From: DRC <dcommander@users.sourceforge.net>
Date: Fri, 18 Feb 2011 01:45:24 +0000 (+0000)
Subject: The SIMD quantization algorithm does not produce correct results with the fast forwar... 
X-Git-Tag: 1.1.0~37
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=72abc29722c11c71375440608efc71f713bb834b;p=libjpeg-turbo

The SIMD quantization algorithm does not produce correct results with the fast forward integer DCT and JPEG qualities >= 98, so for now, use the non-SIMD quantization function under those circumstances.


git-svn-id: svn+ssh://svn.code.sf.net/p/libjpeg-turbo/code/branches/1.1.x@378 632fc199-4ca6-4c93-a231-07263d6284db
---

diff --git a/ChangeLog.txt b/ChangeLog.txt
index 0097c41..14d251a 100644
--- a/ChangeLog.txt
+++ b/ChangeLog.txt
@@ -1,15 +1,19 @@
 1.1.0
 =====
 
-[1] Since the fast integer forward DCT seems to degrade for JPEG qualities
-greater than 95, TurboJPEG/OSS will now automatically use the slow integer
-forward DCT when generating JPEG images of quality 96 or greater.  This
-reduces compression performance by as much as 15% for these high-quality images
-but is necessary to ensure that the images are perceptually lossless.
+[1] The algorithm used by the SIMD quantization function cannot produce correct
+results when the JPEG quality is >= 98 and the fast integer forward DCT is
+used.  Thus, the non-SIMD quantization function is now used for those cases.
 
-[2] Ported jpgtest.cxx to pure C to avoid the need for a C++ compiler.
+[2] Despite the above, the fast integer forward DCT still degrades somewhat for
+JPEG qualities greater than 95, so TurboJPEG/OSS will now automatically use the
+slow integer forward DCT when generating JPEG images of quality 96 or greater.
+This reduces compression performance by as much as 15% for these high-quality
+images but is necessary to ensure that the images are perceptually lossless.
 
-[3] Fixed visual artifacts in grayscale JPEG compression caused by a typo in
+[3] Ported jpgtest.cxx to pure C to avoid the need for a C++ compiler.
+
+[4] Fixed visual artifacts in grayscale JPEG compression caused by a typo in
 the RGB-to-chrominance lookup tables.
 
 
diff --git a/Makefile.am b/Makefile.am
index adfa5c0..ca69a38 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -129,6 +129,8 @@ test: testclean all
 	cmp $(srcdir)/testimgint.jpg testoutint.jpg
 	./cjpeg -dct fast -opt -outfile testoutfst.jpg $(srcdir)/testorig.ppm
 	cmp $(srcdir)/testimgfst.jpg testoutfst.jpg
+	./cjpeg -dct fast -quality 100 -opt -outfile testoutfst100.jpg $(srcdir)/testorig.ppm
+	cmp $(srcdir)/testimgfst100.jpg testoutfst100.jpg
 	./cjpeg -dct float -outfile testoutflt.jpg $(srcdir)/testorig.ppm
 if WITH_SIMD
 	cmp $(srcdir)/testimgflt.jpg testoutflt.jpg
diff --git a/jcdctmgr.c b/jcdctmgr.c
index 156957a..711f9da 100644
--- a/jcdctmgr.c
+++ b/jcdctmgr.c
@@ -4,6 +4,7 @@
  * Copyright (C) 1994-1996, Thomas G. Lane.
  * Copyright (C) 1999-2006, MIYASAKA Masaru.
  * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
+ * Copyright (C) 2011 D. R. Commander
  * This file is part of the Independent JPEG Group's software.
  * For conditions of distribution and use, see the accompanying README file.
  *
@@ -39,6 +40,8 @@ typedef JMETHOD(void, float_quantize_method_ptr,
                 (JCOEFPTR coef_block, FAST_FLOAT * divisors,
                  FAST_FLOAT * workspace));
 
+METHODDEF(void) quantize (JCOEFPTR, DCTELEM *, DCTELEM *);
+
 typedef struct {
   struct jpeg_forward_dct pub;	/* public fields */
 
@@ -160,7 +163,7 @@ flss (UINT16 val)
  * of in a consecutive manner, yet again in order to allow SIMD
  * routines.
  */
-LOCAL(void)
+LOCAL(int)
 compute_reciprocal (UINT16 divisor, DCTELEM * dtbl)
 {
   UDCTELEM2 fq, fr;
@@ -189,6 +192,9 @@ compute_reciprocal (UINT16 divisor, DCTELEM * dtbl)
   dtbl[DCTSIZE2 * 1] = (DCTELEM) c;       /* correction + roundfactor */
   dtbl[DCTSIZE2 * 2] = (DCTELEM) (1 << (sizeof(DCTELEM)*8*2 - r));  /* scale */
   dtbl[DCTSIZE2 * 3] = (DCTELEM) r - sizeof(DCTELEM)*8; /* shift */
+
+  if(r <= 16) return 0;
+  else return 1;
 }
 
 /*
@@ -232,7 +238,9 @@ start_pass_fdctmgr (j_compress_ptr cinfo)
       }
       dtbl = fdct->divisors[qtblno];
       for (i = 0; i < DCTSIZE2; i++) {
-	compute_reciprocal(qtbl->quantval[i] << 3, &dtbl[i]);
+	if(!compute_reciprocal(qtbl->quantval[i] << 3, &dtbl[i])
+	  && fdct->quantize == jsimd_quantize)
+	  fdct->quantize = quantize;
       }
       break;
 #endif
@@ -266,10 +274,12 @@ start_pass_fdctmgr (j_compress_ptr cinfo)
 	}
 	dtbl = fdct->divisors[qtblno];
 	for (i = 0; i < DCTSIZE2; i++) {
-	  compute_reciprocal(
+	  if(!compute_reciprocal(
 	    DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
 				  (INT32) aanscales[i]),
-		    CONST_BITS-3), &dtbl[i]);
+		    CONST_BITS-3), &dtbl[i])
+	    && fdct->quantize == jsimd_quantize)
+	    fdct->quantize = quantize;
 	}
       }
       break;
diff --git a/testimgfst100.jpg b/testimgfst100.jpg
new file mode 100644
index 0000000..36d9b75
Binary files /dev/null and b/testimgfst100.jpg differ