Bump version to 2.1.

more logical flow of control if trellis_quant enabled

CMakeLists.txt

@@ -5,7 +5,7 @@
 cmake_minimum_required(VERSION 2.6)
 
 project(libmozjpeg C)
-set(VERSION 2.0pre)
+set(VERSION 2.1)
 
 if(MINGW OR CYGWIN)
   execute_process(COMMAND "date" "+%Y%m%d" OUTPUT_VARIABLE BUILD)
@@ -264,7 +264,7 @@ set_property(TARGET jpegtran-static PROPERTY COMPILE_FLAGS "-DUSE_SETMODE")
 
 add_executable(rdjpgcom rdjpgcom.c)
 
-add_executable(wrjpgcom rdjpgcom.c)
+add_executable(wrjpgcom wrjpgcom.c)
 
 
 #

ChangeLog.txt

@@ -23,6 +23,14 @@ and is out of scope for a codec library.
 [5] The TurboJPEG API can now be used to compress JPEG images from YUV planar
 source images.
 
+[7] Improved the accuracy and performance of the non-SIMD implementation of the
+floating point inverse DCT (using code borrowed from libjpeg v8a and later.)
+The accuracy of this implementation now matches the accuracy of the SSE/SSE2
+implementation.  Note, however, that the floating point DCT/IDCT algorithms are
+mainly a legacy feature.  They generally do not produce significantly better
+accuracy than the slow integer DCT/IDCT algorithms, and they are quite a bit
+slower.
+
 
 1.3.1
 =====

README-turbo.txt

@@ -419,10 +419,25 @@ details.
 
 For the most part, libjpeg-turbo should produce identical output to libjpeg
 v6b.  The one exception to this is when using the floating point DCT/IDCT, in
-which case the outputs of libjpeg v6b and libjpeg-turbo are not guaranteed to
-be identical (the accuracy of the floating point DCT/IDCT is constant when
-using libjpeg-turbo's SIMD extensions, but otherwise, it can depend heavily on
-the compiler and compiler settings.)
+which case the outputs of libjpeg v6b and libjpeg-turbo can differ for the
+following reasons:
+
+-- The SSE/SSE2 floating point DCT implementation in libjpeg-turbo is ever so
+   slightly more accurate than the implementation in libjpeg v6b, but not by
+   any amount perceptible to human vision (generally in the range of 0.01 to
+   0.08 dB gain in PNSR.)
+-- When not using the SIMD extensions, libjpeg-turbo uses the more accurate
+   (and slightly faster) floating point IDCT algorithm introduced in libjpeg
+   v8a as opposed to the algorithm used in libjpeg v6b.  It should be noted,
+   however, that this algorithm basically brings the accuracy of the floating
+   point IDCT in line with the accuracy of the slow integer IDCT.  The floating
+   point DCT/IDCT algorithms are mainly a legacy feature, and they do not
+   produce significantly more accuracy than the slow integer algorithms (to put
+   numbers on this, the typical difference in PNSR between the two algorithms
+   is less than 0.10 dB, whereas changing the quality level by 1 in the upper
+   range of the quality scale is typically more like a 1.0 dB difference.)
+-- When not using the SIMD extensions, then the accuracy of the floating point
+   DCT/IDCT can depend on the compiler and compiler settings.
 
 While libjpeg-turbo does emulate the libjpeg v8 API/ABI, under the hood, it is
 still using the same algorithms as libjpeg v6b, so there are several specific
@@ -430,16 +445,14 @@ cases in which libjpeg-turbo cannot be expected to produce the same output as
 libjpeg v8:
 
 -- When decompressing using scaling factors of 1/2 and 1/4, because libjpeg v8
-   implements those scaling algorithms a bit differently than libjpeg v6b does,
-   and libjpeg-turbo's SIMD extensions are based on the libjpeg v6b behavior.
+   implements those scaling algorithms differently than libjpeg v6b does, and
+   libjpeg-turbo's SIMD extensions are based on the libjpeg v6b behavior.
 
 -- When using chrominance subsampling, because libjpeg v8 implements this
    with its DCT/IDCT scaling algorithms rather than with a separate
-   downsampling/upsampling algorithm.
-
--- When using the floating point IDCT, for the reasons stated above and also
-   because the floating point IDCT algorithm was modified in libjpeg v8a to
-   improve accuracy.
+   downsampling/upsampling algorithm.  In our testing, the subsampled/upsampled
+   output of libjpeg v8 is less accurate than that of libjpeg v6b for this
+   reason.
 
 -- When decompressing using a scaling factor > 1 and merged (AKA "non-fancy" or
    "non-smooth") chrominance upsampling, because libjpeg v8 does not support

README.md

@@ -7,6 +7,8 @@ The idea is to reduce transfer times for JPEGs on the Web, thus reducing page lo
 
 'mozjpeg' is not intended to be a general JPEG library replacement. It makes tradeoffs that are intended to benefit Web use cases and focuses solely on improving encoding. It is best used as part of a Web encoding workflow. For a general JPEG library (e.g. your system libjpeg), especially if you care about decoding, we recommend libjpeg-turbo.
 
-For more information, see the project announcement:
+More information:
 
-https://blog.mozilla.org/research/2014/03/05/introducing-the-mozjpeg-project/
+* [Version 1.0 Announcement](https://blog.mozilla.org/research/2014/03/05/introducing-the-mozjpeg-project/)
+* [Version 2.0 Announcement](https://blog.mozilla.org/research/2014/07/15/mozilla-advances-jpeg-encoding-with-mozjpeg-2-0/)
+* [Mailing List](https://lists.mozilla.org/listinfo/dev-mozjpeg)</a>

cjpeg.1

@@ -1,4 +1,4 @@
-.TH CJPEG 1 "18 January 2013"
+.TH CJPEG 1 "11 May 2014"
 .SH NAME
 cjpeg \- compress an image file to a JPEG file
 .SH SYNOPSIS
@@ -166,14 +166,25 @@ Use integer DCT method (default).
 .TP
 .B \-dct fast
 Use fast integer DCT (less accurate).
+In libjpeg-turbo, the fast method is generally about 5-15% faster than the int
+method when using the x86/x86-64 SIMD extensions (results may vary with other
+SIMD implementations, or when using libjpeg-turbo without SIMD extensions.)
+For quality levels of 90 and below, there should be little or no perceptible
+difference between the two algorithms.  For quality levels above 90, however,
+the difference between the fast and the int methods becomes more pronounced.
+With quality=97, for instance, the fast method incurs generally about a 1-3 dB
+loss (in PSNR) relative to the int method, but this can be larger for some
+images.  Do not use the fast method with quality levels above 97.  The
+algorithm often degenerates at quality=98 and above and can actually produce a
+more lossy image than if lower quality levels had been used.
 .TP
 .B \-dct float
 Use floating-point DCT method.
-The float method is very slightly more accurate than the int method, but is
-much slower unless your machine has very fast floating-point hardware.  Also
-note that results of the floating-point method may vary slightly across
-machines, while the integer methods should give the same results everywhere.
-The fast integer method is much less accurate than the other two.
+The float method is mostly a legacy feature.  It does not produce significantly
+more accurate results than the int method, and it is much slower.  The float
+method may also give different results on different machines due to varying
+roundoff behavior, whereas the integer methods should give the same results on
+all machines.
 .TP
 .BI \-restart " N"
 Emit a JPEG restart marker every N MCU rows, or every N MCU blocks if "B" is

cjpeg.c

@@ -168,6 +168,7 @@ usage (void)
 #ifdef C_PROGRESSIVE_SUPPORTED
   fprintf(stderr, "  -progressive   Create progressive JPEG file (enabled by default)\n");
 #endif
+  fprintf(stderr, "  -baseline      Create baseline JPEG file (disable progressive coding)\n");
 #ifdef TARGA_SUPPORTED
   fprintf(stderr, "  -targa         Input file is Targa format (usually not needed)\n");
 #endif
@@ -206,7 +207,6 @@ usage (void)
 #endif
   fprintf(stderr, "  -verbose  or  -debug   Emit debug output\n");
   fprintf(stderr, "Switches for wizards:\n");
-  fprintf(stderr, "  -baseline      Force baseline quantization tables\n");
   fprintf(stderr, "  -qtables file  Use quantization tables given in file\n");
   fprintf(stderr, "  -qslots N[,...]    Set component quantization tables\n");
   fprintf(stderr, "  -sample HxV[,...]  Set component sampling factors\n");
@@ -279,11 +279,17 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
     } else if (keymatch(arg, "baseline", 1)) {
       /* Force baseline-compatible output (8-bit quantizer values). */
       force_baseline = TRUE;
+      /* Disable multiple scans */
+      simple_progressive = FALSE;
+      cinfo->num_scans = 0;
+      cinfo->scan_info = NULL;
 
     } else if (keymatch(arg, "dct", 2)) {
       /* Select DCT algorithm. */
-      if (++argn >= argc)	/* advance to next argument */
+      if (++argn >= argc) { /* advance to next argument */
+        fprintf(stderr, "%s: missing argument for dct\n", progname);
 	usage();
+      }
       if (keymatch(argv[argn], "int", 1)) {
 	cinfo->dct_method = JDCT_ISLOW;
       } else if (keymatch(argv[argn], "fast", 2)) {
@@ -291,6 +297,7 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
       } else if (keymatch(argv[argn], "float", 2)) {
 	cinfo->dct_method = JDCT_FLOAT;
       } else
+        fprintf(stderr, "%s: invalid argument for dct\n", progname);
 	usage();
 
     } else if (keymatch(arg, "debug", 1) || keymatch(arg, "verbose", 1)) {
@@ -314,7 +321,7 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
     } else if (keymatch(arg, "flat", 4)) {
       cinfo->use_flat_quant_tbl = TRUE;
       jpeg_set_quality(cinfo, 75, TRUE);
-      
+
     } else if (keymatch(arg, "grayscale", 2) || keymatch(arg, "greyscale",2)) {
       /* Force a monochrome JPEG file to be generated. */
       jpeg_set_colorspace(cinfo, JCS_GRAYSCALE);
@@ -327,12 +334,12 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
       if (++argn >= argc)	/* advance to next argument */
 	usage();
       cinfo->lambda_log_scale1 = atof(argv[argn]);
-      
+
     } else if (keymatch(arg, "lambda2", 7)) {
       if (++argn >= argc)	/* advance to next argument */
 	usage();
       cinfo->lambda_log_scale2 = atof(argv[argn]);
-      
+
     } else if (keymatch(arg, "maxmemory", 3)) {
       /* Maximum memory in Kb (or Mb with 'm'). */
       long lval;
@@ -348,7 +355,7 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
 
     } else if (keymatch(arg, "multidcscan", 3)) {
       cinfo->one_dc_scan = FALSE;
-      
+
     } else if (keymatch(arg, "optimize", 1) || keymatch(arg, "optimise", 1)) {
       /* Enable entropy parm optimization. */
 #ifdef ENTROPY_OPT_SUPPORTED
@@ -361,8 +368,10 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
 
     } else if (keymatch(arg, "outfile", 4)) {
       /* Set output file name. */
-      if (++argn >= argc)	/* advance to next argument */
+      if (++argn >= argc)	{ /* advance to next argument */
+        fprintf(stderr, "%s: missing argument for outfile\n", progname);
 	usage();
+      }
       outfilename = argv[argn];	/* save it away for later use */
 
     } else if (keymatch(arg, "progressive", 1)) {
@@ -388,8 +397,10 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
 
     } else if (keymatch(arg, "quality", 1)) {
       /* Quality ratings (quantization table scaling factors). */
-      if (++argn >= argc)	/* advance to next argument */
+      if (++argn >= argc)	{ /* advance to next argument */
+        fprintf(stderr, "%s: missing argument for quality\n", progname);
 	usage();
+      }
       qualityarg = argv[argn];
 
     } else if (keymatch(arg, "qslots", 2)) {
@@ -505,6 +516,7 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
       jpeg_set_quality(cinfo, 75, TRUE);
       
     } else {
+      fprintf(stderr, "%s: unknown option '%s'\n", progname, arg);
       usage();			/* bogus switch */
     }
   }
@@ -516,20 +528,26 @@ parse_switches (j_compress_ptr cinfo, int argc, char **argv,
     /* Set quantization tables for selected quality. */
     /* Some or all may be overridden if -qtables is present. */
     if (qualityarg != NULL)	/* process -quality if it was present */
-      if (! set_quality_ratings(cinfo, qualityarg, force_baseline))
+      if (! set_quality_ratings(cinfo, qualityarg, force_baseline)) {
+        fprintf(stderr, "%s: can't set quality ratings\n", progname);
 	usage();
+      }
 
     if (qtablefile != NULL)	/* process -qtables if it was present */
-      if (! read_quant_tables(cinfo, qtablefile, force_baseline))
+      if (! read_quant_tables(cinfo, qtablefile, force_baseline)) {
+        fprintf(stderr, "%s: can't read qtable file\n", progname);
 	usage();
+      }
 
     if (qslotsarg != NULL)	/* process -qslots if it was present */
       if (! set_quant_slots(cinfo, qslotsarg))
 	usage();
 
     if (samplearg != NULL)	/* process -sample if it was present */
-      if (! set_sample_factors(cinfo, samplearg))
+      if (! set_sample_factors(cinfo, samplearg)) {
+        fprintf(stderr, "%s: can't set sample factors\n", progname);
 	usage();
+      }
 
 #ifdef C_PROGRESSIVE_SUPPORTED
     if (simple_progressive)	/* process -progressive; -scans can override */

configure.ac

@@ -2,7 +2,7 @@
 # Process this file with autoconf to produce a configure script.
 
 AC_PREREQ([2.56])
-AC_INIT([libmozjpeg], [2.0pre])
+AC_INIT([libmozjpeg], [2.1])
 BUILD=`date +%Y%m%d`
 
 AM_INIT_AUTOMAKE([-Wall foreign dist-bzip2])
@@ -20,7 +20,7 @@ AC_PROG_CC
 AM_PROG_CC_C_O
 AM_PROG_AS
 AC_PROG_INSTALL
-AM_PROG_AR
+m4_ifdef([AM_PROG_AR], [AM_PROG_AR])
 AC_PROG_LIBTOOL
 AC_PROG_LN_S
 

djpeg.1

@@ -1,4 +1,4 @@
-.TH DJPEG 1 "18 January 2013"
+.TH DJPEG 1 "11 May 2014"
 .SH NAME
 djpeg \- decompress a JPEG file to an image file
 .SH SYNOPSIS
@@ -115,14 +115,28 @@ Use integer DCT method (default).
 .TP
 .B \-dct fast
 Use fast integer DCT (less accurate).
+In libjpeg-turbo, the fast method is generally about 5-15% faster than the int
+method when using the x86/x86-64 SIMD extensions (results may vary with other
+SIMD implementations, or when using libjpeg-turbo without SIMD extensions.)  If
+the JPEG image was compressed using a quality level of 85 or below, then there
+should be little or no perceptible difference between the two algorithms.  When
+decompressing images that were compressed using quality levels above 85,
+however, the difference between the fast and int methods becomes more
+pronounced.  With images compressed using quality=97, for instance, the fast
+method incurs generally about a 4-6 dB loss (in PSNR) relative to the int
+method, but this can be larger for some images.  If you can avoid it, do not
+use the fast method when decompressing images that were compressed using
+quality levels above 97.  The algorithm often degenerates for such images and
+can actually produce a more lossy output image than if the JPEG image had been
+compressed using lower quality levels.
 .TP
 .B \-dct float
 Use floating-point DCT method.
-The float method is very slightly more accurate than the int method, but is
-much slower unless your machine has very fast floating-point hardware.  Also
-note that results of the floating-point method may vary slightly across
-machines, while the integer methods should give the same results everywhere.
-The fast integer method is much less accurate than the other two.
+The float method is mostly a legacy feature.  It does not produce significantly
+more accurate results than the int method, and it is much slower.  The float
+method may also give different results on different machines due to varying
+roundoff behavior, whereas the integer methods should give the same results on
+all machines.
 .TP
 .B \-dither fs
 Use Floyd-Steinberg dithering in color quantization.

jcapistd.c

@@ -46,7 +46,7 @@ jpeg_start_compress (j_compress_ptr cinfo, boolean write_all_tables)
     jpeg_suppress_tables(cinfo, FALSE);	/* mark all tables to be written */
 
   /* setting up scan optimisation pattern failed, disable scan optimisation */
-  if (cinfo->num_scans_luma == 0)
+  if (cinfo->num_scans_luma == 0 || cinfo->scan_info == NULL || cinfo->num_scans == 0)
     cinfo->optimize_scans = FALSE;
   
   /* (Re)initialize error mgr and destination modules */

jcdctmgr.c

@@ -543,6 +543,8 @@ forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
   FAST_FLOAT * divisors = fdct->float_divisors[compptr->quant_tbl_no];
   FAST_FLOAT * workspace;
   JDIMENSION bi;
+  float v;
+  int x;
 
 
   /* Make sure the compiler doesn't look up these every pass */
@@ -572,10 +574,10 @@ forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
       };
 
       for (i = 0; i < DCTSIZE2; i++) {
-        float v = workspace[i];
+        v = workspace[i];
         v /= aanscalefactor[i%8];
         v /= aanscalefactor[i/8];
-        int x = (v >= 0.0) ? (int)(v + 0.5) : (int)(v - 0.5);
+        x = (v >= 0.0) ? (int)(v + 0.5) : (int)(v - 0.5);
         dst[bi][i] = x;
       }
     }
@@ -588,9 +590,7 @@ forward_DCT_float (j_compress_ptr cinfo, jpeg_component_info * compptr,
 #endif /* DCT_FLOAT_SUPPORTED */
 
 #include "jchuff.h"
-
-static unsigned char jpeg_nbits_table[65536];
-static int jpeg_nbits_table_init = 0;
+#include "jpeg_nbits_table.h"
 
 static const float jpeg_lambda_weights_flat[64] = {
   1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f, 1.0f,
@@ -637,6 +637,10 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *actbl, JBLOCKROW coef_bloc
   int has_eob;
   float cost_all_zeros;
   float best_cost_skip;
+  float cost;
+  int zero_run;
+  int run_bits;
+  int rate;
 
   Ss = cinfo->Ss;
   Se = cinfo->Se;
@@ -654,15 +658,6 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *actbl, JBLOCKROW coef_bloc
     requires_eob[0] = 0;
   }
   
-  if(!jpeg_nbits_table_init) {
-    for(i = 0; i < 65536; i++) {
-      int nbits = 0, temp = i;
-      while (temp) {temp >>= 1;  nbits++;}
-      jpeg_nbits_table[i] = nbits;
-    }
-    jpeg_nbits_table_init = 1;
-  }
-
   norm = 0.0;
   for (i = 1; i < DCTSIZE2; i++) {
     norm += qtbl->quantval[i] * qtbl->quantval[i];
@@ -725,11 +720,11 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *actbl, JBLOCKROW coef_bloc
         if (j != Ss-1 && coef_blocks[bi][zz] == 0)
           continue;
         
-        int zero_run = i - 1 - j;
+        zero_run = i - 1 - j;
         if ((zero_run >> 4) && actbl->ehufsi[0xf0] == 0)
           continue;
         
-        int run_bits = (zero_run >> 4) * actbl->ehufsi[0xf0];
+        run_bits = (zero_run >> 4) * actbl->ehufsi[0xf0];
         zero_run &= 15;
 
         for (k = 0; k < num_candidates; k++) {
@@ -737,8 +732,8 @@ quantize_trellis(j_compress_ptr cinfo, c_derived_tbl *actbl, JBLOCKROW coef_bloc
           if (coef_bits == 0)
             continue;
           
-          int rate = coef_bits + candidate_bits[k] + run_bits;
-          float cost = rate + candidate_dist[k];
+          rate = coef_bits + candidate_bits[k] + run_bits;
+          cost = rate + candidate_dist[k];
           cost += accumulated_zero_dist[i-1] - accumulated_zero_dist[j] + accumulated_cost[j];
           
           if (cost < accumulated_cost[i]) {

jchuff.c

@@ -22,8 +22,7 @@
 #include "jchuff.h"		/* Declarations shared with jcphuff.c */
 #include <limits.h>
 
-static unsigned char jpeg_nbits_table[65536];
-static int jpeg_nbits_table_init = 0;
+#include "jpeg_nbits_table.h"
 
 #ifndef min
  #define min(a,b) ((a)<(b)?(a):(b))
@@ -271,15 +270,6 @@ jpeg_make_c_derived_tbl (j_compress_ptr cinfo, boolean isDC, int tblno,
     dtbl->ehufco[i] = huffcode[p];
     dtbl->ehufsi[i] = huffsize[p];
   }
-
-  if(!jpeg_nbits_table_init) {
-    for(i = 0; i < 65536; i++) {
-      int nbits = 0, temp = i;
-      while (temp) {temp >>= 1;  nbits++;}
-      jpeg_nbits_table[i] = nbits;
-    }
-    jpeg_nbits_table_init = 1;
-  }
 }
 
 

jcmaster.c

@@ -933,11 +933,10 @@ jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
   else
     master->total_passes = cinfo->num_scans;
   
+  master->pass_number_scan_opt_base = 0;
   if (cinfo->trellis_quant) {
-    if (cinfo->progressive_mode)
-      master->total_passes += ((cinfo->use_scans_in_trellis) ? 4 : 2) * cinfo->num_components * cinfo->trellis_num_loops;
-    else
-      master->total_passes += 1;
+    master->pass_number_scan_opt_base = ((cinfo->use_scans_in_trellis) ? 4 : 2) * cinfo->num_components * cinfo->trellis_num_loops;
+    master->total_passes += master->pass_number_scan_opt_base;
   }
   
   if (cinfo->optimize_scans) {
@@ -947,9 +946,4 @@ jinit_c_master_control (j_compress_ptr cinfo, boolean transcode_only)
     for (i = 0; i < cinfo->num_scans; i++)
       master->scan_buffer[i] = NULL;
   }
-  
-  if (cinfo->trellis_quant)
-    master->pass_number_scan_opt_base = ((cinfo->use_scans_in_trellis) ? 4 : 2) * cinfo->num_components * cinfo->trellis_num_loops;
-  else
-    master->pass_number_scan_opt_base = 0;
 }

jidctflt.c

@@ -1,9 +1,12 @@
 /*
  * jidctflt.c
  *
+ * This file was part of the Independent JPEG Group's software:
  * Copyright (C) 1994-1998, Thomas G. Lane.
- * This file is part of the Independent JPEG Group's software.
- * For conditions of distribution and use, see the accompanying README file.
+ * Modified 2010 by Guido Vollbeding.
+ * libjpeg-turbo Modifications:
+ * Copyright (C) 2014, D. R. Commander.
+  * For conditions of distribution and use, see the accompanying README file.
  *
  * This file contains a floating-point implementation of the
  * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
@@ -76,10 +79,10 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
   FLOAT_MULT_TYPE * quantptr;
   FAST_FLOAT * wsptr;
   JSAMPROW outptr;
-  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
+  JSAMPLE *range_limit = cinfo->sample_range_limit;
   int ctr;
   FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */
-  SHIFT_TEMPS
+  #define _0_125 ((FLOAT_MULT_TYPE)0.125)
 
   /* Pass 1: process columns from input, store into work array. */
 
@@ -101,7 +104,8 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
 	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
 	inptr[DCTSIZE*7] == 0) {
       /* AC terms all zero */
-      FAST_FLOAT dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
+      FAST_FLOAT dcval = DEQUANTIZE(inptr[DCTSIZE*0],
+                                    quantptr[DCTSIZE*0] * _0_125);
       
       wsptr[DCTSIZE*0] = dcval;
       wsptr[DCTSIZE*1] = dcval;
@@ -120,10 +124,10 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     
     /* Even part */
 
-    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
-    tmp1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
-    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
-    tmp3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
+    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0] * _0_125);
+    tmp1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2] * _0_125);
+    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4] * _0_125);
+    tmp3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6] * _0_125);
 
     tmp10 = tmp0 + tmp2;	/* phase 3 */
     tmp11 = tmp0 - tmp2;
@@ -138,10 +142,10 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     
     /* Odd part */
 
-    tmp4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
-    tmp5 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
-    tmp6 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
-    tmp7 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
+    tmp4 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1] * _0_125);
+    tmp5 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3] * _0_125);
+    tmp6 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5] * _0_125);
+    tmp7 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7] * _0_125);
 
     z13 = tmp6 + tmp5;		/* phase 6 */
     z10 = tmp6 - tmp5;
@@ -152,12 +156,12 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */
 
     z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
-    tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
-    tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
+    tmp10 = z5 - z12 * ((FAST_FLOAT) 1.082392200); /* 2*(c2-c6) */
+    tmp12 = z5 - z10 * ((FAST_FLOAT) 2.613125930); /* 2*(c2+c6) */
 
     tmp6 = tmp12 - tmp7;	/* phase 2 */
     tmp5 = tmp11 - tmp6;
-    tmp4 = tmp10 + tmp5;
+    tmp4 = tmp10 - tmp5;
 
     wsptr[DCTSIZE*0] = tmp0 + tmp7;
     wsptr[DCTSIZE*7] = tmp0 - tmp7;
@@ -165,8 +169,8 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     wsptr[DCTSIZE*6] = tmp1 - tmp6;
     wsptr[DCTSIZE*2] = tmp2 + tmp5;
     wsptr[DCTSIZE*5] = tmp2 - tmp5;
-    wsptr[DCTSIZE*4] = tmp3 + tmp4;
-    wsptr[DCTSIZE*3] = tmp3 - tmp4;
+    wsptr[DCTSIZE*3] = tmp3 + tmp4;
+    wsptr[DCTSIZE*4] = tmp3 - tmp4;
 
     inptr++;			/* advance pointers to next column */
     quantptr++;
@@ -174,7 +178,6 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
   }
   
   /* Pass 2: process rows from work array, store into output array. */
-  /* Note that we must descale the results by a factor of 8 == 2**3. */
 
   wsptr = workspace;
   for (ctr = 0; ctr < DCTSIZE; ctr++) {
@@ -187,8 +190,10 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     
     /* Even part */
 
-    tmp10 = wsptr[0] + wsptr[4];
-    tmp11 = wsptr[0] - wsptr[4];
+    /* Apply signed->unsigned and prepare float->int conversion */
+    z5 = wsptr[0] + ((FAST_FLOAT) CENTERJSAMPLE + (FAST_FLOAT) 0.5);
+    tmp10 = z5 + wsptr[4];
+    tmp11 = z5 - wsptr[4];
 
     tmp13 = wsptr[2] + wsptr[6];
     tmp12 = (wsptr[2] - wsptr[6]) * ((FAST_FLOAT) 1.414213562) - tmp13;
@@ -209,31 +214,23 @@ jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
     tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562);
 
     z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
-    tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
-    tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */
+    tmp10 = z5 - z12 * ((FAST_FLOAT) 1.082392200); /* 2*(c2-c6) */
+    tmp12 = z5 - z10 * ((FAST_FLOAT) 2.613125930); /* 2*(c2+c6) */
 
     tmp6 = tmp12 - tmp7;
     tmp5 = tmp11 - tmp6;
-    tmp4 = tmp10 + tmp5;
+    tmp4 = tmp10 - tmp5;
 
-    /* Final output stage: scale down by a factor of 8 and range-limit */
+    /* Final output stage: float->int conversion and range-limit */
 
-    outptr[0] = range_limit[(int) DESCALE((INT32) (tmp0 + tmp7), 3)
-			    & RANGE_MASK];
-    outptr[7] = range_limit[(int) DESCALE((INT32) (tmp0 - tmp7), 3)
-			    & RANGE_MASK];
-    outptr[1] = range_limit[(int) DESCALE((INT32) (tmp1 + tmp6), 3)
-			    & RANGE_MASK];
-    outptr[6] = range_limit[(int) DESCALE((INT32) (tmp1 - tmp6), 3)
-			    & RANGE_MASK];
-    outptr[2] = range_limit[(int) DESCALE((INT32) (tmp2 + tmp5), 3)
-			    & RANGE_MASK];
-    outptr[5] = range_limit[(int) DESCALE((INT32) (tmp2 - tmp5), 3)
-			    & RANGE_MASK];
-    outptr[4] = range_limit[(int) DESCALE((INT32) (tmp3 + tmp4), 3)
-			    & RANGE_MASK];
-    outptr[3] = range_limit[(int) DESCALE((INT32) (tmp3 - tmp4), 3)
-			    & RANGE_MASK];
+    outptr[0] = range_limit[((int) (tmp0 + tmp7)) & RANGE_MASK];
+    outptr[7] = range_limit[((int) (tmp0 - tmp7)) & RANGE_MASK];
+    outptr[1] = range_limit[((int) (tmp1 + tmp6)) & RANGE_MASK];
+    outptr[6] = range_limit[((int) (tmp1 - tmp6)) & RANGE_MASK];
+    outptr[2] = range_limit[((int) (tmp2 + tmp5)) & RANGE_MASK];
+    outptr[5] = range_limit[((int) (tmp2 - tmp5)) & RANGE_MASK];
+    outptr[3] = range_limit[((int) (tmp3 + tmp4)) & RANGE_MASK];
+    outptr[4] = range_limit[((int) (tmp3 - tmp4)) & RANGE_MASK];
     
     wsptr += DCTSIZE;		/* advance pointer to next row */
   }

libjpeg.txt

@@ -3,7 +3,7 @@ USING THE IJG JPEG LIBRARY
 This file was part of the Independent JPEG Group's software:
 Copyright (C) 1994-2011, Thomas G. Lane, Guido Vollbeding.
 Modifications:
-Copyright (C) 2010, D. R. Commander.
+Copyright (C) 2010, 2014, D. R. Commander.
 For conditions of distribution and use, see the accompanying README file.
 
 
@@ -886,14 +886,23 @@ J_DCT_METHOD dct_method
 		JDCT_FLOAT: floating-point method
 		JDCT_DEFAULT: default method (normally JDCT_ISLOW)
 		JDCT_FASTEST: fastest method (normally JDCT_IFAST)
-	The FLOAT method is very slightly more accurate than the ISLOW method,
-	but may give different results on different machines due to varying
-	roundoff behavior.  The integer methods should give the same results
-	on all machines.  On machines with sufficiently fast FP hardware, the
-	floating-point method may also be the fastest.  The IFAST method is
-	considerably less accurate than the other two; its use is not
-	recommended if high quality is a concern.  JDCT_DEFAULT and
-	JDCT_FASTEST are macros configurable by each installation.
+        In libjpeg-turbo, JDCT_IFAST is generally about 5-15% faster than
+        JDCT_ISLOW when using the x86/x86-64 SIMD extensions (results may vary
+        with other SIMD implementations, or when using libjpeg-turbo without
+        SIMD extensions.)  For quality levels of 90 and below, there should be
+        little or no perceptible difference between the two algorithms.  For
+        quality levels above 90, however, the difference between JDCT_IFAST and
+        JDCT_ISLOW becomes more pronounced.  With quality=97, for instance,
+        JDCT_IFAST incurs generally about a 1-3 dB loss (in PSNR) relative to
+        JDCT_ISLOW, but this can be larger for some images.  Do not use
+        JDCT_IFAST with quality levels above 97.  The algorithm often
+        degenerates at quality=98 and above and can actually produce a more
+        lossy image than if lower quality levels had been used.  JDCT_FLOAT is
+        mostly a legacy feature.  It does not produce significantly more
+        accurate results than the ISLOW method, and it is much slower.  The
+        FLOAT method may also give different results on different machines due
+        to varying roundoff behavior, whereas the integer methods should give
+        the same results on all machines.
 
 J_COLOR_SPACE jpeg_color_space
 int num_components
@@ -1170,8 +1179,32 @@ int actual_number_of_colors
 Additional decompression parameters that the application may set include:
 
 J_DCT_METHOD dct_method
-	Selects the algorithm used for the DCT step.  Choices are the same
-	as described above for compression.
+        Selects the algorithm used for the DCT step.  Choices are:
+                JDCT_ISLOW: slow but accurate integer algorithm
+                JDCT_IFAST: faster, less accurate integer method
+                JDCT_FLOAT: floating-point method
+                JDCT_DEFAULT: default method (normally JDCT_ISLOW)
+                JDCT_FASTEST: fastest method (normally JDCT_IFAST)
+        In libjpeg-turbo, JDCT_IFAST is generally about 5-15% faster than
+        JDCT_ISLOW when using the x86/x86-64 SIMD extensions (results may vary
+        with other SIMD implementations, or when using libjpeg-turbo without
+        SIMD extensions.)  If the JPEG image was compressed using a quality
+        level of 85 or below, then there should be little or no perceptible
+        difference between the two algorithms.  When decompressing images that
+        were compressed using quality levels above 85, however, the difference
+        between JDCT_IFAST and JDCT_ISLOW becomes more pronounced.  With images
+        compressed using quality=97, for instance, JDCT_IFAST incurs generally
+        about a 4-6 dB loss (in PSNR) relative to JDCT_ISLOW, but this can be
+        larger for some images.  If you can avoid it, do not use JDCT_IFAST
+        when decompressing images that were compressed using quality levels
+        above 97.  The algorithm often degenerates for such images and can
+        actually produce a more lossy output image than if the JPEG image had
+        been compressed using lower quality levels.  JDCT_FLOAT is mostly a
+        legacy feature.  It does not produce significantly more accurate
+        results than the ISLOW method, and it is much slower.  The FLOAT method
+        may also give different results on different machines due to varying
+        roundoff behavior, whereas the integer methods should give the same
+        results on all machines.
 
 boolean do_fancy_upsampling
 	If TRUE, do careful upsampling of chroma components.  If FALSE,

usage.txt

@@ -172,13 +172,28 @@ Switches for advanced users:
 	-dct int	Use integer DCT method (default).
 	-dct fast	Use fast integer DCT (less accurate).
 	-dct float	Use floating-point DCT method.
-			The float method is very slightly more accurate than
-			the int method, but is much slower unless your machine
-			has very fast floating-point hardware.  Also note that
-			results of the floating-point method may vary slightly
-			across machines, while the integer methods should give
-			the same results everywhere.  The fast integer method
-			is much less accurate than the other two.
+                        In libjpeg-turbo, the fast method is generally about
+                        5-15% faster than the int method when using the
+                        x86/x86-64 SIMD extensions (results may vary with other
+                        SIMD implementations, or when using libjpeg-turbo
+                        without SIMD extensions.)  For quality levels of 90 and
+                        below, there should be little or no perceptible
+                        difference between the two algorithms.  For quality
+                        levels above 90, however, the difference between
+                        the fast and the int methods becomes more pronounced.
+                        With quality=97, for instance, the fast method incurs
+                        generally about a 1-3 dB loss (in PSNR) relative to
+                        the int method, but this can be larger for some images.
+                        Do not use the fast method with quality levels above
+                        97.  The algorithm often degenerates at quality=98 and
+                        above and can actually produce a more lossy image than
+                        if lower quality levels had been used.  The float
+                        method is mostly a legacy feature.  It does not produce
+                        significantly more accurate results than the int
+                        method, and it is much slower.  The float method may
+                        also give different results on different machines due
+                        to varying roundoff behavior, whereas the integer
+                        methods should give the same results on all machines.
 
 	-restart N	Emit a JPEG restart marker every N MCU rows, or every
 			N MCU blocks if "B" is attached to the number.
@@ -296,13 +311,32 @@ Switches for advanced users:
 	-dct int	Use integer DCT method (default).
 	-dct fast	Use fast integer DCT (less accurate).
 	-dct float	Use floating-point DCT method.
-			The float method is very slightly more accurate than
-			the int method, but is much slower unless your machine
-			has very fast floating-point hardware.  Also note that
-			results of the floating-point method may vary slightly
-			across machines, while the integer methods should give
-			the same results everywhere.  The fast integer method
-			is much less accurate than the other two.
+                        In libjpeg-turbo, the fast method is generally about
+                        5-15% faster than the int method when using the
+                        x86/x86-64 SIMD extensions (results may vary with other
+                        SIMD implementations, or when using libjpeg-turbo
+                        without SIMD extensions.)  If the JPEG image was
+                        compressed using a quality level of 85 or below, then
+                        there should be little or no perceptible difference
+                        between the two algorithms.  When decompressing images
+                        that were compressed using quality levels above 85,
+                        however, the difference between the fast and int
+                        methods becomes more pronounced.  With images
+                        compressed using quality=97, for instance, the fast
+                        method incurs generally about a 4-6 dB loss (in PSNR)
+                        relative to the int method, but this can be larger for
+                        some images.  If you can avoid it, do not use the fast
+                        method when decompressing images that were compressed
+                        using quality levels above 97.  The algorithm often
+                        degenerates for such images and can actually produce
+                        a more lossy output image than if the JPEG image had
+                        been compressed using lower quality levels.  The float
+                        method is mostly a legacy feature.  It does not produce
+                        significantly more accurate results than the int
+                        method, and it is much slower.  The float method may
+                        also give different results on different machines due
+                        to varying roundoff behavior, whereas the integer
+                        methods should give the same results on all machines.
 
 	-dither fs	Use Floyd-Steinberg dithering in color quantization.
 	-dither ordered	Use ordered dithering in color quantization.
@@ -381,12 +415,6 @@ When producing a color-quantized image, "-onepass -dither ordered" is fast but
 much lower quality than the default behavior.  "-dither none" may give
 acceptable results in two-pass mode, but is seldom tolerable in one-pass mode.
 
-If you are fortunate enough to have very fast floating point hardware,
-"-dct float" may be even faster than "-dct fast".  But on most machines
-"-dct float" is slower than "-dct int"; in this case it is not worth using,
-because its theoretical accuracy advantage is too small to be significant
-in practice.
-
 Two-pass color quantization requires a good deal of memory; on MS-DOS machines
 it may run out of memory even with -maxmemory 0.  In that case you can still
 decompress, with some loss of image quality, by specifying -onepass for