1 Background 2 ========== 3 4 libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2, 5 AVX2, NEON, AltiVec) to accelerate baseline JPEG compression and decompression 6 on x86, x86-64, ARM, and PowerPC systems, as well as progressive JPEG 7 compression on x86 and x86-64 systems. On such systems, libjpeg-turbo is 8 generally 2-6x as fast as libjpeg, all else being equal. On other types of 9 systems, libjpeg-turbo can still outperform libjpeg by a significant amount, by 10 virtue of its highly-optimized Huffman coding routines. In many cases, the 11 performance of libjpeg-turbo rivals that of proprietary high-speed JPEG codecs. 12 13 libjpeg-turbo implements both the traditional libjpeg API as well as the less 14 powerful but more straightforward TurboJPEG API. libjpeg-turbo also features 15 colorspace extensions that allow it to compress from/decompress to 32-bit and 16 big-endian pixel buffers (RGBX, XBGR, etc.), as well as a full-featured Java 17 interface. 18 19 libjpeg-turbo was originally based on libjpeg/SIMD, an MMX-accelerated 20 derivative of libjpeg v6b developed by Miyasaka Masaru. The TigerVNC and 21 VirtualGL projects made numerous enhancements to the codec in 2009, and in 22 early 2010, libjpeg-turbo spun off into an independent project, with the goal 23 of making high-speed JPEG compression/decompression technology available to a 24 broader range of users and developers. 25 26 27 License 28 ======= 29 30 libjpeg-turbo is covered by three compatible BSD-style open source licenses. 31 Refer to [LICENSE.md](LICENSE.md) for a roll-up of license terms. 32 33 34 Building libjpeg-turbo 35 ====================== 36 37 Refer to [BUILDING.md](BUILDING.md) for complete instructions. 38 39 40 Using libjpeg-turbo 41 =================== 42 43 libjpeg-turbo includes two APIs that can be used to compress and decompress 44 JPEG images: 45 46 - **TurboJPEG API**<br> 47 This API provides an easy-to-use interface for compressing and decompressing 48 JPEG images in memory. It also provides some functionality that would not be 49 straightforward to achieve using the underlying libjpeg API, such as 50 generating planar YUV images and performing multiple simultaneous lossless 51 transforms on an image. The Java interface for libjpeg-turbo is written on 52 top of the TurboJPEG API. The TurboJPEG API is recommended for first-time 53 users of libjpeg-turbo. Refer to [tjexample.c](tjexample.c) and 54 [TJExample.java](java/TJExample.java) for examples of its usage and to 55 <http://libjpeg-turbo.org/Documentation/Documentation> for API documentation. 56 57 - **libjpeg API**<br> 58 This is the de facto industry-standard API for compressing and decompressing 59 JPEG images. It is more difficult to use than the TurboJPEG API but also 60 more powerful. The libjpeg API implementation in libjpeg-turbo is both 61 API/ABI-compatible and mathematically compatible with libjpeg v6b. It can 62 also optionally be configured to be API/ABI-compatible with libjpeg v7 and v8 63 (see below.) Refer to [cjpeg.c](cjpeg.c) and [djpeg.c](djpeg.c) for examples 64 of its usage and to [libjpeg.txt](libjpeg.txt) for API documentation. 65 66 There is no significant performance advantage to either API when both are used 67 to perform similar operations. 68 69 Colorspace Extensions 70 --------------------- 71 72 libjpeg-turbo includes extensions that allow JPEG images to be compressed 73 directly from (and decompressed directly to) buffers that use BGR, BGRX, 74 RGBX, XBGR, and XRGB pixel ordering. This is implemented with ten new 75 colorspace constants: 76 77 JCS_EXT_RGB /* red/green/blue */ 78 JCS_EXT_RGBX /* red/green/blue/x */ 79 JCS_EXT_BGR /* blue/green/red */ 80 JCS_EXT_BGRX /* blue/green/red/x */ 81 JCS_EXT_XBGR /* x/blue/green/red */ 82 JCS_EXT_XRGB /* x/red/green/blue */ 83 JCS_EXT_RGBA /* red/green/blue/alpha */ 84 JCS_EXT_BGRA /* blue/green/red/alpha */ 85 JCS_EXT_ABGR /* alpha/blue/green/red */ 86 JCS_EXT_ARGB /* alpha/red/green/blue */ 87 88 Setting `cinfo.in_color_space` (compression) or `cinfo.out_color_space` 89 (decompression) to one of these values will cause libjpeg-turbo to read the 90 red, green, and blue values from (or write them to) the appropriate position in 91 the pixel when compressing from/decompressing to an RGB buffer. 92 93 Your application can check for the existence of these extensions at compile 94 time with: 95 96 #ifdef JCS_EXTENSIONS 97 98 At run time, attempting to use these extensions with a libjpeg implementation 99 that does not support them will result in a "Bogus input colorspace" error. 100 Applications can trap this error in order to test whether run-time support is 101 available for the colorspace extensions. 102 103 When using the RGBX, BGRX, XBGR, and XRGB colorspaces during decompression, the 104 X byte is undefined, and in order to ensure the best performance, libjpeg-turbo 105 can set that byte to whatever value it wishes. If an application expects the X 106 byte to be used as an alpha channel, then it should specify `JCS_EXT_RGBA`, 107 `JCS_EXT_BGRA`, `JCS_EXT_ABGR`, or `JCS_EXT_ARGB`. When these colorspace 108 constants are used, the X byte is guaranteed to be 0xFF, which is interpreted 109 as opaque. 110 111 Your application can check for the existence of the alpha channel colorspace 112 extensions at compile time with: 113 114 #ifdef JCS_ALPHA_EXTENSIONS 115 116 [jcstest.c](jcstest.c), located in the libjpeg-turbo source tree, demonstrates 117 how to check for the existence of the colorspace extensions at compile time and 118 run time. 119 120 libjpeg v7 and v8 API/ABI Emulation 121 ----------------------------------- 122 123 With libjpeg v7 and v8, new features were added that necessitated extending the 124 compression and decompression structures. Unfortunately, due to the exposed 125 nature of those structures, extending them also necessitated breaking backward 126 ABI compatibility with previous libjpeg releases. Thus, programs that were 127 built to use libjpeg v7 or v8 did not work with libjpeg-turbo, since it is 128 based on the libjpeg v6b code base. Although libjpeg v7 and v8 are not 129 as widely used as v6b, enough programs (including a few Linux distros) made 130 the switch that there was a demand to emulate the libjpeg v7 and v8 ABIs 131 in libjpeg-turbo. It should be noted, however, that this feature was added 132 primarily so that applications that had already been compiled to use libjpeg 133 v7+ could take advantage of accelerated baseline JPEG encoding/decoding 134 without recompiling. libjpeg-turbo does not claim to support all of the 135 libjpeg v7+ features, nor to produce identical output to libjpeg v7+ in all 136 cases (see below.) 137 138 By passing an argument of `--with-jpeg7` or `--with-jpeg8` to `configure`, or 139 an argument of `-DWITH_JPEG7=1` or `-DWITH_JPEG8=1` to `cmake`, you can build a 140 version of libjpeg-turbo that emulates the libjpeg v7 or v8 ABI, so that 141 programs that are built against libjpeg v7 or v8 can be run with libjpeg-turbo. 142 The following section describes which libjpeg v7+ features are supported and 143 which aren't. 144 145 ### Support for libjpeg v7 and v8 Features 146 147 #### Fully supported 148 149 - **libjpeg: IDCT scaling extensions in decompressor**<br> 150 libjpeg-turbo supports IDCT scaling with scaling factors of 1/8, 1/4, 3/8, 151 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 1/4 152 and 1/2 are SIMD-accelerated.) 153 154 - **libjpeg: Arithmetic coding** 155 156 - **libjpeg: In-memory source and destination managers**<br> 157 See notes below. 158 159 - **cjpeg: Separate quality settings for luminance and chrominance**<br> 160 Note that the libpjeg v7+ API was extended to accommodate this feature only 161 for convenience purposes. It has always been possible to implement this 162 feature with libjpeg v6b (see rdswitch.c for an example.) 163 164 - **cjpeg: 32-bit BMP support** 165 166 - **cjpeg: `-rgb` option** 167 168 - **jpegtran: Lossless cropping** 169 170 - **jpegtran: `-perfect` option** 171 172 - **jpegtran: Forcing width/height when performing lossless crop** 173 174 - **rdjpgcom: `-raw` option** 175 176 - **rdjpgcom: Locale awareness** 177 178 179 #### Not supported 180 181 NOTE: As of this writing, extensive research has been conducted into the 182 usefulness of DCT scaling as a means of data reduction and SmartScale as a 183 means of quality improvement. The reader is invited to peruse the research at 184 <http://www.libjpeg-turbo.org/About/SmartScale> and draw his/her own conclusions, 185 but it is the general belief of our project that these features have not 186 demonstrated sufficient usefulness to justify inclusion in libjpeg-turbo. 187 188 - **libjpeg: DCT scaling in compressor**<br> 189 `cinfo.scale_num` and `cinfo.scale_denom` are silently ignored. 190 There is no technical reason why DCT scaling could not be supported when 191 emulating the libjpeg v7+ API/ABI, but without the SmartScale extension (see 192 below), only scaling factors of 1/2, 8/15, 4/7, 8/13, 2/3, 8/11, 4/5, and 193 8/9 would be available, which is of limited usefulness. 194 195 - **libjpeg: SmartScale**<br> 196 `cinfo.block_size` is silently ignored. 197 SmartScale is an extension to the JPEG format that allows for DCT block 198 sizes other than 8x8. Providing support for this new format would be 199 feasible (particularly without full acceleration.) However, until/unless 200 the format becomes either an official industry standard or, at minimum, an 201 accepted solution in the community, we are hesitant to implement it, as 202 there is no sense of whether or how it might change in the future. It is 203 our belief that SmartScale has not demonstrated sufficient usefulness as a 204 lossless format nor as a means of quality enhancement, and thus our primary 205 interest in providing this feature would be as a means of supporting 206 additional DCT scaling factors. 207 208 - **libjpeg: Fancy downsampling in compressor**<br> 209 `cinfo.do_fancy_downsampling` is silently ignored. 210 This requires the DCT scaling feature, which is not supported. 211 212 - **jpegtran: Scaling**<br> 213 This requires both the DCT scaling and SmartScale features, which are not 214 supported. 215 216 - **Lossless RGB JPEG files**<br> 217 This requires the SmartScale feature, which is not supported. 218 219 ### What About libjpeg v9? 220 221 libjpeg v9 introduced yet another field to the JPEG compression structure 222 (`color_transform`), thus making the ABI backward incompatible with that of 223 libjpeg v8. This new field was introduced solely for the purpose of supporting 224 lossless SmartScale encoding. Furthermore, there was actually no reason to 225 extend the API in this manner, as the color transform could have just as easily 226 been activated by way of a new JPEG colorspace constant, thus preserving 227 backward ABI compatibility. 228 229 Our research (see link above) has shown that lossless SmartScale does not 230 generally accomplish anything that can't already be accomplished better with 231 existing, standard lossless formats. Therefore, at this time it is our belief 232 that there is not sufficient technical justification for software projects to 233 upgrade from libjpeg v8 to libjpeg v9, and thus there is not sufficient 234 technical justification for us to emulate the libjpeg v9 ABI. 235 236 In-Memory Source/Destination Managers 237 ------------------------------------- 238 239 By default, libjpeg-turbo 1.3 and later includes the `jpeg_mem_src()` and 240 `jpeg_mem_dest()` functions, even when not emulating the libjpeg v8 API/ABI. 241 Previously, it was necessary to build libjpeg-turbo from source with libjpeg v8 242 API/ABI emulation in order to use the in-memory source/destination managers, 243 but several projects requested that those functions be included when emulating 244 the libjpeg v6b API/ABI as well. This allows the use of those functions by 245 programs that need them, without breaking ABI compatibility for programs that 246 don't, and it allows those functions to be provided in the "official" 247 libjpeg-turbo binaries. 248 249 Those who are concerned about maintaining strict conformance with the libjpeg 250 v6b or v7 API can pass an argument of `--without-mem-srcdst` to `configure` or 251 an argument of `-DWITH_MEM_SRCDST=0` to `cmake` prior to building 252 libjpeg-turbo. This will restore the pre-1.3 behavior, in which 253 `jpeg_mem_src()` and `jpeg_mem_dest()` are only included when emulating the 254 libjpeg v8 API/ABI. 255 256 On Un*x systems, including the in-memory source/destination managers changes 257 the dynamic library version from 62.1.0 to 62.2.0 if using libjpeg v6b API/ABI 258 emulation and from 7.1.0 to 7.2.0 if using libjpeg v7 API/ABI emulation. 259 260 Note that, on most Un*x systems, the dynamic linker will not look for a 261 function in a library until that function is actually used. Thus, if a program 262 is built against libjpeg-turbo 1.3+ and uses `jpeg_mem_src()` or 263 `jpeg_mem_dest()`, that program will not fail if run against an older version 264 of libjpeg-turbo or against libjpeg v7- until the program actually tries to 265 call `jpeg_mem_src()` or `jpeg_mem_dest()`. Such is not the case on Windows. 266 If a program is built against the libjpeg-turbo 1.3+ DLL and uses 267 `jpeg_mem_src()` or `jpeg_mem_dest()`, then it must use the libjpeg-turbo 1.3+ 268 DLL at run time. 269 270 Both cjpeg and djpeg have been extended to allow testing the in-memory 271 source/destination manager functions. See their respective man pages for more 272 details. 273 274 275 Mathematical Compatibility 276 ========================== 277 278 For the most part, libjpeg-turbo should produce identical output to libjpeg 279 v6b. The one exception to this is when using the floating point DCT/IDCT, in 280 which case the outputs of libjpeg v6b and libjpeg-turbo can differ for the 281 following reasons: 282 283 - The SSE/SSE2 floating point DCT implementation in libjpeg-turbo is ever so 284 slightly more accurate than the implementation in libjpeg v6b, but not by 285 any amount perceptible to human vision (generally in the range of 0.01 to 286 0.08 dB gain in PNSR.) 287 288 - When not using the SIMD extensions, libjpeg-turbo uses the more accurate 289 (and slightly faster) floating point IDCT algorithm introduced in libjpeg 290 v8a as opposed to the algorithm used in libjpeg v6b. It should be noted, 291 however, that this algorithm basically brings the accuracy of the floating 292 point IDCT in line with the accuracy of the slow integer IDCT. The floating 293 point DCT/IDCT algorithms are mainly a legacy feature, and they do not 294 produce significantly more accuracy than the slow integer algorithms (to put 295 numbers on this, the typical difference in PNSR between the two algorithms 296 is less than 0.10 dB, whereas changing the quality level by 1 in the upper 297 range of the quality scale is typically more like a 1.0 dB difference.) 298 299 - If the floating point algorithms in libjpeg-turbo are not implemented using 300 SIMD instructions on a particular platform, then the accuracy of the 301 floating point DCT/IDCT can depend on the compiler settings. 302 303 While libjpeg-turbo does emulate the libjpeg v8 API/ABI, under the hood it is 304 still using the same algorithms as libjpeg v6b, so there are several specific 305 cases in which libjpeg-turbo cannot be expected to produce the same output as 306 libjpeg v8: 307 308 - When decompressing using scaling factors of 1/2 and 1/4, because libjpeg v8 309 implements those scaling algorithms differently than libjpeg v6b does, and 310 libjpeg-turbo's SIMD extensions are based on the libjpeg v6b behavior. 311 312 - When using chrominance subsampling, because libjpeg v8 implements this 313 with its DCT/IDCT scaling algorithms rather than with a separate 314 downsampling/upsampling algorithm. In our testing, the subsampled/upsampled 315 output of libjpeg v8 is less accurate than that of libjpeg v6b for this 316 reason. 317 318 - When decompressing using a scaling factor > 1 and merged (AKA "non-fancy" or 319 "non-smooth") chrominance upsampling, because libjpeg v8 does not support 320 merged upsampling with scaling factors > 1. 321 322 323 Performance Pitfalls 324 ==================== 325 326 Restart Markers 327 --------------- 328 329 The optimized Huffman decoder in libjpeg-turbo does not handle restart markers 330 in a way that makes the rest of the libjpeg infrastructure happy, so it is 331 necessary to use the slow Huffman decoder when decompressing a JPEG image that 332 has restart markers. This can cause the decompression performance to drop by 333 as much as 20%, but the performance will still be much greater than that of 334 libjpeg. Many consumer packages, such as PhotoShop, use restart markers when 335 generating JPEG images, so images generated by those programs will experience 336 this issue. 337 338 Fast Integer Forward DCT at High Quality Levels 339 ----------------------------------------------- 340 341 The algorithm used by the SIMD-accelerated quantization function cannot produce 342 correct results whenever the fast integer forward DCT is used along with a JPEG 343 quality of 98-100. Thus, libjpeg-turbo must use the non-SIMD quantization 344 function in those cases. This causes performance to drop by as much as 40%. 345 It is therefore strongly advised that you use the slow integer forward DCT 346 whenever encoding images with a JPEG quality of 98 or higher. 347