• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1Background
2==========
3
4libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2,
5NEON, AltiVec) to accelerate baseline JPEG compression and decompression on
6x86, x86-64, ARM, and PowerPC systems.  On such systems, libjpeg-turbo is
7generally 2-6x as fast as libjpeg, all else being equal.  On other types of
8systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by
9virtue of its highly-optimized Huffman coding routines.  In many cases, the
10performance of libjpeg-turbo rivals that of proprietary high-speed JPEG codecs.
11
12libjpeg-turbo implements both the traditional libjpeg API as well as the less
13powerful but more straightforward TurboJPEG API.  libjpeg-turbo also features
14colorspace extensions that allow it to compress from/decompress to 32-bit and
15big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java
16interface.
17
18libjpeg-turbo was originally based on libjpeg/SIMD, an MMX-accelerated
19derivative of libjpeg v6b developed by Miyasaka Masaru.  The TigerVNC and
20VirtualGL projects made numerous enhancements to the codec in 2009, and in
21early 2010, libjpeg-turbo spun off into an independent project, with the goal
22of making high-speed JPEG compression/decompression technology available to a
23broader range of users and developers.
24
25
26License
27=======
28
29libjpeg-turbo is covered by three compatible BSD-style open source licenses.
30Refer to [LICENSE.md](LICENSE.md) for a roll-up of license terms.
31
32
33Building libjpeg-turbo
34======================
35
36Refer to [BUILDING.md](BUILDING.md) for complete instructions.
37
38
39Using libjpeg-turbo
40===================
41
42libjpeg-turbo includes two APIs that can be used to compress and decompress
43JPEG images:
44
45- **TurboJPEG API**<br>
46  This API provides an easy-to-use interface for compressing and decompressing
47  JPEG images in memory.  It also provides some functionality that would not be
48  straightforward to achieve using the underlying libjpeg API, such as
49  generating planar YUV images and performing multiple simultaneous lossless
50  transforms on an image.  The Java interface for libjpeg-turbo is written on
51  top of the TurboJPEG API.
52
53- **libjpeg API**<br>
54  This is the de facto industry-standard API for compressing and decompressing
55  JPEG images.  It is more difficult to use than the TurboJPEG API but also
56  more powerful.  The libjpeg API implementation in libjpeg-turbo is both
57  API/ABI-compatible and mathematically compatible with libjpeg v6b.  It can
58  also optionally be configured to be API/ABI-compatible with libjpeg v7 and v8
59  (see below.)
60
61There is no significant performance advantage to either API when both are used
62to perform similar operations.
63
64Colorspace Extensions
65---------------------
66
67libjpeg-turbo includes extensions that allow JPEG images to be compressed
68directly from (and decompressed directly to) buffers that use BGR, BGRX,
69RGBX, XBGR, and XRGB pixel ordering.  This is implemented with ten new
70colorspace constants:
71
72    JCS_EXT_RGB   /* red/green/blue */
73    JCS_EXT_RGBX  /* red/green/blue/x */
74    JCS_EXT_BGR   /* blue/green/red */
75    JCS_EXT_BGRX  /* blue/green/red/x */
76    JCS_EXT_XBGR  /* x/blue/green/red */
77    JCS_EXT_XRGB  /* x/red/green/blue */
78    JCS_EXT_RGBA  /* red/green/blue/alpha */
79    JCS_EXT_BGRA  /* blue/green/red/alpha */
80    JCS_EXT_ABGR  /* alpha/blue/green/red */
81    JCS_EXT_ARGB  /* alpha/red/green/blue */
82
83Setting `cinfo.in_color_space` (compression) or `cinfo.out_color_space`
84(decompression) to one of these values will cause libjpeg-turbo to read the
85red, green, and blue values from (or write them to) the appropriate position in
86the pixel when compressing from/decompressing to an RGB buffer.
87
88Your application can check for the existence of these extensions at compile
89time with:
90
91    #ifdef JCS_EXTENSIONS
92
93At run time, attempting to use these extensions with a libjpeg implementation
94that does not support them will result in a "Bogus input colorspace" error.
95Applications can trap this error in order to test whether run-time support is
96available for the colorspace extensions.
97
98When using the RGBX, BGRX, XBGR, and XRGB colorspaces during decompression, the
99X byte is undefined, and in order to ensure the best performance, libjpeg-turbo
100can set that byte to whatever value it wishes.  If an application expects the X
101byte to be used as an alpha channel, then it should specify `JCS_EXT_RGBA`,
102`JCS_EXT_BGRA`, `JCS_EXT_ABGR`, or `JCS_EXT_ARGB`.  When these colorspace
103constants are used, the X byte is guaranteed to be 0xFF, which is interpreted
104as opaque.
105
106Your application can check for the existence of the alpha channel colorspace
107extensions at compile time with:
108
109    #ifdef JCS_ALPHA_EXTENSIONS
110
111[jcstest.c](jcstest.c), located in the libjpeg-turbo source tree, demonstrates
112how to check for the existence of the colorspace extensions at compile time and
113run time.
114
115libjpeg v7 and v8 API/ABI Emulation
116-----------------------------------
117
118With libjpeg v7 and v8, new features were added that necessitated extending the
119compression and decompression structures.  Unfortunately, due to the exposed
120nature of those structures, extending them also necessitated breaking backward
121ABI compatibility with previous libjpeg releases.  Thus, programs that were
122built to use libjpeg v7 or v8 did not work with libjpeg-turbo, since it is
123based on the libjpeg v6b code base.  Although libjpeg v7 and v8 are not
124as widely used as v6b, enough programs (including a few Linux distros) made
125the switch that there was a demand to emulate the libjpeg v7 and v8 ABIs
126in libjpeg-turbo.  It should be noted, however, that this feature was added
127primarily so that applications that had already been compiled to use libjpeg
128v7+ could take advantage of accelerated baseline JPEG encoding/decoding
129without recompiling.  libjpeg-turbo does not claim to support all of the
130libjpeg v7+ features, nor to produce identical output to libjpeg v7+ in all
131cases (see below.)
132
133By passing an argument of `--with-jpeg7` or `--with-jpeg8` to `configure`, or
134an argument of `-DWITH_JPEG7=1` or `-DWITH_JPEG8=1` to `cmake`, you can build a
135version of libjpeg-turbo that emulates the libjpeg v7 or v8 ABI, so that
136programs that are built against libjpeg v7 or v8 can be run with libjpeg-turbo.
137The following section describes which libjpeg v7+ features are supported and
138which aren't.
139
140### Support for libjpeg v7 and v8 Features
141
142#### Fully supported
143
144- **libjpeg: IDCT scaling extensions in decompressor**<br>
145  libjpeg-turbo supports IDCT scaling with scaling factors of 1/8, 1/4, 3/8,
146  1/2, 5/8, 3/4, 7/8, 9/8, 5/4, 11/8, 3/2, 13/8, 7/4, 15/8, and 2/1 (only 1/4
147  and 1/2 are SIMD-accelerated.)
148
149- **libjpeg: Arithmetic coding**
150
151- **libjpeg: In-memory source and destination managers**<br>
152  See notes below.
153
154- **cjpeg: Separate quality settings for luminance and chrominance**<br>
155  Note that the libpjeg v7+ API was extended to accommodate this feature only
156  for convenience purposes.  It has always been possible to implement this
157  feature with libjpeg v6b (see rdswitch.c for an example.)
158
159- **cjpeg: 32-bit BMP support**
160
161- **cjpeg: `-rgb` option**
162
163- **jpegtran: Lossless cropping**
164
165- **jpegtran: `-perfect` option**
166
167- **jpegtran: Forcing width/height when performing lossless crop**
168
169- **rdjpgcom: `-raw` option**
170
171- **rdjpgcom: Locale awareness**
172
173
174#### Not supported
175
176NOTE:  As of this writing, extensive research has been conducted into the
177usefulness of DCT scaling as a means of data reduction and SmartScale as a
178means of quality improvement.  The reader is invited to peruse the research at
179<http://www.libjpeg-turbo.org/About/SmartScale> and draw his/her own conclusions,
180but it is the general belief of our project that these features have not
181demonstrated sufficient usefulness to justify inclusion in libjpeg-turbo.
182
183- **libjpeg: DCT scaling in compressor**<br>
184  `cinfo.scale_num` and `cinfo.scale_denom` are silently ignored.
185  There is no technical reason why DCT scaling could not be supported when
186  emulating the libjpeg v7+ API/ABI, but without the SmartScale extension (see
187  below), only scaling factors of 1/2, 8/15, 4/7, 8/13, 2/3, 8/11, 4/5, and
188  8/9 would be available, which is of limited usefulness.
189
190- **libjpeg: SmartScale**<br>
191  `cinfo.block_size` is silently ignored.
192  SmartScale is an extension to the JPEG format that allows for DCT block
193  sizes other than 8x8.  Providing support for this new format would be
194  feasible (particularly without full acceleration.)  However, until/unless
195  the format becomes either an official industry standard or, at minimum, an
196  accepted solution in the community, we are hesitant to implement it, as
197  there is no sense of whether or how it might change in the future.  It is
198  our belief that SmartScale has not demonstrated sufficient usefulness as a
199  lossless format nor as a means of quality enhancement, and thus our primary
200  interest in providing this feature would be as a means of supporting
201  additional DCT scaling factors.
202
203- **libjpeg: Fancy downsampling in compressor**<br>
204  `cinfo.do_fancy_downsampling` is silently ignored.
205  This requires the DCT scaling feature, which is not supported.
206
207- **jpegtran: Scaling**<br>
208  This requires both the DCT scaling and SmartScale features, which are not
209  supported.
210
211- **Lossless RGB JPEG files**<br>
212  This requires the SmartScale feature, which is not supported.
213
214### What About libjpeg v9?
215
216libjpeg v9 introduced yet another field to the JPEG compression structure
217(`color_transform`), thus making the ABI backward incompatible with that of
218libjpeg v8.  This new field was introduced solely for the purpose of supporting
219lossless SmartScale encoding.  Furthermore, there was actually no reason to
220extend the API in this manner, as the color transform could have just as easily
221been activated by way of a new JPEG colorspace constant, thus preserving
222backward ABI compatibility.
223
224Our research (see link above) has shown that lossless SmartScale does not
225generally accomplish anything that can't already be accomplished better with
226existing, standard lossless formats.  Therefore, at this time it is our belief
227that there is not sufficient technical justification for software projects to
228upgrade from libjpeg v8 to libjpeg v9, and thus there is not sufficient
229technical justification for us to emulate the libjpeg v9 ABI.
230
231In-Memory Source/Destination Managers
232-------------------------------------
233
234By default, libjpeg-turbo 1.3 and later includes the `jpeg_mem_src()` and
235`jpeg_mem_dest()` functions, even when not emulating the libjpeg v8 API/ABI.
236Previously, it was necessary to build libjpeg-turbo from source with libjpeg v8
237API/ABI emulation in order to use the in-memory source/destination managers,
238but several projects requested that those functions be included when emulating
239the libjpeg v6b API/ABI as well.  This allows the use of those functions by
240programs that need them, without breaking ABI compatibility for programs that
241don't, and it allows those functions to be provided in the "official"
242libjpeg-turbo binaries.
243
244Those who are concerned about maintaining strict conformance with the libjpeg
245v6b or v7 API can pass an argument of `--without-mem-srcdst` to `configure` or
246an argument of `-DWITH_MEM_SRCDST=0` to `cmake` prior to building
247libjpeg-turbo.  This will restore the pre-1.3 behavior, in which
248`jpeg_mem_src()` and `jpeg_mem_dest()` are only included when emulating the
249libjpeg v8 API/ABI.
250
251On Un*x systems, including the in-memory source/destination managers changes
252the dynamic library version from 62.1.0 to 62.2.0 if using libjpeg v6b API/ABI
253emulation and from 7.1.0 to 7.2.0 if using libjpeg v7 API/ABI emulation.
254
255Note that, on most Un*x systems, the dynamic linker will not look for a
256function in a library until that function is actually used.  Thus, if a program
257is built against libjpeg-turbo 1.3+ and uses `jpeg_mem_src()` or
258`jpeg_mem_dest()`, that program will not fail if run against an older version
259of libjpeg-turbo or against libjpeg v7- until the program actually tries to
260call `jpeg_mem_src()` or `jpeg_mem_dest()`.  Such is not the case on Windows.
261If a program is built against the libjpeg-turbo 1.3+ DLL and uses
262`jpeg_mem_src()` or `jpeg_mem_dest()`, then it must use the libjpeg-turbo 1.3+
263DLL at run time.
264
265Both cjpeg and djpeg have been extended to allow testing the in-memory
266source/destination manager functions.  See their respective man pages for more
267details.
268
269
270Mathematical Compatibility
271==========================
272
273For the most part, libjpeg-turbo should produce identical output to libjpeg
274v6b.  The one exception to this is when using the floating point DCT/IDCT, in
275which case the outputs of libjpeg v6b and libjpeg-turbo can differ for the
276following reasons:
277
278- The SSE/SSE2 floating point DCT implementation in libjpeg-turbo is ever so
279  slightly more accurate than the implementation in libjpeg v6b, but not by
280  any amount perceptible to human vision (generally in the range of 0.01 to
281  0.08 dB gain in PNSR.)
282
283- When not using the SIMD extensions, libjpeg-turbo uses the more accurate
284  (and slightly faster) floating point IDCT algorithm introduced in libjpeg
285  v8a as opposed to the algorithm used in libjpeg v6b.  It should be noted,
286  however, that this algorithm basically brings the accuracy of the floating
287  point IDCT in line with the accuracy of the slow integer IDCT.  The floating
288  point DCT/IDCT algorithms are mainly a legacy feature, and they do not
289  produce significantly more accuracy than the slow integer algorithms (to put
290  numbers on this, the typical difference in PNSR between the two algorithms
291  is less than 0.10 dB, whereas changing the quality level by 1 in the upper
292  range of the quality scale is typically more like a 1.0 dB difference.)
293
294- If the floating point algorithms in libjpeg-turbo are not implemented using
295  SIMD instructions on a particular platform, then the accuracy of the
296  floating point DCT/IDCT can depend on the compiler settings.
297
298While libjpeg-turbo does emulate the libjpeg v8 API/ABI, under the hood it is
299still using the same algorithms as libjpeg v6b, so there are several specific
300cases in which libjpeg-turbo cannot be expected to produce the same output as
301libjpeg v8:
302
303- When decompressing using scaling factors of 1/2 and 1/4, because libjpeg v8
304  implements those scaling algorithms differently than libjpeg v6b does, and
305  libjpeg-turbo's SIMD extensions are based on the libjpeg v6b behavior.
306
307- When using chrominance subsampling, because libjpeg v8 implements this
308  with its DCT/IDCT scaling algorithms rather than with a separate
309  downsampling/upsampling algorithm.  In our testing, the subsampled/upsampled
310  output of libjpeg v8 is less accurate than that of libjpeg v6b for this
311  reason.
312
313- When decompressing using a scaling factor > 1 and merged (AKA "non-fancy" or
314  "non-smooth") chrominance upsampling, because libjpeg v8 does not support
315  merged upsampling with scaling factors > 1.
316
317
318Performance Pitfalls
319====================
320
321Restart Markers
322---------------
323
324The optimized Huffman decoder in libjpeg-turbo does not handle restart markers
325in a way that makes the rest of the libjpeg infrastructure happy, so it is
326necessary to use the slow Huffman decoder when decompressing a JPEG image that
327has restart markers.  This can cause the decompression performance to drop by
328as much as 20%, but the performance will still be much greater than that of
329libjpeg.  Many consumer packages, such as PhotoShop, use restart markers when
330generating JPEG images, so images generated by those programs will experience
331this issue.
332
333Fast Integer Forward DCT at High Quality Levels
334-----------------------------------------------
335
336The algorithm used by the SIMD-accelerated quantization function cannot produce
337correct results whenever the fast integer forward DCT is used along with a JPEG
338quality of 98-100.  Thus, libjpeg-turbo must use the non-SIMD quantization
339function in those cases.  This causes performance to drop by as much as 40%.
340It is therefore strongly advised that you use the slow integer forward DCT
341whenever encoding images with a JPEG quality of 98 or higher.
342