
|
|
|
|
nMaxWidth = (w > maxRectWidth) ? maxRectWidth : w;
|
|
|
|
|
nMaxRows = maxRectSize / nMaxWidth;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Try to find large solid-color areas and send them separately. */
|
|
|
|
|
|
|
|
|
|
for (dy = y; dy < y + h; dy += MAX_SPLIT_TILE_SIZE) {
|
|
|
|
|
|
|
|
|
|
/* If a rectangle becomes too large, send its upper part now. */
|
|
|
|
|
|
|
|
|
|
if (dy - y >= nMaxRows) {
|
|
|
|
|
if (!SendRectSimple(cl, x, y, w, nMaxRows))
|
|
|
|
|
return 0;
|
|
|
|
|
y += nMaxRows;
|
|
|
|
|
h -= nMaxRows;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
dh = (dy + MAX_SPLIT_TILE_SIZE <= y + h) ?
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
MAX_SPLIT_TILE_SIZE : (y + h - dy);
|
|
|
|
|
|
|
|
|
|
for (dx = x; dx < x + w; dx += MAX_SPLIT_TILE_SIZE) {
|
|
|
|
|
|
|
|
|
|
dw = (dx + MAX_SPLIT_TILE_SIZE <= x + w) ?
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
MAX_SPLIT_TILE_SIZE : (x + w - dx);
|
|
|
|
|
|
|
|
|
|
if (CheckSolidTile(cl, dx, dy, dw, dh, &colorValue, FALSE)) {
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (subsampLevel == TJ_GRAYSCALE && qualityLevel != -1) {
|
|
|
|
|
uint32_t r = (colorValue >> 16) & 0xFF;
|
|
|
|
|
uint32_t g = (colorValue >> 8) & 0xFF;
|
|
|
|
|
uint32_t b = (colorValue) & 0xFF;
|
|
|
|
|
double y = (0.257 * (double)r) + (0.504 * (double)g)
|
|
|
|
|
+ (0.098 * (double)b) + 16.;
|
|
|
|
|
colorValue = (int)y + (((int)y) << 8) + (((int)y) << 16);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Get dimensions of solid-color area. */
|
|
|
|
|
|
|
|
|
|
FindBestSolidArea(cl, dx, dy, w - (dx - x), h - (dy - y),
|
|
|
|
|
colorValue, &w_best, &h_best);
|
|
|
|
|
|
|
|
|
|
/* Make sure a solid rectangle is large enough
|
|
|
|
|
(or the whole rectangle is of the same color). */
|
|
|
|
|
|
|
|
|
|
if ( w_best * h_best != w * h &&
|
|
|
|
|
w_best * h_best < MIN_SOLID_SUBRECT_SIZE )
|
|
|
|
|
continue;
|
|
|
|
|
|
|
|
|
|
/* Try to extend solid rectangle to maximum size. */
|
|
|
|
|
|
|
|
|
|
x_best = dx; y_best = dy;
|
|
|
|
|
ExtendSolidArea(cl, x, y, w, h, colorValue,
|
|
|
|
|
&x_best, &y_best, &w_best, &h_best);
|
|
|
|
|
|
|
|
|
|
/* Send rectangles at top and left to solid-color area. */
|
|
|
|
|
|
|
|
|
|
if ( y_best != y &&
|
|
|
|
|
!SendRectSimple(cl, x, y, w, y_best-y) )
|
|
|
|
|
return FALSE;
|
|
|
|
|
if ( x_best != x &&
|
|
|
|
|
!SendRectEncodingTight(cl, x, y_best,
|
|
|
|
|
x_best-x, h_best) )
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
|
|
/* Send solid-color rectangle. */
|
|
|
|
|
|
|
|
|
|
if (!SendTightHeader(cl, x_best, y_best, w_best, h_best))
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
|
|
fbptr = (cl->scaledScreen->frameBuffer +
|
|
|
|
|
(cl->scaledScreen->paddedWidthInBytes * y_best) +
|
|
|
|
|
(x_best * (cl->scaledScreen->bitsPerPixel / 8)));
|
|
|
|
|
|
|
|
|
|
(*cl->translateFn)(cl->translateLookupTable, &cl->screen->serverFormat,
|
|
|
|
|
&cl->format, fbptr, tightBeforeBuf,
|
|
|
|
|
cl->scaledScreen->paddedWidthInBytes, 1, 1);
|
|
|
|
|
|
|
|
|
|
if (!SendSolidRect(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
|
|
/* Send remaining rectangles (at right and bottom). */
|
|
|
|
|
|
|
|
|
|
if ( x_best + w_best != x + w &&
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
!SendRectEncodingTight(cl, x_best + w_best, y_best,
|
|
|
|
|
w - (x_best-x) - w_best, h_best) )
|
|
|
|
|
return FALSE;
|
|
|
|
|
if ( y_best + h_best != y + h &&
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
!SendRectEncodingTight(cl, x, y_best + h_best,
|
|
|
|
|
w, h - (y_best-y) - h_best) )
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
|
|
/* Return after all recursive calls are done. */
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* No suitable solid-color rectangles found. */
|
|
|
|
|
|
|
|
|
|
return SendRectSimple(cl, x, y, w, h);
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
FindBestSolidArea(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h,
|
|
|
|
|
uint32_t colorValue,
|
|
|
|
|
int *w_ptr,
|
|
|
|
|
int *h_ptr)
|
|
|
|
|
{
|
|
|
|
|
int dx, dy, dw, dh;
|
|
|
|
|
int w_prev;
|
|
|
|
|
int w_best = 0, h_best = 0;
|
|
|
|
|
|
|
|
|
|
w_prev = w;
|
|
|
|
|
|
|
|
|
|
for (dy = y; dy < y + h; dy += MAX_SPLIT_TILE_SIZE) {
|
|
|
|
|
|
|
|
|
|
dh = (dy + MAX_SPLIT_TILE_SIZE <= y + h) ?
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
MAX_SPLIT_TILE_SIZE : (y + h - dy);
|
|
|
|
|
dw = (w_prev > MAX_SPLIT_TILE_SIZE) ?
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
MAX_SPLIT_TILE_SIZE : w_prev;
|
|
|
|
|
|
|
|
|
|
if (!CheckSolidTile(cl, x, dy, dw, dh, &colorValue, TRUE))
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
for (dx = x + dw; dx < x + w_prev;) {
|
|
|
|
|
dw = (dx + MAX_SPLIT_TILE_SIZE <= x + w_prev) ?
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
MAX_SPLIT_TILE_SIZE : (x + w_prev - dx);
|
|
|
|
|
if (!CheckSolidTile(cl, dx, dy, dw, dh, &colorValue, TRUE))
|
|
|
|
|
break;
|
|
|
|
|
dx += dw;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
w_prev = dx - x;
|
|
|
|
|
if (w_prev * (dy + dh - y) > w_best * h_best) {
|
|
|
|
|
w_best = w_prev;
|
|
|
|
|
h_best = dy + dh - y;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
*w_ptr = w_best;
|
|
|
|
|
*h_ptr = h_best;
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
ExtendSolidArea(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h,
|
|
|
|
|
uint32_t colorValue,
|
|
|
|
|
int *x_ptr,
|
|
|
|
|
int *y_ptr,
|
|
|
|
|
int *w_ptr,
|
|
|
|
|
int *h_ptr)
|
|
|
|
|
{
|
|
|
|
|
int cx, cy;
|
|
|
|
|
|
|
|
|
|
/* Try to extend the area upwards. */
|
|
|
|
|
for ( cy = *y_ptr - 1;
|
|
|
|
|
cy >= y && CheckSolidTile(cl, *x_ptr, cy, *w_ptr, 1, &colorValue, TRUE);
|
|
|
|
|
cy-- );
|
|
|
|
|
*h_ptr += *y_ptr - (cy + 1);
|
|
|
|
|
*y_ptr = cy + 1;
|
|
|
|
|
|
|
|
|
|
/* ... downwards. */
|
|
|
|
|
for ( cy = *y_ptr + *h_ptr;
|
|
|
|
|
cy < y + h &&
|
|
|
|
|
CheckSolidTile(cl, *x_ptr, cy, *w_ptr, 1, &colorValue, TRUE);
|
|
|
|
|
cy++ );
|
|
|
|
|
*h_ptr += cy - (*y_ptr + *h_ptr);
|
|
|
|
|
|
|
|
|
|
/* ... to the left. */
|
|
|
|
|
for ( cx = *x_ptr - 1;
|
|
|
|
|
cx >= x && CheckSolidTile(cl, cx, *y_ptr, 1, *h_ptr, &colorValue, TRUE);
|
|
|
|
|
cx-- );
|
|
|
|
|
*w_ptr += *x_ptr - (cx + 1);
|
|
|
|
|
*x_ptr = cx + 1;
|
|
|
|
|
|
|
|
|
|
/* ... to the right. */
|
|
|
|
|
for ( cx = *x_ptr + *w_ptr;
|
|
|
|
|
cx < x + w &&
|
|
|
|
|
CheckSolidTile(cl, cx, *y_ptr, 1, *h_ptr, &colorValue, TRUE);
|
|
|
|
|
cx++ );
|
|
|
|
|
*w_ptr += cx - (*x_ptr + *w_ptr);
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Check if a rectangle is all of the same color. If needSameColor is
|
|
|
|
|
* set to non-zero, then also check that its color equals to the
|
|
|
|
|
* *colorPtr value. The result is 1 if the test is successfull, and in
|
|
|
|
|
* that case new color will be stored in *colorPtr.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
static rfbBool CheckSolidTile(rfbClientPtr cl, int x, int y, int w, int h, uint32_t* colorPtr, rfbBool needSameColor)
|
|
|
|
|
{
|
|
|
|
|
switch(cl->screen->serverFormat.bitsPerPixel) {
|
|
|
|
|
case 32:
|
|
|
|
|
return CheckSolidTile32(cl, x, y, w, h, colorPtr, needSameColor);
|
|
|
|
|
case 16:
|
|
|
|
|
return CheckSolidTile16(cl, x, y, w, h, colorPtr, needSameColor);
|
|
|
|
|
default:
|
|
|
|
|
return CheckSolidTile8(cl, x, y, w, h, colorPtr, needSameColor);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
#define DEFINE_CHECK_SOLID_FUNCTION(bpp) \
|
|
|
|
|
\
|
|
|
|
|
static rfbBool \
|
|
|
|
|
CheckSolidTile##bpp(rfbClientPtr cl, int x, int y, int w, int h, \
|
|
|
|
|
uint32_t* colorPtr, rfbBool needSameColor) \
|
|
|
|
|
{ \
|
|
|
|
|
uint##bpp##_t *fbptr; \
|
|
|
|
|
uint##bpp##_t colorValue; \
|
|
|
|
|
int dx, dy; \
|
|
|
|
|
\
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
fbptr = (uint##bpp##_t *)&cl->scaledScreen->frameBuffer \
|
|
|
|
|
[y * cl->scaledScreen->paddedWidthInBytes + x * (bpp/8)]; \
|
|
|
|
|
\
|
|
|
|
|
colorValue = *fbptr; \
|
|
|
|
|
if (needSameColor && (uint32_t)colorValue != *colorPtr) \
|
|
|
|
|
return FALSE; \
|
|
|
|
|
\
|
|
|
|
|
for (dy = 0; dy < h; dy++) { \
|
|
|
|
|
for (dx = 0; dx < w; dx++) { \
|
|
|
|
|
if (colorValue != fbptr[dx]) \
|
|
|
|
|
return FALSE; \
|
|
|
|
|
} \
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
fbptr = (uint##bpp##_t *)((uint8_t *)fbptr \
|
|
|
|
|
+ cl->scaledScreen->paddedWidthInBytes); \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
*colorPtr = (uint32_t)colorValue; \
|
|
|
|
|
return TRUE; \
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
DEFINE_CHECK_SOLID_FUNCTION(8)
|
|
|
|
|
DEFINE_CHECK_SOLID_FUNCTION(16)
|
|
|
|
|
DEFINE_CHECK_SOLID_FUNCTION(32)
|
|
|
|
|
|
|
|
|
|
static rfbBool
|
|
|
|
|
SendRectSimple(rfbClientPtr cl, int x, int y, int w, int h)
|
|
|
|
|
{
|
|
|
|
|
int maxBeforeSize, maxAfterSize;
|
|
|
|
|
int maxRectSize, maxRectWidth;
|
|
|
|
|
int subrectMaxWidth, subrectMaxHeight;
|
|
|
|
|
int dx, dy;
|
|
|
|
|
int rw, rh;
|
|
|
|
|
|
|
|
|
|
maxRectSize = tightConf[compressLevel].maxRectSize;
|
|
|
|
|
maxRectWidth = tightConf[compressLevel].maxRectWidth;
|
|
|
|
|
|
|
|
|
|
maxBeforeSize = maxRectSize * (cl->format.bitsPerPixel / 8);
|
|
|
|
|
maxAfterSize = maxBeforeSize + (maxBeforeSize + 99) / 100 + 12;
|
|
|
|
|
|
|
|
|
|
if (tightBeforeBufSize < maxBeforeSize) {
|
|
|
|
|
tightBeforeBufSize = maxBeforeSize;
|
|
|
|
|
if (tightBeforeBuf == NULL)
|
|
|
|
|
tightBeforeBuf = (char *)malloc(tightBeforeBufSize);
|
|
|
|
|
else
|
|
|
|
|
tightBeforeBuf = (char *)realloc(tightBeforeBuf,
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
tightBeforeBufSize);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (tightAfterBufSize < maxAfterSize) {
|
|
|
|
|
tightAfterBufSize = maxAfterSize;
|
|
|
|
|
if (tightAfterBuf == NULL)
|
|
|
|
|
tightAfterBuf = (char *)malloc(tightAfterBufSize);
|
|
|
|
|
else
|
|
|
|
|
tightAfterBuf = (char *)realloc(tightAfterBuf,
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
tightAfterBufSize);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (w > maxRectWidth || w * h > maxRectSize) {
|
|
|
|
|
subrectMaxWidth = (w > maxRectWidth) ? maxRectWidth : w;
|
|
|
|
|
subrectMaxHeight = maxRectSize / subrectMaxWidth;
|
|
|
|
|
|
|
|
|
|
for (dy = 0; dy < h; dy += subrectMaxHeight) {
|
|
|
|
|
for (dx = 0; dx < w; dx += maxRectWidth) {
|
|
|
|
|
rw = (dx + maxRectWidth < w) ? maxRectWidth : w - dx;
|
|
|
|
|
rh = (dy + subrectMaxHeight < h) ? subrectMaxHeight : h - dy;
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (!SendSubrect(cl, x + dx, y + dy, rw, rh))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
if (!SendSubrect(cl, x, y, w, h))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static rfbBool
|
|
|
|
|
SendSubrect(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
char *fbptr;
|
|
|
|
|
rfbBool success = FALSE;
|
|
|
|
|
|
|
|
|
|
/* Send pending data if there is more than 128 bytes. */
|
|
|
|
|
if (cl->ublen > 128) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (!SendTightHeader(cl, x, y, w, h))
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
fbptr = (cl->scaledScreen->frameBuffer
|
|
|
|
|
+ (cl->scaledScreen->paddedWidthInBytes * y)
|
|
|
|
|
+ (x * (cl->scaledScreen->bitsPerPixel / 8)));
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (subsampLevel == TJ_GRAYSCALE && qualityLevel != -1)
|
|
|
|
|
return SendJpegRect(cl, x, y, w, h, qualityLevel);
|
|
|
|
|
|
|
|
|
|
paletteMaxColors = w * h / tightConf[compressLevel].idxMaxColorsDivisor;
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if(qualityLevel != -1)
|
|
|
|
|
paletteMaxColors = tightConf[compressLevel].palMaxColorsWithJPEG;
|
|
|
|
|
if ( paletteMaxColors < 2 &&
|
|
|
|
|
w * h >= tightConf[compressLevel].monoMinRectSize ) {
|
|
|
|
|
paletteMaxColors = 2;
|
|
|
|
|
}
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
if (cl->format.bitsPerPixel == cl->screen->serverFormat.bitsPerPixel &&
|
|
|
|
|
cl->format.redMax == cl->screen->serverFormat.redMax &&
|
|
|
|
|
cl->format.greenMax == cl->screen->serverFormat.greenMax &&
|
|
|
|
|
cl->format.blueMax == cl->screen->serverFormat.blueMax &&
|
|
|
|
|
cl->format.bitsPerPixel >= 16) {
|
|
|
|
|
|
|
|
|
|
/* This is so we can avoid translating the pixels when compressing
|
|
|
|
|
with JPEG, since it is unnecessary */
|
|
|
|
|
switch (cl->format.bitsPerPixel) {
|
|
|
|
|
case 16:
|
|
|
|
|
FastFillPalette16(cl, (uint16_t *)fbptr, w,
|
|
|
|
|
cl->scaledScreen->paddedWidthInBytes / 2, h);
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
FastFillPalette32(cl, (uint32_t *)fbptr, w,
|
|
|
|
|
cl->scaledScreen->paddedWidthInBytes / 4, h);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if(paletteNumColors != 0 || qualityLevel == -1) {
|
|
|
|
|
(*cl->translateFn)(cl->translateLookupTable,
|
|
|
|
|
&cl->screen->serverFormat, &cl->format, fbptr,
|
|
|
|
|
tightBeforeBuf,
|
|
|
|
|
cl->scaledScreen->paddedWidthInBytes, w, h);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
else {
|
|
|
|
|
(*cl->translateFn)(cl->translateLookupTable, &cl->screen->serverFormat,
|
|
|
|
|
&cl->format, fbptr, tightBeforeBuf,
|
|
|
|
|
cl->scaledScreen->paddedWidthInBytes, w, h);
|
|
|
|
|
|
|
|
|
|
switch (cl->format.bitsPerPixel) {
|
|
|
|
|
case 8:
|
|
|
|
|
FillPalette8(w * h);
|
|
|
|
|
break;
|
|
|
|
|
case 16:
|
|
|
|
|
FillPalette16(w * h);
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
FillPalette32(w * h);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
switch (paletteNumColors) {
|
|
|
|
|
case 0:
|
|
|
|
|
/* Truecolor image */
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (qualityLevel != -1) {
|
|
|
|
|
success = SendJpegRect(cl, x, y, w, h, qualityLevel);
|
|
|
|
|
} else {
|
|
|
|
|
success = SendFullColorRect(cl, x, y, w, h);
|
|
|
|
|
}
|
|
|
|
|
break;
|
|
|
|
|
case 1:
|
|
|
|
|
/* Solid rectangle */
|
|
|
|
|
success = SendSolidRect(cl);
|
|
|
|
|
break;
|
|
|
|
|
case 2:
|
|
|
|
|
/* Two-color rectangle */
|
|
|
|
|
success = SendMonoRect(cl, x, y, w, h);
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
/* Up to 256 different colors */
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
success = SendIndexedRect(cl, x, y, w, h);
|
|
|
|
|
}
|
|
|
|
|
return success;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static rfbBool
|
|
|
|
|
SendTightHeader(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
rfbFramebufferUpdateRectHeader rect;
|
|
|
|
|
|
|
|
|
|
if (cl->ublen + sz_rfbFramebufferUpdateRectHeader > UPDATE_BUF_SIZE) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
rect.r.x = Swap16IfLE(x);
|
|
|
|
|
rect.r.y = Swap16IfLE(y);
|
|
|
|
|
rect.r.w = Swap16IfLE(w);
|
|
|
|
|
rect.r.h = Swap16IfLE(h);
|
|
|
|
|
rect.encoding = Swap32IfLE(cl->tightEncoding);
|
|
|
|
|
|
|
|
|
|
memcpy(&cl->updateBuf[cl->ublen], (char *)&rect,
|
|
|
|
|
sz_rfbFramebufferUpdateRectHeader);
|
|
|
|
|
cl->ublen += sz_rfbFramebufferUpdateRectHeader;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
rfbStatRecordEncodingSent(cl, cl->tightEncoding,
|
|
|
|
|
sz_rfbFramebufferUpdateRectHeader,
|
|
|
|
|
sz_rfbFramebufferUpdateRectHeader
|
|
|
|
|
+ w * (cl->format.bitsPerPixel / 8) * h);
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Subencoding implementations.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
static rfbBool
|
|
|
|
|
SendSolidRect(rfbClientPtr cl)
|
|
|
|
|
{
|
|
|
|
|
int len;
|
|
|
|
|
|
|
|
|
|
if (usePixelFormat24) {
|
|
|
|
|
Pack24(cl, tightBeforeBuf, &cl->format, 1);
|
|
|
|
|
len = 3;
|
|
|
|
|
} else
|
|
|
|
|
len = cl->format.bitsPerPixel / 8;
|
|
|
|
|
|
|
|
|
|
if (cl->ublen + 1 + len > UPDATE_BUF_SIZE) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (char)(rfbTightFill << 4);
|
|
|
|
|
memcpy (&cl->updateBuf[cl->ublen], tightBeforeBuf, len);
|
|
|
|
|
cl->ublen += len;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, len + 1);
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static rfbBool
|
|
|
|
|
SendMonoRect(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
int streamId = 1;
|
|
|
|
|
int paletteLen, dataLen;
|
|
|
|
|
|
|
|
|
|
#ifdef LIBVNCSERVER_HAVE_LIBPNG
|
|
|
|
|
if (CanSendPngRect(cl, w, h)) {
|
|
|
|
|
/* TODO: setup palette maybe */
|
|
|
|
|
return SendPngRect(cl, x, y, w, h);
|
|
|
|
|
/* TODO: destroy palette maybe */
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
if ( cl->ublen + TIGHT_MIN_TO_COMPRESS + 6 +
|
|
|
|
|
2 * cl->format.bitsPerPixel / 8 > UPDATE_BUF_SIZE ) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Prepare tight encoding header. */
|
|
|
|
|
dataLen = (w + 7) / 8;
|
|
|
|
|
dataLen *= h;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (tightConf[compressLevel].monoZlibLevel == 0 &&
|
|
|
|
|
cl->tightEncoding != rfbEncodingTightPng)
|
|
|
|
|
cl->updateBuf[cl->ublen++] =
|
|
|
|
|
(char)((rfbTightNoZlib | rfbTightExplicitFilter) << 4);
|
|
|
|
|
else
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (streamId | rfbTightExplicitFilter) << 4;
|
|
|
|
|
cl->updateBuf[cl->ublen++] = rfbTightFilterPalette;
|
|
|
|
|
cl->updateBuf[cl->ublen++] = 1;
|
|
|
|
|
|
|
|
|
|
/* Prepare palette, convert image. */
|
|
|
|
|
switch (cl->format.bitsPerPixel) {
|
|
|
|
|
|
|
|
|
|
case 32:
|
|
|
|
|
EncodeMonoRect32((uint8_t *)tightBeforeBuf, w, h);
|
|
|
|
|
|
|
|
|
|
((uint32_t *)tightAfterBuf)[0] = monoBackground;
|
|
|
|
|
((uint32_t *)tightAfterBuf)[1] = monoForeground;
|
|
|
|
|
if (usePixelFormat24) {
|
|
|
|
|
Pack24(cl, tightAfterBuf, &cl->format, 2);
|
|
|
|
|
paletteLen = 6;
|
|
|
|
|
} else
|
|
|
|
|
paletteLen = 8;
|
|
|
|
|
|
|
|
|
|
memcpy(&cl->updateBuf[cl->ublen], tightAfterBuf, paletteLen);
|
|
|
|
|
cl->ublen += paletteLen;
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 3 + paletteLen);
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case 16:
|
|
|
|
|
EncodeMonoRect16((uint8_t *)tightBeforeBuf, w, h);
|
|
|
|
|
|
|
|
|
|
((uint16_t *)tightAfterBuf)[0] = (uint16_t)monoBackground;
|
|
|
|
|
((uint16_t *)tightAfterBuf)[1] = (uint16_t)monoForeground;
|
|
|
|
|
|
|
|
|
|
memcpy(&cl->updateBuf[cl->ublen], tightAfterBuf, 4);
|
|
|
|
|
cl->ublen += 4;
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 7);
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
default:
|
|
|
|
|
EncodeMonoRect8((uint8_t *)tightBeforeBuf, w, h);
|
|
|
|
|
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (char)monoBackground;
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (char)monoForeground;
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 5);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return CompressData(cl, streamId, dataLen,
|
|
|
|
|
tightConf[compressLevel].monoZlibLevel,
|
|
|
|
|
Z_DEFAULT_STRATEGY);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static rfbBool
|
|
|
|
|
SendIndexedRect(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
int streamId = 2;
|
|
|
|
|
int i, entryLen;
|
|
|
|
|
|
|
|
|
|
#ifdef LIBVNCSERVER_HAVE_LIBPNG
|
|
|
|
|
if (CanSendPngRect(cl, w, h)) {
|
|
|
|
|
return SendPngRect(cl, x, y, w, h);
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
if ( cl->ublen + TIGHT_MIN_TO_COMPRESS + 6 +
|
|
|
|
|
paletteNumColors * cl->format.bitsPerPixel / 8 >
|
|
|
|
|
UPDATE_BUF_SIZE ) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Prepare tight encoding header. */
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (tightConf[compressLevel].idxZlibLevel == 0 &&
|
|
|
|
|
cl->tightEncoding != rfbEncodingTightPng)
|
|
|
|
|
cl->updateBuf[cl->ublen++] =
|
|
|
|
|
(char)((rfbTightNoZlib | rfbTightExplicitFilter) << 4);
|
|
|
|
|
else
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (streamId | rfbTightExplicitFilter) << 4;
|
|
|
|
|
cl->updateBuf[cl->ublen++] = rfbTightFilterPalette;
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (char)(paletteNumColors - 1);
|
|
|
|
|
|
|
|
|
|
/* Prepare palette, convert image. */
|
|
|
|
|
switch (cl->format.bitsPerPixel) {
|
|
|
|
|
|
|
|
|
|
case 32:
|
|
|
|
|
EncodeIndexedRect32((uint8_t *)tightBeforeBuf, w * h);
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < paletteNumColors; i++) {
|
|
|
|
|
((uint32_t *)tightAfterBuf)[i] =
|
|
|
|
|
palette.entry[i].listNode->rgb;
|
|
|
|
|
}
|
|
|
|
|
if (usePixelFormat24) {
|
|
|
|
|
Pack24(cl, tightAfterBuf, &cl->format, paletteNumColors);
|
|
|
|
|
entryLen = 3;
|
|
|
|
|
} else
|
|
|
|
|
entryLen = 4;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
memcpy(&cl->updateBuf[cl->ublen], tightAfterBuf,
|
|
|
|
|
paletteNumColors * entryLen);
|
|
|
|
|
cl->ublen += paletteNumColors * entryLen;
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding,
|
|
|
|
|
3 + paletteNumColors * entryLen);
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
case 16:
|
|
|
|
|
EncodeIndexedRect16((uint8_t *)tightBeforeBuf, w * h);
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < paletteNumColors; i++) {
|
|
|
|
|
((uint16_t *)tightAfterBuf)[i] =
|
|
|
|
|
(uint16_t)palette.entry[i].listNode->rgb;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
memcpy(&cl->updateBuf[cl->ublen], tightAfterBuf, paletteNumColors * 2);
|
|
|
|
|
cl->ublen += paletteNumColors * 2;
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding,
|
|
|
|
|
3 + paletteNumColors * 2);
|
|
|
|
|
break;
|
|
|
|
|
|
|
|
|
|
default:
|
|
|
|
|
return FALSE; /* Should never happen. */
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return CompressData(cl, streamId, w * h,
|
|
|
|
|
tightConf[compressLevel].idxZlibLevel,
|
|
|
|
|
Z_DEFAULT_STRATEGY);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static rfbBool
|
|
|
|
|
SendFullColorRect(rfbClientPtr cl,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int w,
|
|
|
|
|
int h)
|
|
|
|
|
{
|
|
|
|
|
int streamId = 0;
|
|
|
|
|
int len;
|
|
|
|
|
|
|
|
|
|
#ifdef LIBVNCSERVER_HAVE_LIBPNG
|
|
|
|
|
if (CanSendPngRect(cl, w, h)) {
|
|
|
|
|
return SendPngRect(cl, x, y, w, h);
|
|
|
|
|
}
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
if (cl->ublen + TIGHT_MIN_TO_COMPRESS + 1 > UPDATE_BUF_SIZE) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (tightConf[compressLevel].rawZlibLevel == 0 &&
|
|
|
|
|
cl->tightEncoding != rfbEncodingTightPng)
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (char)(rfbTightNoZlib << 4);
|
|
|
|
|
else
|
|
|
|
|
cl->updateBuf[cl->ublen++] = 0x00; /* stream id = 0, no flushing, no filter */
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 1);
|
|
|
|
|
|
|
|
|
|
if (usePixelFormat24) {
|
|
|
|
|
Pack24(cl, tightBeforeBuf, &cl->format, w * h);
|
|
|
|
|
len = 3;
|
|
|
|
|
} else
|
|
|
|
|
len = cl->format.bitsPerPixel / 8;
|
|
|
|
|
|
|
|
|
|
return CompressData(cl, streamId, w * h * len,
|
|
|
|
|
tightConf[compressLevel].rawZlibLevel,
|
|
|
|
|
Z_DEFAULT_STRATEGY);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static rfbBool
|
|
|
|
|
CompressData(rfbClientPtr cl,
|
|
|
|
|
int streamId,
|
|
|
|
|
int dataLen,
|
|
|
|
|
int zlibLevel,
|
|
|
|
|
int zlibStrategy)
|
|
|
|
|
{
|
|
|
|
|
z_streamp pz;
|
|
|
|
|
int err;
|
|
|
|
|
|
|
|
|
|
if (dataLen < TIGHT_MIN_TO_COMPRESS) {
|
|
|
|
|
memcpy(&cl->updateBuf[cl->ublen], tightBeforeBuf, dataLen);
|
|
|
|
|
cl->ublen += dataLen;
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, dataLen);
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (zlibLevel == 0)
|
|
|
|
|
return SendCompressedData (cl, tightBeforeBuf, dataLen);
|
|
|
|
|
|
|
|
|
|
pz = &cl->zsStruct[streamId];
|
|
|
|
|
|
|
|
|
|
/* Initialize compression stream if needed. */
|
|
|
|
|
if (!cl->zsActive[streamId]) {
|
|
|
|
|
pz->zalloc = Z_NULL;
|
|
|
|
|
pz->zfree = Z_NULL;
|
|
|
|
|
pz->opaque = Z_NULL;
|
|
|
|
|
|
|
|
|
|
err = deflateInit2 (pz, zlibLevel, Z_DEFLATED, MAX_WBITS,
|
|
|
|
|
MAX_MEM_LEVEL, zlibStrategy);
|
|
|
|
|
if (err != Z_OK)
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
|
|
cl->zsActive[streamId] = TRUE;
|
|
|
|
|
cl->zsLevel[streamId] = zlibLevel;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Prepare buffer pointers. */
|
|
|
|
|
pz->next_in = (Bytef *)tightBeforeBuf;
|
|
|
|
|
pz->avail_in = dataLen;
|
|
|
|
|
pz->next_out = (Bytef *)tightAfterBuf;
|
|
|
|
|
pz->avail_out = tightAfterBufSize;
|
|
|
|
|
|
|
|
|
|
/* Change compression parameters if needed. */
|
|
|
|
|
if (zlibLevel != cl->zsLevel[streamId]) {
|
|
|
|
|
if (deflateParams (pz, zlibLevel, zlibStrategy) != Z_OK) {
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
cl->zsLevel[streamId] = zlibLevel;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Actual compression. */
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (deflate(pz, Z_SYNC_FLUSH) != Z_OK ||
|
|
|
|
|
pz->avail_in != 0 || pz->avail_out == 0) {
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
return SendCompressedData(cl, tightAfterBuf,
|
|
|
|
|
tightAfterBufSize - pz->avail_out);
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static rfbBool SendCompressedData(rfbClientPtr cl, char *buf,
|
|
|
|
|
int compressedLen)
|
|
|
|
|
{
|
|
|
|
|
int i, portionLen;
|
|
|
|
|
|
|
|
|
|
cl->updateBuf[cl->ublen++] = compressedLen & 0x7F;
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 1);
|
|
|
|
|
if (compressedLen > 0x7F) {
|
|
|
|
|
cl->updateBuf[cl->ublen-1] |= 0x80;
|
|
|
|
|
cl->updateBuf[cl->ublen++] = compressedLen >> 7 & 0x7F;
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 1);
|
|
|
|
|
if (compressedLen > 0x3FFF) {
|
|
|
|
|
cl->updateBuf[cl->ublen-1] |= 0x80;
|
|
|
|
|
cl->updateBuf[cl->ublen++] = compressedLen >> 14 & 0xFF;
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 1);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
portionLen = UPDATE_BUF_SIZE;
|
|
|
|
|
for (i = 0; i < compressedLen; i += portionLen) {
|
|
|
|
|
if (i + portionLen > compressedLen) {
|
|
|
|
|
portionLen = compressedLen - i;
|
|
|
|
|
}
|
|
|
|
|
if (cl->ublen + portionLen > UPDATE_BUF_SIZE) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
memcpy(&cl->updateBuf[cl->ublen], &buf[i], portionLen);
|
|
|
|
|
cl->ublen += portionLen;
|
|
|
|
|
}
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, compressedLen);
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Code to determine how many different colors used in rectangle.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
FillPalette8(int count)
|
|
|
|
|
{
|
|
|
|
|
uint8_t *data = (uint8_t *)tightBeforeBuf;
|
|
|
|
|
uint8_t c0, c1;
|
|
|
|
|
int i, n0, n1;
|
|
|
|
|
|
|
|
|
|
paletteNumColors = 0;
|
|
|
|
|
|
|
|
|
|
c0 = data[0];
|
|
|
|
|
for (i = 1; i < count && data[i] == c0; i++);
|
|
|
|
|
if (i == count) {
|
|
|
|
|
paletteNumColors = 1;
|
|
|
|
|
return; /* Solid rectangle */
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (paletteMaxColors < 2)
|
|
|
|
|
return;
|
|
|
|
|
|
|
|
|
|
n0 = i;
|
|
|
|
|
c1 = data[i];
|
|
|
|
|
n1 = 0;
|
|
|
|
|
for (i++; i < count; i++) {
|
|
|
|
|
if (data[i] == c0) {
|
|
|
|
|
n0++;
|
|
|
|
|
} else if (data[i] == c1) {
|
|
|
|
|
n1++;
|
|
|
|
|
} else
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
if (i == count) {
|
|
|
|
|
if (n0 > n1) {
|
|
|
|
|
monoBackground = (uint32_t)c0;
|
|
|
|
|
monoForeground = (uint32_t)c1;
|
|
|
|
|
} else {
|
|
|
|
|
monoBackground = (uint32_t)c1;
|
|
|
|
|
monoForeground = (uint32_t)c0;
|
|
|
|
|
}
|
|
|
|
|
paletteNumColors = 2; /* Two colors */
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
#define DEFINE_FILL_PALETTE_FUNCTION(bpp) \
|
|
|
|
|
\
|
|
|
|
|
static void \
|
|
|
|
|
FillPalette##bpp(int count) { \
|
|
|
|
|
uint##bpp##_t *data = (uint##bpp##_t *)tightBeforeBuf; \
|
|
|
|
|
uint##bpp##_t c0, c1, ci; \
|
|
|
|
|
int i, n0, n1, ni; \
|
|
|
|
|
\
|
|
|
|
|
c0 = data[0]; \
|
|
|
|
|
for (i = 1; i < count && data[i] == c0; i++); \
|
|
|
|
|
if (i >= count) { \
|
|
|
|
|
paletteNumColors = 1; /* Solid rectangle */ \
|
|
|
|
|
return; \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
if (paletteMaxColors < 2) { \
|
|
|
|
|
paletteNumColors = 0; /* Full-color encoding preferred */ \
|
|
|
|
|
return; \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
n0 = i; \
|
|
|
|
|
c1 = data[i]; \
|
|
|
|
|
n1 = 0; \
|
|
|
|
|
for (i++; i < count; i++) { \
|
|
|
|
|
ci = data[i]; \
|
|
|
|
|
if (ci == c0) { \
|
|
|
|
|
n0++; \
|
|
|
|
|
} else if (ci == c1) { \
|
|
|
|
|
n1++; \
|
|
|
|
|
} else \
|
|
|
|
|
break; \
|
|
|
|
|
} \
|
|
|
|
|
if (i >= count) { \
|
|
|
|
|
if (n0 > n1) { \
|
|
|
|
|
monoBackground = (uint32_t)c0; \
|
|
|
|
|
monoForeground = (uint32_t)c1; \
|
|
|
|
|
} else { \
|
|
|
|
|
monoBackground = (uint32_t)c1; \
|
|
|
|
|
monoForeground = (uint32_t)c0; \
|
|
|
|
|
} \
|
|
|
|
|
paletteNumColors = 2; /* Two colors */ \
|
|
|
|
|
return; \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
PaletteReset(); \
|
|
|
|
|
PaletteInsert (c0, (uint32_t)n0, bpp); \
|
|
|
|
|
PaletteInsert (c1, (uint32_t)n1, bpp); \
|
|
|
|
|
\
|
|
|
|
|
ni = 1; \
|
|
|
|
|
for (i++; i < count; i++) { \
|
|
|
|
|
if (data[i] == ci) { \
|
|
|
|
|
ni++; \
|
|
|
|
|
} else { \
|
|
|
|
|
if (!PaletteInsert (ci, (uint32_t)ni, bpp)) \
|
|
|
|
|
return; \
|
|
|
|
|
ci = data[i]; \
|
|
|
|
|
ni = 1; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
PaletteInsert (ci, (uint32_t)ni, bpp); \
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
DEFINE_FILL_PALETTE_FUNCTION(16)
|
|
|
|
|
DEFINE_FILL_PALETTE_FUNCTION(32)
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
#define DEFINE_FAST_FILL_PALETTE_FUNCTION(bpp) \
|
|
|
|
|
\
|
|
|
|
|
static void \
|
|
|
|
|
FastFillPalette##bpp(rfbClientPtr cl, uint##bpp##_t *data, int w, \
|
|
|
|
|
int pitch, int h) \
|
|
|
|
|
{ \
|
|
|
|
|
uint##bpp##_t c0, c1, ci, mask, c0t, c1t, cit; \
|
|
|
|
|
int i, j, i2 = 0, j2, n0, n1, ni; \
|
|
|
|
|
\
|
|
|
|
|
if (cl->translateFn != rfbTranslateNone) { \
|
|
|
|
|
mask = cl->screen->serverFormat.redMax \
|
|
|
|
|
<< cl->screen->serverFormat.redShift; \
|
|
|
|
|
mask |= cl->screen->serverFormat.greenMax \
|
|
|
|
|
<< cl->screen->serverFormat.greenShift; \
|
|
|
|
|
mask |= cl->screen->serverFormat.blueMax \
|
|
|
|
|
<< cl->screen->serverFormat.blueShift; \
|
|
|
|
|
} else mask = ~0; \
|
|
|
|
|
\
|
|
|
|
|
c0 = data[0] & mask; \
|
|
|
|
|
for (j = 0; j < h; j++) { \
|
|
|
|
|
for (i = 0; i < w; i++) { \
|
|
|
|
|
if ((data[j * pitch + i] & mask) != c0) \
|
|
|
|
|
goto done; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
done: \
|
|
|
|
|
if (j >= h) { \
|
|
|
|
|
paletteNumColors = 1; /* Solid rectangle */ \
|
|
|
|
|
return; \
|
|
|
|
|
} \
|
|
|
|
|
if (paletteMaxColors < 2) { \
|
|
|
|
|
paletteNumColors = 0; /* Full-color encoding preferred */ \
|
|
|
|
|
return; \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
n0 = j * w + i; \
|
|
|
|
|
c1 = data[j * pitch + i] & mask; \
|
|
|
|
|
n1 = 0; \
|
|
|
|
|
i++; if (i >= w) {i = 0; j++;} \
|
|
|
|
|
for (j2 = j; j2 < h; j2++) { \
|
|
|
|
|
for (i2 = i; i2 < w; i2++) { \
|
|
|
|
|
ci = data[j2 * pitch + i2] & mask; \
|
|
|
|
|
if (ci == c0) { \
|
|
|
|
|
n0++; \
|
|
|
|
|
} else if (ci == c1) { \
|
|
|
|
|
n1++; \
|
|
|
|
|
} else \
|
|
|
|
|
goto done2; \
|
|
|
|
|
} \
|
|
|
|
|
i = 0; \
|
|
|
|
|
} \
|
|
|
|
|
done2: \
|
|
|
|
|
(*cl->translateFn)(cl->translateLookupTable, \
|
|
|
|
|
&cl->screen->serverFormat, &cl->format, \
|
|
|
|
|
(char *)&c0, (char *)&c0t, bpp/8, 1, 1); \
|
|
|
|
|
(*cl->translateFn)(cl->translateLookupTable, \
|
|
|
|
|
&cl->screen->serverFormat, &cl->format, \
|
|
|
|
|
(char *)&c1, (char *)&c1t, bpp/8, 1, 1); \
|
|
|
|
|
if (j2 >= h) { \
|
|
|
|
|
if (n0 > n1) { \
|
|
|
|
|
monoBackground = (uint32_t)c0t; \
|
|
|
|
|
monoForeground = (uint32_t)c1t; \
|
|
|
|
|
} else { \
|
|
|
|
|
monoBackground = (uint32_t)c1t; \
|
|
|
|
|
monoForeground = (uint32_t)c0t; \
|
|
|
|
|
} \
|
|
|
|
|
paletteNumColors = 2; /* Two colors */ \
|
|
|
|
|
return; \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
PaletteReset(); \
|
|
|
|
|
PaletteInsert (c0t, (uint32_t)n0, bpp); \
|
|
|
|
|
PaletteInsert (c1t, (uint32_t)n1, bpp); \
|
|
|
|
|
\
|
|
|
|
|
ni = 1; \
|
|
|
|
|
i2++; if (i2 >= w) {i2 = 0; j2++;} \
|
|
|
|
|
for (j = j2; j < h; j++) { \
|
|
|
|
|
for (i = i2; i < w; i++) { \
|
|
|
|
|
if ((data[j * pitch + i] & mask) == ci) { \
|
|
|
|
|
ni++; \
|
|
|
|
|
} else { \
|
|
|
|
|
(*cl->translateFn)(cl->translateLookupTable, \
|
|
|
|
|
&cl->screen->serverFormat, \
|
|
|
|
|
&cl->format, (char *)&ci, \
|
|
|
|
|
(char *)&cit, bpp/8, 1, 1); \
|
|
|
|
|
if (!PaletteInsert (cit, (uint32_t)ni, bpp)) \
|
|
|
|
|
return; \
|
|
|
|
|
ci = data[j * pitch + i] & mask; \
|
|
|
|
|
ni = 1; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
i2 = 0; \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
(*cl->translateFn)(cl->translateLookupTable, \
|
|
|
|
|
&cl->screen->serverFormat, &cl->format, \
|
|
|
|
|
(char *)&ci, (char *)&cit, bpp/8, 1, 1); \
|
|
|
|
|
PaletteInsert (cit, (uint32_t)ni, bpp); \
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
DEFINE_FAST_FILL_PALETTE_FUNCTION(16)
|
|
|
|
|
DEFINE_FAST_FILL_PALETTE_FUNCTION(32)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Functions to operate with palette structures.
|
|
|
|
|
*/
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
#define HASH_FUNC16(rgb) ((int)((((rgb) >> 8) + (rgb)) & 0xFF))
|
|
|
|
|
#define HASH_FUNC32(rgb) ((int)((((rgb) >> 16) + ((rgb) >> 8)) & 0xFF))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
PaletteReset(void)
|
|
|
|
|
{
|
|
|
|
|
paletteNumColors = 0;
|
|
|
|
|
memset(palette.hash, 0, 256 * sizeof(COLOR_LIST *));
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
PaletteInsert(uint32_t rgb,
|
|
|
|
|
int numPixels,
|
|
|
|
|
int bpp)
|
|
|
|
|
{
|
|
|
|
|
COLOR_LIST *pnode;
|
|
|
|
|
COLOR_LIST *prev_pnode = NULL;
|
|
|
|
|
int hash_key, idx, new_idx, count;
|
|
|
|
|
|
|
|
|
|
hash_key = (bpp == 16) ? HASH_FUNC16(rgb) : HASH_FUNC32(rgb);
|
|
|
|
|
|
|
|
|
|
pnode = palette.hash[hash_key];
|
|
|
|
|
|
|
|
|
|
while (pnode != NULL) {
|
|
|
|
|
if (pnode->rgb == rgb) {
|
|
|
|
|
/* Such palette entry already exists. */
|
|
|
|
|
new_idx = idx = pnode->idx;
|
|
|
|
|
count = palette.entry[idx].numPixels + numPixels;
|
|
|
|
|
if (new_idx && palette.entry[new_idx-1].numPixels < count) {
|
|
|
|
|
do {
|
|
|
|
|
palette.entry[new_idx] = palette.entry[new_idx-1];
|
|
|
|
|
palette.entry[new_idx].listNode->idx = new_idx;
|
|
|
|
|
new_idx--;
|
|
|
|
|
}
|
|
|
|
|
while (new_idx && palette.entry[new_idx-1].numPixels < count);
|
|
|
|
|
palette.entry[new_idx].listNode = pnode;
|
|
|
|
|
pnode->idx = new_idx;
|
|
|
|
|
}
|
|
|
|
|
palette.entry[new_idx].numPixels = count;
|
|
|
|
|
return paletteNumColors;
|
|
|
|
|
}
|
|
|
|
|
prev_pnode = pnode;
|
|
|
|
|
pnode = pnode->next;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Check if palette is full. */
|
|
|
|
|
if (paletteNumColors == 256 || paletteNumColors == paletteMaxColors) {
|
|
|
|
|
paletteNumColors = 0;
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Move palette entries with lesser pixel counts. */
|
|
|
|
|
for ( idx = paletteNumColors;
|
|
|
|
|
idx > 0 && palette.entry[idx-1].numPixels < numPixels;
|
|
|
|
|
idx-- ) {
|
|
|
|
|
palette.entry[idx] = palette.entry[idx-1];
|
|
|
|
|
palette.entry[idx].listNode->idx = idx;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* Add new palette entry into the freed slot. */
|
|
|
|
|
pnode = &palette.list[paletteNumColors];
|
|
|
|
|
if (prev_pnode != NULL) {
|
|
|
|
|
prev_pnode->next = pnode;
|
|
|
|
|
} else {
|
|
|
|
|
palette.hash[hash_key] = pnode;
|
|
|
|
|
}
|
|
|
|
|
pnode->next = NULL;
|
|
|
|
|
pnode->idx = idx;
|
|
|
|
|
pnode->rgb = rgb;
|
|
|
|
|
palette.entry[idx].listNode = pnode;
|
|
|
|
|
palette.entry[idx].numPixels = numPixels;
|
|
|
|
|
|
|
|
|
|
return (++paletteNumColors);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Converting 32-bit color samples into 24-bit colors.
|
|
|
|
|
* Should be called only when redMax, greenMax and blueMax are 255.
|
|
|
|
|
* Color components assumed to be byte-aligned.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
static void Pack24(rfbClientPtr cl,
|
|
|
|
|
char *buf,
|
|
|
|
|
rfbPixelFormat *fmt,
|
|
|
|
|
int count)
|
|
|
|
|
{
|
|
|
|
|
uint32_t *buf32;
|
|
|
|
|
uint32_t pix;
|
|
|
|
|
int r_shift, g_shift, b_shift;
|
|
|
|
|
|
|
|
|
|
buf32 = (uint32_t *)buf;
|
|
|
|
|
|
|
|
|
|
if (!cl->screen->serverFormat.bigEndian == !fmt->bigEndian) {
|
|
|
|
|
r_shift = fmt->redShift;
|
|
|
|
|
g_shift = fmt->greenShift;
|
|
|
|
|
b_shift = fmt->blueShift;
|
|
|
|
|
} else {
|
|
|
|
|
r_shift = 24 - fmt->redShift;
|
|
|
|
|
g_shift = 24 - fmt->greenShift;
|
|
|
|
|
b_shift = 24 - fmt->blueShift;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
while (count--) {
|
|
|
|
|
pix = *buf32++;
|
|
|
|
|
*buf++ = (char)(pix >> r_shift);
|
|
|
|
|
*buf++ = (char)(pix >> g_shift);
|
|
|
|
|
*buf++ = (char)(pix >> b_shift);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Converting truecolor samples into palette indices.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#define DEFINE_IDX_ENCODE_FUNCTION(bpp) \
|
|
|
|
|
\
|
|
|
|
|
static void \
|
|
|
|
|
EncodeIndexedRect##bpp(uint8_t *buf, int count) { \
|
|
|
|
|
COLOR_LIST *pnode; \
|
|
|
|
|
uint##bpp##_t *src; \
|
|
|
|
|
uint##bpp##_t rgb; \
|
|
|
|
|
int rep = 0; \
|
|
|
|
|
\
|
|
|
|
|
src = (uint##bpp##_t *) buf; \
|
|
|
|
|
\
|
|
|
|
|
while (count--) { \
|
|
|
|
|
rgb = *src++; \
|
|
|
|
|
while (count && *src == rgb) { \
|
|
|
|
|
rep++, src++, count--; \
|
|
|
|
|
} \
|
|
|
|
|
pnode = palette.hash[HASH_FUNC##bpp(rgb)]; \
|
|
|
|
|
while (pnode != NULL) { \
|
|
|
|
|
if ((uint##bpp##_t)pnode->rgb == rgb) { \
|
|
|
|
|
*buf++ = (uint8_t)pnode->idx; \
|
|
|
|
|
while (rep) { \
|
|
|
|
|
*buf++ = (uint8_t)pnode->idx; \
|
|
|
|
|
rep--; \
|
|
|
|
|
} \
|
|
|
|
|
break; \
|
|
|
|
|
} \
|
|
|
|
|
pnode = pnode->next; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
DEFINE_IDX_ENCODE_FUNCTION(16)
|
|
|
|
|
DEFINE_IDX_ENCODE_FUNCTION(32)
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
|
|
|
|
|
#define DEFINE_MONO_ENCODE_FUNCTION(bpp) \
|
|
|
|
|
\
|
|
|
|
|
static void \
|
|
|
|
|
EncodeMonoRect##bpp(uint8_t *buf, int w, int h) { \
|
|
|
|
|
uint##bpp##_t *ptr; \
|
|
|
|
|
uint##bpp##_t bg; \
|
|
|
|
|
unsigned int value, mask; \
|
|
|
|
|
int aligned_width; \
|
|
|
|
|
int x, y, bg_bits; \
|
|
|
|
|
\
|
|
|
|
|
ptr = (uint##bpp##_t *) buf; \
|
|
|
|
|
bg = (uint##bpp##_t) monoBackground; \
|
|
|
|
|
aligned_width = w - w % 8; \
|
|
|
|
|
\
|
|
|
|
|
for (y = 0; y < h; y++) { \
|
|
|
|
|
for (x = 0; x < aligned_width; x += 8) { \
|
|
|
|
|
for (bg_bits = 0; bg_bits < 8; bg_bits++) { \
|
|
|
|
|
if (*ptr++ != bg) \
|
|
|
|
|
break; \
|
|
|
|
|
} \
|
|
|
|
|
if (bg_bits == 8) { \
|
|
|
|
|
*buf++ = 0; \
|
|
|
|
|
continue; \
|
|
|
|
|
} \
|
|
|
|
|
mask = 0x80 >> bg_bits; \
|
|
|
|
|
value = mask; \
|
|
|
|
|
for (bg_bits++; bg_bits < 8; bg_bits++) { \
|
|
|
|
|
mask >>= 1; \
|
|
|
|
|
if (*ptr++ != bg) { \
|
|
|
|
|
value |= mask; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
*buf++ = (uint8_t)value; \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
mask = 0x80; \
|
|
|
|
|
value = 0; \
|
|
|
|
|
if (x >= w) \
|
|
|
|
|
continue; \
|
|
|
|
|
\
|
|
|
|
|
for (; x < w; x++) { \
|
|
|
|
|
if (*ptr++ != bg) { \
|
|
|
|
|
value |= mask; \
|
|
|
|
|
} \
|
|
|
|
|
mask >>= 1; \
|
|
|
|
|
} \
|
|
|
|
|
*buf++ = (uint8_t)value; \
|
|
|
|
|
} \
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
DEFINE_MONO_ENCODE_FUNCTION(8)
|
|
|
|
|
DEFINE_MONO_ENCODE_FUNCTION(16)
|
|
|
|
|
DEFINE_MONO_ENCODE_FUNCTION(32)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
* JPEG compression stuff.
|
|
|
|
|
*/
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
static rfbBool
|
|
|
|
|
SendJpegRect(rfbClientPtr cl, int x, int y, int w, int h, int quality)
|
|
|
|
|
{
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
unsigned char *srcbuf;
|
|
|
|
|
int ps = cl->screen->serverFormat.bitsPerPixel / 8;
|
|
|
|
|
int subsamp = subsampLevel2tjsubsamp[subsampLevel];
|
|
|
|
|
unsigned long size = 0;
|
|
|
|
|
int flags = 0, pitch;
|
|
|
|
|
unsigned char *tmpbuf = NULL;
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (cl->screen->serverFormat.bitsPerPixel == 8)
|
|
|
|
|
return SendFullColorRect(cl, x, y, w, h);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (ps < 2) {
|
|
|
|
|
rfbLog("Error: JPEG requires 16-bit, 24-bit, or 32-bit pixel format.\n");
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (!j) {
|
|
|
|
|
if ((j = tjInitCompress()) == NULL) {
|
|
|
|
|
rfbLog("JPEG Error: %s\n", tjGetErrorStr());
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (tightAfterBufSize < TJBUFSIZE(w, h)) {
|
|
|
|
|
if (tightAfterBuf == NULL)
|
|
|
|
|
tightAfterBuf = (char *)malloc(TJBUFSIZE(w, h));
|
|
|
|
|
else
|
|
|
|
|
tightAfterBuf = (char *)realloc(tightAfterBuf,
|
|
|
|
|
TJBUFSIZE(w, h));
|
|
|
|
|
if (!tightAfterBuf) {
|
|
|
|
|
rfbLog("Memory allocation failure!\n");
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
tightAfterBufSize = TJBUFSIZE(w, h);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (ps == 2) {
|
|
|
|
|
uint16_t *srcptr, pix;
|
|
|
|
|
unsigned char *dst;
|
|
|
|
|
int inRed, inGreen, inBlue, i, j;
|
|
|
|
|
|
|
|
|
|
if((tmpbuf = (unsigned char *)malloc(w * h * 3)) == NULL)
|
|
|
|
|
rfbLog("Memory allocation failure!\n");
|
|
|
|
|
srcptr = (uint16_t *)&cl->scaledScreen->frameBuffer
|
|
|
|
|
[y * cl->scaledScreen->paddedWidthInBytes + x * ps];
|
|
|
|
|
dst = tmpbuf;
|
|
|
|
|
for(j = 0; j < h; j++) {
|
|
|
|
|
uint16_t *srcptr2 = srcptr;
|
|
|
|
|
unsigned char *dst2 = dst;
|
|
|
|
|
for (i = 0; i < w; i++) {
|
|
|
|
|
pix = *srcptr2++;
|
|
|
|
|
inRed = (int) (pix >> cl->screen->serverFormat.redShift
|
|
|
|
|
& cl->screen->serverFormat.redMax);
|
|
|
|
|
inGreen = (int) (pix >> cl->screen->serverFormat.greenShift
|
|
|
|
|
& cl->screen->serverFormat.greenMax);
|
|
|
|
|
inBlue = (int) (pix >> cl->screen->serverFormat.blueShift
|
|
|
|
|
& cl->screen->serverFormat.blueMax);
|
|
|
|
|
*dst2++ = (uint8_t)((inRed * 255
|
|
|
|
|
+ cl->screen->serverFormat.redMax / 2)
|
|
|
|
|
/ cl->screen->serverFormat.redMax);
|
|
|
|
|
*dst2++ = (uint8_t)((inGreen * 255
|
|
|
|
|
+ cl->screen->serverFormat.greenMax / 2)
|
|
|
|
|
/ cl->screen->serverFormat.greenMax);
|
|
|
|
|
*dst2++ = (uint8_t)((inBlue * 255
|
|
|
|
|
+ cl->screen->serverFormat.blueMax / 2)
|
|
|
|
|
/ cl->screen->serverFormat.blueMax);
|
|
|
|
|
}
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
srcptr += cl->scaledScreen->paddedWidthInBytes / ps;
|
|
|
|
|
dst += w * 3;
|
|
|
|
|
}
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
srcbuf = tmpbuf;
|
|
|
|
|
pitch = w * 3;
|
|
|
|
|
ps = 3;
|
|
|
|
|
} else {
|
|
|
|
|
if (cl->screen->serverFormat.bigEndian && ps == 4)
|
|
|
|
|
flags |= TJ_ALPHAFIRST;
|
|
|
|
|
if (cl->screen->serverFormat.redShift == 16
|
|
|
|
|
&& cl->screen->serverFormat.blueShift == 0)
|
|
|
|
|
flags |= TJ_BGR;
|
|
|
|
|
if (cl->screen->serverFormat.bigEndian)
|
|
|
|
|
flags ^= TJ_BGR;
|
|
|
|
|
pitch = cl->scaledScreen->paddedWidthInBytes;
|
|
|
|
|
srcbuf = (unsigned char *)&cl->scaledScreen->frameBuffer
|
|
|
|
|
[y * pitch + x * ps];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (tjCompress(j, srcbuf, w, pitch, h, ps, (unsigned char *)tightAfterBuf,
|
|
|
|
|
&size, subsamp, quality, flags) == -1) {
|
|
|
|
|
rfbLog("JPEG Error: %s\n", tjGetErrorStr());
|
|
|
|
|
if (tmpbuf) {
|
|
|
|
|
free(tmpbuf);
|
|
|
|
|
tmpbuf = NULL;
|
|
|
|
|
}
|
|
|
|
|
return 0;
|
|
|
|
|
}
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
if (tmpbuf) {
|
|
|
|
|
free(tmpbuf);
|
|
|
|
|
tmpbuf = NULL;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if (cl->ublen + TIGHT_MIN_TO_COMPRESS + 1 > UPDATE_BUF_SIZE) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (char)(rfbTightJpeg << 4);
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 1);
|
|
|
|
|
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
return SendCompressedData(cl, tightAfterBuf, (int)size);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
PrepareRowForImg(rfbClientPtr cl,
|
|
|
|
|
uint8_t *dst,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int count)
|
|
|
|
|
{
|
|
|
|
|
if (cl->screen->serverFormat.bitsPerPixel == 32) {
|
|
|
|
|
if ( cl->screen->serverFormat.redMax == 0xFF &&
|
|
|
|
|
cl->screen->serverFormat.greenMax == 0xFF &&
|
|
|
|
|
cl->screen->serverFormat.blueMax == 0xFF ) {
|
|
|
|
|
PrepareRowForImg24(cl, dst, x, y, count);
|
|
|
|
|
} else {
|
|
|
|
|
PrepareRowForImg32(cl, dst, x, y, count);
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
/* 16 bpp assumed. */
|
|
|
|
|
PrepareRowForImg16(cl, dst, x, y, count);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void
|
|
|
|
|
PrepareRowForImg24(rfbClientPtr cl,
|
|
|
|
|
uint8_t *dst,
|
|
|
|
|
int x,
|
|
|
|
|
int y,
|
|
|
|
|
int count)
|
|
|
|
|
{
|
|
|
|
|
uint32_t *fbptr;
|
|
|
|
|
uint32_t pix;
|
|
|
|
|
|
|
|
|
|
fbptr = (uint32_t *)
|
|
|
|
|
&cl->scaledScreen->frameBuffer[y * cl->scaledScreen->paddedWidthInBytes + x * 4];
|
|
|
|
|
|
|
|
|
|
while (count--) {
|
|
|
|
|
pix = *fbptr++;
|
|
|
|
|
*dst++ = (uint8_t)(pix >> cl->screen->serverFormat.redShift);
|
|
|
|
|
*dst++ = (uint8_t)(pix >> cl->screen->serverFormat.greenShift);
|
|
|
|
|
*dst++ = (uint8_t)(pix >> cl->screen->serverFormat.blueShift);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
#define DEFINE_JPEG_GET_ROW_FUNCTION(bpp) \
|
|
|
|
|
\
|
|
|
|
|
static void \
|
|
|
|
|
PrepareRowForImg##bpp(rfbClientPtr cl, uint8_t *dst, int x, int y, int count) { \
|
|
|
|
|
uint##bpp##_t *fbptr; \
|
|
|
|
|
uint##bpp##_t pix; \
|
|
|
|
|
int inRed, inGreen, inBlue; \
|
|
|
|
|
\
|
|
|
|
|
fbptr = (uint##bpp##_t *) \
|
|
|
|
|
&cl->scaledScreen->frameBuffer[y * cl->scaledScreen->paddedWidthInBytes + \
|
|
|
|
|
x * (bpp / 8)]; \
|
|
|
|
|
\
|
|
|
|
|
while (count--) { \
|
|
|
|
|
pix = *fbptr++; \
|
|
|
|
|
\
|
|
|
|
|
inRed = (int) \
|
|
|
|
|
(pix >> cl->screen->serverFormat.redShift & cl->screen->serverFormat.redMax); \
|
|
|
|
|
inGreen = (int) \
|
|
|
|
|
(pix >> cl->screen->serverFormat.greenShift & cl->screen->serverFormat.greenMax); \
|
|
|
|
|
inBlue = (int) \
|
|
|
|
|
(pix >> cl->screen->serverFormat.blueShift & cl->screen->serverFormat.blueMax); \
|
|
|
|
|
\
|
|
|
|
|
*dst++ = (uint8_t)((inRed * 255 + cl->screen->serverFormat.redMax / 2) / \
|
|
|
|
|
cl->screen->serverFormat.redMax); \
|
|
|
|
|
*dst++ = (uint8_t)((inGreen * 255 + cl->screen->serverFormat.greenMax / 2) / \
|
|
|
|
|
cl->screen->serverFormat.greenMax); \
|
|
|
|
|
*dst++ = (uint8_t)((inBlue * 255 + cl->screen->serverFormat.blueMax / 2) / \
|
|
|
|
|
cl->screen->serverFormat.blueMax); \
|
|
|
|
|
} \
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
DEFINE_JPEG_GET_ROW_FUNCTION(16)
|
|
|
|
|
DEFINE_JPEG_GET_ROW_FUNCTION(32)
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* PNG compression stuff.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#ifdef LIBVNCSERVER_HAVE_LIBPNG
|
|
|
|
|
|
|
|
|
|
static TLS int pngDstDataLen = 0;
|
|
|
|
|
|
|
|
|
|
static rfbBool CanSendPngRect(rfbClientPtr cl, int w, int h) {
|
|
|
|
|
if (cl->tightEncoding != rfbEncodingTightPng) {
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if ( cl->screen->serverFormat.bitsPerPixel == 8 ||
|
|
|
|
|
cl->format.bitsPerPixel == 8) {
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
return TRUE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void pngWriteData(png_structp png_ptr, png_bytep data,
|
|
|
|
|
png_size_t length)
|
|
|
|
|
{
|
|
|
|
|
#if 0
|
|
|
|
|
rfbClientPtr cl = png_get_io_ptr(png_ptr);
|
|
|
|
|
|
|
|
|
|
buffer_reserve(&vs->tight.png, vs->tight.png.offset + length);
|
|
|
|
|
memcpy(vs->tight.png.buffer + vs->tight.png.offset, data, length);
|
|
|
|
|
#endif
|
|
|
|
|
memcpy(tightAfterBuf + pngDstDataLen, data, length);
|
|
|
|
|
|
|
|
|
|
pngDstDataLen += length;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void pngFlushData(png_structp png_ptr)
|
|
|
|
|
{
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
static void *pngMalloc(png_structp png_ptr, png_size_t size)
|
|
|
|
|
{
|
|
|
|
|
return malloc(size);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void pngFree(png_structp png_ptr, png_voidp ptr)
|
|
|
|
|
{
|
|
|
|
|
free(ptr);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static rfbBool SendPngRect(rfbClientPtr cl, int x, int y, int w, int h) {
|
|
|
|
|
/* rfbLog(">> SendPngRect x:%d, y:%d, w:%d, h:%d\n", x, y, w, h); */
|
|
|
|
|
|
|
|
|
|
png_byte color_type;
|
|
|
|
|
png_structp png_ptr;
|
|
|
|
|
png_infop info_ptr;
|
|
|
|
|
png_colorp png_palette = NULL;
|
|
|
|
|
int level = tightPngConf[cl->tightCompressLevel].png_zlib_level;
|
|
|
|
|
int filters = tightPngConf[cl->tightCompressLevel].png_filters;
|
|
|
|
|
uint8_t *buf;
|
|
|
|
|
int dy;
|
|
|
|
|
|
|
|
|
|
pngDstDataLen = 0;
|
|
|
|
|
|
|
|
|
|
png_ptr = png_create_write_struct_2(PNG_LIBPNG_VER_STRING, NULL, NULL, NULL,
|
|
|
|
|
NULL, pngMalloc, pngFree);
|
|
|
|
|
|
|
|
|
|
if (png_ptr == NULL)
|
|
|
|
|
return FALSE;
|
|
|
|
|
|
|
|
|
|
info_ptr = png_create_info_struct(png_ptr);
|
|
|
|
|
|
|
|
|
|
if (info_ptr == NULL) {
|
|
|
|
|
png_destroy_write_struct(&png_ptr, NULL);
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
png_set_write_fn(png_ptr, (void *) cl, pngWriteData, pngFlushData);
|
|
|
|
|
png_set_compression_level(png_ptr, level);
|
|
|
|
|
png_set_filter(png_ptr, PNG_FILTER_TYPE_DEFAULT, filters);
|
|
|
|
|
|
|
|
|
|
#if 0
|
|
|
|
|
/* TODO: */
|
|
|
|
|
if (palette) {
|
|
|
|
|
color_type = PNG_COLOR_TYPE_PALETTE;
|
|
|
|
|
} else {
|
|
|
|
|
color_type = PNG_COLOR_TYPE_RGB;
|
|
|
|
|
}
|
|
|
|
|
#else
|
|
|
|
|
color_type = PNG_COLOR_TYPE_RGB;
|
|
|
|
|
#endif
|
|
|
|
|
png_set_IHDR(png_ptr, info_ptr, w, h,
|
|
|
|
|
8, color_type, PNG_INTERLACE_NONE,
|
|
|
|
|
PNG_COMPRESSION_TYPE_DEFAULT, PNG_FILTER_TYPE_DEFAULT);
|
|
|
|
|
|
|
|
|
|
#if 0
|
|
|
|
|
if (color_type == PNG_COLOR_TYPE_PALETTE) {
|
|
|
|
|
struct palette_cb_priv priv;
|
|
|
|
|
|
|
|
|
|
png_palette = pngMalloc(png_ptr, sizeof(*png_palette) *
|
|
|
|
|
palette_size(palette));
|
|
|
|
|
|
|
|
|
|
priv.vs = vs;
|
|
|
|
|
priv.png_palette = png_palette;
|
|
|
|
|
palette_iter(palette, write_png_palette, &priv);
|
|
|
|
|
|
|
|
|
|
png_set_PLTE(png_ptr, info_ptr, png_palette, palette_size(palette));
|
|
|
|
|
|
|
|
|
|
offset = vs->tight.tight.offset;
|
|
|
|
|
if (vs->clientds.pf.bytes_per_pixel == 4) {
|
|
|
|
|
tight_encode_indexed_rect32(vs->tight.tight.buffer, w * h, palette);
|
|
|
|
|
} else {
|
|
|
|
|
tight_encode_indexed_rect16(vs->tight.tight.buffer, w * h, palette);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
buffer_reserve(&vs->tight.png, 2048);
|
|
|
|
|
#endif
|
|
|
|
|
|
|
|
|
|
png_write_info(png_ptr, info_ptr);
|
|
|
|
|
buf = malloc(w * 3);
|
|
|
|
|
for (dy = 0; dy < h; dy++)
|
|
|
|
|
{
|
|
|
|
|
#if 0
|
|
|
|
|
if (color_type == PNG_COLOR_TYPE_PALETTE) {
|
|
|
|
|
memcpy(buf, vs->tight.tight.buffer + (dy * w), w);
|
|
|
|
|
} else {
|
|
|
|
|
PrepareRowForImg(cl, buf, x, y + dy, w);
|
|
|
|
|
}
|
|
|
|
|
#else
|
|
|
|
|
PrepareRowForImg(cl, buf, x, y + dy, w);
|
|
|
|
|
#endif
|
|
|
|
|
png_write_row(png_ptr, buf);
|
|
|
|
|
}
|
|
|
|
|
free(buf);
|
|
|
|
|
|
|
|
|
|
png_write_end(png_ptr, NULL);
|
|
|
|
|
|
|
|
|
|
if (color_type == PNG_COLOR_TYPE_PALETTE) {
|
|
|
|
|
pngFree(png_ptr, png_palette);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
png_destroy_write_struct(&png_ptr, &info_ptr);
|
|
|
|
|
|
|
|
|
|
/* done v */
|
|
|
|
|
|
|
|
|
|
if (cl->ublen + TIGHT_MIN_TO_COMPRESS + 1 > UPDATE_BUF_SIZE) {
|
|
|
|
|
if (!rfbSendUpdateBuf(cl))
|
|
|
|
|
return FALSE;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
cl->updateBuf[cl->ublen++] = (char)(rfbTightPng << 4);
|
|
|
|
|
rfbStatRecordEncodingSentAdd(cl, cl->tightEncoding, 1);
|
|
|
|
|
|
|
|
|
|
/* rfbLog("<< SendPngRect\n"); */
|
Replace TightVNC encoder with TurboVNC encoder. This patch is the result of further research and discussion that revealed the following:
-- TightPng encoding and the rfbTightNoZlib extension need not conflict. Since
TightPng is a separate encoding type, not supported by TurboVNC-compatible
viewers, then the rfbTightNoZlib extension can be used solely whenever the
encoding type is Tight and disabled with the encoding type is TightPng.
-- In the TightVNC encoder, compression levels above 5 are basically useless.
On the set of 20 low-level datasets that were used to design the TurboVNC
encoder (these include the eight 2D application captures that were also used
when designing the TightVNC encoder, as well as 12 3D application captures
provided by the VirtualGL Project--
see http://www.virtualgl.org/pmwiki/uploads/About/tighttoturbo.pdf), moving
from Compression Level (CL) 5 to CL 9 in the TightVNC encoder did not
increase the compression ratio of any datasets more than 10%, and the
compression ratio only increased by more than 5% on four of them. The
compression ratio actually decreased a few percent on five of them. In
exchange for this paltry increase in compression ratio, the CPU usage, on
average, went up by a factor of 5. Thus, for all intents and purposes,
TightVNC CL 5 provides the "best useful compression" for that encoder.
-- TurboVNC's best compression level (CL 2) compresses 3D and video workloads
significantly more "tightly" than TightVNC CL 5 (~70% better, in the
aggregate) but does not quite achieve the same level of compression with 2D
workloads (~20% worse, in the aggregate.) This decrease in compression ratio
may or may not be noticeable, since many of the datasets it affects are not
performance-critical (such as the console output of a compilation, etc.)
However, for peace of mind, it was still desirable to have a mode that
compressed with equal "tightness" to TightVNC CL 5, since we proposed to
replace that encoder entirely.
-- A new mode was discovered in the TurboVNC encoder that produces, in the
aggregate, similar compression ratios on 2D datasets as TightVNC CL 5. That
new mode involves using Zlib level 7 (the same level used by TightVNC CL 5)
but setting the "palette threshold" to 256, so that indexed color encoding
is used whenever possible. This mode reduces bandwidth only marginally
(typically 10-20%) relative to TurboVNC CL 2 on low-color workloads, in
exchange for nearly doubling CPU usage, and it does not benefit high-color
workloads at all (since those are usually encoded with JPEG.) However, it
provides a means of reproducing the same "tightness" as the TightVNC
encoder on 2D workloads without sacrificing any compression for 3D/video
workloads, and without using any more CPU time than necessary.
-- The TurboVNC encoder still performs as well or better than the TightVNC
encoder when plain libjpeg is used instead of libjpeg-turbo.
Specific notes follow:
common/turbojpeg.c common/turbojpeg.h:
Added code to emulate the libjpeg-turbo colorspace extensions, so that the
TurboJPEG wrapper can be used with plain libjpeg as well. This required
updating the TurboJPEG wrapper to the latest code from libjpeg-turbo 1.2.0,
mainly because the TurboJPEG 1.2 API handles pixel formats in a much cleaner
way, which made the conversion code easier to write. It also eases the
maintenance to have the wrapper synced as much as possible with the upstream
code base (so I can merge any relevant bug fixes that are discovered upstream.)
The libvncserver version of the TurboJPEG wrapper is a "lite" version,
containing only the JPEG compression/decompression code and not the lossless
transform, YUV encoding/decoding, and dynamic buffer allocation features from
TurboJPEG 1.2.
configure.ac:
Removed the --with-turbovnc option. configure still checks for the presence of
libjpeg-turbo, but only for the purposes of printing a performance warning if
it isn't available.
rfb/rfb.h:
Fix a bug introduced with the initial TurboVNC encoder patch. We cannot use
tightQualityLevel for the TurboVNC 1-100 quality level, because
tightQualityLevel is also used by ZRLE. Thus, a new parameter
(turboQualityLevel) was created.
rfb/rfbproto.h:
Remove TurboVNC-specific #ifdefs and language
libvncserver/rfbserver.c:
Remove TurboVNC-specific #ifdefs. Fix afore-mentioned tightQualityLevel bug.
libvncserver/tight.c:
Replaced the TightVNC encoder with the TurboVNC encoder. Relative to the
initial TurboVNC encoder patch, this patch also:
-- Adds TightPng support to the TurboVNC encoder
-- Adds the afore-mentioned low-bandwidth mode, which is mapped externally to
Compression Level 9
test/*:
Included TJUnitTest (a regression test for the TurboJPEG wrapper) as well as
TJBench (a benchmark for same.) These are useful for ensuring that the wrapper
still functions correctly and performantly if it needs to be modified for
whatever reason. Both of these programs are derived from libjpeg-turbo 1.2.0.
As with the TurboJPEG wrapper, they do not contain the more advanced features
of TurboJPEG 1.2, such as YUV encoding/decoding and lossless transforms.
13 years ago
|
|
|
return SendCompressedData(cl, tightAfterBuf, pngDstDataLen);
|
|
|
|
|
}
|
|
|
|
|
#endif
|
