1 ******************************************************************************* 2 ** Background 3 ******************************************************************************* 4 5 libjpeg-turbo is a JPEG image codec that uses SIMD instructions (MMX, SSE2, 6 NEON) to accelerate baseline JPEG compression and decompression on x86, x86-64, 7 and ARM systems. On such systems, libjpeg-turbo is generally 2-4x as fast as 8 libjpeg, all else being equal. On other types of systems, libjpeg-turbo can 9 still outperform libjpeg by a significant amount, by virtue of its 10 highly-optimized Huffman coding routines. In many cases, the performance of 11 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 ******************************************************************************* 28 ** License 29 ******************************************************************************* 30 31 libjpeg-turbo is covered by three compatible BSD-style open source licenses. 32 Refer to LICENSE.txt for a roll-up of license terms. 33 34 35 ******************************************************************************* 36 ** Using libjpeg-turbo 37 ******************************************************************************* 38 39 libjpeg-turbo includes two APIs that can be used to compress and decompress 40 JPEG images: 41 42 TurboJPEG API: This API provides an easy-to-use interface for compressing 43 and decompressing JPEG images in memory. It also provides some functionality 44 that would not be straightforward to achieve using the underlying libjpeg 45 API, such as generating planar YUV images and performing multiple 46 simultaneous lossless transforms on an image. The Java interface for 47 libjpeg-turbo is written on top of the TurboJPEG API. 48 49 libjpeg API: This is the de facto industry-standard API for compressing and 50 decompressing JPEG images. It is more difficult to use than the TurboJPEG 51 API but also more powerful. The libjpeg API implementation in libjpeg-turbo 52 is both API/ABI-compatible and mathematically compatible with libjpeg v6b. 53 It can also optionally be configured to be API/ABI-compatible with libjpeg v7 54 and v8 (see below.) 55 56 There is no significant performance advantage to either API when both are used 57 to perform similar operations. 58 59 ===================== 60 Colorspace Extensions 61 ===================== 62 63 libjpeg-turbo includes extensions that allow JPEG images to be compressed 64 directly from (and decompressed directly to) buffers that use BGR, BGRX, 65 RGBX, XBGR, and XRGB pixel ordering. This is implemented with ten new 66 colorspace constants: 67 68 JCS_EXT_RGB /* red/green/blue */ 69 JCS_EXT_RGBX /* red/green/blue/x */ 70 JCS_EXT_BGR /* blue/green/red */ 71 JCS_EXT_BGRX /* blue/green/red/x */ 72 JCS_EXT_XBGR /* x/blue/green/red */ 73 JCS_EXT_XRGB /* x/red/green/blue */ 74 JCS_EXT_RGBA /* red/green/blue/alpha */ 75 JCS_EXT_BGRA /* blue/green/red/alpha */ 76 JCS_EXT_ABGR /* alpha/blue/green/red */ 77 JCS_EXT_ARGB /* alpha/red/green/blue */ 78 79 Setting cinfo.in_color_space (compression) or cinfo.out_color_space 80 (decompression) to one of these values will cause libjpeg-turbo to read the 81 red, green, and blue values from (or write them to) the appropriate position in 82 the pixel when compressing from/decompressing to an RGB buffer. 83 84 Your application can check for the existence of these extensions at compile 85 time with: 86 87 #ifdef JCS_EXTENSIONS 88 89 At run time, attempting to use these extensions with a libjpeg implementation 90 that does not support them will result in a "Bogus input colorspace" error. 91 Applications can trap this error in order to test whether run-time support is 92 available for the colorspace extensions. 93 94 When using the RGBX, BGRX, XBGR, and XRGB colorspaces during decompression, the 95 X byte is undefined, and in order to ensure the best performance, libjpeg-turbo 96 can set that byte to whatever value it wishes. If an application expects the X 97 byte to be used as an alpha channel, then it should specify JCS_EXT_RGBA, 98 JCS_EXT_BGRA, JCS_EXT_ABGR, or JCS_EXT_ARGB. When these colorspace constants 99 are used, the X byte is guaranteed to be 0xFF, which is interpreted as opaque. 100 101 Your application can check for the existence of the alpha channel colorspace 102 extensions at compile time with: 103 104 #ifdef JCS_ALPHA_EXTENSIONS 105 106 jcstest.c, located in the libjpeg-turbo source tree, demonstrates how to check 107 for the existence of the colorspace extensions at compile time and run time. 108 109 =================================== 110 libjpeg v7 and v8 API/ABI Emulation 111 =================================== 112 113 With libjpeg v7 and v8, new features were added that necessitated extending the 114 compression and decompression structures. Unfortunately, due to the exposed 115 nature of those structures, extending them also necessitated breaking backward 116 ABI compatibility with previous libjpeg releases. Thus, programs that were 117 built to use libjpeg v7 or v8 did not work with libjpeg-turbo, since it is 118 based on the libjpeg v6b code base. Although libjpeg v7 and v8 are not 119 as widely used as v6b, enough programs (including a few Linux distros) made 120 the switch that there was a demand to emulate the libjpeg v7 and v8 ABIs 121 in libjpeg-turbo. It should be noted, however, that this feature was added 122 primarily so that applications that had already been compiled to use libjpeg 123 v7+ could take advantage of accelerated baseline JPEG encoding/decoding 124 without recompiling. libjpeg-turbo does not claim to support all of the 125 libjpeg v7+ features, nor to produce identical output to libjpeg v7+ in all 126 cases (see below.) 127 128 By passing an argument of --with-jpeg7 or --with-jpeg8 to configure, or an 129 argument of -DWITH_JPEG7=1 or -DWITH_JPEG8=1 to cmake, you can build a version 130 of libjpeg-turbo that emulates the libjpeg v7 or v8 ABI, so that programs 131 that are built against libjpeg v7 or v8 can be run with libjpeg-turbo. The 132 following section describes which libjpeg v7+ features are supported and which 133 aren't. 134 135 Support for libjpeg v7 and v8 Features: 136 --------------------------------------- 137 138 Fully supported: 139 140 -- libjpeg: IDCT scaling extensions in decompressor 141 libjpeg-turbo supports IDCT scaling with scaling factors of 1/8, 1/4, 3/8, 142 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 1/4 143 and 1/2 are SIMD-accelerated.) 144 145 -- libjpeg: arithmetic coding 146 147 -- libjpeg: In-memory source and destination managers 148 See notes below. 149 150 -- cjpeg: Separate quality settings for luminance and chrominance 151 Note that the libpjeg v7+ API was extended to accommodate this feature only 152 for convenience purposes. It has always been possible to implement this 153 feature with libjpeg v6b (see rdswitch.c for an example.) 154 155 -- cjpeg: 32-bit BMP support 156 157 -- cjpeg: -rgb option 158 159 -- jpegtran: lossless cropping 160 161 -- jpegtran: -perfect option 162 163 -- jpegtran: forcing width/height when performing lossless crop 164 165 -- rdjpgcom: -raw option 166 167 -- rdjpgcom: locale awareness 168 169 170 Not supported: 171 172 NOTE: As of this writing, extensive research has been conducted into the 173 usefulness of DCT scaling as a means of data reduction and SmartScale as a 174 means of quality improvement. The reader is invited to peruse the research at 175 http://www.libjpeg-turbo.org/About/SmartScale and draw his/her own conclusions, 176 but it is the general belief of our project that these features have not 177 demonstrated sufficient usefulness to justify inclusion in libjpeg-turbo. 178 179 -- libjpeg: DCT scaling in compressor 180 cinfo.scale_num and cinfo.scale_denom are silently ignored. 181 There is no technical reason why DCT scaling could not be supported when 182 emulating the libjpeg v7+ API/ABI, but without the SmartScale extension (see 183 below), only scaling factors of 1/2, 8/15, 4/7, 8/13, 2/3, 8/11, 4/5, and 184 8/9 would be available, which is of limited usefulness. 185 186 -- libjpeg: SmartScale 187 cinfo.block_size is silently ignored. 188 SmartScale is an extension to the JPEG format that allows for DCT block 189 sizes other than 8x8. Providing support for this new format would be 190 feasible (particularly without full acceleration.) However, until/unless 191 the format becomes either an official industry standard or, at minimum, an 192 accepted solution in the community, we are hesitant to implement it, as 193 there is no sense of whether or how it might change in the future. It is 194 our belief that SmartScale has not demonstrated sufficient usefulness as a 195 lossless format nor as a means of quality enhancement, and thus, our primary 196 interest in providing this feature would be as a means of supporting 197 additional DCT scaling factors. 198 199 -- libjpeg: Fancy downsampling in compressor 200 cinfo.do_fancy_downsampling is silently ignored. 201 This requires the DCT scaling feature, which is not supported. 202 203 -- jpegtran: Scaling 204 This requires both the DCT scaling and SmartScale features, which are not 205 supported. 206 207 -- Lossless RGB JPEG files 208 This requires the SmartScale feature, which is not supported. 209 210 What About libjpeg v9? 211 ---------------------- 212 213 libjpeg v9 introduced yet another field to the JPEG compression structure 214 (color_transform), thus making the ABI backward incompatible with that of 215 libjpeg v8. This new field was introduced solely for the purpose of supporting 216 lossless SmartScale encoding. Further, there was actually no reason to extend 217 the API in this manner, as the color transform could have just as easily been 218 activated by way of a new JPEG colorspace constant, thus preserving backward 219 ABI compatibility. 220 221 Our research (see link above) has shown that lossless SmartScale does not 222 generally accomplish anything that can't already be accomplished better with 223 existing, standard lossless formats. Thus, at this time, it is our belief that 224 there is not sufficient technical justification for software to upgrade from 225 libjpeg v8 to libjpeg v9, and therefore, not sufficient technical justification 226 for us to emulate the libjpeg v9 ABI. 227 228 ===================================== 229 In-Memory Source/Destination Managers 230 ===================================== 231 232 By default, libjpeg-turbo 1.3 and later includes the jpeg_mem_src() and 233 jpeg_mem_dest() functions, even when not emulating the libjpeg v8 API/ABI. 234 Previously, it was necessary to build libjpeg-turbo from source with libjpeg v8 235 API/ABI emulation in order to use the in-memory source/destination managers, 236 but several projects requested that those functions be included when emulating 237 the libjpeg v6b API/ABI as well. This allows the use of those functions by 238 programs that need them without breaking ABI compatibility for programs that 239 don't, and it allows those functions to be provided in the "official" 240 libjpeg-turbo binaries. 241 242 Those who are concerned about maintaining strict conformance with the libjpeg 243 v6b or v7 API can pass an argument of --without-mem-srcdst to configure or 244 an argument of -DWITH_MEM_SRCDST=0 to CMake prior to building libjpeg-turbo. 245 This will restore the pre-1.3 behavior, in which jpeg_mem_src() and 246 jpeg_mem_dest() are only included when emulating the libjpeg v8 API/ABI. 247 248 On Un*x systems, including the in-memory source/destination managers changes 249 the dynamic library version from 62.0.0 to 62.1.0 if using libjpeg v6b API/ABI 250 emulation and from 7.0.0 to 7.1.0 if using libjpeg v7 API/ABI emulation. 251 252 Note that, on most Un*x systems, the dynamic linker will not look for a 253 function in a library until that function is actually used. Thus, if a program 254 is built against libjpeg-turbo 1.3+ and uses jpeg_mem_src() or jpeg_mem_dest(), 255 that program will not fail if run against an older version of libjpeg-turbo or 256 against libjpeg v7- until the program actually tries to call jpeg_mem_src() or 257 jpeg_mem_dest(). Such is not the case on Windows. If a program is built 258 against the libjpeg-turbo 1.3+ DLL and uses jpeg_mem_src() or jpeg_mem_dest(), 259 then it must use the libjpeg-turbo 1.3+ DLL at run time. 260 261 Both cjpeg and djpeg have been extended to allow testing the in-memory 262 source/destination manager functions. See their respective man pages for more 263 details. 264 265 266 ******************************************************************************* 267 ** Mathematical Compatibility 268 ******************************************************************************* 269 270 For the most part, libjpeg-turbo should produce identical output to libjpeg 271 v6b. The one exception to this is when using the floating point DCT/IDCT, in 272 which case the outputs of libjpeg v6b and libjpeg-turbo can differ for the 273 following reasons: 274 275 -- The SSE/SSE2 floating point DCT implementation in libjpeg-turbo is ever so 276 slightly more accurate than the implementation in libjpeg v6b, but not by 277 any amount perceptible to human vision (generally in the range of 0.01 to 278 0.08 dB gain in PNSR.) 279 -- When not using the SIMD extensions, libjpeg-turbo uses the more accurate 280 (and slightly faster) floating point IDCT algorithm introduced in libjpeg 281 v8a as opposed to the algorithm used in libjpeg v6b. It should be noted, 282 however, that this algorithm basically brings the accuracy of the floating 283 point IDCT in line with the accuracy of the slow integer IDCT. The floating 284 point DCT/IDCT algorithms are mainly a legacy feature, and they do not 285 produce significantly more accuracy than the slow integer algorithms (to put 286 numbers on this, the typical difference in PNSR between the two algorithms 287 is less than 0.10 dB, whereas changing the quality level by 1 in the upper 288 range of the quality scale is typically more like a 1.0 dB difference.) 289 -- If the floating point algorithms in libjpeg-turbo are not implemented using 290 SIMD instructions on a particular platform, then the accuracy of the 291 floating point DCT/IDCT can depend on the compiler settings. 292 293 While libjpeg-turbo does emulate the libjpeg v8 API/ABI, under the hood, it is 294 still using the same algorithms as libjpeg v6b, so there are several specific 295 cases in which libjpeg-turbo cannot be expected to produce the same output as 296 libjpeg v8: 297 298 -- When decompressing using scaling factors of 1/2 and 1/4, because libjpeg v8 299 implements those scaling algorithms differently than libjpeg v6b does, and 300 libjpeg-turbo's SIMD extensions are based on the libjpeg v6b behavior. 301 302 -- When using chrominance subsampling, because libjpeg v8 implements this 303 with its DCT/IDCT scaling algorithms rather than with a separate 304 downsampling/upsampling algorithm. In our testing, the subsampled/upsampled 305 output of libjpeg v8 is less accurate than that of libjpeg v6b for this 306 reason. 307 308 -- When decompressing using a scaling factor > 1 and merged (AKA "non-fancy" or 309 "non-smooth") chrominance upsampling, because libjpeg v8 does not support 310 merged upsampling with scaling factors > 1. 311 312 313 ******************************************************************************* 314 ** Performance Pitfalls 315 ******************************************************************************* 316 317 =============== 318 Restart Markers 319 =============== 320 321 The optimized Huffman decoder in libjpeg-turbo does not handle restart markers 322 in a way that makes the rest of the libjpeg infrastructure happy, so it is 323 necessary to use the slow Huffman decoder when decompressing a JPEG image that 324 has restart markers. This can cause the decompression performance to drop by 325 as much as 20%, but the performance will still be much greater than that of 326 libjpeg. Many consumer packages, such as PhotoShop, use restart markers when 327 generating JPEG images, so images generated by those programs will experience 328 this issue. 329 330 =============================================== 331 Fast Integer Forward DCT at High Quality Levels 332 =============================================== 333 334 The algorithm used by the SIMD-accelerated quantization function cannot produce 335 correct results whenever the fast integer forward DCT is used along with a JPEG 336 quality of 98-100. Thus, libjpeg-turbo must use the non-SIMD quantization 337 function in those cases. This causes performance to drop by as much as 40%. 338 It is therefore strongly advised that you use the slow integer forward DCT 339 whenever encoding images with a JPEG quality of 98 or higher. 340
