From d20666a228af58e97b755fbb8c7f86f4f49ed482 Mon Sep 17 00:00:00 2001 From: Glenn Randers-Pehrson Date: Sat, 1 Sep 2007 15:12:50 -0500 Subject: [PATCH] Imported from libpng-1.2.20rc4.tar --- ANNOUNCE | 31 +- CHANGES | 3 + INSTALL | 27 +- KNOWNBUG | 2 +- LICENSE | 4 +- Makefile.am | 4 +- Makefile.in | 26 +- README | 14 +- Y2KINFO | 4 +- configure | 121 +- configure.ac | 15 +- libpng-1.2.20rc3.txt => libpng-1.2.20rc4.txt | 184 +- libpng.3 | 198 +- libpngpf.3 | 4 +- png.5 | 2 +- png.c | 52 +- png.h | 28 +- pngconf.h | 79 +- pngerror.c | 2 +- pnggccrd.c | 6043 +----------------- pngget.c | 77 +- pngpread.c | 7 +- pngread.c | 6 - pngrutil.c | 1053 +-- pngset.c | 51 +- pngtest.c | 2 +- pngvcrd.c | 3918 +----------- pngwrite.c | 16 - scripts/CMakeLists.txt | 21 +- scripts/libpng-config-head.in | 2 +- scripts/libpng.pc-configure.in | 2 +- scripts/libpng.pc.in | 2 +- scripts/makefile.32sunu | 2 +- scripts/makefile.64sunu | 2 +- scripts/makefile.aix | 2 +- scripts/makefile.beos | 2 +- scripts/makefile.cygwin | 21 +- scripts/makefile.darwin | 2 +- scripts/makefile.dec | 2 +- scripts/makefile.elf | 2 +- scripts/makefile.gcmmx | 12 +- scripts/makefile.hp64 | 2 +- scripts/makefile.hpgcc | 2 +- scripts/makefile.hpux | 2 +- scripts/makefile.intel | 8 +- scripts/makefile.linux | 2 +- scripts/makefile.mingw | 9 +- scripts/makefile.ne12bsd | 8 +- scripts/makefile.netbsd | 8 +- scripts/makefile.nommx | 2 +- scripts/makefile.openbsd | 4 +- scripts/makefile.sco | 2 +- scripts/makefile.sggcc | 2 +- scripts/makefile.sgi | 2 +- scripts/makefile.so9 | 2 +- scripts/makefile.solaris | 2 +- scripts/makefile.solaris-x86 | 5 +- scripts/makefile.vcawin32 | 6 +- scripts/pngos2.def | 2 +- scripts/pngw32.def | 2 +- 60 files changed, 204 insertions(+), 11915 deletions(-) rename libpng-1.2.20rc3.txt => libpng-1.2.20rc4.txt (94%) diff --git a/ANNOUNCE b/ANNOUNCE index 2c813538..0a7187c6 100644 --- a/ANNOUNCE +++ b/ANNOUNCE @@ -1,5 +1,5 @@ -Libpng 1.2.20rc3 - August 30, 2007 +Libpng 1.2.20rc4 - September 1, 2007 This is not intended to be a public release. It will be replaced within a few weeks by a public version or by another test version. @@ -9,32 +9,32 @@ Files available for download: Source files with LF line endings (for Unix/Linux) and with a "configure" script - libpng-1.2.20rc3.tar.gz - libpng-1.2.20rc3.tar.bz2 + libpng-1.2.20rc4.tar.gz + libpng-1.2.20rc4.tar.bz2 Source files with LF line endings (for Unix/Linux) without the "configure" script - libpng-1.2.20rc3-no-config.tar.gz - libpng-1.2.20rc3-no-config.tar.bz2 + libpng-1.2.20rc4-no-config.tar.gz + libpng-1.2.20rc4-no-config.tar.bz2 Source files with CRLF line endings (for Windows), without the "configure" script - lp1220r03.zip - lp1220r03.tar.bz2 + lp1220r04.zip + lp1220r04.tar.bz2 Project files - libpng-1.2.20rc3-project-netware.zip - libpng-1.2.20rc3-project-wince.zip + libpng-1.2.20rc4-project-netware.zip + libpng-1.2.20rc4-project-wince.zip Other information: - libpng-1.2.20rc3-README.txt - libpng-1.2.20rc3-KNOWNBUGS.txt - libpng-1.2.20rc3-LICENSE.txt - libpng-1.2.20rc3-Y2K-compliance.txt + libpng-1.2.20rc4-README.txt + libpng-1.2.20rc4-KNOWNBUGS.txt + libpng-1.2.20rc4-LICENSE.txt + libpng-1.2.20rc4-Y2K-compliance.txt Changes since the last public release (1.2.19): @@ -62,13 +62,16 @@ version 1.2.20rc2 [August 27, 2007] Revised #ifdefs to ensure one and only one of pnggccrd.c, pngvcrd.c, or part of pngrutil.c is selected. -version 1.2.20rc3 [August 30, 2007] +version 1.2.20rc3 [September 1, 2007] Remove a little more code in pngwutil.c when PNG_NO_WRITE_FILTER is selected. Added /D _CRT_SECURE_NO_WARNINGS to visual6c and visualc71 projects. Compile png_mmx_support() in png.c even when PNG_NO_MMX_CODE is defined. Restored a "superfluous" #ifdef that was removed from 1.2.20rc2 pnggccrd.c, breaking the png_mmx_support() function. +version 1.2.20rc4 [September 1, 2007] + Removed Intel contributions (MMX, Optimized C). + Send comments/corrections/commendations to png-mng-implement at lists.sf.net (subscription required; visit diff --git a/CHANGES b/CHANGES index e09eeb34..ffb954e2 100644 --- a/CHANGES +++ b/CHANGES @@ -1926,6 +1926,9 @@ version 1.2.20rc3 [August 30, 2007] Restored a "superfluous" #ifdef that was removed from 1.2.20rc2 pnggccrd.c, breaking the png_mmx_support() function. +version 1.2.20rc4 [September 1, 2007] + Removed Intel contributions (MMX, Optimized C). + Send comments/corrections/commendations to png-mng-implement at lists.sf.net (subscription required; visit https://lists.sourceforge.net/lists/listinfo/png-mng-implement diff --git a/INSTALL b/INSTALL index ed8b5132..75f58e81 100644 --- a/INSTALL +++ b/INSTALL @@ -1,5 +1,5 @@ -Installing libpng version 1.2.20rc3 - August 30, 2007 +Installing libpng version 1.2.20rc4 - September 1, 2007 On Unix/Linux and similar systems, you can simply type @@ -44,7 +44,7 @@ to have access to the zlib.h and zconf.h include files that correspond to the version of zlib that's installed. You can rename the directories that you downloaded (they -might be called "libpng-1.2.20rc3" or "lpng109" and "zlib-1.2.1" +might be called "libpng-1.2.20rc4" or "lpng109" and "zlib-1.2.1" or "zlib121") so that you have directories called "zlib" and "libpng". Your directory structure should look like this: @@ -101,15 +101,9 @@ include CMakeLists.txt => "cmake" script makefile.std => Generic UNIX makefile (cc, creates static libpng.a) makefile.elf => Linux/ELF makefile symbol versioning, - gcc, creates libpng12.so.0.1.2.20rc3) + gcc, creates libpng12.so.0.1.2.20rc4) makefile.linux => Linux/ELF makefile - (gcc, creates libpng12.so.0.1.2.20rc3) - makefile.gcmmx => Linux/ELF makefile - (gcc, creates libpng12.so.0.1.2.20rc3, - uses assembler code tuned for Intel MMX platform) - makefile.nommx => Linux/ELF makefile - (gcc, creates libpng12.so.0.1.2.20rc3 - does not use Intel MMX assembler code) + (gcc, creates libpng12.so.0.1.2.20rc4) makefile.gcc => Generic makefile (gcc, creates static libpng.a) makefile.knr => Archaic UNIX Makefile that converts files with ansi2knr (Requires ansi2knr.c from @@ -131,14 +125,14 @@ include makefile.openbsd => OpenBSD makefile makefile.sgi => Silicon Graphics IRIX makefile (cc, creates static lib) makefile.sggcc => Silicon Graphics (gcc, - creates libpng12.so.0.1.2.20rc3) + creates libpng12.so.0.1.2.20rc4) makefile.sunos => Sun makefile makefile.solaris => Solaris 2.X makefile (gcc, - creates libpng12.so.0.1.2.20rc3) + creates libpng12.so.0.1.2.20rc4) makefile.solaris-x86 => Solaris/intelMMX 2.X makefile (gcc, - creates libpng12.so.0.1.2.20rc3) + creates libpng12.so.0.1.2.20rc4) makefile.so9 => Solaris 9 makefile (gcc, - creates libpng12.so.0.1.2.20rc3) + creates libpng12.so.0.1.2.20rc4) makefile.32sunu => Sun Ultra 32-bit makefile makefile.64sunu => Sun Ultra 64-bit makefile makefile.sco => For SCO OSr5 ELF and Unixware 7 with Native cc @@ -154,10 +148,7 @@ include makefile.tc3 => Turbo C 3.0 makefile makefile.dj2 => DJGPP 2 makefile makefile.msc => Microsoft C makefile - makefile.vcawin32 => makefile for Microsoft Visual C++ 5.0 and later (uses - assembler code tuned for Intel MMX platform) - makefile.vcwin32 => makefile for Microsoft Visual C++ 4.0 and later (does - not use assembler code) + makefile.vcwin32 => makefile for Microsoft Visual C++ 4.0 and later makefile.os2 => OS/2 Makefile (gcc and emx, requires pngos2.def) pngos2.def => OS/2 module definition file used by makefile.os2 makefile.watcom => Watcom 10a+ Makefile, 32-bit flat memory model diff --git a/KNOWNBUG b/KNOWNBUG index 02451232..b1a946ce 100644 --- a/KNOWNBUG +++ b/KNOWNBUG @@ -1,5 +1,5 @@ -Known bugs in libpng version 1.2.20rc3 +Known bugs in libpng version 1.2.20rc4 1. April 22, 2001: pnggccrd.c has been reported to crash on NetBSD when reading interlaced PNG files, when assembler code is enabled but running diff --git a/LICENSE b/LICENSE index 9f9b1221..15011523 100644 --- a/LICENSE +++ b/LICENSE @@ -8,7 +8,7 @@ COPYRIGHT NOTICE, DISCLAIMER, and LICENSE: If you modify libpng you may insert additional notices immediately following this sentence. -libpng versions 1.2.6, August 15, 2004, through 1.2.20rc3, August 30, 2007, are +libpng versions 1.2.6, August 15, 2004, through 1.2.20rc4, September 1, 2007, are Copyright (c) 2004, 2006-2007 Glenn Randers-Pehrson, and are distributed according to the same disclaimer and license as libpng-1.2.5 with the following individual added to the list of Contributing Authors @@ -106,4 +106,4 @@ certification mark of the Open Source Initiative. Glenn Randers-Pehrson glennrp at users.sourceforge.net -August 30, 2007 +September 1, 2007 diff --git a/Makefile.am b/Makefile.am index de7dd311..549c2c7b 100644 --- a/Makefile.am +++ b/Makefile.am @@ -33,7 +33,7 @@ lib_LTLIBRARIES=libpng12.la @compatlib@ EXTRA_LTLIBRARIES= libpng.la libpng12_la_SOURCES = png.c pngset.c pngget.c pngrutil.c pngtrans.c pngwutil.c \ pngread.c pngrio.c pngwio.c pngwrite.c pngrtran.c \ - pngwtran.c pngmem.c pngerror.c pngpread.c pnggccrd.c \ + pngwtran.c pngmem.c pngerror.c pngpread.c \ png.h pngconf.h libpng_la_SOURCES = $(libpng12_la_SOURCES) @@ -83,7 +83,7 @@ EXTRA_DIST= \ ${srcdir}/contrib/pngsuite/* \ ${srcdir}/contrib/visupng/* \ $(TESTS) \ - example.c libpng.txt pngvcrd.c + example.c libpng.txt CLEANFILES= pngout.png libpng12.pc libpng12-config libpng.vers \ libpng.sym diff --git a/Makefile.in b/Makefile.in index 41913eba..a56378ad 100644 --- a/Makefile.in +++ b/Makefile.in @@ -87,8 +87,7 @@ am__objects_1 = libpng_la-png.lo libpng_la-pngset.lo \ libpng_la-pngread.lo libpng_la-pngrio.lo libpng_la-pngwio.lo \ libpng_la-pngwrite.lo libpng_la-pngrtran.lo \ libpng_la-pngwtran.lo libpng_la-pngmem.lo \ - libpng_la-pngerror.lo libpng_la-pngpread.lo \ - libpng_la-pnggccrd.lo + libpng_la-pngerror.lo libpng_la-pngpread.lo am_libpng_la_OBJECTS = $(am__objects_1) libpng_la_OBJECTS = $(am_libpng_la_OBJECTS) libpng12_la_LIBADD = @@ -99,7 +98,7 @@ am_libpng12_la_OBJECTS = libpng12_la-png.lo libpng12_la-pngset.lo \ libpng12_la-pngwio.lo libpng12_la-pngwrite.lo \ libpng12_la-pngrtran.lo libpng12_la-pngwtran.lo \ libpng12_la-pngmem.lo libpng12_la-pngerror.lo \ - libpng12_la-pngpread.lo libpng12_la-pnggccrd.lo + libpng12_la-pngpread.lo libpng12_la_OBJECTS = $(am_libpng12_la_OBJECTS) am_pngtest_OBJECTS = pngtest.$(OBJEXT) pngtest_OBJECTS = $(am_pngtest_OBJECTS) @@ -184,7 +183,6 @@ INSTALL_STRIP_PROGRAM = @INSTALL_STRIP_PROGRAM@ LDFLAGS = @LDFLAGS@ LIBOBJS = @LIBOBJS@ LIBPNG_DEFINES = @LIBPNG_DEFINES@ -LIBPNG_NO_MMX = @LIBPNG_NO_MMX@ LIBS = @LIBS@ LIBTOOL = @LIBTOOL@ LN_S = @LN_S@ @@ -286,7 +284,7 @@ lib_LTLIBRARIES = libpng12.la @compatlib@ EXTRA_LTLIBRARIES = libpng.la libpng12_la_SOURCES = png.c pngset.c pngget.c pngrutil.c pngtrans.c pngwutil.c \ pngread.c pngrio.c pngwio.c pngwrite.c pngrtran.c \ - pngwtran.c pngmem.c pngerror.c pngpread.c pnggccrd.c \ + pngwtran.c pngmem.c pngerror.c pngpread.c \ png.h pngconf.h libpng_la_SOURCES = $(libpng12_la_SOURCES) @@ -322,7 +320,7 @@ EXTRA_DIST = \ ${srcdir}/contrib/pngsuite/* \ ${srcdir}/contrib/visupng/* \ $(TESTS) \ - example.c libpng.txt pngvcrd.c + example.c libpng.txt CLEANFILES = pngout.png libpng12.pc libpng12-config libpng.vers \ libpng.sym @@ -455,7 +453,6 @@ distclean-compile: @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng12_la-png.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng12_la-pngerror.Plo@am__quote@ -@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng12_la-pnggccrd.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng12_la-pngget.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng12_la-pngmem.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng12_la-pngpread.Plo@am__quote@ @@ -471,7 +468,6 @@ distclean-compile: @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng12_la-pngwutil.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng_la-png.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng_la-pngerror.Plo@am__quote@ -@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng_la-pnggccrd.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng_la-pngget.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng_la-pngmem.Plo@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpng_la-pngpread.Plo@am__quote@ @@ -613,13 +609,6 @@ libpng_la-pngpread.lo: pngpread.c @AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@ @am__fastdepCC_FALSE@ $(LIBTOOL) --mode=compile --tag=CC $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libpng_la_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o libpng_la-pngpread.lo `test -f 'pngpread.c' || echo '$(srcdir)/'`pngpread.c -libpng_la-pnggccrd.lo: pnggccrd.c -@am__fastdepCC_TRUE@ if $(LIBTOOL) --mode=compile --tag=CC $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libpng_la_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT libpng_la-pnggccrd.lo -MD -MP -MF "$(DEPDIR)/libpng_la-pnggccrd.Tpo" -c -o libpng_la-pnggccrd.lo `test -f 'pnggccrd.c' || echo '$(srcdir)/'`pnggccrd.c; \ -@am__fastdepCC_TRUE@ then mv -f "$(DEPDIR)/libpng_la-pnggccrd.Tpo" "$(DEPDIR)/libpng_la-pnggccrd.Plo"; else rm -f "$(DEPDIR)/libpng_la-pnggccrd.Tpo"; exit 1; fi -@AMDEP_TRUE@@am__fastdepCC_FALSE@ source='pnggccrd.c' object='libpng_la-pnggccrd.lo' libtool=yes @AMDEPBACKSLASH@ -@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@ -@am__fastdepCC_FALSE@ $(LIBTOOL) --mode=compile --tag=CC $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libpng_la_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o libpng_la-pnggccrd.lo `test -f 'pnggccrd.c' || echo '$(srcdir)/'`pnggccrd.c - libpng12_la-png.lo: png.c @am__fastdepCC_TRUE@ if $(LIBTOOL) --mode=compile --tag=CC $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libpng12_la_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT libpng12_la-png.lo -MD -MP -MF "$(DEPDIR)/libpng12_la-png.Tpo" -c -o libpng12_la-png.lo `test -f 'png.c' || echo '$(srcdir)/'`png.c; \ @am__fastdepCC_TRUE@ then mv -f "$(DEPDIR)/libpng12_la-png.Tpo" "$(DEPDIR)/libpng12_la-png.Plo"; else rm -f "$(DEPDIR)/libpng12_la-png.Tpo"; exit 1; fi @@ -725,13 +714,6 @@ libpng12_la-pngpread.lo: pngpread.c @AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@ @am__fastdepCC_FALSE@ $(LIBTOOL) --mode=compile --tag=CC $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libpng12_la_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o libpng12_la-pngpread.lo `test -f 'pngpread.c' || echo '$(srcdir)/'`pngpread.c -libpng12_la-pnggccrd.lo: pnggccrd.c -@am__fastdepCC_TRUE@ if $(LIBTOOL) --mode=compile --tag=CC $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libpng12_la_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -MT libpng12_la-pnggccrd.lo -MD -MP -MF "$(DEPDIR)/libpng12_la-pnggccrd.Tpo" -c -o libpng12_la-pnggccrd.lo `test -f 'pnggccrd.c' || echo '$(srcdir)/'`pnggccrd.c; \ -@am__fastdepCC_TRUE@ then mv -f "$(DEPDIR)/libpng12_la-pnggccrd.Tpo" "$(DEPDIR)/libpng12_la-pnggccrd.Plo"; else rm -f "$(DEPDIR)/libpng12_la-pnggccrd.Tpo"; exit 1; fi -@AMDEP_TRUE@@am__fastdepCC_FALSE@ source='pnggccrd.c' object='libpng12_la-pnggccrd.lo' libtool=yes @AMDEPBACKSLASH@ -@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@ -@am__fastdepCC_FALSE@ $(LIBTOOL) --mode=compile --tag=CC $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libpng12_la_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS) -c -o libpng12_la-pnggccrd.lo `test -f 'pnggccrd.c' || echo '$(srcdir)/'`pnggccrd.c - mostlyclean-libtool: -rm -f *.lo diff --git a/README b/README index dec2d45d..6ae9daf7 100644 --- a/README +++ b/README @@ -1,4 +1,4 @@ -README for libpng version 1.2.20rc3 - August 30, 2007 (shared library 12.0) +README for libpng version 1.2.20rc4 - September 1, 2007 (shared library 12.0) See the note about version numbers near the top of png.h See INSTALL for instructions on how to install libpng. @@ -190,11 +190,11 @@ Files in this distribution: descrip.mms => VMS makefile for MMS or MMK makefile.std => Generic UNIX makefile (cc, creates static libpng.a) makefile.elf => Linux/ELF makefile symbol versioning, - gcc, creates libpng12.so.0.1.2.20rc3) + gcc, creates libpng12.so.0.1.2.20rc4) makefile.linux => Linux/ELF makefile - (gcc, creates libpng12.so.0.1.2.20rc3) + (gcc, creates libpng12.so.0.1.2.20rc4) makefile.gcmmx => Linux/ELF makefile - (gcc, creates libpng12.so.0.1.2.20rc3, + (gcc, creates libpng12.so.0.1.2.20rc4, uses assembler code tuned for Intel MMX platform) makefile.gcc => Generic makefile (gcc, creates static libpng.a) makefile.knr => Archaic UNIX Makefile that converts files with @@ -216,12 +216,12 @@ Files in this distribution: makefile.openbsd => OpenBSD makefile makefile.sgi => Silicon Graphics IRIX (cc, creates static lib) makefile.sggcc => Silicon Graphics - (gcc, creates libpng12.so.0.1.2.20rc3) + (gcc, creates libpng12.so.0.1.2.20rc4) makefile.sunos => Sun makefile makefile.solaris => Solaris 2.X makefile - (gcc, creates libpng12.so.0.1.2.20rc3) + (gcc, creates libpng12.so.0.1.2.20rc4) makefile.so9 => Solaris 9 makefile - (gcc, creates libpng12.so.0.1.2.20rc3) + (gcc, creates libpng12.so.0.1.2.20rc4) makefile.32sunu => Sun Ultra 32-bit makefile makefile.64sunu => Sun Ultra 64-bit makefile makefile.sco => For SCO OSr5 ELF and Unixware 7 with Native cc diff --git a/Y2KINFO b/Y2KINFO index af33e0ed..52cb27b1 100644 --- a/Y2KINFO +++ b/Y2KINFO @@ -1,13 +1,13 @@ Y2K compliance in libpng: ========================= - August 30, 2007 + September 1, 2007 Since the PNG Development group is an ad-hoc body, we can't make an official declaration. This is your unofficial assurance that libpng from version 0.71 and - upward through 1.2.20rc3 are Y2K compliant. It is my belief that earlier + upward through 1.2.20rc4 are Y2K compliant. It is my belief that earlier versions were also Y2K compliant. Libpng only has three year fields. One is a 2-byte unsigned integer diff --git a/configure b/configure index c60ff403..bfaf6366 100755 --- a/configure +++ b/configure @@ -1,6 +1,6 @@ #! /bin/sh # Guess values for system-dependent variables and create Makefiles. -# Generated by GNU Autoconf 2.61 for libpng 1.2.20rc3. +# Generated by GNU Autoconf 2.61 for libpng 1.2.20rc4. # # Report bugs to . # @@ -728,8 +728,8 @@ SHELL=${CONFIG_SHELL-/bin/sh} # Identity of this package. PACKAGE_NAME='libpng' PACKAGE_TARNAME='libpng' -PACKAGE_VERSION='1.2.20rc3' -PACKAGE_STRING='libpng 1.2.20rc3' +PACKAGE_VERSION='1.2.20rc4' +PACKAGE_STRING='libpng 1.2.20rc4' PACKAGE_BUGREPORT='png-mng-implement@lists.sourceforge.net' ac_unique_file="pngget.c" @@ -876,7 +876,6 @@ LIBTOOL POW_LIB LIBOBJS LIBPNG_DEFINES -LIBPNG_NO_MMX HAVE_LD_VERSION_SCRIPT_TRUE HAVE_LD_VERSION_SCRIPT_FALSE PNGLIB_VERSION @@ -1405,7 +1404,7 @@ if test "$ac_init_help" = "long"; then # Omit some internal or obsolete options to make the list less imposing. # This message is too long to be a string in the A/UX 3.1 sh. cat <<_ACEOF -\`configure' configures libpng 1.2.20rc3 to adapt to many kinds of systems. +\`configure' configures libpng 1.2.20rc4 to adapt to many kinds of systems. Usage: $0 [OPTION]... [VAR=VALUE]... @@ -1475,7 +1474,7 @@ fi if test -n "$ac_init_help"; then case $ac_init_help in - short | recursive ) echo "Configuration of libpng 1.2.20rc3:";; + short | recursive ) echo "Configuration of libpng 1.2.20rc4:";; esac cat <<\_ACEOF @@ -1585,7 +1584,7 @@ fi test -n "$ac_init_help" && exit $ac_status if $ac_init_version; then cat <<\_ACEOF -libpng configure 1.2.20rc3 +libpng configure 1.2.20rc4 generated by GNU Autoconf 2.61 Copyright (C) 1992, 1993, 1994, 1995, 1996, 1998, 1999, 2000, 2001, @@ -1599,7 +1598,7 @@ cat >config.log <<_ACEOF This file contains any messages produced by compilers while running configure, to aid debugging if configure makes a mistake. -It was created by libpng $as_me 1.2.20rc3, which was +It was created by libpng $as_me 1.2.20rc4, which was generated by GNU Autoconf 2.61. Invocation command line was $ $0 $@ @@ -2269,7 +2268,7 @@ fi # Define the identity of the package. PACKAGE='libpng' - VERSION='1.2.20rc3' + VERSION='1.2.20rc4' cat >>confdefs.h <<_ACEOF @@ -2440,7 +2439,7 @@ fi -PNGLIB_VERSION=1.2.20rc3 +PNGLIB_VERSION=1.2.20rc4 PNGLIB_MAJOR=1 PNGLIB_MINOR=2 PNGLIB_RELEASE=20 @@ -4788,7 +4787,7 @@ ia64-*-hpux*) ;; *-*-irix6*) # Find out which ABI we are using. - echo '#line 4791 "configure"' > conftest.$ac_ext + echo '#line 4790 "configure"' > conftest.$ac_ext if { (eval echo "$as_me:$LINENO: \"$ac_compile\"") >&5 (eval $ac_compile) 2>&5 ac_status=$? @@ -7301,11 +7300,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:7304: $lt_compile\"" >&5) + (eval echo "\"\$as_me:7303: $lt_compile\"" >&5) (eval "$lt_compile" 2>conftest.err) ac_status=$? cat conftest.err >&5 - echo "$as_me:7308: \$? = $ac_status" >&5 + echo "$as_me:7307: \$? = $ac_status" >&5 if (exit $ac_status) && test -s "$ac_outfile"; then # The compiler can only warn and ignore the option if not recognized # So say no if there are warnings other than the usual output. @@ -7591,11 +7590,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:7594: $lt_compile\"" >&5) + (eval echo "\"\$as_me:7593: $lt_compile\"" >&5) (eval "$lt_compile" 2>conftest.err) ac_status=$? cat conftest.err >&5 - echo "$as_me:7598: \$? = $ac_status" >&5 + echo "$as_me:7597: \$? = $ac_status" >&5 if (exit $ac_status) && test -s "$ac_outfile"; then # The compiler can only warn and ignore the option if not recognized # So say no if there are warnings other than the usual output. @@ -7695,11 +7694,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:7698: $lt_compile\"" >&5) + (eval echo "\"\$as_me:7697: $lt_compile\"" >&5) (eval "$lt_compile" 2>out/conftest.err) ac_status=$? cat out/conftest.err >&5 - echo "$as_me:7702: \$? = $ac_status" >&5 + echo "$as_me:7701: \$? = $ac_status" >&5 if (exit $ac_status) && test -s out/conftest2.$ac_objext then # The compiler can only warn and ignore the option if not recognized @@ -10044,7 +10043,7 @@ else lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2 lt_status=$lt_dlunknown cat > conftest.$ac_ext < conftest.$ac_ext <&5) + (eval echo "\"\$as_me:12566: $lt_compile\"" >&5) (eval "$lt_compile" 2>conftest.err) ac_status=$? cat conftest.err >&5 - echo "$as_me:12571: \$? = $ac_status" >&5 + echo "$as_me:12570: \$? = $ac_status" >&5 if (exit $ac_status) && test -s "$ac_outfile"; then # The compiler can only warn and ignore the option if not recognized # So say no if there are warnings other than the usual output. @@ -12668,11 +12667,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:12671: $lt_compile\"" >&5) + (eval echo "\"\$as_me:12670: $lt_compile\"" >&5) (eval "$lt_compile" 2>out/conftest.err) ac_status=$? cat out/conftest.err >&5 - echo "$as_me:12675: \$? = $ac_status" >&5 + echo "$as_me:12674: \$? = $ac_status" >&5 if (exit $ac_status) && test -s out/conftest2.$ac_objext then # The compiler can only warn and ignore the option if not recognized @@ -14230,11 +14229,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:14233: $lt_compile\"" >&5) + (eval echo "\"\$as_me:14232: $lt_compile\"" >&5) (eval "$lt_compile" 2>conftest.err) ac_status=$? cat conftest.err >&5 - echo "$as_me:14237: \$? = $ac_status" >&5 + echo "$as_me:14236: \$? = $ac_status" >&5 if (exit $ac_status) && test -s "$ac_outfile"; then # The compiler can only warn and ignore the option if not recognized # So say no if there are warnings other than the usual output. @@ -14334,11 +14333,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:14337: $lt_compile\"" >&5) + (eval echo "\"\$as_me:14336: $lt_compile\"" >&5) (eval "$lt_compile" 2>out/conftest.err) ac_status=$? cat out/conftest.err >&5 - echo "$as_me:14341: \$? = $ac_status" >&5 + echo "$as_me:14340: \$? = $ac_status" >&5 if (exit $ac_status) && test -s out/conftest2.$ac_objext then # The compiler can only warn and ignore the option if not recognized @@ -16521,11 +16520,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:16524: $lt_compile\"" >&5) + (eval echo "\"\$as_me:16523: $lt_compile\"" >&5) (eval "$lt_compile" 2>conftest.err) ac_status=$? cat conftest.err >&5 - echo "$as_me:16528: \$? = $ac_status" >&5 + echo "$as_me:16527: \$? = $ac_status" >&5 if (exit $ac_status) && test -s "$ac_outfile"; then # The compiler can only warn and ignore the option if not recognized # So say no if there are warnings other than the usual output. @@ -16811,11 +16810,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:16814: $lt_compile\"" >&5) + (eval echo "\"\$as_me:16813: $lt_compile\"" >&5) (eval "$lt_compile" 2>conftest.err) ac_status=$? cat conftest.err >&5 - echo "$as_me:16818: \$? = $ac_status" >&5 + echo "$as_me:16817: \$? = $ac_status" >&5 if (exit $ac_status) && test -s "$ac_outfile"; then # The compiler can only warn and ignore the option if not recognized # So say no if there are warnings other than the usual output. @@ -16915,11 +16914,11 @@ else -e 's:.*FLAGS}\{0,1\} :&$lt_compiler_flag :; t' \ -e 's: [^ ]*conftest\.: $lt_compiler_flag&:; t' \ -e 's:$: $lt_compiler_flag:'` - (eval echo "\"\$as_me:16918: $lt_compile\"" >&5) + (eval echo "\"\$as_me:16917: $lt_compile\"" >&5) (eval "$lt_compile" 2>out/conftest.err) ac_status=$? cat out/conftest.err >&5 - echo "$as_me:16922: \$? = $ac_status" >&5 + echo "$as_me:16921: \$? = $ac_status" >&5 if (exit $ac_status) && test -s out/conftest2.$ac_objext then # The compiler can only warn and ignore the option if not recognized @@ -20713,55 +20712,6 @@ fi LIBPNG_DEFINES=-DPNG_CONFIGURE_LIBPNG -{ echo "$as_me:$LINENO: checking if assembler code in pnggccrd.c can be compiled without PNG_NO_MMX_CODE" >&5 -echo $ECHO_N "checking if assembler code in pnggccrd.c can be compiled without PNG_NO_MMX_CODE... $ECHO_C" >&6; } -cat >conftest.$ac_ext <<_ACEOF -/* confdefs.h. */ -_ACEOF -cat confdefs.h >>conftest.$ac_ext -cat >>conftest.$ac_ext <<_ACEOF -/* end confdefs.h. */ -#include "$srcdir/pnggccrd.c" -int -main () -{ -return 0; - ; - return 0; -} -_ACEOF -rm -f conftest.$ac_objext -if { (ac_try="$ac_compile" -case "(($ac_try" in - *\"* | *\`* | *\\*) ac_try_echo=\$ac_try;; - *) ac_try_echo=$ac_try;; -esac -eval "echo \"\$as_me:$LINENO: $ac_try_echo\"") >&5 - (eval "$ac_compile") 2>conftest.er1 - ac_status=$? - grep -v '^ *+' conftest.er1 >conftest.err - rm -f conftest.er1 - cat conftest.err >&5 - echo "$as_me:$LINENO: \$? = $ac_status" >&5 - (exit $ac_status); } && { - test -z "$ac_c_werror_flag" || - test ! -s conftest.err - } && test -s conftest.$ac_objext; then - { echo "$as_me:$LINENO: result: yes" >&5 -echo "${ECHO_T}yes" >&6; } - LIBPNG_NO_MMX="" -else - echo "$as_me: failed program was:" >&5 -sed 's/^/| /' conftest.$ac_ext >&5 - - { echo "$as_me:$LINENO: result: no" >&5 -echo "${ECHO_T}no" >&6; } - LIBPNG_NO_MMX=-DPNG_NO_MMX_CODE -fi - -rm -f core conftest.err conftest.$ac_objext conftest.$ac_ext -LIBPNG_DEFINES=$LIBPNG_DEFINES\ $LIBPNG_NO_MMX - { echo "$as_me:$LINENO: checking if libraries can be versioned" >&5 @@ -21282,7 +21232,7 @@ exec 6>&1 # report actual input values of CONFIG_FILES etc. instead of their # values after options handling. ac_log=" -This file was extended by libpng $as_me 1.2.20rc3, which was +This file was extended by libpng $as_me 1.2.20rc4, which was generated by GNU Autoconf 2.61. Invocation command line was CONFIG_FILES = $CONFIG_FILES @@ -21335,7 +21285,7 @@ Report bugs to ." _ACEOF cat >>$CONFIG_STATUS <<_ACEOF ac_cs_version="\\ -libpng config.status 1.2.20rc3 +libpng config.status 1.2.20rc4 configured by $0, generated by GNU Autoconf 2.61, with options \\"`echo "$ac_configure_args" | sed 's/^ //; s/[\\""\`\$]/\\\\&/g'`\\" @@ -21663,7 +21613,6 @@ LIBTOOL!$LIBTOOL$ac_delim POW_LIB!$POW_LIB$ac_delim LIBOBJS!$LIBOBJS$ac_delim LIBPNG_DEFINES!$LIBPNG_DEFINES$ac_delim -LIBPNG_NO_MMX!$LIBPNG_NO_MMX$ac_delim HAVE_LD_VERSION_SCRIPT_TRUE!$HAVE_LD_VERSION_SCRIPT_TRUE$ac_delim HAVE_LD_VERSION_SCRIPT_FALSE!$HAVE_LD_VERSION_SCRIPT_FALSE$ac_delim PNGLIB_VERSION!$PNGLIB_VERSION$ac_delim @@ -21676,7 +21625,7 @@ compatlib!$compatlib$ac_delim LTLIBOBJS!$LTLIBOBJS$ac_delim _ACEOF - if test `sed -n "s/.*$ac_delim\$/X/p" conf$$subs.sed | grep -c X` = 21; then + if test `sed -n "s/.*$ac_delim\$/X/p" conf$$subs.sed | grep -c X` = 20; then break elif $ac_last_try; then { { echo "$as_me:$LINENO: error: could not make $CONFIG_STATUS" >&5 diff --git a/configure.ac b/configure.ac index c7a98e2c..244952c5 100644 --- a/configure.ac +++ b/configure.ac @@ -18,12 +18,12 @@ AC_PREREQ(2.59) dnl Version number stuff here: -AC_INIT([libpng], [1.2.20rc3], [png-mng-implement@lists.sourceforge.net]) +AC_INIT([libpng], [1.2.20rc4], [png-mng-implement@lists.sourceforge.net]) AM_INIT_AUTOMAKE dnl stop configure from automagically running automake AM_MAINTAINER_MODE -PNGLIB_VERSION=1.2.20rc3 +PNGLIB_VERSION=1.2.20rc4 PNGLIB_MAJOR=1 PNGLIB_MINOR=2 PNGLIB_RELEASE=20 @@ -59,18 +59,7 @@ AC_CHECK_FUNCS([pow], , AC_CHECK_LIB(m, pow, , AC_ERROR([cannot find pow])) ) AC_CHECK_LIB(z, zlibVersion, , AC_ERROR([zlib not installed])) LIBPNG_DEFINES=-DPNG_CONFIGURE_LIBPNG -AC_MSG_CHECKING( - [if assembler code in pnggccrd.c can be compiled without PNG_NO_MMX_CODE]) -AC_TRY_COMPILE( - [#include "$srcdir/pnggccrd.c"], - [return 0;], - AC_MSG_RESULT(yes) - LIBPNG_NO_MMX="", - AC_MSG_RESULT(no) - LIBPNG_NO_MMX=-DPNG_NO_MMX_CODE) -LIBPNG_DEFINES=$LIBPNG_DEFINES\ $LIBPNG_NO_MMX AC_SUBST(LIBPNG_DEFINES) -AC_SUBST(LIBPNG_NO_MMX) AC_MSG_CHECKING([if libraries can be versioned]) GLD=`$LD --help < /dev/null 2>/dev/null | grep version-script` diff --git a/libpng-1.2.20rc3.txt b/libpng-1.2.20rc4.txt similarity index 94% rename from libpng-1.2.20rc3.txt rename to libpng-1.2.20rc4.txt index ce051241..ba4d9ab7 100644 --- a/libpng-1.2.20rc3.txt +++ b/libpng-1.2.20rc4.txt @@ -1,6 +1,6 @@ libpng.txt - A description on how to use and modify libpng - libpng version 1.2.20rc3 - August 30, 2007 + libpng version 1.2.20rc4 - September 1, 2007 Updated and distributed by Glenn Randers-Pehrson Copyright (c) 1998-2007 Glenn Randers-Pehrson @@ -86,9 +86,7 @@ Libpng is thread safe, provided the threads are using different instances of the structures. Each thread should have its own png_struct and png_info instances, and thus its own image. Libpng does not protect itself against two threads using the -same instance of a structure. Note: thread safety may be defeated -by use of some of the MMX assembler code in pnggccrd.c, which is only -compiled when the user defines PNG_THREAD_UNSAFE_OK. +same instance of a structure. II. Structures @@ -2355,17 +2353,13 @@ For a more compact example of writing a PNG image, see the file example.c. V. Modifying/Customizing libpng: -There are three issues here. The first is changing how libpng does +There are two issues here. The first is changing how libpng does standard things like memory allocation, input/output, and error handling. The second deals with more complicated things like adding new chunks, adding new transformations, and generally changing how libpng works. Both of those are compile-time issues; that is, they are generally determined at the time the code is written, and there is rarely a need -to provide the user with a means of changing them. The third is a -run-time issue: choosing between and/or tuning one or more alternate -versions of computationally intensive routines; specifically, optimized -assembly-language (and therefore compiler- and platform-dependent) -versions. +to provide the user with a means of changing them. Memory allocation, input/output, and error handling @@ -2726,172 +2720,6 @@ When PNG_DEBUG = 1, the macros are defined, but only png_debug statements having level = 0 will be printed. There aren't any such statements in this version of libpng, but if you insert some they will be printed. -VI. Runtime optimization - -A new feature in libpng 1.2.0 is the ability to dynamically switch between -standard and optimized versions of some routines. Currently these are -limited to three computationally intensive tasks when reading PNG files: -decoding row filters, expanding interlacing, and combining interlaced or -transparent row data with previous row data. Currently the optimized -versions are available only for x86 (Intel, AMD, etc.) platforms with -MMX support, though this may change in future versions. (For example, -the non-MMX assembler optimizations for zlib might become similarly -runtime-selectable in future releases, in which case libpng could be -extended to support them. Alternatively, the compile-time choice of -floating-point versus integer routines for gamma correction might become -runtime-selectable.) - -Because such optimizations tend to be very platform- and compiler-dependent, -both in how they are written and in how they perform, the new runtime code -in libpng has been written to allow programs to query, enable, and disable -either specific optimizations or all such optimizations. For example, to -enable all possible optimizations (bearing in mind that some "optimizations" -may actually run more slowly in rare cases): - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) - png_uint_32 mask, flags; - - flags = png_get_asm_flags(png_ptr); - mask = png_get_asm_flagmask(PNG_SELECT_READ | PNG_SELECT_WRITE); - png_set_asm_flags(png_ptr, flags | mask); - #endif - -To enable only optimizations relevant to reading PNGs, use PNG_SELECT_READ -by itself when calling png_get_asm_flagmask(); similarly for optimizing -only writing. To disable all optimizations: - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) - flags = png_get_asm_flags(png_ptr); - mask = png_get_asm_flagmask(PNG_SELECT_READ | PNG_SELECT_WRITE); - png_set_asm_flags(png_ptr, flags & ~mask); - #endif - -To enable or disable only MMX-related features, use png_get_mmx_flagmask() -in place of png_get_asm_flagmask(). The mmx version takes one additional -parameter: - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) - int selection = PNG_SELECT_READ | PNG_SELECT_WRITE; - int compilerID; - - mask = png_get_mmx_flagmask(selection, &compilerID); - #endif - -On return, compilerID will indicate which version of the MMX assembler -optimizations was compiled. Currently two flavors exist: Microsoft -Visual C++ (compilerID == 1) and GNU C (a.k.a. gcc/gas, compilerID == 2). -On non-x86 platforms or on systems compiled without MMX optimizations, a -value of -1 is used. - -Note that both png_get_asm_flagmask() and png_get_mmx_flagmask() return -all valid, settable optimization bits for the version of the library that's -currently in use. In the case of shared (dynamically linked) libraries, -this may include optimizations that did not exist at the time the code was -written and compiled. It is also possible, of course, to enable only known, -specific optimizations; for example: - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) && \ - defined(PNG_USE_PNGGCCRD) && defined(PNG_ASSEMBLER_CODE_SUPPORTED) - flags = PNG_ASM_FLAG_MMX_READ_COMBINE_ROW \ - | PNG_ASM_FLAG_MMX_READ_INTERLACE \ - | PNG_ASM_FLAG_MMX_READ_FILTER_SUB \ - | PNG_ASM_FLAG_MMX_READ_FILTER_UP \ - | PNG_ASM_FLAG_MMX_READ_FILTER_AVG \ - | PNG_ASM_FLAG_MMX_READ_FILTER_PAETH ; - png_set_asm_flags(png_ptr, flags); - #endif - -This method would enable only the MMX read-optimizations available at the -time of libpng 1.2.0's release, regardless of whether a later version of -the DLL were actually being used. (Also note that these functions did not -exist in versions older than 1.2.0, so any attempt to run a dynamically -linked app on such an older version would fail.) - -To determine whether the processor supports MMX instructions at all, use -the png_mmx_support() function: - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) - mmxsupport = png_mmx_support(); - #endif - -It returns -1 if MMX support is not compiled into libpng, 0 if MMX code -is compiled but MMX is not supported by the processor, or 1 if MMX support -is fully available. Note that png_mmx_support(), png_get_mmx_flagmask(), -and png_get_asm_flagmask() all may be called without allocating and ini- -tializing any PNG structures (for example, as part of a usage screen or -"about" box). - -The following code can be used to prevent an application from using the -thread_unsafe features, even if libpng was built with PNG_THREAD_UNSAFE_OK -defined: - -#if defined(PNG_USE_PNGGCCRD) && defined(PNG_ASSEMBLER_CODE_SUPPORTED) \ - && defined(PNG_THREAD_UNSAFE_OK) - /* Disable thread-unsafe features of pnggccrd */ - if (png_access_version_number() >= 10200) - { - png_uint_32 mmx_disable_mask = 0; - png_uint_32 asm_flags; - - mmx_disable_mask |= ( PNG_ASM_FLAG_MMX_READ_COMBINE_ROW \ - | PNG_ASM_FLAG_MMX_READ_FILTER_SUB \ - | PNG_ASM_FLAG_MMX_READ_FILTER_AVG \ - | PNG_ASM_FLAG_MMX_READ_FILTER_PAETH ); - asm_flags = png_get_asm_flags(png_ptr); - png_set_asm_flags(png_ptr, asm_flags & ~mmx_disable_mask); - } -#endif - -For more extensive examples of runtime querying, enabling and disabling -of optimized features, see contrib/gregbook/readpng2.c in the libpng -source-code distribution. - -It is also possible to disable or enable specific optimization features -at compile time. To disable them entirely, which may result in slower -but smaller and less complex and troublesome code, define - - PNG_NO_ASSEMBLER_CODE - -If you do this, then the run-time code for setting and querying flags -described above, and the assembler code in pnggccrd.c or pngvcrd.c will -not be built. - -If you have disabled the assembler code, you can also disable the -optimized C code, to obtain even slower but smaller code, by defining - - PNG_NO_OPTIMZED_CODE - -To disable only the MMX assembler code, define - - PNG_NO_MMX_CODE - -There are two versions of the MMX code: one for gcc compilers and one for -MSVC compilers. Pngconf.h should automatically detect which you are -using, and it will set either PNG_USE_PNGGCCRD or PNG_USE_PNGVCRD for you. - -Define one or more of the following to disable specific parts of the -assembler code: - - PNG_NO_MMX_COMBINE_ROW - PNG_NO_MMX_READ_INTERLACE - PNG_NO_MMX_READ_FILTER_ROW - -If the latter is not disabled, you can disable one or more of the -individual filter types by defining - - PNG_NO_MMX_FILTER_SUB - PNG_NO_MMX_FILTER_UP - PNG_NO_MMX_FILTER_AVG - PNG_NO_MMX_FILTER_PAETH - -By default, libpng only enables the "sub" filter when gcc-3.x is used -because experiments show that gcc-3.x produces bad code for the others. - -When you disable various MMX features, libpng will still be able to decode -files with those features. It will fall back upon the optimized C code -or the unoptimized C code to decode them. - - VII. MNG support The MNG specification (available at http://www.libpng.org/pub/mng) allows @@ -2968,13 +2796,13 @@ application: IX. Y2K Compliance in libpng -August 30, 2007 +September 1, 2007 Since the PNG Development group is an ad-hoc body, we can't make an official declaration. This is your unofficial assurance that libpng from version 0.71 and -upward through 1.2.20rc3 are Y2K compliant. It is my belief that earlier +upward through 1.2.20rc4 are Y2K compliant. It is my belief that earlier versions were also Y2K compliant. Libpng only has three year fields. One is a 2-byte unsigned integer that diff --git a/libpng.3 b/libpng.3 index d7073413..d80f90bf 100644 --- a/libpng.3 +++ b/libpng.3 @@ -1,6 +1,6 @@ -.TH LIBPNG 3 "August 30, 2007" +.TH LIBPNG 3 "September 1, 2007" .SH NAME -libpng \- Portable Network Graphics (PNG) Reference Library 1.2.20rc3 +libpng \- Portable Network Graphics (PNG) Reference Library 1.2.20rc4 .SH SYNOPSIS \fB #include \fP @@ -410,7 +410,7 @@ Following is a copy of the libpng.txt file that accompanies libpng. .SH LIBPNG.TXT libpng.txt - A description on how to use and modify libpng - libpng version 1.2.20rc3 - August 30, 2007 + libpng version 1.2.20rc4 - September 1, 2007 Updated and distributed by Glenn Randers-Pehrson Copyright (c) 1998-2007 Glenn Randers-Pehrson @@ -496,9 +496,7 @@ Libpng is thread safe, provided the threads are using different instances of the structures. Each thread should have its own png_struct and png_info instances, and thus its own image. Libpng does not protect itself against two threads using the -same instance of a structure. Note: thread safety may be defeated -by use of some of the MMX assembler code in pnggccrd.c, which is only -compiled when the user defines PNG_THREAD_UNSAFE_OK. +same instance of a structure. .SH II. Structures @@ -2765,17 +2763,13 @@ For a more compact example of writing a PNG image, see the file example.c. .SH V. Modifying/Customizing libpng: -There are three issues here. The first is changing how libpng does +There are two issues here. The first is changing how libpng does standard things like memory allocation, input/output, and error handling. The second deals with more complicated things like adding new chunks, adding new transformations, and generally changing how libpng works. Both of those are compile-time issues; that is, they are generally determined at the time the code is written, and there is rarely a need -to provide the user with a means of changing them. The third is a -run-time issue: choosing between and/or tuning one or more alternate -versions of computationally intensive routines; specifically, optimized -assembly-language (and therefore compiler- and platform-dependent) -versions. +to provide the user with a means of changing them. Memory allocation, input/output, and error handling @@ -3136,172 +3130,6 @@ When PNG_DEBUG = 1, the macros are defined, but only png_debug statements having level = 0 will be printed. There aren't any such statements in this version of libpng, but if you insert some they will be printed. -.SH VI. Runtime optimization - -A new feature in libpng 1.2.0 is the ability to dynamically switch between -standard and optimized versions of some routines. Currently these are -limited to three computationally intensive tasks when reading PNG files: -decoding row filters, expanding interlacing, and combining interlaced or -transparent row data with previous row data. Currently the optimized -versions are available only for x86 (Intel, AMD, etc.) platforms with -MMX support, though this may change in future versions. (For example, -the non-MMX assembler optimizations for zlib might become similarly -runtime-selectable in future releases, in which case libpng could be -extended to support them. Alternatively, the compile-time choice of -floating-point versus integer routines for gamma correction might become -runtime-selectable.) - -Because such optimizations tend to be very platform- and compiler-dependent, -both in how they are written and in how they perform, the new runtime code -in libpng has been written to allow programs to query, enable, and disable -either specific optimizations or all such optimizations. For example, to -enable all possible optimizations (bearing in mind that some "optimizations" -may actually run more slowly in rare cases): - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) - png_uint_32 mask, flags; - - flags = png_get_asm_flags(png_ptr); - mask = png_get_asm_flagmask(PNG_SELECT_READ | PNG_SELECT_WRITE); - png_set_asm_flags(png_ptr, flags | mask); - #endif - -To enable only optimizations relevant to reading PNGs, use PNG_SELECT_READ -by itself when calling png_get_asm_flagmask(); similarly for optimizing -only writing. To disable all optimizations: - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) - flags = png_get_asm_flags(png_ptr); - mask = png_get_asm_flagmask(PNG_SELECT_READ | PNG_SELECT_WRITE); - png_set_asm_flags(png_ptr, flags & ~mask); - #endif - -To enable or disable only MMX-related features, use png_get_mmx_flagmask() -in place of png_get_asm_flagmask(). The mmx version takes one additional -parameter: - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) - int selection = PNG_SELECT_READ | PNG_SELECT_WRITE; - int compilerID; - - mask = png_get_mmx_flagmask(selection, &compilerID); - #endif - -On return, compilerID will indicate which version of the MMX assembler -optimizations was compiled. Currently two flavors exist: Microsoft -Visual C++ (compilerID == 1) and GNU C (a.k.a. gcc/gas, compilerID == 2). -On non-x86 platforms or on systems compiled without MMX optimizations, a -value of -1 is used. - -Note that both png_get_asm_flagmask() and png_get_mmx_flagmask() return -all valid, settable optimization bits for the version of the library that's -currently in use. In the case of shared (dynamically linked) libraries, -this may include optimizations that did not exist at the time the code was -written and compiled. It is also possible, of course, to enable only known, -specific optimizations; for example: - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) && \ - defined(PNG_USE_PNGGCCRD) && defined(PNG_ASSEMBLER_CODE_SUPPORTED) - flags = PNG_ASM_FLAG_MMX_READ_COMBINE_ROW \ - | PNG_ASM_FLAG_MMX_READ_INTERLACE \ - | PNG_ASM_FLAG_MMX_READ_FILTER_SUB \ - | PNG_ASM_FLAG_MMX_READ_FILTER_UP \ - | PNG_ASM_FLAG_MMX_READ_FILTER_AVG \ - | PNG_ASM_FLAG_MMX_READ_FILTER_PAETH ; - png_set_asm_flags(png_ptr, flags); - #endif - -This method would enable only the MMX read-optimizations available at the -time of libpng 1.2.0's release, regardless of whether a later version of -the DLL were actually being used. (Also note that these functions did not -exist in versions older than 1.2.0, so any attempt to run a dynamically -linked app on such an older version would fail.) - -To determine whether the processor supports MMX instructions at all, use -the png_mmx_support() function: - - #if defined(PNG_LIBPNG_VER) && (PNG_LIBPNG_VER >= 10200) - mmxsupport = png_mmx_support(); - #endif - -It returns -1 if MMX support is not compiled into libpng, 0 if MMX code -is compiled but MMX is not supported by the processor, or 1 if MMX support -is fully available. Note that png_mmx_support(), png_get_mmx_flagmask(), -and png_get_asm_flagmask() all may be called without allocating and ini- -tializing any PNG structures (for example, as part of a usage screen or -"about" box). - -The following code can be used to prevent an application from using the -thread_unsafe features, even if libpng was built with PNG_THREAD_UNSAFE_OK -defined: - -#if defined(PNG_USE_PNGGCCRD) && defined(PNG_ASSEMBLER_CODE_SUPPORTED) \ - && defined(PNG_THREAD_UNSAFE_OK) - /* Disable thread-unsafe features of pnggccrd */ - if (png_access_version_number() >= 10200) - { - png_uint_32 mmx_disable_mask = 0; - png_uint_32 asm_flags; - - mmx_disable_mask |= ( PNG_ASM_FLAG_MMX_READ_COMBINE_ROW \ - | PNG_ASM_FLAG_MMX_READ_FILTER_SUB \ - | PNG_ASM_FLAG_MMX_READ_FILTER_AVG \ - | PNG_ASM_FLAG_MMX_READ_FILTER_PAETH ); - asm_flags = png_get_asm_flags(png_ptr); - png_set_asm_flags(png_ptr, asm_flags & ~mmx_disable_mask); - } -#endif - -For more extensive examples of runtime querying, enabling and disabling -of optimized features, see contrib/gregbook/readpng2.c in the libpng -source-code distribution. - -It is also possible to disable or enable specific optimization features -at compile time. To disable them entirely, which may result in slower -but smaller and less complex and troublesome code, define - - PNG_NO_ASSEMBLER_CODE - -If you do this, then the run-time code for setting and querying flags -described above, and the assembler code in pnggccrd.c or pngvcrd.c will -not be built. - -If you have disabled the assembler code, you can also disable the -optimized C code, to obtain even slower but smaller code, by defining - - PNG_NO_OPTIMZED_CODE - -To disable only the MMX assembler code, define - - PNG_NO_MMX_CODE - -There are two versions of the MMX code: one for gcc compilers and one for -MSVC compilers. Pngconf.h should automatically detect which you are -using, and it will set either PNG_USE_PNGGCCRD or PNG_USE_PNGVCRD for you. - -Define one or more of the following to disable specific parts of the -assembler code: - - PNG_NO_MMX_COMBINE_ROW - PNG_NO_MMX_READ_INTERLACE - PNG_NO_MMX_READ_FILTER_ROW - -If the latter is not disabled, you can disable one or more of the -individual filter types by defining - - PNG_NO_MMX_FILTER_SUB - PNG_NO_MMX_FILTER_UP - PNG_NO_MMX_FILTER_AVG - PNG_NO_MMX_FILTER_PAETH - -By default, libpng only enables the "sub" filter when gcc-3.x is used -because experiments show that gcc-3.x produces bad code for the others. - -When you disable various MMX features, libpng will still be able to decode -files with those features. It will fall back upon the optimized C code -or the unoptimized C code to decode them. - - .SH VII. MNG support The MNG specification (available at http://www.libpng.org/pub/mng) allows @@ -3378,13 +3206,13 @@ application: .SH IX. Y2K Compliance in libpng -August 30, 2007 +September 1, 2007 Since the PNG Development group is an ad-hoc body, we can't make an official declaration. This is your unofficial assurance that libpng from version 0.71 and -upward through 1.2.20rc3 are Y2K compliant. It is my belief that earlier +upward through 1.2.20rc4 are Y2K compliant. It is my belief that earlier versions were also Y2K compliant. Libpng only has three year fields. One is a 2-byte unsigned integer that @@ -3579,8 +3407,8 @@ the first widely used release: 1.0.27 10 10027 10.so.0.27[.0] 1.2.19 13 10219 12.so.0.19[.0] 1.2.20beta01-04 13 10220 12.so.0.20[.0] - 1.0.28rc1-2 10 10028 10.so.0.28[.0] - 1.2.20rc1-2 13 10220 12.so.0.20[.0] + 1.0.28rc1-4 10 10028 10.so.0.28[.0] + 1.2.20rc1-4 13 10220 12.so.0.20[.0] 1.0.28 10 10028 10.so.0.28[.0] 1.2.20 13 10220 12.so.0.20[.0] @@ -3638,7 +3466,7 @@ possible without all of you. Thanks to Frank J. T. Wojcik for helping with the documentation. -Libpng version 1.2.20rc3 - August 30, 2007: +Libpng version 1.2.20rc4 - September 1, 2007: Initially created in 1995 by Guy Eric Schalnat, then of Group 42, Inc. Currently maintained by Glenn Randers-Pehrson (glennrp at users.sourceforge.net). @@ -3659,7 +3487,7 @@ included in the libpng distribution, the latter shall prevail.) If you modify libpng you may insert additional notices immediately following this sentence. -libpng versions 1.2.6, August 15, 2004, through 1.2.20rc3, August 30, 2007, are +libpng versions 1.2.6, August 15, 2004, through 1.2.20rc4, September 1, 2007, are Copyright (c) 2004,2006-2007 Glenn Randers-Pehrson, and are distributed according to the same disclaimer and license as libpng-1.2.5 with the following individual added to the list of Contributing Authors @@ -3758,7 +3586,7 @@ certification mark of the Open Source Initiative. Glenn Randers-Pehrson glennrp at users.sourceforge.net -August 30, 2007 +September 1, 2007 .\" end of man page diff --git a/libpngpf.3 b/libpngpf.3 index 1aea5a8d..06fddfa3 100644 --- a/libpngpf.3 +++ b/libpngpf.3 @@ -1,6 +1,6 @@ -.TH LIBPNGPF 3 "August 30, 2007" +.TH LIBPNGPF 3 "September 1, 2007" .SH NAME -libpng \- Portable Network Graphics (PNG) Reference Library 1.2.20rc3 +libpng \- Portable Network Graphics (PNG) Reference Library 1.2.20rc4 (private functions) .SH SYNOPSIS \fB#include \fP diff --git a/png.5 b/png.5 index fadb858d..317b7aeb 100644 --- a/png.5 +++ b/png.5 @@ -1,4 +1,4 @@ -.TH PNG 5 "August 30, 2007" +.TH PNG 5 "September 1, 2007" .SH NAME png \- Portable Network Graphics (PNG) format .SH DESCRIPTION diff --git a/png.c b/png.c index 54d6f7ab..e6ffcd4d 100644 --- a/png.c +++ b/png.c @@ -13,7 +13,7 @@ #include "png.h" /* Generate a compiler error if there is an old png.h in the search path. */ -typedef version_1_2_20rc3 Your_png_h_is_not_version_1_2_20rc3; +typedef version_1_2_20rc4 Your_png_h_is_not_version_1_2_20rc4; /* Version information for C files. This had better match the version * string defined in png.h. */ @@ -67,11 +67,6 @@ PNG_CONST int FARDATA png_pass_ystart[] = {0, 0, 4, 0, 2, 0, 1}; /* offset to next interlace block in the y direction */ PNG_CONST int FARDATA png_pass_yinc[] = {8, 8, 8, 4, 4, 2, 2}; -/* width of interlace block (used in assembler routines only) */ -#if defined(PNG_HAVE_MMX_COMBINE_ROW) || defined(PNG_OPTIMIZED_CODE_SUPPORTED) -PNG_CONST int FARDATA png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; -#endif - /* Height of interlace block. This is not currently used - if you need * it, uncomment it here and in png.h PNG_CONST int FARDATA png_pass_height[] = {8, 8, 4, 4, 2, 2, 1}; @@ -698,7 +693,7 @@ png_charp PNGAPI png_get_copyright(png_structp png_ptr) { png_ptr = png_ptr; /* silence compiler warning about unused png_ptr */ - return ((png_charp) "\n libpng version 1.2.20rc3 - August 30, 2007\n\ + return ((png_charp) "\n libpng version 1.2.20rc4 - September 1, 2007\n\ Copyright (c) 1998-2007 Glenn Randers-Pehrson\n\ Copyright (c) 1996-1997 Andreas Dilger\n\ Copyright (c) 1995-1996 Guy Eric Schalnat, Group 42, Inc.\n"); @@ -825,57 +820,18 @@ png_access_version_number(void) void /* PRIVATE */ png_init_mmx_flags (png_structp png_ptr) { - if(png_ptr == NULL) return; - png_ptr->mmx_rowbytes_threshold = 0; - png_ptr->mmx_bitdepth_threshold = 0; - -# if (defined(PNG_USE_PNGVCRD) || defined(PNG_USE_PNGGCCRD)) - - png_ptr->asm_flags |= PNG_ASM_FLAG_MMX_SUPPORT_COMPILED; - - if (png_mmx_support() > 0) { - png_ptr->asm_flags |= PNG_ASM_FLAG_MMX_SUPPORT_IN_CPU -# ifdef PNG_HAVE_MMX_COMBINE_ROW - | PNG_ASM_FLAG_MMX_READ_COMBINE_ROW -# endif -# ifdef PNG_HAVE_MMX_READ_INTERLACE - | PNG_ASM_FLAG_MMX_READ_INTERLACE -# endif -# ifndef PNG_HAVE_MMX_READ_FILTER_ROW - ; -# else - | PNG_ASM_FLAG_MMX_READ_FILTER_SUB - | PNG_ASM_FLAG_MMX_READ_FILTER_UP - | PNG_ASM_FLAG_MMX_READ_FILTER_AVG - | PNG_ASM_FLAG_MMX_READ_FILTER_PAETH ; - - png_ptr->mmx_rowbytes_threshold = PNG_MMX_ROWBYTES_THRESHOLD_DEFAULT; - png_ptr->mmx_bitdepth_threshold = PNG_MMX_BITDEPTH_THRESHOLD_DEFAULT; -# endif - } else { - png_ptr->asm_flags &= ~( PNG_ASM_FLAG_MMX_SUPPORT_IN_CPU - | PNG_MMX_READ_FLAGS - | PNG_MMX_WRITE_FLAGS ); - } - -# else /* !(PNGVCRD || PNGGCCRD) */ - - /* clear all MMX flags; no support is compiled in */ - png_ptr->asm_flags &= ~( PNG_MMX_FLAGS ); - -# endif /* ?(PNGVCRD || PNGGCCRD) */ + /* obsolete, to be removed from libpng-1.4.0 */ } #endif /* !(PNG_MMX_CODE_SUPPORTED) */ /* this function was added to libpng 1.2.0 */ -#if !defined(PNG_USE_PNGGCCRD) && !defined(PNG_USE_PNGVCRD) int PNGAPI png_mmx_support(void) { + /* obsolete, to be removed from libpng-1.4.0 */ return -1; } -#endif #endif /* PNG_1_0_X */ #endif /* PNG_READ_SUPPORTED && PNG_ASSEMBLER_CODE_SUPPORTED */ diff --git a/png.h b/png.h index a26a725c..a4aa8d0a 100644 --- a/png.h +++ b/png.h @@ -1,7 +1,7 @@ /* png.h - header file for PNG reference library * - * libpng version 1.2.20rc3 - August 30, 2007 + * libpng version 1.2.20rc4 - September 1, 2007 * Copyright (c) 1998-2007 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) * (Version 0.88 Copyright (c) 1995, 1996 Guy Eric Schalnat, Group 42, Inc.) @@ -9,7 +9,7 @@ * Authors and maintainers: * libpng versions 0.71, May 1995, through 0.88, January 1996: Guy Schalnat * libpng versions 0.89c, June 1996, through 0.96, May 1997: Andreas Dilger - * libpng versions 0.97, January 1998, through 1.2.20rc3 - August 30, 2007: Glenn + * libpng versions 0.97, January 1998, through 1.2.20rc4 - September 1, 2007: Glenn * See also "Contributing Authors", below. * * Note about libpng version numbers: @@ -156,8 +156,8 @@ * 1.0.27 10 10027 10.so.0.27[.0] * 1.2.19 13 10219 12.so.0.19[.0] * 1.2.20beta01-04 13 10220 12.so.0.20[.0] - * 1.0.28rc1-2 10 10028 10.so.0.28[.0] - * 1.2.20rc1-2 13 10220 12.so.0.20[.0] + * 1.0.28rc1-4 10 10028 10.so.0.28[.0] + * 1.2.20rc1-4 13 10220 12.so.0.20[.0] * 1.0.28 10 10028 10.so.0.28[.0] * 1.2.20 13 10220 12.so.0.20[.0] * @@ -189,7 +189,7 @@ * If you modify libpng you may insert additional notices immediately following * this sentence. * - * libpng versions 1.2.6, August 15, 2004, through 1.2.20rc3, August 30, 2007, are + * libpng versions 1.2.6, August 15, 2004, through 1.2.20rc4, September 1, 2007, are * Copyright (c) 2004, 2006-2007 Glenn Randers-Pehrson, and are * distributed according to the same disclaimer and license as libpng-1.2.5 * with the following individual added to the list of Contributing Authors: @@ -301,13 +301,13 @@ * Y2K compliance in libpng: * ========================= * - * August 30, 2007 + * September 1, 2007 * * Since the PNG Development group is an ad-hoc body, we can't make * an official declaration. * * This is your unofficial assurance that libpng from version 0.71 and - * upward through 1.2.20rc3 are Y2K compliant. It is my belief that earlier + * upward through 1.2.20rc4 are Y2K compliant. It is my belief that earlier * versions were also Y2K compliant. * * Libpng only has three year fields. One is a 2-byte unsigned integer @@ -363,9 +363,9 @@ */ /* Version information for png.h - this should match the version in png.c */ -#define PNG_LIBPNG_VER_STRING "1.2.20rc3" +#define PNG_LIBPNG_VER_STRING "1.2.20rc4" #define PNG_HEADER_VERSION_STRING \ - " libpng version 1.2.20rc3 - August 30, 2007\n" + " libpng version 1.2.20rc4 - September 1, 2007\n" #define PNG_LIBPNG_VER_SONUM 0 #define PNG_LIBPNG_VER_DLLNUM 13 @@ -377,7 +377,7 @@ /* This should match the numeric part of the final component of * PNG_LIBPNG_VER_STRING, omitting any leading zero: */ -#define PNG_LIBPNG_VER_BUILD 3 +#define PNG_LIBPNG_VER_BUILD 4 /* Release Status */ #define PNG_LIBPNG_BUILD_ALPHA 1 @@ -504,9 +504,6 @@ PNG_EXPORT_VAR (PNG_CONST int FARDATA) png_pass_ystart[7]; PNG_EXPORT_VAR (PNG_CONST int FARDATA) png_pass_yinc[7]; PNG_EXPORT_VAR (PNG_CONST int FARDATA) png_pass_mask[7]; PNG_EXPORT_VAR (PNG_CONST int FARDATA) png_pass_dsp_mask[7]; -#if defined(PNG_HAVE_MMX_COMBINE_ROW) || defined(PNG_OPTIMIZED_CODE_SUPPORTED) -PNG_EXPORT_VAR (PNG_CONST int FARDATA) png_pass_width[7]; -#endif /* This isn't currently used. If you need it, see png.c for more details. PNG_EXPORT_VAR (PNG_CONST int FARDATA) png_pass_height[7]; */ @@ -1279,8 +1276,7 @@ struct png_struct_def png_size_t current_text_left; /* how much text left to read in input */ png_charp current_text; /* current text chunk buffer */ png_charp current_text_ptr; /* current location in current_text */ -# endif /* PNG_PROGRESSIVE_READ_SUPPORTED && PNG_TEXT_SUPPORTED */ - +# endif /* PNG_TEXT_SUPPORTED */ #endif /* PNG_PROGRESSIVE_READ_SUPPORTED */ #if defined(__TURBOC__) && !defined(_Windows) && !defined(__FLAT__) @@ -1416,7 +1412,7 @@ struct png_struct_def /* This triggers a compiler error in png.c, if png.c and png.h * do not agree upon the version number. */ -typedef png_structp version_1_2_20rc3; +typedef png_structp version_1_2_20rc4; typedef png_struct FAR * FAR * png_structpp; diff --git a/pngconf.h b/pngconf.h index d5ef5e3e..01d75789 100644 --- a/pngconf.h +++ b/pngconf.h @@ -1,7 +1,7 @@ /* pngconf.h - machine configurable file for libpng * - * libpng version 1.2.20rc3 - August 30, 2007 + * libpng version 1.2.20rc4 - September 1, 2007 * For conditions of distribution and use, see copyright notice in png.h * Copyright (c) 1998-2007 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) @@ -600,10 +600,10 @@ #endif /* PNG_READ_TRANSFORMS_SUPPORTED */ #if !defined(PNG_NO_PROGRESSIVE_READ) && \ - !defined(PNG_PROGRESSIVE_READ_NOT_SUPPORTED) /* if you don't do progressive */ -# define PNG_PROGRESSIVE_READ_SUPPORTED /* reading. This is not talking */ -#endif /* about interlacing capability! You'll */ - /* still have interlacing unless you change the following line: */ + !defined(PNG_PROGRESSIVE_READ_SUPPORTED) /* if you don't do progressive */ +# define PNG_PROGRESSIVE_READ_SUPPORTED /* reading. This is not talking */ +#endif /* about interlacing capability! You'll */ + /* still have interlacing unless you change the following line: */ #define PNG_READ_INTERLACING_SUPPORTED /* required in PNG-compliant decoders */ @@ -727,15 +727,8 @@ #endif /* PNG_ASSEMBLER_CODE was enabled by default in version 1.2.0 - * even when PNG_USE_PNGVCRD or PNG_USE_PNGGCCRD is not defined. - * - * PNG_NO_ASSEMBLER_CODE disables use of all assembler code, - * and removes several functions from the API. - * - * PNG_NO_MMX_CODE disables the use of MMX code without changing the API. - * When MMX code is off, then optimized C replacement functions are used, - * if PNG_NO_OPTIMIZED_CODE is not enabled. This was added in version - * 1.2.19. + * and removed from version 1.2.20. The following will be removed + * from libpng-1.4.0 */ #if defined(PNG_READ_SUPPORTED) && !defined(PNG_NO_OPTIMIZED_CODE) @@ -762,6 +755,12 @@ # endif # endif +# if (defined(__MWERKS__) && ((__MWERKS__ < 0x0900) || macintosh)) +# if !defined(PNG_MMX_CODE_SUPPORTED) && !defined(PNG_NO_MMX_CODE) +# define PNG_NO_MMX_CODE +# endif +# endif + # if !defined(PNG_MMX_CODE_SUPPORTED) && !defined(PNG_NO_MMX_CODE) # define PNG_MMX_CODE_SUPPORTED # endif @@ -774,14 +773,11 @@ # if !defined(PNG_USE_PNGGCCRD) && defined(PNG_MMX_CODE_SUPPORTED) && \ !defined(PNG_USE_PNGVCRD) # define PNG_USE_PNGGCCRD - /* If you are sure that you don't need thread safety and you are compiling - with PNG_USE_PNGCCRD for an MMX application, you can define this for - faster execution. See pnggccrd.c. # define PNG_THREAD_UNSAFE_OK - */ # endif #endif +/* end of obsolete code to be removed from libpng-1.4.0 */ #if !defined(PNG_1_0_X) #if !defined(PNG_NO_USER_MEM) && !defined(PNG_USER_MEM_SUPPORTED) @@ -1489,53 +1485,6 @@ typedef z_stream FAR * png_zstreamp; # define PNG_ZBUF_SIZE 65536L #endif -#ifdef PNG_READ_SUPPORTED -/* Prior to libpng-1.0.9, this block was in pngasmrd.h */ -#if defined(PNG_INTERNAL) - -#if defined(PNG_USE_PNGGCCRD) || defined(PNG_USE_PNGVCRD) - /* Platform must be Pentium. Makefile must assemble and load - * pnggccrd.c or pngvcrd.c. MMX will be detected at run time and - * used if present. - */ -# ifndef PNG_NO_MMX_COMBINE_ROW -# define PNG_HAVE_MMX_COMBINE_ROW -# endif -# ifndef PNG_NO_MMX_READ_INTERLACE -# define PNG_HAVE_MMX_READ_INTERLACE -# endif -# ifndef PNG_NO_MMX_READ_FILTER_ROW -# define PNG_HAVE_MMX_READ_FILTER_ROW -# ifndef PNG_NO_MMX_FILTER_SUB -# define PNG_MMX_READ_FILTER_SUB_SUPPORTED -# endif -# ifndef PNG_NO_MMX_FILTER_UP -# define PNG_MMX_READ_FILTER_UP_SUPPORTED -# endif -# ifndef PNG_NO_MMX_FILTER_AVG -# define PNG_MMX_READ_FILTER_AVG_SUPPORTED -# endif -# ifndef PNG_NO_MMX_FILTER_PAETH -# define PNG_MMX_READ_FILTER_PAETH_SUPPORTED -# endif -# endif - /* These are the default thresholds before the MMX code kicks in; if either - * rowbytes or bitdepth is below the threshold, plain C code is used. These - * can be overridden at runtime via the png_set_mmx_thresholds() call in - * libpng 1.2.0 and later. The values below were chosen by Intel. - */ -# ifndef PNG_MMX_ROWBYTES_THRESHOLD_DEFAULT -# define PNG_MMX_ROWBYTES_THRESHOLD_DEFAULT 128 /* >= */ -# endif -# ifndef PNG_MMX_BITDEPTH_THRESHOLD_DEFAULT -# define PNG_MMX_BITDEPTH_THRESHOLD_DEFAULT 9 /* >= */ -# endif -#endif /* PNG_USE_PNGGCCRD || PNG_USE_PNGVCRD */ -/* - see pngvcrd.c or pnggccrd.c for info about what is currently enabled */ - -#endif /* PNG_INTERNAL */ -#endif /* PNG_READ_SUPPORTED */ - /* Added at libpng-1.2.8 */ #endif /* PNG_VERSION_INFO_ONLY */ diff --git a/pngerror.c b/pngerror.c index 93ce47b6..2d03c56f 100644 --- a/pngerror.c +++ b/pngerror.c @@ -1,7 +1,7 @@ /* pngerror.c - stub functions for i/o and memory allocation * - * Last changed in libpng 1.2.20 August 30, 2007 + * Last changed in libpng 1.2.20 September 1, 2007 * For conditions of distribution and use, see copyright notice in png.h * Copyright (c) 1998-2007 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) diff --git a/pnggccrd.c b/pnggccrd.c index 37f3f33c..7007ad9f 100644 --- a/pnggccrd.c +++ b/pnggccrd.c @@ -1,6042 +1 @@ - -/* pnggccrd.c - mixed C/assembler version of utilities to read a PNG file - * - * For Intel/AMD x86 or x86-64 CPU (Pentium-MMX or later) and GNU C compiler. - * - * Last changed in libpng 1.2.19 August 19, 2007 - * For conditions of distribution and use, see copyright notice in png.h - * Copyright (c) 1998 Intel Corporation - * Copyright (c) 1999-2002,2007 Greg Roelofs - * Copyright (c) 1998-2007 Glenn Randers-Pehrson - * - * Based on MSVC code contributed by Nirav Chhatrapati, Intel Corp., 1998. - * Interface to libpng contributed by Gilles Vollant, 1999. - * GNU C port by Greg Roelofs, 1999-2001. - * - * References: - * - * http://www.intel.com/drg/pentiumII/appnotes/916/916.htm - * http://www.intel.com/drg/pentiumII/appnotes/923/923.htm - * [Intel's performance analysis of the MMX vs. non-MMX code; - * moved/deleted as of 2006, but text and some graphs still - * available via WayBack Machine at archive.org] - * - * http://www.ibiblio.org/gferg/ldp/GCC-Inline-Assembly-HOWTO.html - * http://sam.zoy.org/blog/2007-04-13-shlib-with-non-pic-code-have-inline-assembly-and-pic-mix-well - * http://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html - * http://gcc.gnu.org/onlinedocs/gcc/Variable-Attributes.html - * http://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html - * AMD64 Architecture Programmer's Manual, volumes 1 and 5 - * [http://www.amd.com/us-en/Processors/TechnicalResources/0,,30_182_739_7044,00.html] - * Intel 64 and IA-32 Software Developer's Manuals - * [http://developer.intel.com/products/processor/manuals/] - * - * png_read_filter_row_mmx_*() were converted in place with intel2gas 1.3.1: - * - * intel2gas -mdI pnggccrd.c.partially-msvc -o pnggccrd.c - * - * and then cleaned up by hand. See http://hermes.terminal.at/intel2gas/ . - * - * NOTE: A sufficiently recent version of GNU as (or as.exe under DOS/Windows) - * is required to assemble the newer asm instructions such as movq. (Version - * 2.5.2l.15 is definitely too old.) See ftp://ftp.gnu.org/pub/gnu/binutils/ . - */ - -/* - * PORTING NOTES AND CHANGELOG (mostly by Greg Roelofs) - * =========================== - * - * 19991006: - * - fixed sign error in post-MMX cleanup code (16- & 32-bit cases) - * - * 19991007: - * - additional optimizations (possible or definite): - * x [DONE] write MMX code for 64-bit case (pixel_bytes == 8) [not tested] - * - write MMX code for 48-bit case (pixel_bytes == 6) - * - figure out what's up with 24-bit case (pixel_bytes == 3): - * why subtract 8 from width_mmx in the pass 4/5 case? - * (only width_mmx case) (near line 2335) - * x [DONE] replace pixel_bytes within each block with the true - * constant value (or are compilers smart enough to do that?) - * - rewrite all MMX interlacing code so it's aligned with - * the *beginning* of the row buffer, not the end. This - * would not only allow one to eliminate half of the memory - * writes for odd passes (that is, pass == odd), it may also - * eliminate some unaligned-data-access exceptions (assuming - * there's a penalty for not aligning 64-bit accesses on - * 64-bit boundaries). The only catch is that the "leftover" - * pixel(s) at the end of the row would have to be saved, - * but there are enough unused MMX registers in every case, - * so this is not a problem. A further benefit is that the - * post-MMX cleanup code (C code) in at least some of the - * cases could be done within the assembler block. - * x [DONE] the "v3 v2 v1 v0 v7 v6 v5 v4" comments are confusing, - * inconsistent, and don't match the MMX Programmer's Reference - * Manual conventions anyway. They should be changed to - * "b7 b6 b5 b4 b3 b2 b1 b0," where b0 indicates the byte that - * was lowest in memory (i.e., corresponding to a left pixel) - * and b7 is the byte that was highest (i.e., a right pixel). - * - * 19991016: - * - Brennan's Guide notwithstanding, gcc under Linux does *not* - * want globals prefixed by underscores when referencing them-- - * i.e., if the variable is const4, then refer to it as const4, - * not _const4. This seems to be a djgpp-specific requirement. - * Also, such variables apparently *must* be declared outside - * of functions; neither static nor automatic variables work if - * defined within the scope of a single function, but both - * static and truly global (multi-module) variables work fine. - * - * 19991017: - * - replaced pixel_bytes in each png_memcpy() call with constant value for - * inlining (png_do_read_interlace() "non-MMX/modified C code" block) - * - * 19991023: - * - fixed png_combine_row() non-MMX replication bug (odd passes only?) - * - switched from string-concatenation-with-macros to cleaner method of - * renaming global variables for djgpp--i.e., always use prefixes in - * inlined assembler code (== strings) and conditionally rename the - * variables, not the other way around. Hence _const4, _mask8_0, etc. - * - * 19991024: - * - fixed mmxsupport()/png_do_read_interlace() first-row bug - * This one was severely weird: even though mmxsupport() doesn't touch - * ebx (where "row" pointer was stored), it nevertheless managed to zero - * the register (even in static/non-fPIC code--see below), which in turn - * caused png_do_read_interlace() to return prematurely on the first row of - * interlaced images (i.e., without expanding the interlaced pixels). - * Inspection of the generated assembly code didn't turn up any clues, - * although it did point at a minor optimization (i.e., get rid of - * mmx_supported_local variable and just use eax). Possibly the CPUID - * instruction is more destructive than it looks? (Not yet checked.) - * - "info gcc" was next to useless, so compared fPIC and non-fPIC assembly - * listings... Apparently register spillage has to do with ebx, since - * it's used to index the global offset table. Commenting it out of the - * input-reg lists in png_combine_row() eliminated compiler barfage, so - * ifdef'd with __PIC__ macro: if defined, use a global for unmask - * - * 19991107: - * - verified CPUID clobberage: 12-char string constant ("GenuineIntel", - * "AuthenticAMD", etc.) placed in ebx:ecx:edx. Still need to polish. - * - * 19991120: - * - made "diff" variable (now "_dif") global to simplify conversion of - * filtering routines (running out of regs, sigh). "diff" is still used - * in interlacing routines, however. - * - fixed up both versions of mmxsupport() (ORIG_THAT_USED_TO_CLOBBER_EBX - * macro determines which is used); original not yet tested. - * - * 20000213: - * - when compiling with gcc, be sure to use -fomit-frame-pointer - * - * 20000319: - * - fixed a register-name typo in png_do_read_interlace(), default (MMX) case, - * pass == 4 or 5, that caused visible corruption of interlaced images - * - * 20000623: - * - Various problems were reported with gcc 2.95.2 in the Cygwin environment, - * many of the form "forbidden register 0 (ax) was spilled for class AREG." - * This is explained at http://gcc.gnu.org/fom_serv/cache/23.html, and - * Chuck Wilson supplied a patch involving dummy output registers. See - * http://sourceforge.net/bugs/?func=detailbug&bug_id=108741&group_id=5624 - * for the original (anonymous) SourceForge bug report. - * - * 20000706: - * - Chuck Wilson passed along these remaining gcc 2.95.2 errors: - * pnggccrd.c: In function `png_combine_row': - * pnggccrd.c:525: more than 10 operands in `asm' - * pnggccrd.c:669: more than 10 operands in `asm' - * pnggccrd.c:828: more than 10 operands in `asm' - * pnggccrd.c:994: more than 10 operands in `asm' - * pnggccrd.c:1177: more than 10 operands in `asm' - * They are all the same problem and can be worked around by using the - * global _unmask variable unconditionally, not just in the -fPIC case. - * Reportedly earlier versions of gcc also have the problem with more than - * 10 operands; they just don't report it. Much strangeness ensues, etc. - * - * 20000729: - * - enabled png_read_filter_row_mmx_up() (shortest remaining unconverted - * MMX routine); began converting png_read_filter_row_mmx_sub() - * - to finish remaining sections: - * - clean up indentation and comments - * - preload local variables - * - add output and input regs (order of former determines numerical - * mapping of latter) - * - avoid all usage of ebx (including bx, bh, bl) register [20000823] - * - remove "$" from addressing of Shift and Mask variables [20000823] - * - * 20000731: - * - global union vars causing segfaults in png_read_filter_row_mmx_sub()? - * - * 20000822: - * - ARGH, stupid png_read_filter_row_mmx_sub() segfault only happens with - * shared-library (-fPIC) version! Code works just fine as part of static - * library. Should have tested that sooner. - * ebx is getting clobbered again (explicitly this time); need to save it - * on stack or rewrite asm code to avoid using it altogether. Blargh! - * - * 20000823: - * - first section was trickiest; all remaining sections have ebx -> edx now. - * (-fPIC works again.) Also added missing underscores to various Shift* - * and *Mask* globals and got rid of leading "$" signs. - * - * 20000826: - * - added visual separators to help navigate microscopic printed copies - * (http://pobox.com/~newt/code/gpr-latest.zip, mode 10); started working - * on png_read_filter_row_mmx_avg() - * - * 20000828: - * - finished png_read_filter_row_mmx_avg(): only Paeth left! (930 lines...) - * What the hell, did png_read_filter_row_mmx_paeth(), too. Comments not - * cleaned up/shortened in either routine, but functionality is complete - * and seems to be working fine. - * - * 20000829: - * - ahhh, figured out last(?) bit of gcc/gas asm-fu: if register is listed - * as an input reg (with dummy output variables, etc.), then it *cannot* - * also appear in the clobber list or gcc 2.95.2 will barf. The solution - * is simple enough... - * - * 20000914: - * - bug in png_read_filter_row_mmx_avg(): 16-bit grayscale not handled - * correctly (but 48-bit RGB just fine) - * - * 20000916: - * - fixed bug in png_read_filter_row_mmx_avg(), bpp == 2 case; three errors: - * - "_ShiftBpp.use = 24;" should have been "_ShiftBpp.use = 16;" - * - "_ShiftRem.use = 40;" should have been "_ShiftRem.use = 48;" - * - "psllq _ShiftRem, %%mm2" should have been "psrlq _ShiftRem, %%mm2" - * - * 20010101: - * - added new png_init_mmx_flags() function (here only because it needs to - * call mmxsupport(), which should probably become global png_mmxsupport()); - * modified other MMX routines to run conditionally (png_ptr->asm_flags) - * - * 20010103: - * - renamed mmxsupport() to png_mmx_support(), with auto-set of mmx_supported, - * and made it public; moved png_init_mmx_flags() to png.c as internal func - * - * 20010104: - * - removed dependency on png_read_filter_row_c() (C code already duplicated - * within MMX version of png_read_filter_row()) so no longer necessary to - * compile it into pngrutil.o - * - * 20010310: - * - fixed buffer-overrun bug in png_combine_row() C code (non-MMX) - * - * 20010808: - * - added PNG_THREAD_UNSAFE_OK around code using global variables [GR-P] - * - * 20011124: - * - fixed missing save of Eflag in png_mmx_support() [Maxim Sobolev] - * - * 20020304: - * - eliminated incorrect use of width_mmx in pixel_bytes == 8 case - * - * 20020407: - * - fixed insufficient preservation of ebx register [Sami Farin] - * - * 20040724: - * - more tinkering with clobber list at lines 4529 and 5033 to get it to - * compile with gcc 3.4 [GR-P] - * - * 20040809: - * - added "rim" definitions for CONST4 and CONST6 [GR-P] - * - * 20060303: - * - added "OS2" to list of systems that don't need leading underscores [GR-P] - * - * 20060320: - * - made PIC-compliant [Christian Aichinger] - * - * 20070313: - * - finally applied Giuseppe Ghibò's 64-bit patch of 20060803 (completely - * overlooked Dylan Alex Simon's similar patch of 20060414, oops...) - * - * 20070524: - * - fixed link failure caused by asm-only variables being optimized out - * (identified by Dimitri of Trolltech) with __attribute__((used)), which - * also gets rid of warnings => nuked ugly png_squelch_warnings() hack - * - dropped redundant ifdef - * - moved png_mmx_support() back up where originally intended (as in - * pngvcrd.c), using __attribute__((noinline)) in extra prototype - * - * 20070527: - * - revised png_combine_row() to reuse mask in lieu of external _unmask - * - moved 32-bit (RGBA) case to top of png_combine_row(): most common - * - just about ready to give up on x86-64 -fPIC mode; can't even access 16 - * _mask*_* constants without triggering link error on shared library: - * /usr/bin/ld: pnggccrd.pic.o: relocation R_X86_64_32S against `a local - * symbol' can not be used when making a shared object; recompile with - * -fPIC - * pnggccrd.pic.o: could not read symbols: Bad value - * ("objdump -x pnggccrd.pic.o | grep rodata" to verify) - * [might be able to work around by doing within assembly code whatever - * -fPIC does, but given problems to date, seems like long shot...] - * [relevant ifdefs: __x86_64__ && __PIC__ => C code only] - * - changed #if 0 to #ifdef PNG_CLOBBER_MMX_REGS_SUPPORTED in case gcc ever - * supports MMX regs (%mm0, etc.) in clobber list (not supported by gcc - * 2.7.2.3, 2.91.66 (egcs 1.1.2), 3.x, or 4.1.2) - * - * 20070603: - * - revised png_combine_row() to use @GOTPCREL(%%rip) addressing on _c64 - * struct of _mask*_* constants for x86-64 -fPIC; see sam.zoy.org link - * above for details - * - moved _const4 and _const6 into _c64 struct, renamed to _amask5_3_0 and - * _amask7_1_0, respectively - * - can't figure out how to use _c64._mask*_* vars within asm code, so still - * need single variables for non-x86-64/-fPIC half :-( - * - replaced various __PIC__ ifdefs with *_GOT_ebx macros - * - moved _LBCarryMask and _HBClearMask into _c64 struct - * - conditionally replaced _p*temp variables with %r11d-%r13d (via p*_TEMP - * and CLOBBER_r1*d macros) - * - * 20070604: - * - replaced all _ActiveMask and _ActiveMaskEnd with new _amask*_*_* consts - * (_amask naming convention: numbers of 00-bytes, ff-bytes, 00-bytes) - * - _ActiveMask // (10) // avg/paeth/sub; read-only; consts; movq/pand - * 0x0000000000ffffffLL (bpp 3, avg) _amask5_3_0 - * 0xffffffffffffffffLL (bpp 4, 6, avg) _amask0_8_0 - * 0x000000000000ffffLL (bpp 2, avg) _amask6_2_0 - * 0x0000000000ffffffLL (bpp 3, paeth) _amask5_3_0 - * 0x00000000ffffffffLL (bpp 6, paeth) _amask4_4_0 - * 0x00000000ffffffffLL (bpp 4, paeth) _amask4_4_0 - * 0x00000000ffffffffLL (bpp 8, paeth) _amask4_4_0 - * 0x0000ffffff000000LL (bpp 3, sub) _amask2_3_3 - * 0x00000000ffff0000LL (bpp 2, sub) _amask4_2_2 - * - _ActiveMaskEnd // (1) // paeth only; read-only; const; pand - * 0xffff000000000000LL (bpp 3, paeth) _amask0_2_6 - * - changed all "#if defined(__x86_64__) // later // && defined(__PIC__)" - * lines to "#ifdef PNG_x86_64_USE_GOTPCREL" for easier/safer testing - * - * 20070605: - * - merged PNG_x86_64_USE_GOTPCREL, non-PNG_x86_64_USE_GOTPCREL code via - * *MASK* and LOAD/RESTORE macros - * - * 20070607: - * - replaced all constant instances of _ShiftBpp, _ShiftRem with immediates - * (still have two shared cases in avg, sub routines) - * - * 20070609: - * - replaced remaining instances of _ShiftBpp, _ShiftRem with immediates - * (split sub and avg 4/6-bpp cases into separate blocks) - * - fixed paeth bug due to clobbered r11/r12/r13 regs - * - * 20070610: - * - made global "_dif" variable (avg/paeth/sub routines) local again (now - * "diff"--see 19991120 entry above), using register constraints - * - note that %ebp in clobber list doesn't actually work, at least for 32-bit - * version and gcc 4.1.2; must save and restore manually. (Seems to work - * OK for 64-bit version and gcc 3.4.3, but gcc may not be using ebp/rbp - * in that case.) - * - started replacing direct _MMXLength accesses with register constraints - * - * 20070612: - * - continued replacing direct _MMXLength accesses with register constraints - * - * 20070613: - * - finished replacing direct _MMXLength accesses with register constraints; - * switched to local variable (and renamed back to MMXLength) - * - * 20070614: - * - fixed sub bpp = 1 bug - * - started replacing direct _FullLength accesses with register constraints - * - * 20070615: - * - fixed 64-bit paeth bpp 3 crash bug (misplaced LOAD_GOT_rbp) - * - fixed 64-bit paeth bpp 1/2 and cleanup-block crash bugs (misplaced - * RESTORE_r11_r12_r13) - * - slightly optimized avg/paeth cleanup blocks and paeth bpp 1/2 block - * (save/restore ebx only if needed) - * - continued replacing direct _FullLength accesses with register constraints - * - * 20070616: - * - finished replacing direct _FullLength accesses with register constraints - * (*ugly* conditional clobber-separator macros for avg and paeth, sigh) - * - * 20070618: - * - fixed misplaced PNG_THREAD_UNSAFE_OK endif (was missing LOAD_GOT_rbp/ - * RESTORE_rbp in 32-bit thread-safe case) - * - changed all "ifdef *" to "if defined(*)" [GR-P] - * - * 20070619: - * - rearranged most bitdepth-related case statements to put most frequent - * cases at top (24-bit, 32-bit, 8-bit, rest) - * - * 20070623: - * - cleaned up png_debug() warnings/formatting - * - removed PNG_MMX_CODE_SUPPORTED ifdefs and added outer __GNUC__ ifdef - * (module no longer used by non-x86/non-GCC builds as of libpng 1.2.19) - * - removed single libpng-1.2.x PNG_DEBUG dependency on 1.0.x png_struct - * member (row_buf_size) - * - rearranged pass-related if-blocks in png_do_read_interlace() to put most - * frequent cases (4, 5) at top [GR-P suggestion] - * - * 20070624-29: - * - fixed 64-bit crash bug: pointers -> rsi/rdi, not esi/edi (switched to - * %0/%1/%2/%3/%4 notation; eliminated size suffixes from relevant add/ - * inc/sub/mov instructions; changed dummy vars to pointers) - * - png_combine_row() - * - png_do_read_interlace() - * - png_read_filter_row_mmx_avg() - * - png_read_filter_row_mmx_paeth() - * - png_read_filter_row_mmx_sub() - * - png_read_filter_row_mmx_up() - * - NOTE: this fix makes use of the fact that modifying a 32-bit reg (e.g., - * %%ebx) clears the top half of its corresponding 64-bit reg (%%rbx), so - * it's safe to mix 32-bit operations with 64-bit base/index addressing - * (see new PSI/PAX/PBX/PDX/PBP/etc. "pointer-register" macros); applies - * also to clobber lists - * - * 20070630: - * - cleaned up formatting, macros, minor png_read_filter_row_mmx_sub() 8-bpp - * register-usage inefficiency - * - fixed 32-bit png_do_read_interlace() bug (was using pointer size for - * 64-bit dummy values) - * - * 20070703: - * - added check for (manual) PIC macro to fix OpenBSD crash bug - * - * 20070717: - * - fixed 48-bit png_combine_row() bug (was acting like 32-bit): copy 6 - * bytes per pixel, not 4, and use stride of 6, not 4, in the second loop - * of interlace processing of 48-bit pixels [GR-P] - * - * 20070722: - * - fixed 64-bit png_uint_32 bug with MMXLength/FullLength temp vars - * - * [still broken: tops of all row-filter blocks (input/output constraints); - * shows up on 64-bit dynamic (-fPIC) version with -O2, especially if debug- - * printfs enabled, but at right edge of odd-width images even if disabled] - * - * - * STILL TO DO: - * - fix final thread-unsafe code using stack vars and pointer? (paeth top, - * default, bottom only: default, bottom already 5 reg constraints; could - * replace bpp with pointer and group bpp/patemp/pbtemp/pctemp in array) - * - fix ebp/no-reg-constraint inefficiency (avg/paeth/sub top) - * - test png_do_read_interlace() 64-bit case (pixel_bytes == 8) - * - write MMX code for 48-bit case (pixel_bytes == 6) - * - figure out what's up with 24-bit case (pixel_bytes == 3): - * why subtract 8 from width_mmx in the pass 4/5 case? due to - * odd number of bytes? (only width_mmx case) (near line 2335) - * - rewrite all MMX interlacing code so it's aligned with beginning - * of the row buffer, not the end (see 19991007 for details) - * - add error messages to any remaining bogus default cases - * - enable pixel_depth == 8 cases in png_read_filter_row()? (test speed) - * - try =r, etc., as reg constraints? (would gcc use 64-bit ones on x86-64?) - * - need full, non-graphical, CRC-based test suite... maybe autogenerate - * random data of various height/width/depth, compute CRCs, write (C - * funcs), read (asm/MMX), recompute CRCs, and compare? - * - write true x86-64 version using 128-bit "media instructions", %xmm0-15, - * and extra general-purpose registers - */ - -#if defined(__GNUC__) - -#define PNG_INTERNAL -#include "png.h" - - -/* for some inexplicable reason, gcc 3.3.5 on OpenBSD (and elsewhere?) does - * *not* define __PIC__ when the -fPIC option is used, so we have to rely on - * makefiles and whatnot to define the PIC macro explicitly */ -#if defined(PIC) && !defined(__PIC__) // (this can/should move to pngconf.h) -# define __PIC__ -#endif - -#if defined(PNG_ASSEMBLER_CODE_SUPPORTED) && \ - defined(PNG_USE_PNGGCCRD) && defined(PNG_MMX_CODE_SUPPORTED) - -/* if you want/need full thread-safety on x86-64 even when linking statically, - * comment out the "&& defined(__PIC__)" part here: */ -#if defined(__x86_64__) && defined(__PIC__) -# define PNG_x86_64_USE_GOTPCREL // GOTPCREL => full thread-safety -# define PNG_CLOBBER_x86_64_REGS_SUPPORTED // works as of gcc 3.4.3 ... -#endif - -int PNGAPI png_mmx_support(void); - -#if defined(PNG_USE_LOCAL_ARRAYS) -static PNG_CONST int FARDATA png_pass_start[7] = {0, 4, 0, 2, 0, 1, 0}; -static PNG_CONST int FARDATA png_pass_inc[7] = {8, 8, 4, 4, 2, 2, 1}; -static PNG_CONST int FARDATA png_pass_width[7] = {8, 4, 4, 2, 2, 1, 1}; -#endif - -/* djgpp, Win32, Cygwin, and OS2 add their own underscores to global variables, - * so define them without: */ -#if defined(__DJGPP__) || defined(WIN32) || defined(__CYGWIN__) || \ - defined(__OS2__) || defined(__APPLE__) -# define _mmx_supported mmx_supported -# define _mask8_0 mask8_0 -# define _mask16_1 mask16_1 -# define _mask16_0 mask16_0 -# define _mask24_2 mask24_2 -# define _mask24_1 mask24_1 -# define _mask24_0 mask24_0 -# define _mask32_3 mask32_3 -# define _mask32_2 mask32_2 -# define _mask32_1 mask32_1 -# define _mask32_0 mask32_0 -# define _mask48_5 mask48_5 -# define _mask48_4 mask48_4 -# define _mask48_3 mask48_3 -# define _mask48_2 mask48_2 -# define _mask48_1 mask48_1 -# define _mask48_0 mask48_0 -# define _amask5_3_0 amask5_3_0 -# define _amask7_1_0 amask7_1_0 -# define _LBCarryMask LBCarryMask -# define _HBClearMask HBClearMask -# define _amask0_8_0 amask0_8_0 -# define _amask6_2_0 amask6_2_0 -# define _amask4_4_0 amask4_4_0 -# define _amask0_2_6 amask0_2_6 -# define _amask2_3_3 amask2_3_3 -# define _amask4_2_2 amask4_2_2 -# if defined(PNG_THREAD_UNSAFE_OK) -# define _patemp patemp -# define _pbtemp pbtemp -# define _pctemp pctemp -# endif -#endif // djgpp, Win32, Cygwin, OS2 - - -/* These constants are used in the inlined MMX assembly code. */ - -typedef unsigned long long ull; - -#if defined(PNG_x86_64_USE_GOTPCREL) -static PNG_CONST struct { - //ull _mask_array[26]; - - // png_combine_row() constants: - ull _mask8_0; - ull _mask16_0, _mask16_1; - ull _mask24_0, _mask24_1, _mask24_2; - ull _mask32_0, _mask32_1, _mask32_2, _mask32_3; - ull _mask48_0, _mask48_1, _mask48_2, _mask48_3, _mask48_4, _mask48_5; - - // png_do_read_interlace() constants: - ull _amask5_3_0, _amask7_1_0; // was _const4 and _const6, respectively - - // png_read_filter_row_mmx_avg() constants (also uses _amask5_3_0): - ull _LBCarryMask, _HBClearMask; - ull _amask0_8_0, _amask6_2_0; // was ActiveMask for bpp 4/6 and 2 cases - - // png_read_filter_row_mmx_paeth() constants (also uses _amask5_3_0): - ull _amask4_4_0, _amask0_2_6; // was ActiveMask{,End} for bpp 6/4/8 and 3 - - // png_read_filter_row_mmx_sub() constants: - ull _amask2_3_3, _amask4_2_2; // was ActiveMask for bpp 3 and 2 cases - -} _c64 __attribute__((used, aligned(8))) = { - - // png_combine_row() constants: - 0x0102040810204080LL, // _mask8_0 offset 0 - - 0x1010202040408080LL, // _mask16_0 offset 8 - 0x0101020204040808LL, // _mask16_1 offset 16 - - 0x2020404040808080LL, // _mask24_0 offset 24 - 0x0408080810101020LL, // _mask24_1 offset 32 - 0x0101010202020404LL, // _mask24_2 offset 40 - - 0x4040404080808080LL, // _mask32_0 offset 48 - 0x1010101020202020LL, // _mask32_1 offset 56 - 0x0404040408080808LL, // _mask32_2 offset 64 - 0x0101010102020202LL, // _mask32_3 offset 72 - - 0x4040808080808080LL, // _mask48_0 offset 80 - 0x2020202040404040LL, // _mask48_1 offset 88 - 0x1010101010102020LL, // _mask48_2 offset 96 - 0x0404080808080808LL, // _mask48_3 offset 104 - 0x0202020204040404LL, // _mask48_4 offset 112 - 0x0101010101010202LL, // _mask48_5 offset 120 - - // png_do_read_interlace() constants: - 0x0000000000FFFFFFLL, // _amask5_3_0 offset 128 (bpp 3, avg/paeth) const4 - 0x00000000000000FFLL, // _amask7_1_0 offset 136 const6 - - // png_read_filter_row_mmx_avg() constants: - 0x0101010101010101LL, // _LBCarryMask offset 144 - 0x7F7F7F7F7F7F7F7FLL, // _HBClearMask offset 152 - 0xFFFFFFFFFFFFFFFFLL, // _amask0_8_0 offset 160 (bpp 4/6, avg) - 0x000000000000FFFFLL, // _amask6_2_0 offset 168 (bpp 2, avg) - - // png_read_filter_row_mmx_paeth() constants: - 0x00000000FFFFFFFFLL, // _amask4_4_0 offset 176 (bpp 6/4/8, paeth) - 0xFFFF000000000000LL, // _amask0_2_6 offset 184 (bpp 3, paeth) A.M.End - - // png_read_filter_row_mmx_sub() constants: - 0x0000FFFFFF000000LL, // _amask2_3_3 offset 192 (bpp 3, sub) - 0x00000000FFFF0000LL, // _amask4_2_2 offset 200 (bpp 2, sub) - -}; - -#define MASK8_0 "(%%rbp)" -#define MASK16_0 "8(%%rbp)" -#define MASK16_1 "16(%%rbp)" -#define MASK24_0 "24(%%rbp)" -#define MASK24_1 "32(%%rbp)" -#define MASK24_2 "40(%%rbp)" -#define MASK32_0 "48(%%rbp)" -#define MASK32_1 "56(%%rbp)" -#define MASK32_2 "64(%%rbp)" -#define MASK32_3 "72(%%rbp)" -#define MASK48_0 "80(%%rbp)" -#define MASK48_1 "88(%%rbp)" -#define MASK48_2 "96(%%rbp)" -#define MASK48_3 "104(%%rbp)" -#define MASK48_4 "112(%%rbp)" -#define MASK48_5 "120(%%rbp)" -#define AMASK5_3_0 "128(%%rbp)" -#define AMASK7_1_0 "136(%%rbp)" -#define LB_CARRY_MASK "144(%%rbp)" -#define HB_CLEAR_MASK "152(%%rbp)" -#define AMASK0_8_0 "160(%%rbp)" -#define AMASK6_2_0 "168(%%rbp)" -#define AMASK4_4_0 "176(%%rbp)" -#define AMASK0_2_6 "184(%%rbp)" -#define AMASK2_3_3 "192(%%rbp)" -#define AMASK4_2_2 "200(%%rbp)" - -#else // !PNG_x86_64_USE_GOTPCREL - -static PNG_CONST ull _mask8_0 __attribute__((used, aligned(8))) = 0x0102040810204080LL; - -static PNG_CONST ull _mask16_1 __attribute__((used, aligned(8))) = 0x0101020204040808LL; -static PNG_CONST ull _mask16_0 __attribute__((used, aligned(8))) = 0x1010202040408080LL; - -static PNG_CONST ull _mask24_2 __attribute__((used, aligned(8))) = 0x0101010202020404LL; -static PNG_CONST ull _mask24_1 __attribute__((used, aligned(8))) = 0x0408080810101020LL; -static PNG_CONST ull _mask24_0 __attribute__((used, aligned(8))) = 0x2020404040808080LL; - -static PNG_CONST ull _mask32_3 __attribute__((used, aligned(8))) = 0x0101010102020202LL; -static PNG_CONST ull _mask32_2 __attribute__((used, aligned(8))) = 0x0404040408080808LL; -static PNG_CONST ull _mask32_1 __attribute__((used, aligned(8))) = 0x1010101020202020LL; -static PNG_CONST ull _mask32_0 __attribute__((used, aligned(8))) = 0x4040404080808080LL; - -static PNG_CONST ull _mask48_5 __attribute__((used, aligned(8))) = 0x0101010101010202LL; -static PNG_CONST ull _mask48_4 __attribute__((used, aligned(8))) = 0x0202020204040404LL; -static PNG_CONST ull _mask48_3 __attribute__((used, aligned(8))) = 0x0404080808080808LL; -static PNG_CONST ull _mask48_2 __attribute__((used, aligned(8))) = 0x1010101010102020LL; -static PNG_CONST ull _mask48_1 __attribute__((used, aligned(8))) = 0x2020202040404040LL; -static PNG_CONST ull _mask48_0 __attribute__((used, aligned(8))) = 0x4040808080808080LL; - -// png_do_read_interlace() constants: -static PNG_CONST ull _amask5_3_0 __attribute__((aligned(8))) = 0x0000000000FFFFFFLL; // was _const4 -static PNG_CONST ull _amask7_1_0 __attribute__((aligned(8))) = 0x00000000000000FFLL; // was _const6 - -// png_read_filter_row_mmx_avg() constants: -static PNG_CONST ull _LBCarryMask __attribute__((used, aligned(8))) = 0x0101010101010101LL; -static PNG_CONST ull _HBClearMask __attribute__((used, aligned(8))) = 0x7f7f7f7f7f7f7f7fLL; -static PNG_CONST ull _amask0_8_0 __attribute__((used, aligned(8))) = 0xFFFFFFFFFFFFFFFFLL; -static PNG_CONST ull _amask6_2_0 __attribute__((used, aligned(8))) = 0x000000000000FFFFLL; - -// png_read_filter_row_mmx_paeth() constants: -static PNG_CONST ull _amask4_4_0 __attribute__((used, aligned(8))) = 0x00000000FFFFFFFFLL; -static PNG_CONST ull _amask0_2_6 __attribute__((used, aligned(8))) = 0xFFFF000000000000LL; - -// png_read_filter_row_mmx_sub() constants: -static PNG_CONST ull _amask2_3_3 __attribute__((used, aligned(8))) = 0x0000FFFFFF000000LL; -static PNG_CONST ull _amask4_2_2 __attribute__((used, aligned(8))) = 0x00000000FFFF0000LL; - -#define MASK8_0 "_mask8_0" -#define MASK16_0 "_mask16_0" -#define MASK16_1 "_mask16_1" -#define MASK24_0 "_mask24_0" -#define MASK24_1 "_mask24_1" -#define MASK24_2 "_mask24_2" -#define MASK32_0 "_mask32_0" -#define MASK32_1 "_mask32_1" -#define MASK32_2 "_mask32_2" -#define MASK32_3 "_mask32_3" -#define MASK48_0 "_mask48_0" -#define MASK48_1 "_mask48_1" -#define MASK48_2 "_mask48_2" -#define MASK48_3 "_mask48_3" -#define MASK48_4 "_mask48_4" -#define MASK48_5 "_mask48_5" -#define AMASK5_3_0 "_amask5_3_0" -#define AMASK7_1_0 "_amask7_1_0" -#define LB_CARRY_MASK "_LBCarryMask" -#define HB_CLEAR_MASK "_HBClearMask" -#define AMASK0_8_0 "_amask0_8_0" -#define AMASK6_2_0 "_amask6_2_0" -#define AMASK4_4_0 "_amask4_4_0" -#define AMASK0_2_6 "_amask0_2_6" -#define AMASK2_3_3 "_amask2_3_3" -#define AMASK4_2_2 "_amask4_2_2" - -#endif // ?PNG_x86_64_USE_GOTPCREL - - -#if defined(PNG_HAVE_MMX_READ_FILTER_ROW) || defined(PNG_HAVE_MMX_COMBINE_ROW) - -// this block is specific to png_read_filter_row_mmx_paeth() except for -// LOAD_GOT_rbp and RESTORE_rbp, which are also used in png_combine_row() -#if defined(PNG_x86_64_USE_GOTPCREL) -# define pa_TEMP "%%r11d" -# define pb_TEMP "%%r12d" -# define pc_TEMP "%%r13d" -# if defined(PNG_CLOBBER_x86_64_REGS_SUPPORTED) // works as of gcc 3.4.3 ... -# define SAVE_r11_r12_r13 -# define RESTORE_r11_r12_r13 -# define _CLOBBER_r11_r12_r13 ,"%r11", "%r12", "%r13" -# define CLOBBER_r11_r12_r13 "%r11", "%r12", "%r13" -# else // !PNG_CLOBBER_x86_64_REGS_SUPPORTED -# define SAVE_r11_r12_r13 "pushq %%r11 \n\t" \ - "pushq %%r12 \n\t" \ - "pushq %%r13 \n\t" // "normally 0-extended" -# define RESTORE_r11_r12_r13 "popq %%r13 \n\t" \ - "popq %%r12 \n\t" \ - "popq %%r11 \n\t" -# define _CLOBBER_r11_r12_r13 -# define CLOBBER_r11_r12_r13 -# endif -# define LOAD_GOT_rbp "pushq %%rbp \n\t" \ - "movq _c64@GOTPCREL(%%rip), %%rbp \n\t" -# define RESTORE_rbp "popq %%rbp \n\t" -#else // 32-bit and/or non-PIC -# if defined(PNG_THREAD_UNSAFE_OK) - // These variables are used in png_read_filter_row_mmx_paeth() and would be - // local variables if not for gcc-inline-assembly addressing limitations - // (some apparently related to ELF format, others to CPU type). - // - // WARNING: Their presence defeats the thread-safety of libpng. - static int _patemp __attribute__((used)); - static int _pbtemp __attribute__((used)); - static int _pctemp __attribute__((used)); -# define pa_TEMP "_patemp" -# define pb_TEMP "_pbtemp" // temp variables for -# define pc_TEMP "_pctemp" // Paeth routine -# define SAVE_r11_r12_r13 -# define RESTORE_r11_r12_r13 -# define _CLOBBER_r11_r12_r13 // not using regs => not clobbering -# define CLOBBER_r11_r12_r13 -# endif // PNG_THREAD_UNSAFE_OK -# define LOAD_GOT_rbp -# define RESTORE_rbp -#endif - -#if defined(__x86_64__) -# define SAVE_ebp -# define RESTORE_ebp -# define _CLOBBER_ebp ,"%ebp" -# define CLOBBER_ebp "%ebp" -# define SAVE_FullLength "movl %%eax, %%r15d \n\t" -# define RESTORE_FullLength "movl %%r15d, " // may go into eax or ecx -# if defined(PNG_CLOBBER_x86_64_REGS_SUPPORTED) // works as of gcc 3.4.3 ... -# define SAVE_r15 -# define RESTORE_r15 -# define _CLOBBER_r15 ,"%r15" -# else -# define SAVE_r15 "pushq %%r15 \n\t" -# define RESTORE_r15 "popq %%r15 \n\t" -# define _CLOBBER_r15 -# endif -# define PBP "%%rbp" // regs used for 64-bit -# define PAX "%%rax" // pointers or in -# define PBX "%%rbx" // combination with -# define PCX "%%rcx" // 64-bit pointer-regs -# define PDX "%%rdx" // (base/index pairs, -# define PSI "%%rsi" // add/sub/mov pairs) -# define CLEAR_BOTTOM_3_BITS "and $0xfffffffffffffff8, " -#else -# define SAVE_ebp "pushl %%ebp \n\t" // clobber list doesn't work -# define RESTORE_ebp "popl %%ebp \n\t" // for %ebp on 32-bit; not -# define _CLOBBER_ebp // clear why not -# define CLOBBER_ebp -# define SAVE_FullLength "pushl %%eax \n\t" -# define RESTORE_FullLength "popl " // eax (avg) or ecx (paeth) -# define SAVE_r15 -# define RESTORE_r15 -# define _CLOBBER_r15 -# define PBP "%%ebp" // regs used for or in -# define PAX "%%eax" // combination with -# define PBX "%%ebx" // "normal," 32-bit -# define PCX "%%ecx" // pointers -# define PDX "%%edx" -# define PSI "%%esi" -# define CLEAR_BOTTOM_3_BITS "and $0xfffffff8, " -#endif - -// CLOB_COMMA_ebx_ebp: need comma ONLY if both CLOBBER_ebp and CLOBBER_GOT_ebx -// have values, i.e., only if __x86_64__ AND !__PIC__ -#if defined(__x86_64__) && !defined(__PIC__) -# define CLOB_COMMA_ebx_ebp , // clobbering both ebp and ebx => need comma -#else -# define CLOB_COMMA_ebx_ebp -#endif - -// CLOB_COMMA_ebX_r1X: need comma UNLESS both CLOBBER_ebp and CLOBBER_GOT_ebx -// are empty OR CLOBBER_r11_r12_r13 is empty--i.e., NO comma -// if (!__x86_64__ AND __PIC__) OR !(PNG_x86_64_USE_GOTPCREL -// AND PNG_CLOBBER_x86_64_REGS_SUPPORTED) (double sigh...) -#if (!defined(__x86_64__) && defined(__PIC__)) || \ - !defined(PNG_x86_64_USE_GOTPCREL) || \ - !defined(PNG_CLOBBER_x86_64_REGS_SUPPORTED) -# define CLOB_COMMA_ebX_r1X -#else -# define CLOB_COMMA_ebX_r1X , // clobbering (ebp OR ebx) AND r11_r12_r13 -#endif - -// CLOB_COLON_ebx_ebp: need colon unless CLOBBER_ebp and CLOBBER_GOT_ebx are -// BOTH empty--i.e., NO colon if (!__x86_64__ AND __PIC__) -// CLOB_COLON_ebx_ebp_r1X: if, in addition, CLOBBER_r11_r12_r13 is empty, then -// no colon for Paeth blocks, either--i.e., NO colon -// if !(PNG_x86_64_USE_GOTPCREL AND -// PNG_CLOBBER_x86_64_REGS_SUPPORTED) -#if (!defined(__x86_64__) && defined(__PIC__)) -# define CLOB_COLON_ebx_ebp -# if !(defined(PNG_x86_64_USE_GOTPCREL) && \ - defined(PNG_CLOBBER_x86_64_REGS_SUPPORTED)) -# define CLOB_COLON_ebx_ebp_r1X -# else -# define CLOB_COLON_ebx_ebp_r1X : // clobbering ebp OR ebx OR r11_r12_r13 -# endif -#else -# define CLOB_COLON_ebx_ebp : // clobbering ebp OR ebx -# define CLOB_COLON_ebx_ebp_r1X : // clobbering ebp OR ebx OR r11_r12_r13 -#endif - -#endif // PNG_HAVE_MMX_READ_FILTER_ROW - -#if defined(__PIC__) // macros to save, restore index to Global Offset Table -# if defined(__x86_64__) -# define SAVE_GOT_ebx "pushq %%rbx \n\t" -# define RESTORE_GOT_ebx "popq %%rbx \n\t" -# else -# define SAVE_GOT_ebx "pushl %%ebx \n\t" -# define RESTORE_GOT_ebx "popl %%ebx \n\t" -# endif -# define _CLOBBER_GOT_ebx // explicitly saved, restored => not clobbered -# define CLOBBER_GOT_ebx -#else -# define SAVE_GOT_ebx -# define RESTORE_GOT_ebx -# define _CLOBBER_GOT_ebx ,"%ebx" -# define CLOBBER_GOT_ebx "%ebx" -#endif - -#if defined(PNG_HAVE_MMX_COMBINE_ROW) || defined(PNG_HAVE_MMX_READ_INTERLACE) -# define BPP2 2 -# define BPP3 3 // bytes per pixel (a.k.a. pixel_bytes) -# define BPP4 4 // (defined only to help avoid cut-and-paste errors) -# define BPP6 6 -# define BPP8 8 -#endif - - - -static int _mmx_supported = 2; // 0: no MMX; 1: MMX supported; 2: not tested - -/*===========================================================================*/ -/* */ -/* P N G _ M M X _ S U P P O R T */ -/* */ -/*===========================================================================*/ - -// GRR NOTES: (1) the following code assumes 386 or better (pushfl/popfl) -// (2) all instructions compile with gcc 2.7.2.3 and later -// x (3) the function is moved down here to prevent gcc from -// x inlining it in multiple places and then barfing be- -// x cause the ".NOT_SUPPORTED" label is multiply defined -// [need to retest with gcc 2.7.2.3] - -// GRR 20070524: This declaration apparently is compatible with but supersedes -// the one in png.h; in any case, the generated object file is slightly -// smaller. It is unnecessary with gcc 4.1.2, but gcc 2.x apparently -// replicated the ".NOT_SUPPORTED" label in each location the function was -// inlined, leading to compilation errors due to the "multiply defined" -// label. Old workaround was to leave the function at the end of this -// file; new one (still testing) is to use a gcc-specific function attribute -// to prevent local inlining. -int PNGAPI -png_mmx_support(void) __attribute__((noinline)); - -int PNGAPI -png_mmx_support(void) -{ - int result; -#if defined(PNG_MMX_CODE_SUPPORTED) // superfluous, but what the heck - __asm__ __volatile__ ( -#if defined(__x86_64__) - "pushq %%rbx \n\t" // rbx gets clobbered by CPUID instruction - "pushq %%rcx \n\t" // so does rcx... - "pushq %%rdx \n\t" // ...and rdx (but rcx & rdx safe on Linux) - "pushfq \n\t" // save Eflag to stack - "popq %%rax \n\t" // get Eflag from stack into rax - "movq %%rax, %%rcx \n\t" // make another copy of Eflag in rcx - "xorl $0x200000, %%eax \n\t" // toggle ID bit in Eflag (i.e., bit 21) - "pushq %%rax \n\t" // save modified Eflag back to stack - "popfq \n\t" // restore modified value to Eflag reg - "pushfq \n\t" // save Eflag to stack - "popq %%rax \n\t" // get Eflag from stack - "pushq %%rcx \n\t" // save original Eflag to stack - "popfq \n\t" // restore original Eflag -#else - "pushl %%ebx \n\t" // ebx gets clobbered by CPUID instruction - "pushl %%ecx \n\t" // so does ecx... - "pushl %%edx \n\t" // ...and edx (but ecx & edx safe on Linux) - "pushfl \n\t" // save Eflag to stack - "popl %%eax \n\t" // get Eflag from stack into eax - "movl %%eax, %%ecx \n\t" // make another copy of Eflag in ecx - "xorl $0x200000, %%eax \n\t" // toggle ID bit in Eflag (i.e., bit 21) - "pushl %%eax \n\t" // save modified Eflag back to stack - "popfl \n\t" // restore modified value to Eflag reg - "pushfl \n\t" // save Eflag to stack - "popl %%eax \n\t" // get Eflag from stack - "pushl %%ecx \n\t" // save original Eflag to stack - "popfl \n\t" // restore original Eflag -#endif - "xorl %%ecx, %%eax \n\t" // compare new Eflag with original Eflag - "jz 0f \n\t" // if same, CPUID instr. is not supported - - "xorl %%eax, %%eax \n\t" // set eax to zero -// ".byte 0x0f, 0xa2 \n\t" // CPUID instruction (two-byte opcode) - "cpuid \n\t" // get the CPU identification info - "cmpl $1, %%eax \n\t" // make sure eax return non-zero value - "jl 0f \n\t" // if eax is zero, MMX is not supported - - "xorl %%eax, %%eax \n\t" // set eax to zero and... - "incl %%eax \n\t" // ...increment eax to 1. This pair is - // faster than the instruction "mov eax, 1" - "cpuid \n\t" // get the CPU identification info again - "andl $0x800000, %%edx \n\t" // mask out all bits but MMX bit (23) - "cmpl $0, %%edx \n\t" // 0 = MMX not supported - "jz 0f \n\t" // non-zero = yes, MMX IS supported - - "movl $1, %%eax \n\t" // set return value to 1 - "jmp 1f \n\t" // DONE: have MMX support - - "0: \n\t" // .NOT_SUPPORTED: target label for jump instructions - "movl $0, %%eax \n\t" // set return value to 0 - "1: \n\t" // .RETURN: target label for jump instructions -#if defined(__x86_64__) - "popq %%rdx \n\t" // restore rdx - "popq %%rcx \n\t" // restore rcx - "popq %%rbx \n\t" // restore rbx -#else - "popl %%edx \n\t" // restore edx - "popl %%ecx \n\t" // restore ecx - "popl %%ebx \n\t" // restore ebx -#endif - -// "ret \n\t" // DONE: no MMX support - // (fall through to standard C "ret") - - : "=a" (result) // output list - - : // any variables used on input (none) - - // no clobber list -// , "%ebx", "%ecx", "%edx" // GRR: we handle these manually -// , "memory" // if write to a variable gcc thought was in a reg -// , "cc" // "condition codes" (flag bits) - ); - _mmx_supported = result; -#else - _mmx_supported = 0; -#endif /* PNG_MMX_CODE_SUPPORTED */ - - return _mmx_supported; -} - - -/*===========================================================================*/ -/* */ -/* P N G _ C O M B I N E _ R O W */ -/* */ -/*===========================================================================*/ - -#if defined(PNG_HAVE_MMX_COMBINE_ROW) - -/* Combines the row recently read in with the previous row. - This routine takes care of alpha and transparency if requested. - This routine also handles the two methods of progressive display - of interlaced images, depending on the mask value. - The mask value describes which pixels are to be combined with - the row. The pattern always repeats every 8 pixels, so just 8 - bits are needed. A one indicates the pixel is to be combined; a - zero indicates the pixel is to be skipped. This is in addition - to any alpha or transparency value associated with the pixel. - If you want all pixels to be combined, pass 0xff (255) in mask. */ - -/* Use this routine for the x86 platform - it uses a faster MMX routine - if the machine supports MMX. */ - -void /* PRIVATE */ -png_combine_row(png_structp png_ptr, png_bytep row, int mask) -{ - int dummy_value_a; // fix 'forbidden register spilled' error - int dummy_value_c; - int dummy_value_d; - png_bytep dummy_value_S; - png_bytep dummy_value_D; - - png_debug(1, "in png_combine_row (pnggccrd.c)\n"); - - if (_mmx_supported == 2) { -#if !defined(PNG_1_0_X) - /* this should have happened in png_init_mmx_flags() already */ - png_warning(png_ptr, "asm_flags may not have been initialized"); -#endif - png_mmx_support(); - } - - if (mask == 0xff) - { - png_debug(2,"mask == 0xff: doing single png_memcpy()\n"); - png_memcpy(row, png_ptr->row_buf + 1, - (png_size_t)PNG_ROWBYTES(png_ptr->row_info.pixel_depth,png_ptr->width)); - } - else /* (png_combine_row() is never called with mask == 0) */ - { - switch (png_ptr->row_info.pixel_depth) - { - case 24: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - -#if !defined(PNG_1_0_X) - if (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) -#else - if (_mmx_supported) -#endif - { - png_uint_32 len; - int diff; - - srcptr = png_ptr->row_buf + 1; - dstptr = row; - len = png_ptr->width & ~7; // reduce to multiple of 8 - diff = (int) (png_ptr->width & 7); // amount lost - - __asm__ __volatile__ ( - "not %%edx \n\t" // mask => unmask - "movd %%edx, %%mm7 \n\t" // load bit pattern - "not %%edx \n\t" // unmask => mask for later - "psubb %%mm6, %%mm6 \n\t" // zero mm6 - "punpcklbw %%mm7, %%mm7 \n\t" - "punpcklwd %%mm7, %%mm7 \n\t" - "punpckldq %%mm7, %%mm7 \n\t" // fill reg with 8 masks - - LOAD_GOT_rbp - "movq " MASK24_0 ", %%mm0 \n\t" // _mask24_0 -> mm0 - "movq " MASK24_1 ", %%mm1 \n\t" // _mask24_1 -> mm1 - "movq " MASK24_2 ", %%mm2 \n\t" // _mask24_2 -> mm2 - RESTORE_rbp - - "pand %%mm7, %%mm0 \n\t" - "pand %%mm7, %%mm1 \n\t" - "pand %%mm7, %%mm2 \n\t" - - "pcmpeqb %%mm6, %%mm0 \n\t" - "pcmpeqb %%mm6, %%mm1 \n\t" - "pcmpeqb %%mm6, %%mm2 \n\t" - -// preload "movl len, %%ecx \n\t" // load length of line -// preload "movl srcptr, %3 \n\t" // load source -// preload "movl dstptr, %4 \n\t" // load dest - - "cmpl $0, %%ecx \n\t" - "jz mainloop24end \n\t" - - "mainloop24: \n\t" - "movq (%3), %%mm4 \n\t" - "pand %%mm0, %%mm4 \n\t" - "movq %%mm0, %%mm6 \n\t" - "movq (%4), %%mm7 \n\t" - "pandn %%mm7, %%mm6 \n\t" - "por %%mm6, %%mm4 \n\t" - "movq %%mm4, (%4) \n\t" - - "movq 8(%3), %%mm5 \n\t" - "pand %%mm1, %%mm5 \n\t" - "movq %%mm1, %%mm7 \n\t" - "movq 8(%4), %%mm6 \n\t" - "pandn %%mm6, %%mm7 \n\t" - "por %%mm7, %%mm5 \n\t" - "movq %%mm5, 8(%4) \n\t" - - "movq 16(%3), %%mm6 \n\t" - "pand %%mm2, %%mm6 \n\t" - "movq %%mm2, %%mm4 \n\t" - "movq 16(%4), %%mm7 \n\t" - "pandn %%mm7, %%mm4 \n\t" - "por %%mm4, %%mm6 \n\t" - "movq %%mm6, 16(%4) \n\t" - - "add $24, %3 \n\t" // inc by 24 bytes processed - "add $24, %4 \n\t" - "subl $8, %%ecx \n\t" // dec by 8 pixels processed - - "ja mainloop24 \n\t" - - "mainloop24end: \n\t" -// preload "movl diff, %%ecx \n\t" // (diff is in eax) - "movl %%eax, %%ecx \n\t" - "cmpl $0, %%ecx \n\t" - "jz end24 \n\t" -// preload "movl mask, %%edx \n\t" - "sall $24, %%edx \n\t" // make low byte, high byte - - "secondloop24: \n\t" - "sall %%edx \n\t" // move high bit to CF - "jnc skip24 \n\t" // if CF = 0 - "movw (%3), %%ax \n\t" - "movw %%ax, (%4) \n\t" - "xorl %%eax, %%eax \n\t" - "movb 2(%3), %%al \n\t" - "movb %%al, 2(%4) \n\t" - - "skip24: \n\t" - "add $3, %3 \n\t" - "add $3, %4 \n\t" - "decl %%ecx \n\t" - "jnz secondloop24 \n\t" - - "end24: \n\t" - "EMMS \n\t" // DONE - - : "=a" (dummy_value_a), // output regs (dummy) - "=d" (dummy_value_d), - "=c" (dummy_value_c), - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (diff), // eax // input regs - "1" (mask), // edx - "2" (len), // ecx -// was (unmask) "b" RESERVED // ebx // Global Offset Table idx - "3" (srcptr), // esi/rsi - "4" (dstptr) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - else /* not _mmx_supported - use modified C routine */ - { - register png_uint_32 i; - png_uint_32 initial_val = BPP3 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP3 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP3 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult. of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP3 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP3; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } /* end of else (_mmx_supported) */ - - break; - } /* end 24 bpp */ - - // formerly claimed to be most common case (combining 32-bit RGBA), - // but almost certainly less common than 24-bit RGB case - case 32: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - -#if !defined(PNG_1_0_X) - if (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) -#else - if (_mmx_supported) -#endif - { - png_uint_32 len; - int diff; - - srcptr = png_ptr->row_buf + 1; - dstptr = row; - len = png_ptr->width & ~7; // reduce to multiple of 8 - diff = (int) (png_ptr->width & 7); // amount lost - - __asm__ __volatile__ ( - "not %%edx \n\t" // mask => unmask - "movd %%edx, %%mm7 \n\t" // load bit pattern - "not %%edx \n\t" // unmask => mask for later - "psubb %%mm6, %%mm6 \n\t" // zero mm6 - "punpcklbw %%mm7, %%mm7 \n\t" - "punpcklwd %%mm7, %%mm7 \n\t" - "punpckldq %%mm7, %%mm7 \n\t" // fill reg with 8 masks - - LOAD_GOT_rbp - "movq " MASK32_0 ", %%mm0 \n\t" // _mask32_0 - "movq " MASK32_1 ", %%mm1 \n\t" // _mask32_1 - "movq " MASK32_2 ", %%mm2 \n\t" // _mask32_2 - "movq " MASK32_3 ", %%mm3 \n\t" // _mask32_3 - RESTORE_rbp - - "pand %%mm7, %%mm0 \n\t" - "pand %%mm7, %%mm1 \n\t" - "pand %%mm7, %%mm2 \n\t" - "pand %%mm7, %%mm3 \n\t" - - "pcmpeqb %%mm6, %%mm0 \n\t" - "pcmpeqb %%mm6, %%mm1 \n\t" - "pcmpeqb %%mm6, %%mm2 \n\t" - "pcmpeqb %%mm6, %%mm3 \n\t" - -// preload "movl len, %%ecx \n\t" // load length of line -// preload "movl srcptr, %3 \n\t" // load source -// preload "movl dstptr, %4 \n\t" // load dest - - "cmpl $0, %%ecx \n\t" // lcr - "jz mainloop32end \n\t" - - "mainloop32: \n\t" - "movq (%3), %%mm4 \n\t" - "pand %%mm0, %%mm4 \n\t" - "movq %%mm0, %%mm6 \n\t" - "movq (%4), %%mm7 \n\t" - "pandn %%mm7, %%mm6 \n\t" - "por %%mm6, %%mm4 \n\t" - "movq %%mm4, (%4) \n\t" - - "movq 8(%3), %%mm5 \n\t" - "pand %%mm1, %%mm5 \n\t" - "movq %%mm1, %%mm7 \n\t" - "movq 8(%4), %%mm6 \n\t" - "pandn %%mm6, %%mm7 \n\t" - "por %%mm7, %%mm5 \n\t" - "movq %%mm5, 8(%4) \n\t" - - "movq 16(%3), %%mm6 \n\t" - "pand %%mm2, %%mm6 \n\t" - "movq %%mm2, %%mm4 \n\t" - "movq 16(%4), %%mm7 \n\t" - "pandn %%mm7, %%mm4 \n\t" - "por %%mm4, %%mm6 \n\t" - "movq %%mm6, 16(%4) \n\t" - - "movq 24(%3), %%mm7 \n\t" - "pand %%mm3, %%mm7 \n\t" - "movq %%mm3, %%mm5 \n\t" - "movq 24(%4), %%mm4 \n\t" - "pandn %%mm4, %%mm5 \n\t" - "por %%mm5, %%mm7 \n\t" - "movq %%mm7, 24(%4) \n\t" - - "add $32, %3 \n\t" // inc by 32 bytes processed - "add $32, %4 \n\t" - "subl $8, %%ecx \n\t" // dec by 8 pixels processed - "ja mainloop32 \n\t" - - "mainloop32end: \n\t" -// preload "movl diff, %%ecx \n\t" // (diff is in eax) - "movl %%eax, %%ecx \n\t" - "cmpl $0, %%ecx \n\t" - "jz end32 \n\t" -// preload "movl mask, %%edx \n\t" - "sall $24, %%edx \n\t" // low byte => high byte - - "secondloop32: \n\t" - "sall %%edx \n\t" // move high bit to CF - "jnc skip32 \n\t" // if CF = 0 - "movl (%3), %%eax \n\t" - "movl %%eax, (%4) \n\t" - - "skip32: \n\t" - "add $4, %3 \n\t" - "add $4, %4 \n\t" - "decl %%ecx \n\t" - "jnz secondloop32 \n\t" - - "end32: \n\t" - "EMMS \n\t" // DONE - - : "=a" (dummy_value_a), // output regs (dummy) - "=d" (dummy_value_d), - "=c" (dummy_value_c), - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (diff), // eax // input regs - "1" (mask), // edx - "2" (len), // ecx -// was (unmask) "b" RESERVED // ebx // Global Offset Table idx - "3" (srcptr), // esi/rsi - "4" (dstptr) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - else /* not _mmx_supported - use modified C routine */ - { - register png_uint_32 i; - png_uint_32 initial_val = BPP4 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP4 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP4 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult. of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP4 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP4; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } /* end of else (_mmx_supported) */ - - break; - } /* end 32 bpp */ - - case 8: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - -#if !defined(PNG_1_0_X) - if (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) -#else - if (_mmx_supported) -#endif - { - png_uint_32 len; - int diff; - - srcptr = png_ptr->row_buf + 1; - dstptr = row; - len = png_ptr->width & ~7; // reduce to multiple of 8 - diff = (int) (png_ptr->width & 7); // amount lost - - __asm__ __volatile__ ( - "not %%edx \n\t" // mask => unmask - "movd %%edx, %%mm7 \n\t" // load bit pattern - "not %%edx \n\t" // unmask => mask for later - "psubb %%mm6, %%mm6 \n\t" // zero mm6 - "punpcklbw %%mm7, %%mm7 \n\t" - "punpcklwd %%mm7, %%mm7 \n\t" - "punpckldq %%mm7, %%mm7 \n\t" // fill reg with 8 masks - - LOAD_GOT_rbp - "movq " MASK8_0 ", %%mm0 \n\t" // _mask8_0 -> mm0 - RESTORE_rbp - - "pand %%mm7, %%mm0 \n\t" // nonzero if keep byte - "pcmpeqb %%mm6, %%mm0 \n\t" // zeros->1s, v versa - -// preload "movl len, %%ecx \n\t" // load length of line -// preload "movl srcptr, %3 \n\t" // load source -// preload "movl dstptr, %4 \n\t" // load dest - - "cmpl $0, %%ecx \n\t" // len == 0 ? - "je mainloop8end \n\t" - - "mainloop8: \n\t" - "movq (%3), %%mm4 \n\t" // *srcptr - "pand %%mm0, %%mm4 \n\t" - "movq %%mm0, %%mm6 \n\t" - "pandn (%4), %%mm6 \n\t" // *dstptr - "por %%mm6, %%mm4 \n\t" - "movq %%mm4, (%4) \n\t" - "add $8, %3 \n\t" // inc by 8 bytes processed - "add $8, %4 \n\t" - "subl $8, %%ecx \n\t" // dec by 8 pixels processed - "ja mainloop8 \n\t" - - "mainloop8end: \n\t" -// preload "movl diff, %%ecx \n\t" // (diff is in eax) - "movl %%eax, %%ecx \n\t" - "cmpl $0, %%ecx \n\t" - "jz end8 \n\t" -// preload "movl mask, %%edx \n\t" - "sall $24, %%edx \n\t" // make low byte, high byte - - "secondloop8: \n\t" - "sall %%edx \n\t" // move high bit to CF - "jnc skip8 \n\t" // if CF = 0 - "movb (%3), %%al \n\t" - "movb %%al, (%4) \n\t" - - "skip8: \n\t" - "inc %3 \n\t" - "inc %4 \n\t" - "decl %%ecx \n\t" - "jnz secondloop8 \n\t" - - "end8: \n\t" - "EMMS \n\t" // DONE - - : "=a" (dummy_value_a), // output regs (dummy) - "=d" (dummy_value_d), - "=c" (dummy_value_c), - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (diff), // eax // input regs - "1" (mask), // edx - "2" (len), // ecx -// was (unmask) "b" RESERVED // ebx // Global Offset Table idx - "3" (srcptr), // esi/rsi - "4" (dstptr) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm4", "%mm6", "%mm7" // clobber list -#endif - ); - } - else /* not _mmx_supported - use modified C routine */ - { - register png_uint_32 i; - png_uint_32 initial_val = png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult. of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff /* *BPP1 */ ; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - - } /* end of else (_mmx_supported) */ - - break; - } /* end 8 bpp */ - - case 1: /* png_ptr->row_info.pixel_depth */ - { - png_bytep sp; - png_bytep dp; - int s_inc, s_start, s_end; - int m; - int shift; - png_uint_32 i; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 7; - s_inc = 1; - } - else -#endif - { - s_start = 7; - s_end = 0; - s_inc = -1; - } - - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - int value; - - value = (*sp >> shift) & 0x1; - *dp &= (png_byte)((0x7f7f >> (7 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } /* end 1 bpp */ - - case 2: /* png_ptr->row_info.pixel_depth */ - { - png_bytep sp; - png_bytep dp; - int s_start, s_end, s_inc; - int m; - int shift; - png_uint_32 i; - int value; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 6; - s_inc = 2; - } - else -#endif - { - s_start = 6; - s_end = 0; - s_inc = -2; - } - - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - value = (*sp >> shift) & 0x3; - *dp &= (png_byte)((0x3f3f >> (6 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } /* end 2 bpp */ - - case 4: /* png_ptr->row_info.pixel_depth */ - { - png_bytep sp; - png_bytep dp; - int s_start, s_end, s_inc; - int m; - int shift; - png_uint_32 i; - int value; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 4; - s_inc = 4; - } - else -#endif - { - s_start = 4; - s_end = 0; - s_inc = -4; - } - - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - value = (*sp >> shift) & 0xf; - *dp &= (png_byte)((0xf0f >> (4 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } /* end 4 bpp */ - - case 16: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - -#if !defined(PNG_1_0_X) - if (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) -#else - if (_mmx_supported) -#endif - { - png_uint_32 len; - int diff; - - srcptr = png_ptr->row_buf + 1; - dstptr = row; - len = png_ptr->width & ~7; // reduce to multiple of 8 - diff = (int) (png_ptr->width & 7); // amount lost - - __asm__ __volatile__ ( - "not %%edx \n\t" // mask => unmask - "movd %%edx, %%mm7 \n\t" // load bit pattern - "not %%edx \n\t" // unmask => mask for later - "psubb %%mm6, %%mm6 \n\t" // zero mm6 - "punpcklbw %%mm7, %%mm7 \n\t" - "punpcklwd %%mm7, %%mm7 \n\t" - "punpckldq %%mm7, %%mm7 \n\t" // fill reg with 8 masks - - LOAD_GOT_rbp - "movq " MASK16_0 ", %%mm0 \n\t" // _mask16_0 -> mm0 - "movq " MASK16_1 ", %%mm1 \n\t" // _mask16_1 -> mm1 - RESTORE_rbp - - "pand %%mm7, %%mm0 \n\t" - "pand %%mm7, %%mm1 \n\t" - - "pcmpeqb %%mm6, %%mm0 \n\t" - "pcmpeqb %%mm6, %%mm1 \n\t" - -// preload "movl len, %%ecx \n\t" // load length of line -// preload "movl srcptr, %3 \n\t" // load source -// preload "movl dstptr, %4 \n\t" // load dest - - "cmpl $0, %%ecx \n\t" - "jz mainloop16end \n\t" - - "mainloop16: \n\t" - "movq (%3), %%mm4 \n\t" - "pand %%mm0, %%mm4 \n\t" - "movq %%mm0, %%mm6 \n\t" - "movq (%4), %%mm7 \n\t" - "pandn %%mm7, %%mm6 \n\t" - "por %%mm6, %%mm4 \n\t" - "movq %%mm4, (%4) \n\t" - - "movq 8(%3), %%mm5 \n\t" - "pand %%mm1, %%mm5 \n\t" - "movq %%mm1, %%mm7 \n\t" - "movq 8(%4), %%mm6 \n\t" - "pandn %%mm6, %%mm7 \n\t" - "por %%mm7, %%mm5 \n\t" - "movq %%mm5, 8(%4) \n\t" - - "add $16, %3 \n\t" // inc by 16 bytes processed - "add $16, %4 \n\t" - "subl $8, %%ecx \n\t" // dec by 8 pixels processed - "ja mainloop16 \n\t" - - "mainloop16end: \n\t" -// preload "movl diff, %%ecx \n\t" // (diff is in eax) - "movl %%eax, %%ecx \n\t" - "cmpl $0, %%ecx \n\t" - "jz end16 \n\t" -// preload "movl mask, %%edx \n\t" - "sall $24, %%edx \n\t" // make low byte, high byte - - "secondloop16: \n\t" - "sall %%edx \n\t" // move high bit to CF - "jnc skip16 \n\t" // if CF = 0 - "movw (%3), %%ax \n\t" - "movw %%ax, (%4) \n\t" - - "skip16: \n\t" - "add $2, %3 \n\t" - "add $2, %4 \n\t" - "decl %%ecx \n\t" - "jnz secondloop16 \n\t" - - "end16: \n\t" - "EMMS \n\t" // DONE - - : "=a" (dummy_value_a), // output regs (dummy) - "=d" (dummy_value_d), - "=c" (dummy_value_c), - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (diff), // eax // input regs - "1" (mask), // edx - "2" (len), // ecx -// was (unmask) "b" RESERVED // ebx // Global Offset Table idx - "3" (srcptr), // esi/rsi - "4" (dstptr) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm4" // clobber list - , "%mm5", "%mm6", "%mm7" -#endif - ); - } - else /* not _mmx_supported - use modified C routine */ - { - register png_uint_32 i; - png_uint_32 initial_val = BPP2 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP2 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP2 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult. of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP2 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP2; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } /* end of else (_mmx_supported) */ - - break; - } /* end 16 bpp */ - - case 48: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - -#if !defined(PNG_1_0_X) - if (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) -#else - if (_mmx_supported) -#endif - { - png_uint_32 len; - int diff; - - srcptr = png_ptr->row_buf + 1; - dstptr = row; - len = png_ptr->width & ~7; // reduce to multiple of 8 - diff = (int) (png_ptr->width & 7); // amount lost - - __asm__ __volatile__ ( - "not %%edx \n\t" // mask => unmask - "movd %%edx, %%mm7 \n\t" // load bit pattern - "not %%edx \n\t" // unmask => mask for later - "psubb %%mm6, %%mm6 \n\t" // zero mm6 - "punpcklbw %%mm7, %%mm7 \n\t" - "punpcklwd %%mm7, %%mm7 \n\t" - "punpckldq %%mm7, %%mm7 \n\t" // fill reg with 8 masks - - LOAD_GOT_rbp - "movq " MASK48_0 ", %%mm0 \n\t" // _mask48_0 -> mm0 - "movq " MASK48_1 ", %%mm1 \n\t" // _mask48_1 -> mm1 - "movq " MASK48_2 ", %%mm2 \n\t" // _mask48_2 -> mm2 - "movq " MASK48_3 ", %%mm3 \n\t" // _mask48_3 -> mm3 - "movq " MASK48_4 ", %%mm4 \n\t" // _mask48_4 -> mm4 - "movq " MASK48_5 ", %%mm5 \n\t" // _mask48_5 -> mm5 - RESTORE_rbp - - "pand %%mm7, %%mm0 \n\t" - "pand %%mm7, %%mm1 \n\t" - "pand %%mm7, %%mm2 \n\t" - "pand %%mm7, %%mm3 \n\t" - "pand %%mm7, %%mm4 \n\t" - "pand %%mm7, %%mm5 \n\t" - - "pcmpeqb %%mm6, %%mm0 \n\t" - "pcmpeqb %%mm6, %%mm1 \n\t" - "pcmpeqb %%mm6, %%mm2 \n\t" - "pcmpeqb %%mm6, %%mm3 \n\t" - "pcmpeqb %%mm6, %%mm4 \n\t" - "pcmpeqb %%mm6, %%mm5 \n\t" - -// preload "movl len, %%ecx \n\t" // load length of line -// preload "movl srcptr, %3 \n\t" // load source -// preload "movl dstptr, %4 \n\t" // load dest - - "cmpl $0, %%ecx \n\t" - "jz mainloop48end \n\t" - - "mainloop48: \n\t" - "movq (%3), %%mm7 \n\t" - "pand %%mm0, %%mm7 \n\t" - "movq %%mm0, %%mm6 \n\t" - "pandn (%4), %%mm6 \n\t" - "por %%mm6, %%mm7 \n\t" - "movq %%mm7, (%4) \n\t" - - "movq 8(%3), %%mm6 \n\t" - "pand %%mm1, %%mm6 \n\t" - "movq %%mm1, %%mm7 \n\t" - "pandn 8(%4), %%mm7 \n\t" - "por %%mm7, %%mm6 \n\t" - "movq %%mm6, 8(%4) \n\t" - - "movq 16(%3), %%mm6 \n\t" - "pand %%mm2, %%mm6 \n\t" - "movq %%mm2, %%mm7 \n\t" - "pandn 16(%4), %%mm7 \n\t" - "por %%mm7, %%mm6 \n\t" - "movq %%mm6, 16(%4) \n\t" - - "movq 24(%3), %%mm7 \n\t" - "pand %%mm3, %%mm7 \n\t" - "movq %%mm3, %%mm6 \n\t" - "pandn 24(%4), %%mm6 \n\t" - "por %%mm6, %%mm7 \n\t" - "movq %%mm7, 24(%4) \n\t" - - "movq 32(%3), %%mm6 \n\t" - "pand %%mm4, %%mm6 \n\t" - "movq %%mm4, %%mm7 \n\t" - "pandn 32(%4), %%mm7 \n\t" - "por %%mm7, %%mm6 \n\t" - "movq %%mm6, 32(%4) \n\t" - - "movq 40(%3), %%mm7 \n\t" - "pand %%mm5, %%mm7 \n\t" - "movq %%mm5, %%mm6 \n\t" - "pandn 40(%4), %%mm6 \n\t" - "por %%mm6, %%mm7 \n\t" - "movq %%mm7, 40(%4) \n\t" - - "add $48, %3 \n\t" // inc by 48 bytes processed - "add $48, %4 \n\t" - "subl $8, %%ecx \n\t" // dec by 8 pixels processed - - "ja mainloop48 \n\t" - - "mainloop48end: \n\t" -// preload "movl diff, %%ecx \n\t" // (diff is in eax) - "movl %%eax, %%ecx \n\t" - "cmpl $0, %%ecx \n\t" - "jz end48 \n\t" -// preload "movl mask, %%edx \n\t" - "sall $24, %%edx \n\t" // make low byte, high byte - - "secondloop48: \n\t" - "sall %%edx \n\t" // move high bit to CF - "jnc skip48 \n\t" // if CF = 0 - "movl (%3), %%eax \n\t" - "movl %%eax, (%4) \n\t" - "movw 4(%3), %%ax \n\t" // GR-P bugfix 20070717 - "movw %%ax, 4(%4) \n\t" // GR-P bugfix 20070717 - - "skip48: \n\t" - "add $6, %3 \n\t" // GR-P bugfix 20070717 - "add $6, %4 \n\t" // GR-P bugfix 20070717 - "decl %%ecx \n\t" - "jnz secondloop48 \n\t" - - "end48: \n\t" - "EMMS \n\t" // DONE - - : "=a" (dummy_value_a), // output regs (dummy) - "=d" (dummy_value_d), - "=c" (dummy_value_c), - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (diff), // eax // input regs - "1" (mask), // edx - "2" (len), // ecx -// was (unmask) "b" RESERVED // ebx // Global Offset Table idx - "3" (srcptr), // esi/rsi - "4" (dstptr) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - else /* not _mmx_supported - use modified C routine */ - { - register png_uint_32 i; - png_uint_32 initial_val = BPP6 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP6 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP6 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult. of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP6 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP6; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } /* end of else (_mmx_supported) */ - - break; - } /* end 48 bpp */ - - case 64: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - register png_uint_32 i; - png_uint_32 initial_val = BPP8 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP8 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP8 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult. of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP8 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP8; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - - break; - } /* end 64 bpp */ - - default: /* png_ptr->row_info.pixel_depth != 1,2,4,8,16,24,32,48,64 */ - { - // ERROR: SHOULD NEVER BE REACHED -#if defined(PNG_DEBUG) - png_debug(1, "Internal libpng logic error (GCC " - "png_combine_row() pixel_depth)\n"); -#endif - break; - } - } /* end switch (png_ptr->row_info.pixel_depth) */ - - } /* end if (non-trivial mask) */ - -} /* end png_combine_row() */ - -#endif /* PNG_HAVE_MMX_COMBINE_ROW */ - - - - -/*===========================================================================*/ -/* */ -/* P N G _ D O _ R E A D _ I N T E R L A C E */ -/* */ -/*===========================================================================*/ - -#if defined(PNG_READ_INTERLACING_SUPPORTED) -#if defined(PNG_HAVE_MMX_READ_INTERLACE) - -/* png_do_read_interlace() is called after any 16-bit to 8-bit conversion - * has taken place. [GRR: what other steps come before and/or after?] - */ - -void /* PRIVATE */ -png_do_read_interlace(png_structp png_ptr) -{ - png_row_infop row_info = &(png_ptr->row_info); - png_bytep row = png_ptr->row_buf + 1; - int pass = png_ptr->pass; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - png_uint_32 transformations = png_ptr->transformations; -#endif - - png_debug(1, "in png_do_read_interlace (pnggccrd.c)\n"); - - if (_mmx_supported == 2) { -#if !defined(PNG_1_0_X) - /* this should have happened in png_init_mmx_flags() already */ - png_warning(png_ptr, "asm_flags may not have been initialized"); -#endif - png_mmx_support(); - } - - if (row != NULL && row_info != NULL) - { - png_uint_32 final_width; - - final_width = row_info->width * png_pass_inc[pass]; - - switch (row_info->pixel_depth) - { - case 1: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_byte v; - png_uint_32 i; - int j; - - sp = row + (png_size_t)((row_info->width - 1) >> 3); - dp = row + (png_size_t)((final_width - 1) >> 3); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (int)((row_info->width + 7) & 7); - dshift = (int)((final_width + 7) & 7); - s_start = 7; - s_end = 0; - s_inc = -1; - } - else -#endif - { - sshift = 7 - (int)((row_info->width + 7) & 7); - dshift = 7 - (int)((final_width + 7) & 7); - s_start = 0; - s_end = 7; - s_inc = 1; - } - - for (i = row_info->width; i; i--) - { - v = (png_byte)((*sp >> sshift) & 0x1); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0x7f7f >> (7 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - case 2: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_uint_32 i; - - sp = row + (png_size_t)((row_info->width - 1) >> 2); - dp = row + (png_size_t)((final_width - 1) >> 2); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (png_size_t)(((row_info->width + 3) & 3) << 1); - dshift = (png_size_t)(((final_width + 3) & 3) << 1); - s_start = 6; - s_end = 0; - s_inc = -2; - } - else -#endif - { - sshift = (png_size_t)((3 - ((row_info->width + 3) & 3)) << 1); - dshift = (png_size_t)((3 - ((final_width + 3) & 3)) << 1); - s_start = 0; - s_end = 6; - s_inc = 2; - } - - for (i = row_info->width; i; i--) - { - png_byte v; - int j; - - v = (png_byte)((*sp >> sshift) & 0x3); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0x3f3f >> (6 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - case 4: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_uint_32 i; - - sp = row + (png_size_t)((row_info->width - 1) >> 1); - dp = row + (png_size_t)((final_width - 1) >> 1); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (png_size_t)(((row_info->width + 1) & 1) << 2); - dshift = (png_size_t)(((final_width + 1) & 1) << 2); - s_start = 4; - s_end = 0; - s_inc = -4; - } - else -#endif - { - sshift = (png_size_t)((1 - ((row_info->width + 1) & 1)) << 2); - dshift = (png_size_t)((1 - ((final_width + 1) & 1)) << 2); - s_start = 0; - s_end = 4; - s_inc = 4; - } - - for (i = row_info->width; i; i--) - { - png_byte v; - int j; - - v = (png_byte)((*sp >> sshift) & 0xf); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0xf0f >> (4 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - /*====================================================================*/ - - default: /* 8-bit or larger (this is where the routine is modified) */ - { - png_bytep sptr, dp; - png_uint_32 i; - png_size_t pixel_bytes; - int width = (int)row_info->width; - - pixel_bytes = (row_info->pixel_depth >> 3); - - /* point sptr at the last pixel in the pre-expanded row: */ - sptr = row + (width - 1) * pixel_bytes; - - /* point dp at the last pixel position in the expanded row: */ - dp = row + (final_width - 1) * pixel_bytes; - - /* New code by Nirav Chhatrapati - Intel Corporation */ - -#if !defined(PNG_1_0_X) - if (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_INTERLACE) -#else - if (_mmx_supported) -#endif - { - int dummy_value_c; // fix 'forbidden register spilled' - png_bytep dummy_value_S; - png_bytep dummy_value_D; - png_bytep dummy_value_a; - png_bytep dummy_value_d; - - //-------------------------------------------------------------- - if (pixel_bytes == BPP3) - { - if (((pass == 4) || (pass == 5)) && width) - { - int width_mmx = ((width >> 1) << 1) - 8; // GRR: huh? - if (width_mmx < 0) - width_mmx = 0; - width -= width_mmx; // 8 or 9 pix, 24 or 27 bytes - if (width_mmx) - { - // png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; - // sptr points at last pixel in pre-expanded row - // dp points at last pixel position in expanded row - __asm__ __volatile__ ( - "sub $3, %1 \n\t" - "sub $9, %2 \n\t" - // (png_pass_inc[pass] + 1)*pixel_bytes - - ".loop3_pass4: \n\t" - "movq (%1), %%mm0 \n\t" // x x 5 4 3 2 1 0 - "movq %%mm0, %%mm1 \n\t" // x x 5 4 3 2 1 0 - "movq %%mm0, %%mm2 \n\t" // x x 5 4 3 2 1 0 - "psllq $24, %%mm0 \n\t" // 4 3 2 1 0 z z z - "pand (%3), %%mm1 \n\t" // z z z z z 2 1 0 - "psrlq $24, %%mm2 \n\t" // z z z x x 5 4 3 - "por %%mm1, %%mm0 \n\t" // 4 3 2 1 0 2 1 0 - "movq %%mm2, %%mm3 \n\t" // z z z x x 5 4 3 - "psllq $8, %%mm2 \n\t" // z z x x 5 4 3 z - "movq %%mm0, (%2) \n\t" - "psrlq $16, %%mm3 \n\t" // z z z z z x x 5 - "pand (%4), %%mm3 \n\t" // z z z z z z z 5 - "por %%mm3, %%mm2 \n\t" // z z x x 5 4 3 5 - "sub $6, %1 \n\t" - "movd %%mm2, 8(%2) \n\t" - "sub $12, %2 \n\t" - "subl $2, %%ecx \n\t" - "jnz .loop3_pass4 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D), - "=a" (dummy_value_a), - "=d" (dummy_value_d) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp), // edi/rdi -#if defined(PNG_x86_64_USE_GOTPCREL) // formerly _const4 and _const6: - "3" (&_c64._amask5_3_0), // (0x0000000000FFFFFFLL) - "4" (&_c64._amask7_1_0) // (0x00000000000000FFLL) -#else - "3" (&_amask5_3_0), // eax (0x0000000000FFFFFFLL) - "4" (&_amask7_1_0) // edx (0x00000000000000FFLL) -#endif - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list - , "%mm2", "%mm3" -#endif - ); - } - - sptr -= width_mmx*BPP3; - dp -= width_mmx*2*BPP3; - for (i = width; i; i--) - { - png_byte v[8]; - int j; - - png_memcpy(v, sptr, BPP3); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, BPP3); - dp -= BPP3; - } - sptr -= BPP3; - } - } - else if (((pass == 2) || (pass == 3)) && width) - { - __asm__ __volatile__ ( - "sub $9, %2 \n\t" - // (png_pass_inc[pass] - 1)*pixel_bytes - - ".loop3_pass2: \n\t" - "movd (%1), %%mm0 \n\t" // x x x x x 2 1 0 - "pand (%3), %%mm0 \n\t" // z z z z z 2 1 0 - "movq %%mm0, %%mm1 \n\t" // z z z z z 2 1 0 - "psllq $16, %%mm0 \n\t" // z z z 2 1 0 z z - "movq %%mm0, %%mm2 \n\t" // z z z 2 1 0 z z - "psllq $24, %%mm0 \n\t" // 2 1 0 z z z z z - "psrlq $8, %%mm1 \n\t" // z z z z z z 2 1 - "por %%mm2, %%mm0 \n\t" // 2 1 0 2 1 0 z z - "por %%mm1, %%mm0 \n\t" // 2 1 0 2 1 0 2 1 - "movq %%mm0, 4(%2) \n\t" - "psrlq $16, %%mm0 \n\t" // z z 2 1 0 2 1 0 - "sub $3, %1 \n\t" - "movd %%mm0, (%2) \n\t" - "sub $12, %2 \n\t" - "decl %%ecx \n\t" - "jnz .loop3_pass2 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D), - "=a" (dummy_value_a) - - : "0" (width), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp), // edi/rdi -#if defined(PNG_x86_64_USE_GOTPCREL) // formerly _const4: - "3" (&_c64._amask5_3_0) // (0x0000000000FFFFFFLL) -#else - "3" (&_amask5_3_0) // eax (0x0000000000FFFFFFLL) -#endif - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2" // clobber list -#endif - ); - } - else if (width) // && ((pass == 0) || (pass == 1)) - { - __asm__ __volatile__ ( - "sub $21, %2 \n\t" - // (png_pass_inc[pass] - 1)*pixel_bytes - - ".loop3_pass0: \n\t" - "movd (%1), %%mm0 \n\t" // x x x x x 2 1 0 - "pand (%3), %%mm0 \n\t" // z z z z z 2 1 0 - "movq %%mm0, %%mm1 \n\t" // z z z z z 2 1 0 - "psllq $16, %%mm0 \n\t" // z z z 2 1 0 z z - "movq %%mm0, %%mm2 \n\t" // z z z 2 1 0 z z - "psllq $24, %%mm0 \n\t" // 2 1 0 z z z z z - "psrlq $8, %%mm1 \n\t" // z z z z z z 2 1 - "por %%mm2, %%mm0 \n\t" // 2 1 0 2 1 0 z z - "por %%mm1, %%mm0 \n\t" // 2 1 0 2 1 0 2 1 - "movq %%mm0, %%mm3 \n\t" // 2 1 0 2 1 0 2 1 - "psllq $16, %%mm0 \n\t" // 0 2 1 0 2 1 z z - "movq %%mm3, %%mm4 \n\t" // 2 1 0 2 1 0 2 1 - "punpckhdq %%mm0, %%mm3 \n\t" // 0 2 1 0 2 1 0 2 - "movq %%mm4, 16(%2) \n\t" - "psrlq $32, %%mm0 \n\t" // z z z z 0 2 1 0 - "movq %%mm3, 8(%2) \n\t" - "punpckldq %%mm4, %%mm0 \n\t" // 1 0 2 1 0 2 1 0 - "sub $3, %1 \n\t" - "movq %%mm0, (%2) \n\t" - "sub $24, %2 \n\t" - "decl %%ecx \n\t" - "jnz .loop3_pass0 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D), - "=a" (dummy_value_a) - - : "0" (width), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp), // edi/rdi -#if defined(PNG_x86_64_USE_GOTPCREL) // formerly _const4: - "3" (&_c64._amask5_3_0) // (0x0000000000FFFFFFLL) -#else - "3" (&_amask5_3_0) // eax (0x0000000000FFFFFFLL) -#endif - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2" // clobber list - , "%mm3", "%mm4" -#endif - ); - } - } /* end of pixel_bytes == 3 */ - - //-------------------------------------------------------------- - else if (pixel_bytes == BPP4) - { - if (((pass == 4) || (pass == 5)) && width) - { - int width_mmx = ((width >> 1) << 1) ; - width -= width_mmx; // 0,1 pixels => 0,4 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $4, %1 \n\t" - "sub $12, %2 \n\t" - - ".loop4_pass4: \n\t" - "movq (%1), %%mm0 \n\t" // 7 6 5 4 3 2 1 0 - "movq %%mm0, %%mm1 \n\t" // 7 6 5 4 3 2 1 0 - "punpckldq %%mm0, %%mm0 \n\t" // 3 2 1 0 3 2 1 0 - "punpckhdq %%mm1, %%mm1 \n\t" // 7 6 5 4 7 6 5 4 - "movq %%mm0, (%2) \n\t" - "sub $8, %1 \n\t" - "movq %%mm1, 8(%2) \n\t" - "sub $16, %2 \n\t" - "subl $2, %%ecx \n\t" - "jnz .loop4_pass4 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - - sptr -= (width_mmx*BPP4 - BPP4); // sign fixed - dp -= (width_mmx*2*BPP4 - BPP4); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= BPP4; - png_memcpy(v, sptr, BPP4); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= BPP4; - png_memcpy(dp, v, BPP4); - } - } - } - else if (((pass == 2) || (pass == 3)) && width) - { - int width_mmx = ((width >> 1) << 1); - width -= width_mmx; // 0,1 pixels => 0,4 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $4, %1 \n\t" - "sub $28, %2 \n\t" - - ".loop4_pass2: \n\t" - "movq (%1), %%mm0 \n\t" // 7 6 5 4 3 2 1 0 - "movq %%mm0, %%mm1 \n\t" // 7 6 5 4 3 2 1 0 - "punpckldq %%mm0, %%mm0 \n\t" // 3 2 1 0 3 2 1 0 - "punpckhdq %%mm1, %%mm1 \n\t" // 7 6 5 4 7 6 5 4 - "movq %%mm0, (%2) \n\t" - "movq %%mm0, 8(%2) \n\t" - "movq %%mm1, 16(%2) \n\t" - "movq %%mm1, 24(%2) \n\t" - "sub $8, %1 \n\t" - "sub $32, %2 \n\t" - "subl $2, %%ecx \n\t" - "jnz .loop4_pass2 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - - sptr -= (width_mmx*4 - 4); // sign fixed - dp -= (width_mmx*16 - 4); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 4; - png_memcpy(v, sptr, 4); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 4; - png_memcpy(dp, v, 4); - } - } - } - else if (width) // && ((pass == 0) || (pass == 1)) - { - int width_mmx = ((width >> 1) << 1); - width -= width_mmx; // 0,1 pixels => 0,4 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $4, %1 \n\t" - "sub $60, %2 \n\t" - - ".loop4_pass0: \n\t" - "movq (%1), %%mm0 \n\t" // 7 6 5 4 3 2 1 0 - "movq %%mm0, %%mm1 \n\t" // 7 6 5 4 3 2 1 0 - "punpckldq %%mm0, %%mm0 \n\t" // 3 2 1 0 3 2 1 0 - "punpckhdq %%mm1, %%mm1 \n\t" // 7 6 5 4 7 6 5 4 - "movq %%mm0, (%2) \n\t" - "movq %%mm0, 8(%2) \n\t" - "movq %%mm0, 16(%2) \n\t" - "movq %%mm0, 24(%2) \n\t" - "movq %%mm1, 32(%2) \n\t" - "movq %%mm1, 40(%2) \n\t" - "movq %%mm1, 48(%2) \n\t" - "sub $8, %1 \n\t" - "movq %%mm1, 56(%2) \n\t" - "sub $64, %2 \n\t" - "subl $2, %%ecx \n\t" - "jnz .loop4_pass0 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - - sptr -= (width_mmx*4 - 4); // sign fixed - dp -= (width_mmx*32 - 4); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 4; - png_memcpy(v, sptr, 4); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 4; - png_memcpy(dp, v, 4); - } - } - } - } /* end of pixel_bytes == 4 */ - - //-------------------------------------------------------------- - else if (pixel_bytes == 1) - { - if (((pass == 4) || (pass == 5)) && width) - { - int width_mmx = ((width >> 3) << 3); - width -= width_mmx; // 0-3 pixels => 0-3 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $7, %1 \n\t" - "sub $15, %2 \n\t" - - ".loop1_pass4: \n\t" - "movq (%1), %%mm0 \n\t" // 7 6 5 4 3 2 1 0 - "movq %%mm0, %%mm1 \n\t" // 7 6 5 4 3 2 1 0 - "punpcklbw %%mm0, %%mm0 \n\t" // 3 3 2 2 1 1 0 0 - "punpckhbw %%mm1, %%mm1 \n\t" // 7 7 6 6 5 5 4 4 - "movq %%mm1, 8(%2) \n\t" - "sub $8, %1 \n\t" - "movq %%mm0, (%2) \n\t" - "sub $16, %2 \n\t" - "subl $8, %%ecx \n\t" - "jnz .loop1_pass4 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - - sptr -= width_mmx; - dp -= width_mmx*2; - for (i = width; i; i--) - { - int j; - - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp-- = *sptr; - } - --sptr; - } - } - else if (((pass == 2) || (pass == 3)) && width) - { - int width_mmx = ((width >> 2) << 2); - width -= width_mmx; // 0-3 pixels => 0-3 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $3, %1 \n\t" - "sub $15, %2 \n\t" - - ".loop1_pass2: \n\t" - "movd (%1), %%mm0 \n\t" // x x x x 3 2 1 0 - "punpcklbw %%mm0, %%mm0 \n\t" // 3 3 2 2 1 1 0 0 - "movq %%mm0, %%mm1 \n\t" // 3 3 2 2 1 1 0 0 - "punpcklwd %%mm0, %%mm0 \n\t" // 1 1 1 1 0 0 0 0 - "punpckhwd %%mm1, %%mm1 \n\t" // 3 3 3 3 2 2 2 2 - "movq %%mm0, (%2) \n\t" - "sub $4, %1 \n\t" - "movq %%mm1, 8(%2) \n\t" - "sub $16, %2 \n\t" - "subl $4, %%ecx \n\t" - "jnz .loop1_pass2 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - - sptr -= width_mmx; - dp -= width_mmx*4; - for (i = width; i; i--) - { - int j; - - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp-- = *sptr; - } - --sptr; - } - } - else if (width) // && ((pass == 0) || (pass == 1)) - { - int width_mmx = ((width >> 2) << 2); - width -= width_mmx; // 0-3 pixels => 0-3 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $3, %1 \n\t" - "sub $31, %2 \n\t" - - ".loop1_pass0: \n\t" - "movd (%1), %%mm0 \n\t" // x x x x 3 2 1 0 - "movq %%mm0, %%mm1 \n\t" // x x x x 3 2 1 0 - "punpcklbw %%mm0, %%mm0 \n\t" // 3 3 2 2 1 1 0 0 - "movq %%mm0, %%mm2 \n\t" // 3 3 2 2 1 1 0 0 - "punpcklwd %%mm0, %%mm0 \n\t" // 1 1 1 1 0 0 0 0 - "movq %%mm0, %%mm3 \n\t" // 1 1 1 1 0 0 0 0 - "punpckldq %%mm0, %%mm0 \n\t" // 0 0 0 0 0 0 0 0 - "punpckhdq %%mm3, %%mm3 \n\t" // 1 1 1 1 1 1 1 1 - "movq %%mm0, (%2) \n\t" - "punpckhwd %%mm2, %%mm2 \n\t" // 3 3 3 3 2 2 2 2 - "movq %%mm3, 8(%2) \n\t" - "movq %%mm2, %%mm4 \n\t" // 3 3 3 3 2 2 2 2 - "punpckldq %%mm2, %%mm2 \n\t" // 2 2 2 2 2 2 2 2 - "punpckhdq %%mm4, %%mm4 \n\t" // 3 3 3 3 3 3 3 3 - "movq %%mm2, 16(%2) \n\t" - "sub $4, %1 \n\t" - "movq %%mm4, 24(%2) \n\t" - "sub $32, %2 \n\t" - "subl $4, %%ecx \n\t" - "jnz .loop1_pass0 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2" // clobber list - , "%mm3", "%mm4" -#endif - ); - } - - sptr -= width_mmx; - dp -= width_mmx*8; - for (i = width; i; i--) - { - int j; - - /* I simplified this part in version 1.0.4e - * here and in several other instances where - * pixel_bytes == 1 -- GR-P - * - * Original code: - * - * png_byte v[8]; - * png_memcpy(v, sptr, pixel_bytes); - * for (j = 0; j < png_pass_inc[pass]; j++) - * { - * png_memcpy(dp, v, pixel_bytes); - * dp -= pixel_bytes; - * } - * sptr -= pixel_bytes; - * - * Replacement code is in the next three lines: - */ - - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp-- = *sptr; - } - --sptr; - } - } - } /* end of pixel_bytes == 1 */ - - //-------------------------------------------------------------- - else if (pixel_bytes == BPP2) - { - if (((pass == 4) || (pass == 5)) && width) - { - int width_mmx = ((width >> 1) << 1) ; - width -= width_mmx; // 0,1 pixels => 0,2 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $2, %1 \n\t" - "sub $6, %2 \n\t" - - ".loop2_pass4: \n\t" - "movd (%1), %%mm0 \n\t" // x x x x 3 2 1 0 - "punpcklwd %%mm0, %%mm0 \n\t" // 3 2 3 2 1 0 1 0 - "sub $4, %1 \n\t" - "movq %%mm0, (%2) \n\t" - "sub $8, %2 \n\t" - "subl $2, %%ecx \n\t" - "jnz .loop2_pass4 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0" // clobber list -#endif - ); - } - - sptr -= (width_mmx*BPP2 - BPP2); // sign fixed - dp -= (width_mmx*2*BPP2 - BPP2); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= BPP2; - png_memcpy(v, sptr, BPP2); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= BPP2; - png_memcpy(dp, v, BPP2); - } - } - } - else if (((pass == 2) || (pass == 3)) && width) - { - int width_mmx = ((width >> 1) << 1) ; - width -= width_mmx; // 0,1 pixels => 0,2 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $2, %1 \n\t" - "sub $14, %2 \n\t" - - ".loop2_pass2: \n\t" - "movd (%1), %%mm0 \n\t" // x x x x 3 2 1 0 - "punpcklwd %%mm0, %%mm0 \n\t" // 3 2 3 2 1 0 1 0 - "movq %%mm0, %%mm1 \n\t" // 3 2 3 2 1 0 1 0 - "punpckldq %%mm0, %%mm0 \n\t" // 1 0 1 0 1 0 1 0 - "punpckhdq %%mm1, %%mm1 \n\t" // 3 2 3 2 3 2 3 2 - "movq %%mm0, (%2) \n\t" - "sub $4, %1 \n\t" - "movq %%mm1, 8(%2) \n\t" - "sub $16, %2 \n\t" - "subl $2, %%ecx \n\t" - "jnz .loop2_pass2 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - - sptr -= (width_mmx*2 - 2); // sign fixed - dp -= (width_mmx*8 - 2); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 2; - png_memcpy(v, sptr, 2); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 2; - png_memcpy(dp, v, 2); - } - } - } - else if (width) // && ((pass == 0) || (pass == 1)) - { - int width_mmx = ((width >> 1) << 1); - width -= width_mmx; // 0,1 pixels => 0,2 bytes - if (width_mmx) - { - __asm__ __volatile__ ( - "sub $2, %1 \n\t" - "sub $30, %2 \n\t" - - ".loop2_pass0: \n\t" - "movd (%1), %%mm0 \n\t" // x x x x 3 2 1 0 - "punpcklwd %%mm0, %%mm0 \n\t" // 3 2 3 2 1 0 1 0 - "movq %%mm0, %%mm1 \n\t" // 3 2 3 2 1 0 1 0 - "punpckldq %%mm0, %%mm0 \n\t" // 1 0 1 0 1 0 1 0 - "punpckhdq %%mm1, %%mm1 \n\t" // 3 2 3 2 3 2 3 2 - "movq %%mm0, (%2) \n\t" - "movq %%mm0, 8(%2) \n\t" - "movq %%mm1, 16(%2) \n\t" - "sub $4, %1 \n\t" - "movq %%mm1, 24(%2) \n\t" - "sub $32, %2 \n\t" - "subl $2, %%ecx \n\t" - "jnz .loop2_pass0 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width_mmx), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - - sptr -= (width_mmx*2 - 2); // sign fixed - dp -= (width_mmx*16 - 2); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 2; - png_memcpy(v, sptr, 2); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 2; - png_memcpy(dp, v, 2); - } - } - } - } /* end of pixel_bytes == 2 */ - - //-------------------------------------------------------------- - else if (pixel_bytes == BPP8) - { -// GRR TEST: should work, but needs testing (special 64-bit version of rpng2?) - // GRR NOTE: no need to combine passes here! - if (((pass == 4) || (pass == 5)) && width) - { - // source is 8-byte RRGGBBAA - // dest is 16-byte RRGGBBAA RRGGBBAA - __asm__ __volatile__ ( - "sub $8, %2 \n\t" // start of last block - - ".loop8_pass4: \n\t" - "movq (%1), %%mm0 \n\t" // 7 6 5 4 3 2 1 0 - "movq %%mm0, (%2) \n\t" - "sub $8, %1 \n\t" - "movq %%mm0, 8(%2) \n\t" - "sub $16, %2 \n\t" - "decl %%ecx \n\t" - "jnz .loop8_pass4 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0" // clobber list -#endif - ); - } - else if (((pass == 2) || (pass == 3)) && width) - { - // source is 8-byte RRGGBBAA - // dest is 32-byte RRGGBBAA RRGGBBAA RRGGBBAA RRGGBBAA - // (recall that expansion is _in place_: sptr and dp - // both point at locations within same row buffer) - __asm__ __volatile__ ( - "sub $24, %2 \n\t" // start of last block - - ".loop8_pass2: \n\t" - "movq (%1), %%mm0 \n\t" // 7 6 5 4 3 2 1 0 - "movq %%mm0, (%2) \n\t" - "movq %%mm0, 8(%2) \n\t" - "movq %%mm0, 16(%2) \n\t" - "sub $8, %1 \n\t" - "movq %%mm0, 24(%2) \n\t" - "sub $32, %2 \n\t" - "decl %%ecx \n\t" - "jnz .loop8_pass2 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0" // clobber list -#endif - ); - } - else if (width) // && ((pass == 0) || (pass == 1)) - { - // source is 8-byte RRGGBBAA - // dest is 64-byte RRGGBBAA RRGGBBAA RRGGBBAA RRGGBBAA ... - __asm__ __volatile__ ( - "sub $56, %2 \n\t" // start of last block - - ".loop8_pass0: \n\t" - "movq (%1), %%mm0 \n\t" // 7 6 5 4 3 2 1 0 - "movq %%mm0, (%2) \n\t" - "movq %%mm0, 8(%2) \n\t" - "movq %%mm0, 16(%2) \n\t" - "movq %%mm0, 24(%2) \n\t" - "movq %%mm0, 32(%2) \n\t" - "movq %%mm0, 40(%2) \n\t" - "movq %%mm0, 48(%2) \n\t" - "sub $8, %1 \n\t" - "movq %%mm0, 56(%2) \n\t" - "sub $64, %2 \n\t" - "decl %%ecx \n\t" - "jnz .loop8_pass0 \n\t" - "EMMS \n\t" // DONE - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D) - - : "0" (width), // ecx // input regs - "1" (sptr), // esi/rsi - "2" (dp) // edi/rdi - -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0" // clobber list -#endif - ); - } - } /* end of pixel_bytes == 8 */ - - //-------------------------------------------------------------- - else if (pixel_bytes == BPP6) // why no MMX for this case? - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, BPP6); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, BPP6); - dp -= BPP6; - } - sptr -= BPP6; - } - } /* end of pixel_bytes == 6 */ - - //-------------------------------------------------------------- - else - { - // ERROR: SHOULD NEVER BE REACHED -#if defined(PNG_DEBUG) - png_debug(1, "Internal libpng logic error (GCC " - "png_do_read_interlace() _mmx_supported)\n"); -#endif - } - - } // end of _mmx_supported ======================================== - - else /* MMX not supported: use modified C code - takes advantage - * of inlining of png_memcpy for a constant */ - { - if (pixel_bytes == BPP3) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, BPP3); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, BPP3); - dp -= BPP3; - } - sptr -= BPP3; - } - } - else if (pixel_bytes == BPP4) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, BPP4); - for (j = 0; j < png_pass_inc[pass]; j++) - { -#if defined(PNG_DEBUG) && defined(PNG_1_0_X) // row_buf_size gone in 1.2.x - if (dp < row || dp+3 > row+png_ptr->row_buf_size) - { - printf("dp out of bounds: row=%10p, dp=%10p, " - "rp=%10p\n", row, dp, row+png_ptr->row_buf_size); - printf("row_buf_size=%lu\n", png_ptr->row_buf_size); - } -#endif - png_memcpy(dp, v, BPP4); - dp -= BPP4; - } - sptr -= BPP4; - } - } - else if (pixel_bytes == 1) - { - for (i = width; i; i--) - { - int j; - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp-- = *sptr; - } - --sptr; - } - } - else if (pixel_bytes == BPP2) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, BPP2); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, BPP2); - dp -= BPP2; - } - sptr -= BPP2; - } - } - else if (pixel_bytes == BPP6) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, BPP6); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, BPP6); - dp -= BPP6; - } - sptr -= BPP6; - } - } - else if (pixel_bytes == BPP8) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, BPP8); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, BPP8); - dp -= BPP8; - } - sptr -= BPP8; - } - } - else - { - // ERROR: SHOULD NEVER BE REACHED -#if defined(PNG_DEBUG) - png_debug(1, "Internal libpng logic error (GCC " - "png_do_read_interlace() !_mmx_supported)\n"); -#endif - } - - } /* end if (MMX not supported) */ - break; - } /* end default (8-bit or larger) */ - } /* end switch (row_info->pixel_depth) */ - - row_info->width = final_width; - - row_info->rowbytes = PNG_ROWBYTES(row_info->pixel_depth,final_width); - } - -} /* end png_do_read_interlace() */ - -#endif /* PNG_HAVE_MMX_READ_INTERLACE */ -#endif /* PNG_READ_INTERLACING_SUPPORTED */ - - - -#if defined(PNG_HAVE_MMX_READ_FILTER_ROW) -#if defined(PNG_MMX_READ_FILTER_AVG_SUPPORTED) - -//===========================================================================// -// // -// P N G _ R E A D _ F I L T E R _ R O W _ M M X _ A V G // -// // -//===========================================================================// - -// Optimized code for PNG Average filter decoder - -static void /* PRIVATE */ -png_read_filter_row_mmx_avg(png_row_infop row_info, png_bytep row, - png_bytep prev_row) -{ - unsigned FullLength, MMXLength; // png_uint_32 is actually 64-bit on x86-64 - int bpp; - int dummy_value_a; - int dummy_value_c; // fix 'forbidden register 2 (cx) was spilled' error - int dummy_value_d; - png_bytep dummy_value_S; - png_bytep dummy_value_D; - int diff; // __attribute__((used)); - - bpp = (row_info->pixel_depth + 7) >> 3; // calc number of bytes per pixel - FullLength = row_info->rowbytes; // number of bytes to filter - - __asm__ __volatile__ ( - "avg_top: \n\t" - SAVE_GOT_ebx - SAVE_r15 - SAVE_ebp - // initialize address pointers and offset -//pre "movl row, %5 \n\t" // edi/rdi: ptr to Avg(x) - "xorl %%ebx, %%ebx \n\t" // ebx: x -//pre "movl prev_row, %4 \n\t" // esi/rsi: ptr to Prior(x) - "mov %5, " PDX " \n\t" // copy of row ptr... -//pre "subl bpp, " PDX " \n\t" // (bpp is preloaded into ecx) - "sub " PCX "," PDX " \n\t" // edx/rdx: ptr to Raw(x-bpp) -//pre "movl FullLength, %%eax \n\t" // bring in via eax... - SAVE_FullLength // ...but store for later use - "xorl %%eax, %%eax \n\t" - - // Compute the Raw value for the first bpp bytes - // Raw(x) = Avg(x) + (Prior(x)/2) - "avg_rlp: \n\t" - "movb (%4," PBX ",), %%al \n\t" // load al with Prior(x) - "incl %%ebx \n\t" - "shrb %%al \n\t" // divide by 2 - "addb -1(%5," PBX ",), %%al \n\t" // add Avg(x); -1 to offset inc ebx -//pre "cmpl bpp, %%ebx \n\t" // (bpp is preloaded into ecx) - "cmpl %%ecx, %%ebx \n\t" - "movb %%al, -1(%5," PBX ",) \n\t" // write Raw(x); -1 to offset inc ebx - "jb avg_rlp \n\t" // mov does not affect flags - - // get # of bytes to alignment (32-bit mask _would_ be good enough - // [computing delta], but 32-bit ops are zero-extended on 64-bit, argh) - // (if swapped edx and ebp, could do 8-bit or 16-bit mask...FIXME?) - "mov %5, " PBP " \n\t" // take start of row - "add " PBX "," PBP " \n\t" // add bpp - "add $0xf, " PBP " \n\t" // add 7+8 to incr past alignment bdry -// "andl $0xfffffff8, %%ebp \n\t" // mask to alignment boundary (32-bit!) - CLEAR_BOTTOM_3_BITS PBP "\n\t" // mask to alignment boundary - "sub %5, " PBP " \n\t" // subtract row ptr again => ebp = - "jz avg_go \n\t" // target value of ebx at alignment - - "xorl %%ecx, %%ecx \n\t" - - // fix alignment - // Compute the Raw value for the bytes up to the alignment boundary - // Raw(x) = Avg(x) + ((Raw(x-bpp) + Prior(x))/2) - "avg_lp1: \n\t" - "xorl %%eax, %%eax \n\t" - "movb (%4," PBX ",), %%cl \n\t" // load cl with Prior(x) - "movb (" PDX "," PBX ",), %%al \n\t" // load al with Raw(x-bpp) - "addw %%cx, %%ax \n\t" - "incl %%ebx \n\t" - "shrw %%ax \n\t" // divide by 2 - "addb -1(%5," PBX ",), %%al \n\t" // add Avg(x); -1 to offset inc ebx - "cmpl %%ebp, %%ebx \n\t" // check if at alignment boundary - "movb %%al, -1(%5," PBX ",) \n\t" // write Raw(x); -1 to offset inc ebx - "jb avg_lp1 \n\t" // repeat until at alignment boundary - - "avg_go: \n\t" - RESTORE_FullLength "%%eax \n\t" // FullLength -> eax - "movl %%eax, %%ecx \n\t" // copy -> ecx - "subl %%ebx, %%eax \n\t" // subtract alignment fix - "andl $0x00000007, %%eax \n\t" // calc bytes over mult of 8 - "subl %%eax, %%ecx \n\t" // sub over-bytes from original length -//out "movl %%ecx, MMXLength \n\t" - "movl %%ebp, %%eax \n\t" // ebp = diff, but no reg constraint(?) - RESTORE_ebp // (could swap ebp and edx functions) - RESTORE_r15 - RESTORE_GOT_ebx - -// "There is no way for you to specify that an input operand is modified -// without also specifying it as an output operand." [makes sense] - -// "Unless an output operand has the `&' constraint modifier, GCC may -// allocate it in the same register as an unrelated input operand, on the -// assumption the inputs are consumed before the outputs are produced." -// [trying to _force_ this] - -// "`=' Means that this operand is write-only for this instruction: -// the previous value is discarded and replaced by output data." -// [operand == variable name, presumably] - - // output regs - // these are operands 0-1 (originally 0-3): - : "=c" (MMXLength), // %0 -> %0 - "=a" (diff) // %3 -> %1 -// "=S" (dummy_value_S), // %1 -> GONE -// "=D" (dummy_value_D), // %2 -> GONE - - // input regs - // these are operands 2-5 (originally 4-7); two of their constraints say - // they must go in same places as operands 0-1 (originally 0-3) above: - : "0" (bpp), // %4 -> %2 ecx - "1" (FullLength), // %7 -> %3 eax - "S" (prev_row), // %5 -> %4 esi/rsi - "D" (row) // %6 -> %5 edi/rdi - - : "%edx" // clobber list - _CLOBBER_r15 - _CLOBBER_ebp - _CLOBBER_GOT_ebx - ); - - // now do the math for the rest of the row - switch (bpp) - { - case 3: - { -// _ShiftBpp = 24; // == 3 * 8 -// _ShiftRem = 40; // == 64 - 24 - - __asm__ __volatile__ ( - // re-init address pointers and offset - LOAD_GOT_rbp - "movq " AMASK5_3_0 ", %%mm7 \n\t" // _amask5_3_0 -> mm7 -// preload "movl diff, %%ecx \n\t" // ecx: x = offset to - // alignment boundary - "movq " LB_CARRY_MASK ", %%mm5 \n\t" // [interleave for parallel.?] -// preload "movl row, %1 \n\t" // edi: Avg(x) - "movq " HB_CLEAR_MASK ", %%mm4 \n\t" // _HBClearMask -> mm4 -// preload "movl prev_row, %0 \n\t" // esi: Prior(x) - RESTORE_rbp - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm2 \n\t"// load previous aligned 8 bytes - // (correct pos. in loop below) - "avg_3lp: \n\t" - "movq (%1," PCX ",), %%mm0 \n\t" // load mm0 with Avg(x) - "movq %%mm5, %%mm3 \n\t" - "psrlq $40, %%mm2 \n\t" // correct position Raw(x-bpp) - // data - "movq (%0," PCX ",), %%mm1 \n\t" // load mm1 with Prior(x) - "movq %%mm7, %%mm6 \n\t" - "pand %%mm1, %%mm3 \n\t" // get lsb for each prevrow byte - "psrlq $1, %%mm1 \n\t" // divide prev_row bytes by 2 - "pand %%mm4, %%mm1 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm0 \n\t" // add (Prev_row/2) to Avg for - // each byte - // add 1st active group (Raw(x-bpp)/2) to average with LBCarry - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both lsb's were == 1 - // (valid only for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to Raw(x-bpp)/2 - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 1 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to - // Avg for each Active byte - // add 2nd active group (Raw(x-bpp)/2) to average with _LBCarry - "psllq $24, %%mm6 \n\t" // shift the mm6 mask to cover - // bytes 3-5 - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "psllq $24, %%mm2 \n\t" // shift data to pos. correctly - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both lsb's were == 1 - // (valid only for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to Raw(x-bpp)/2 - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 2 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to - // Avg for each Active byte - - // add 3rd active group (Raw(x-bpp)/2) to average with _LBCarry - "psllq $24, %%mm6 \n\t" // shift mm6 mask to cover last - // two bytes - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "psllq $24, %%mm2 \n\t" // shift data to pos. correctly - // Data need be shifted only once here to - // get the correct x-bpp offset. - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both - // lsb's were == 1 (only valid for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to Raw(x-bpp)/2 - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 2 - // bytes to add to Avg - "addl $8, %%ecx \n\t" - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to - // Avg for each Active byte - // now ready to write back to memory - "movq %%mm0, -8(%1," PCX ",) \n\t" - // move updated Raw(x) to use as Raw(x-bpp) for next loop - "cmpl %%eax, %%ecx \n\t" // MMXLength - "movq %%mm0, %%mm2 \n\t" // mov updated Raw(x) to mm2 - "jb avg_3lp \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 3 bpp - - case 4: // formerly shared with 6 bpp case via _ShiftBpp and _ShiftRem, - { // but loop uses all 8 MMX regs, and psrlq/psllq require 64-bit - // mem (PIC/.so problems), MMX reg (none left), or immediate -// _ShiftBpp = bpp << 3; // 32 (psllq) -// _ShiftRem = 64 - _ShiftBpp; // 32 (psrlq) - - __asm__ __volatile__ ( - LOAD_GOT_rbp - "movq " HB_CLEAR_MASK ", %%mm4 \n\t" // _HBClearMask -> mm4 - "movq " LB_CARRY_MASK ", %%mm5 \n\t" // _LBCarryMask -> mm5 - // re-init address pointers and offset -// preload "movl diff, %%ecx \n\t" // ecx: x = offset to - // alignment boundary - "movq " AMASK0_8_0 ", %%mm7 \n\t" // _amask0_8_0 -> mm7 - RESTORE_rbp - - // ... and clear all bytes except for 1st active group -// preload "movl row, %1 \n\t" // edi: Avg(x) - "psrlq $32, %%mm7 \n\t" // was _ShiftRem -// preload "movl prev_row, %0 \n\t" // esi: Prior(x) - "movq %%mm7, %%mm6 \n\t" - "psllq $32, %%mm6 \n\t" // mask for 2nd active group - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm2 \n\t" // load previous aligned 8 bytes - // (we correct pos. in loop below) - "avg_4lp: \n\t" - "movq (%1," PCX ",), %%mm0 \n\t" - "psrlq $32, %%mm2 \n\t" // shift data to pos. correctly - "movq (%0," PCX ",), %%mm1 \n\t" - // add (Prev_row/2) to average - "movq %%mm5, %%mm3 \n\t" - "pand %%mm1, %%mm3 \n\t" // get lsb for each prev_row byte - "psrlq $1, %%mm1 \n\t" // divide prev_row bytes by 2 - "pand %%mm4, %%mm1 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm0 \n\t" // add (Prev_row/2) to Avg for - // each byte - // add 1st active group (Raw(x-bpp)/2) to average with _LBCarry - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both - // lsb's were == 1 (only valid for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to (Raw(x-bpp)/2) - // for each byte - "pand %%mm7, %%mm2 \n\t" // leave only Active Group 1 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to Avg - // for each Active - // byte - // add 2nd active group (Raw(x-bpp)/2) to average with _LBCarry - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "psllq $32, %%mm2 \n\t" // shift data to pos. correctly - "addl $8, %%ecx \n\t" - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both - // lsb's were == 1 (only valid for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to (Raw(x-bpp)/2) - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 2 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to - // Avg for each Active byte - "cmpl %%eax, %%ecx \n\t" // MMXLength - // now ready to write back to memory - "movq %%mm0, -8(%1," PCX ",) \n\t" - // prep Raw(x-bpp) for next loop - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "jb avg_4lp \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 4 bpp - - case 1: - { - __asm__ __volatile__ ( - // re-init address pointers and offset -// preload "movl diff, %%ecx \n\t" // ecx: x = offset to align. bdry -// preload "movl row, %1 \n\t" // edi/rdi: Avg(x) -// preload "movl FullLength, %%eax \n\t" - "cmpl %%eax, %%ecx \n\t" // test if offset at end of array - "jnb avg_1end \n\t" - - SAVE_ebp - - // do Avg decode for remaining bytes -// preload "movl prev_row, %0 \n\t" // esi/rsi: Prior(x) - "mov %1, " PBP " \n\t" // copy of row pointer... - "dec " PBP " \n\t" // ebp/rbp: Raw(x-bpp) - "xorl %%edx, %%edx \n\t" // zero edx before using dl & dx - // in loop below - SAVE_GOT_ebx - - "avg_1lp: \n\t" - // Raw(x) = Avg(x) + ((Raw(x-bpp) + Prior(x))/2) - "xorl %%ebx, %%ebx \n\t" - "movb (%0," PCX ",), %%dl \n\t" // load dl with Prior(x) - "movb (" PBP "," PCX ",), %%bl \n\t" // load bl with Raw(x-bpp) - "addw %%dx, %%bx \n\t" - "incl %%ecx \n\t" - "shrw %%bx \n\t" // divide by 2 - "addb -1(%1," PCX ",), %%bl \n\t" // add Avg(x); -1 to offset - // inc ecx - "cmpl %%eax, %%ecx \n\t" // check if at end of array - "movb %%bl, -1(%1," PCX ",) \n\t" // write back Raw(x); - // mov does not affect flags; -1 to offset inc ecx - "jb avg_1lp \n\t" - - RESTORE_GOT_ebx - RESTORE_ebp - - "avg_1end: \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (FullLength) // eax - - : "%edx" // clobber list - _CLOBBER_GOT_ebx - _CLOBBER_ebp - ); - } - return; // end 1 bpp - - case 2: - { -// _ShiftBpp = 16; // == 2 * 8 -// _ShiftRem = 48; // == 64 - _ShiftBpp - - __asm__ __volatile__ ( - LOAD_GOT_rbp - // load (former) _ActiveMask - "movq " AMASK6_2_0 ", %%mm7 \n\t" // _amask6_2_0 -> mm7 - // re-init address pointers and offset -// preload "movl diff, %%ecx \n\t" // ecx: x = offset to - // alignment boundary - "movq " LB_CARRY_MASK ", %%mm5 \n\t" // _LBCarryMask -> mm5 -// preload "movl row, %1 \n\t" // edi: Avg(x) - "movq " HB_CLEAR_MASK ", %%mm4 \n\t" // _HBClearMask -> mm4 -// preload "movl prev_row, %0 \n\t" // esi: Prior(x) - RESTORE_rbp - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm2 \n\t" // load previous aligned 8 bytes - // (we correct pos. in loop below) - "avg_2lp: \n\t" - "movq (%1," PCX ",), %%mm0 \n\t" - "psrlq $48, %%mm2 \n\t" // shift data to pos. correctly - "movq (%0," PCX ",), %%mm1 \n\t" // (GRR BUGFIX: was psllq) - // add (Prev_row/2) to average - "movq %%mm5, %%mm3 \n\t" - "pand %%mm1, %%mm3 \n\t" // get lsb for each prev_row byte - "psrlq $1, %%mm1 \n\t" // divide prev_row bytes by 2 - "pand %%mm4, %%mm1 \n\t" // clear invalid bit 7 of each - // byte - "movq %%mm7, %%mm6 \n\t" - "paddb %%mm1, %%mm0 \n\t" // add (Prev_row/2) to Avg for - // each byte - - // add 1st active group (Raw(x-bpp)/2) to average with _LBCarry - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both - // lsb's were == 1 (only valid - // for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to (Raw(x-bpp)/2) - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 1 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to Avg - // for each Active byte - - // add 2nd active group (Raw(x-bpp)/2) to average with _LBCarry - "psllq $16, %%mm6 \n\t" // shift the mm6 mask to cover - // bytes 2 & 3 - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "psllq $16, %%mm2 \n\t" // shift data to pos. correctly - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both - // lsb's were == 1 (only valid - // for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to (Raw(x-bpp)/2) - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 2 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to - // Avg for each Active byte - - // add 3rd active group (Raw(x-bpp)/2) to average with _LBCarry - "psllq $16, %%mm6 \n\t" // shift the mm6 mask to cover - // bytes 4 & 5 - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "psllq $16, %%mm2 \n\t" // shift data to pos. correctly - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both lsb's were == 1 - // (only valid for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to (Raw(x-bpp)/2) - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 2 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to - // Avg for each Active byte - - // add 4th active group (Raw(x-bpp)/2) to average with _LBCarry - "psllq $16, %%mm6 \n\t" // shift the mm6 mask to cover - // bytes 6 & 7 - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "psllq $16, %%mm2 \n\t" // shift data to pos. correctly - "addl $8, %%ecx \n\t" - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both - // lsb's were == 1 (only valid - // for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to (Raw(x-bpp)/2) - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 2 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to - // Avg for each Active byte - "cmpl %%eax, %%ecx \n\t" // MMXLength - // now ready to write back to memory - "movq %%mm0, -8(%1," PCX ",) \n\t" - // prep Raw(x-bpp) for next loop - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "jb avg_2lp \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 2 bpp - - case 6: // formerly shared with 4 bpp case (see comments there) - { -// _ShiftBpp = bpp << 3; // 48 (psllq) -// _ShiftRem = 64 - _ShiftBpp; // 16 (psrlq) - - __asm__ __volatile__ ( - LOAD_GOT_rbp - "movq " HB_CLEAR_MASK ", %%mm4 \n\t" // _HBClearMask -> mm4 - "movq " LB_CARRY_MASK ", %%mm5 \n\t" // _LBCarryMask -> mm5 - // re-init address pointers and offset -// preload "movl diff, %%ecx \n\t" // ecx: x = offset to - // alignment boundary - "movq " AMASK0_8_0 ", %%mm7 \n\t" // _amask0_8_0 -> mm7 - RESTORE_rbp - - // ... and clear all bytes except for 1st active group -// preload "movl row, %1 \n\t" // edi: Avg(x) - "psrlq $16, %%mm7 \n\t" -// preload "movl prev_row, %0 \n\t" // esi: Prior(x) - "movq %%mm7, %%mm6 \n\t" - "psllq $48, %%mm6 \n\t" // mask for 2nd active group - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm2 \n\t" // load previous aligned 8 bytes - // (we correct pos. in loop below) - "avg_6lp: \n\t" - "movq (%1," PCX ",), %%mm0 \n\t" - "psrlq $16, %%mm2 \n\t" // shift data to pos. correctly - "movq (%0," PCX ",), %%mm1 \n\t" - // add (Prev_row/2) to average - "movq %%mm5, %%mm3 \n\t" - "pand %%mm1, %%mm3 \n\t" // get lsb for each prev_row byte - "psrlq $1, %%mm1 \n\t" // divide prev_row bytes by 2 - "pand %%mm4, %%mm1 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm0 \n\t" // add (Prev_row/2) to Avg for - // each byte - // add 1st active group (Raw(x-bpp)/2) to average with _LBCarry - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both - // lsb's were == 1 (only valid for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to (Raw(x-bpp)/2) - // for each byte - "pand %%mm7, %%mm2 \n\t" // leave only Active Group 1 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to Avg - // for each Active - // byte - // add 2nd active group (Raw(x-bpp)/2) to average with _LBCarry - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "psllq $48, %%mm2 \n\t" // shift data to pos. correctly - "addl $8, %%ecx \n\t" - "movq %%mm3, %%mm1 \n\t" // now use mm1 for getting - // LBCarrys - "pand %%mm2, %%mm1 \n\t" // get LBCarrys for each byte - // where both - // lsb's were == 1 (only valid for active group) - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7 of each - // byte - "paddb %%mm1, %%mm2 \n\t" // add LBCarrys to (Raw(x-bpp)/2) - // for each byte - "pand %%mm6, %%mm2 \n\t" // leave only Active Group 2 - // bytes to add to Avg - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) + LBCarrys to - // Avg for each Active byte - "cmpl %%eax, %%ecx \n\t" // MMXLength - // now ready to write back to memory - "movq %%mm0, -8(%1," PCX ",) \n\t" - // prep Raw(x-bpp) for next loop - "movq %%mm0, %%mm2 \n\t" // mov updated Raws to mm2 - "jb avg_6lp \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 6 bpp - - case 8: - { - __asm__ __volatile__ ( - // re-init address pointers and offset -// preload "movl diff, %%ecx \n\t" // ecx: x = offset to - // alignment boundary - LOAD_GOT_rbp - "movq " LB_CARRY_MASK ", %%mm5 \n\t" // [interleave for parallel.?] -// preload "movl row, %1 \n\t" // edi: Avg(x) - "movq " HB_CLEAR_MASK ", %%mm4 \n\t" // _HBClearMask -> mm4 -// preload "movl prev_row, %0 \n\t" // esi: Prior(x) - RESTORE_rbp - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm2 \n\t" // load previous aligned 8 bytes - // (NO NEED to correct pos. in loop below) - - "avg_8lp: \n\t" - "movq (%1," PCX ",), %%mm0 \n\t" - "movq %%mm5, %%mm3 \n\t" - "movq (%0," PCX ",), %%mm1 \n\t" - "addl $8, %%ecx \n\t" - "pand %%mm1, %%mm3 \n\t" // get lsb for each prev_row byte - "psrlq $1, %%mm1 \n\t" // divide prev_row bytes by 2 - "pand %%mm2, %%mm3 \n\t" // get LBCarrys for each byte - // where both lsb's were == 1 - "psrlq $1, %%mm2 \n\t" // divide raw bytes by 2 - "pand %%mm4, %%mm1 \n\t" // clear invalid bit 7, each byte - "paddb %%mm3, %%mm0 \n\t" // add LBCarrys to Avg, each byte - "pand %%mm4, %%mm2 \n\t" // clear invalid bit 7, each byte - "paddb %%mm1, %%mm0 \n\t" // add (Prev_row/2) to Avg, each - "paddb %%mm2, %%mm0 \n\t" // add (Raw/2) to Avg for each - "cmpl %%eax, %%ecx \n\t" // MMXLength - "movq %%mm0, -8(%1," PCX ",) \n\t" - "movq %%mm0, %%mm2 \n\t" // reuse as Raw(x-bpp) - "jb avg_8lp \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2" // clobber list - , "%mm3", "%mm4", "%mm5" -#endif - ); - } - break; // end 8 bpp - - default: // bpp != 1,2,3,4,6,8: doesn't exist - { - // ERROR: SHOULD NEVER BE REACHED -#if defined(PNG_DEBUG) - png_debug(1, "Internal libpng logic error (GCC " - "png_read_filter_row_mmx_avg())\n"); -#endif - } - break; - - } // end switch (bpp) - - __asm__ __volatile__ ( - // MMX acceleration complete; now do clean-up - // check if any remaining bytes left to decode -//pre "movl FullLength, %%edx \n\t" -//pre "movl MMXLength, %%eax \n\t" // eax: x == offset bytes after MMX -//pre "movl row, %2 \n\t" // edi: Avg(x) - "cmpl %%edx, %%eax \n\t" // test if offset at end of array - "jnb avg_end \n\t" - - SAVE_ebp - - // do Avg decode for remaining bytes -//pre "movl prev_row, %1 \n\t" // esi: Prior(x) - "mov %2, " PBP " \n\t" // copy of row pointer... -//pre "subl bpp, " PBP " \n\t" // (bpp is preloaded into ecx) - "sub " PCX "," PBP " \n\t" // ebp: Raw(x-bpp) - "xorl %%ecx, %%ecx \n\t" // zero ecx before using cl & cx below - - SAVE_GOT_ebx - - "avg_lp2: \n\t" - // Raw(x) = Avg(x) + ((Raw(x-bpp) + Prior(x))/2) - "xorl %%ebx, %%ebx \n\t" - "movb (%1," PAX ",), %%cl \n\t" // load cl with Prior(x) - "movb (" PBP "," PAX ",), %%bl \n\t" // load bl with Raw(x-bpp) - "addw %%cx, %%bx \n\t" - "incl %%eax \n\t" - "shrw %%bx \n\t" // divide by 2 - "addb -1(%2," PAX ",), %%bl \n\t" // add Avg(x); -1 to offset inc eax - "cmpl %%edx, %%eax \n\t" // check if at end of array - "movb %%bl, -1(%2," PAX ",) \n\t" // write back Raw(x) [mov does not - "jb avg_lp2 \n\t" // affect flags; -1 to offset inc eax] - - RESTORE_GOT_ebx - RESTORE_ebp - - "avg_end: \n\t" - "EMMS \n\t" // end MMX; prep for poss. FP instrs. - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D), - "=a" (dummy_value_a), - "=d" (dummy_value_d) - - : "0" (bpp), // ecx // input regs - "1" (prev_row), // esi/rsi - "2" (row), // edi/rdi - "3" (MMXLength), // eax - "4" (FullLength) // edx - - CLOB_COLON_ebx_ebp // clobber list - CLOBBER_GOT_ebx - CLOB_COMMA_ebx_ebp - CLOBBER_ebp - ); - -} /* end png_read_filter_row_mmx_avg() */ - -#endif /* PNG_MMX_READ_FILTER_AVG_SUPPORTED */ - - - -#if defined(PNG_MMX_READ_FILTER_PAETH_SUPPORTED) -#if defined(PNG_x86_64_USE_GOTPCREL) || defined(PNG_THREAD_UNSAFE_OK) - -//===========================================================================// -// // -// P N G _ R E A D _ F I L T E R _ R O W _ M M X _ P A E T H // -// // -//===========================================================================// - -// Optimized code for PNG Paeth filter decoder - -static void /* PRIVATE */ -png_read_filter_row_mmx_paeth(png_row_infop row_info, png_bytep row, - png_bytep prev_row) -{ - unsigned FullLength, MMXLength; // png_uint_32 is actually 64-bit on x86-64 - int bpp; - int dummy_value_a; - int dummy_value_c; // fix 'forbidden register 2 (cx) was spilled' error - int dummy_value_d; - png_charp dummy_value_S; - png_charp dummy_value_D; - int diff; // __attribute__((used)); - - bpp = (row_info->pixel_depth + 7) >> 3; // calc number of bytes per pixel - FullLength = row_info->rowbytes; // number of bytes to filter - - __asm__ __volatile__ ( - SAVE_GOT_ebx - SAVE_r15 - SAVE_ebp -//pre "movl row, %2 \n\t" // edi/rdi - "xorl %%ebx, %%ebx \n\t" // ebx: x offset -//pre "movl prev_row, %1 \n\t" // esi/rsi - "xorl %%edx, %%edx \n\t" // edx: x-bpp offset -//pre "movl FullLength, %%eax \n\t" // bring in via eax... - SAVE_FullLength // ...but store for later use - "xorl %%eax, %%eax \n\t" - - // Compute the Raw value for the first bpp bytes - // Note: the formula works out to be always - // Paeth(x) = Raw(x) + Prior(x) where x < bpp - "paeth_rlp: \n\t" - "movb (%2," PBX ",), %%al \n\t" - "addb (%1," PBX ",), %%al \n\t" - "incl %%ebx \n\t" -//pre "cmpl bpp, %%ebx \n\t" (bpp is preloaded into ecx) - "cmpl %%ecx, %%ebx \n\t" - "movb %%al, -1(%2," PBX ",) \n\t" - "jb paeth_rlp \n\t" - - // get # of bytes to alignment (note: computing _delta_ of two pointers, - // so hereafter %%ebp is sufficient even on 64-bit) - "mov %2, " PBP " \n\t" // take start of row - "add " PBX "," PBP " \n\t" // add bpp - "add $0xf, " PBP " \n\t" // add 7+8 to incr past alignment bdry -// "andl $0xfffffff8, %%ebp \n\t" // mask to alignment boundary (32-bit!) - CLEAR_BOTTOM_3_BITS PBP "\n\t" // mask to alignment boundary - "sub %2, " PBP " \n\t" // subtract row ptr again => ebp = - "jz paeth_go \n\t" // target value of ebx at alignment - - "xorl %%ecx, %%ecx \n\t" - - SAVE_r11_r12_r13 - - // fix alignment - "paeth_lp1: \n\t" - "xorl %%eax, %%eax \n\t" - // pav = p - a = (a + b - c) - a = b - c - "movb (%1," PBX ",), %%al \n\t" // load Prior(x) into al - "movb (%1," PDX ",), %%cl \n\t" // load Prior(x-bpp) into cl - "subl %%ecx, %%eax \n\t" // subtract Prior(x-bpp) - "movl %%eax, " pa_TEMP " \n\t" // Save pav for later use - "xorl %%eax, %%eax \n\t" - // pbv = p - b = (a + b - c) - b = a - c - "movb (%2," PDX ",), %%al \n\t" // load Raw(x-bpp) into al - "subl %%ecx, %%eax \n\t" // subtract Prior(x-bpp) - "movl %%eax, %%ecx \n\t" - // pcv = p - c = (a + b - c) - c = (a - c) + (b - c) = pav + pbv - "addl " pa_TEMP ", %%eax \n\t" // pcv = pav + pbv - // pc = abs(pcv) - "testl $0x80000000, %%eax \n\t" - "jz paeth_pca \n\t" - "negl %%eax \n\t" // reverse sign of neg values - - "paeth_pca: \n\t" - "movl %%eax, " pc_TEMP " \n\t" // save pc for later use - // pb = abs(pbv) - "testl $0x80000000, %%ecx \n\t" - "jz paeth_pba \n\t" - "negl %%ecx \n\t" // reverse sign of neg values - - "paeth_pba: \n\t" - "movl %%ecx, " pb_TEMP " \n\t" // save pb for later use - // pa = abs(pav) - "movl " pa_TEMP ", %%eax \n\t" - "testl $0x80000000, %%eax \n\t" - "jz paeth_paa \n\t" - "negl %%eax \n\t" // reverse sign of neg values - - "paeth_paa: \n\t" - "movl %%eax, " pa_TEMP " \n\t" // save pa for later use - // test if pa <= pb - "cmpl %%ecx, %%eax \n\t" - "jna paeth_abb \n\t" - // pa > pb; now test if pb <= pc - "cmpl " pc_TEMP ", %%ecx \n\t" - "jna paeth_bbc \n\t" - // pb > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - "movb (%1," PDX ",), %%cl \n\t" // load Prior(x-bpp) into cl - "jmp paeth_paeth \n\t" - - "paeth_bbc: \n\t" - // pb <= pc; Raw(x) = Paeth(x) + Prior(x) - "movb (%1," PBX ",), %%cl \n\t" // load Prior(x) into cl - "jmp paeth_paeth \n\t" - - "paeth_abb: \n\t" - // pa <= pb; now test if pa <= pc - "cmpl " pc_TEMP ", %%eax \n\t" - "jna paeth_abc \n\t" - // pa > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - "movb (%1," PDX ",), %%cl \n\t" // load Prior(x-bpp) into cl - "jmp paeth_paeth \n\t" - - "paeth_abc: \n\t" - // pa <= pc; Raw(x) = Paeth(x) + Raw(x-bpp) - "movb (%2," PDX ",), %%cl \n\t" // load Raw(x-bpp) into cl - - "paeth_paeth: \n\t" - "incl %%ebx \n\t" - "incl %%edx \n\t" - // Raw(x) = (Paeth(x) + Paeth_Predictor( a, b, c )) mod 256 - "addb %%cl, -1(%2," PBX ",) \n\t" - "cmpl %%ebp, %%ebx \n\t" - "jb paeth_lp1 \n\t" - - RESTORE_r11_r12_r13 - - "paeth_go: \n\t" - RESTORE_FullLength "%%ecx \n\t" // FullLength -> ecx - "movl %%ecx, %%eax \n\t" - "subl %%ebx, %%eax \n\t" // subtract alignment fix - "andl $0x00000007, %%eax \n\t" // calc bytes over mult of 8 - "subl %%eax, %%ecx \n\t" // drop over bytes from original length -//out "movl %%ecx, MMXLength \n\t" - "movl %%ebp, %%eax \n\t" // ebp = diff, but no reg constraint(?) - RESTORE_ebp // (could swap ebp and edx functions) - RESTORE_r15 - RESTORE_GOT_ebx - - : "=c" (MMXLength), // output regs - "=S" (dummy_value_S), - "=D" (dummy_value_D), - "=a" (diff) - - : "0" (bpp), // ecx // input regs - "1" (prev_row), // esi/rsi - "2" (row), // edi/rdi - "3" (FullLength) // eax - - : "%edx" // clobber list - _CLOBBER_r11_r12_r13 - _CLOBBER_r15 - _CLOBBER_ebp - _CLOBBER_GOT_ebx - ); - - // now do the math for the rest of the row - switch (bpp) - { - case 3: - { -// _ShiftBpp = 24; // == bpp * 8 -// _ShiftRem = 40; // == 64 - _ShiftBpp - - __asm__ __volatile__ ( - LOAD_GOT_rbp -// preload "movl diff, %%ecx \n\t" -// preload "movl row, %1 \n\t" // edi/rdi -// preload "movl prev_row, %0 \n\t" // esi/rsi - "pxor %%mm0, %%mm0 \n\t" - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm1 \n\t" - "paeth_3lp: \n\t" - "psrlq $40, %%mm1 \n\t" // shift last 3 bytes to 1st - // 3 bytes - "movq (%0," PCX ",), %%mm2 \n\t" // load b=Prior(x) - "punpcklbw %%mm0, %%mm1 \n\t" // unpack High bytes of a - "movq -8(%0," PCX ",), %%mm3 \n\t" // prep c=Prior(x-bpp) bytes - "punpcklbw %%mm0, %%mm2 \n\t" // unpack High bytes of b - "psrlq $40, %%mm3 \n\t" // shift last 3 bytes to 1st - // 3 bytes - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - "punpcklbw %%mm0, %%mm3 \n\t" // unpack High bytes of c - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - "psubw %%mm3, %%mm4 \n\t" - "pxor %%mm7, %%mm7 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "movq %%mm4, %%mm6 \n\t" - "psubw %%mm3, %%mm5 \n\t" - - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm4, %%mm0 \n\t" // create mask pav bytes < 0 - "paddw %%mm5, %%mm6 \n\t" - "pand %%mm4, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "pcmpgtw %%mm5, %%mm7 \n\t" // create mask pbv bytes < 0 - "psubw %%mm0, %%mm4 \n\t" - "pand %%mm5, %%mm7 \n\t" // only pbv bytes < 0 in mm0 - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm7, %%mm5 \n\t" - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "pxor %%mm1, %%mm1 \n\t" - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "packuswb %%mm1, %%mm7 \n\t" - "movq (%0," PCX ",), %%mm3 \n\t" // load c=Prior(x-bpp) - "pand " AMASK5_3_0 ", %%mm7 \n\t" // _amask5_3_0 (was _ActiveMask) - "movq %%mm3, %%mm2 \n\t" // load b=Prior(x) step 1 - "paddb (%1," PCX ",), %%mm7 \n\t" // add Paeth predictor + Raw(x) - "punpcklbw %%mm0, %%mm3 \n\t" // unpack High bytes of c - "movq %%mm7, (%1," PCX ",) \n\t" // write back updated value - "movq %%mm7, %%mm1 \n\t" // now mm1 will be used as - // Raw(x-bpp) - // now do Paeth for 2nd set of bytes (3-5) - "psrlq $24, %%mm2 \n\t" // load b=Prior(x) step 2 - "punpcklbw %%mm0, %%mm1 \n\t" // unpack High bytes of a - "pxor %%mm7, %%mm7 \n\t" - "punpcklbw %%mm0, %%mm2 \n\t" // unpack High bytes of b - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - "psubw %%mm3, %%mm5 \n\t" - "psubw %%mm3, %%mm4 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = - // pav + pbv = pbv + pav - "movq %%mm5, %%mm6 \n\t" - "paddw %%mm4, %%mm6 \n\t" - - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm5, %%mm0 \n\t" // create mask pbv bytes < 0 - "pcmpgtw %%mm4, %%mm7 \n\t" // create mask pav bytes < 0 - "pand %%mm5, %%mm0 \n\t" // only pbv bytes < 0 in mm0 - "pand %%mm4, %%mm7 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm0, %%mm5 \n\t" - "psubw %%mm7, %%mm4 \n\t" - "psubw %%mm0, %%mm5 \n\t" - "psubw %%mm7, %%mm4 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "movq (%0," PCX ",), %%mm2 \n\t" // load b=Prior(x) - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "pxor %%mm1, %%mm1 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "packuswb %%mm1, %%mm7 \n\t" - "movq %%mm2, %%mm3 \n\t" // load c=Prior(x-bpp) step 1 - "pand " AMASK5_3_0 ", %%mm7 \n\t" // _amask5_3_0 (was _ActiveMask) - "punpckhbw %%mm0, %%mm2 \n\t" // unpack High bytes of b - "psllq $24, %%mm7 \n\t" // shift bytes to 2nd group of - // 3 bytes - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - "paddb (%1," PCX ",), %%mm7 \n\t" // add Paeth predictor + Raw(x) - "psllq $24, %%mm3 \n\t" // load c=Prior(x-bpp) step 2 - "movq %%mm7, (%1," PCX ",) \n\t" // write back updated value - "movq %%mm7, %%mm1 \n\t" - "punpckhbw %%mm0, %%mm3 \n\t" // unpack High bytes of c - "psllq $24, %%mm1 \n\t" // shift bytes (was _ShiftBpp) - // now mm1 will be used as Raw(x-bpp) - // now do Paeth for 3rd, and final, set of bytes (6-7) - "pxor %%mm7, %%mm7 \n\t" - "punpckhbw %%mm0, %%mm1 \n\t" // unpack High bytes of a - "psubw %%mm3, %%mm4 \n\t" - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "movq %%mm4, %%mm6 \n\t" - "psubw %%mm3, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "paddw %%mm5, %%mm6 \n\t" - - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm4, %%mm0 \n\t" // create mask pav bytes < 0 - "pcmpgtw %%mm5, %%mm7 \n\t" // create mask pbv bytes < 0 - "pand %%mm4, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "pand %%mm5, %%mm7 \n\t" // only pbv bytes < 0 in mm0 - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm1, %%mm1 \n\t" - "packuswb %%mm7, %%mm1 \n\t" - // step ecx to next set of 8 bytes and repeat loop til done - "addl $8, %%ecx \n\t" - "pand " AMASK0_2_6 ", %%mm1 \n\t" // _amask0_2_6 (_ActiveMaskEnd) - "paddb -8(%1," PCX ",), %%mm1 \n\t" // add Paeth predictor + Raw(x) - "cmpl %%eax, %%ecx \n\t" // MMXLength - "pxor %%mm0, %%mm0 \n\t" // pxor does not affect flags - "movq %%mm1, -8(%1," PCX ",) \n\t" // write back updated value - // mm1 will be used as Raw(x-bpp) next loop - // mm3 ready to be used as Prior(x-bpp) next loop - "jb paeth_3lp \n\t" - RESTORE_rbp - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 3 bpp - - case 4: - { - __asm__ __volatile__ ( -// preload "movl diff, %%ecx \n\t" -// preload "movl row, %1 \n\t" // edi/rdi -// preload "movl prev_row, %0 \n\t" // esi/rsi - "pxor %%mm0, %%mm0 \n\t" - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm1 \n\t" // only time should need to read - // a=Raw(x-bpp) bytes - "paeth_4lp: \n\t" - // do first set of 4 bytes - "movq -8(%0," PCX ",), %%mm3 \n\t" // read c=Prior(x-bpp) bytes - "punpckhbw %%mm0, %%mm1 \n\t" // unpack Low bytes of a - "movq (%0," PCX ",), %%mm2 \n\t" // load b=Prior(x) - "punpcklbw %%mm0, %%mm2 \n\t" // unpack High bytes of b - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - "punpckhbw %%mm0, %%mm3 \n\t" // unpack High bytes of c - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - "psubw %%mm3, %%mm4 \n\t" - "pxor %%mm7, %%mm7 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "movq %%mm4, %%mm6 \n\t" - "psubw %%mm3, %%mm5 \n\t" - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm4, %%mm0 \n\t" // create mask pav bytes < 0 - "paddw %%mm5, %%mm6 \n\t" - "pand %%mm4, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "pcmpgtw %%mm5, %%mm7 \n\t" // create mask pbv bytes < 0 - "psubw %%mm0, %%mm4 \n\t" - "pand %%mm5, %%mm7 \n\t" // only pbv bytes < 0 in mm0 - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm7, %%mm5 \n\t" - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "pxor %%mm1, %%mm1 \n\t" - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "packuswb %%mm1, %%mm7 \n\t" - "movq (%0," PCX ",), %%mm3 \n\t" // load c=Prior(x-bpp) - LOAD_GOT_rbp - "pand " AMASK4_4_0 ", %%mm7 \n\t" // _amask4_4_0 (was _ActiveMask) - RESTORE_rbp - "movq %%mm3, %%mm2 \n\t" // load b=Prior(x) step 1 - "paddb (%1," PCX ",), %%mm7 \n\t" // add Paeth predictor + Raw(x) - "punpcklbw %%mm0, %%mm3 \n\t" // unpack High bytes of c - "movq %%mm7, (%1," PCX ",) \n\t" // write back updated value - "movq %%mm7, %%mm1 \n\t" // now mm1 will be used as - // Raw(x-bpp) - // do second set of 4 bytes - "punpckhbw %%mm0, %%mm2 \n\t" // unpack Low bytes of b - "punpcklbw %%mm0, %%mm1 \n\t" // unpack Low bytes of a - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - "psubw %%mm3, %%mm4 \n\t" - "pxor %%mm7, %%mm7 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "movq %%mm4, %%mm6 \n\t" - "psubw %%mm3, %%mm5 \n\t" - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm4, %%mm0 \n\t" // create mask pav bytes < 0 - "paddw %%mm5, %%mm6 \n\t" - "pand %%mm4, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "pcmpgtw %%mm5, %%mm7 \n\t" // create mask pbv bytes < 0 - "psubw %%mm0, %%mm4 \n\t" - "pand %%mm5, %%mm7 \n\t" // only pbv bytes < 0 in mm0 - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm7, %%mm5 \n\t" - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "pxor %%mm1, %%mm1 \n\t" - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "pxor %%mm1, %%mm1 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm0, %%mm0 \n\t" - // step ecx to next set of 8 bytes and repeat loop til done - "addl $8, %%ecx \n\t" - "packuswb %%mm7, %%mm1 \n\t" - "paddb -8(%1," PCX ",), %%mm1 \n\t" // add predictor with Raw(x) - "cmpl %%eax, %%ecx \n\t" // MMXLength - "movq %%mm1, -8(%1," PCX ",) \n\t" // write back updated value - // mm1 will be used as Raw(x-bpp) next loop - "jb paeth_4lp \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 4 bpp - - case 1: - case 2: - { - __asm__ __volatile__ ( -// preload "movl diff, %%eax \n\t" // eax: x = offset to align. bdry -// preload "movl FullLength, %%edx \n\t" - "cmpl %%edx, %%eax \n\t" - "jnb paeth_dend \n\t" - - SAVE_ebp - -// preload "movl row, %2 \n\t" // edi/rdi - // do Paeth decode for remaining bytes -// preload "movl prev_row, %1 \n\t" // esi/rsi - "movl %%eax, %%ebp \n\t" -// preload "subl bpp, %%ebp \n\t" // (bpp is preloaded into ecx) - "subl %%ecx, %%ebp \n\t" // ebp = eax - bpp - "xorl %%ecx, %%ecx \n\t" // zero ecx before using cl & cx - - SAVE_GOT_ebx - SAVE_r11_r12_r13 - - "paeth_dlp: \n\t" - "xorl %%ebx, %%ebx \n\t" - // pav = p - a = (a + b - c) - a = b - c - "movb (%1," PAX ",), %%bl \n\t" // load Prior(x) into bl - "movb (%1," PBP ",), %%cl \n\t" // load Prior(x-bpp) into cl - "subl %%ecx, %%ebx \n\t" // subtract Prior(x-bpp) - "movl %%ebx, " pa_TEMP " \n\t" // Save pav for later use - "xorl %%ebx, %%ebx \n\t" - // pbv = p - b = (a + b - c) - b = a - c - "movb (%2," PBP ",), %%bl \n\t" // load Raw(x-bpp) into bl - "subl %%ecx, %%ebx \n\t" // subtract Prior(x-bpp) - "movl %%ebx, %%ecx \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "addl " pa_TEMP ", %%ebx \n\t" // pcv = pav + pbv - // pc = abs(pcv) - "testl $0x80000000, %%ebx \n\t" - "jz paeth_dpca \n\t" - "negl %%ebx \n\t" // reverse sign of neg values - - "paeth_dpca: \n\t" - "movl %%ebx, " pc_TEMP " \n\t" // save pc for later use - // pb = abs(pbv) - "testl $0x80000000, %%ecx \n\t" - "jz paeth_dpba \n\t" - "negl %%ecx \n\t" // reverse sign of neg values - - "paeth_dpba: \n\t" - "movl %%ecx, " pb_TEMP " \n\t" // save pb for later use - // pa = abs(pav) - "movl " pa_TEMP ", %%ebx \n\t" - "testl $0x80000000, %%ebx \n\t" - "jz paeth_dpaa \n\t" - "negl %%ebx \n\t" // reverse sign of neg values - - "paeth_dpaa: \n\t" - "movl %%ebx, " pa_TEMP " \n\t" // save pa for later use - // test if pa <= pb - "cmpl %%ecx, %%ebx \n\t" - "jna paeth_dabb \n\t" - // pa > pb; now test if pb <= pc - "cmpl " pc_TEMP ", %%ecx \n\t" - "jna paeth_dbbc \n\t" - // pb > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - "movb (%1," PBP ",), %%cl \n\t" // load Prior(x-bpp) into cl - "jmp paeth_dpaeth \n\t" - - "paeth_dbbc: \n\t" - // pb <= pc; Raw(x) = Paeth(x) + Prior(x) - "movb (%1," PAX ",), %%cl \n\t" // load Prior(x) into cl - "jmp paeth_dpaeth \n\t" - - "paeth_dabb: \n\t" - // pa <= pb; now test if pa <= pc - "cmpl " pc_TEMP ", %%ebx \n\t" - "jna paeth_dabc \n\t" - // pa > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - "movb (%1," PBP ",), %%cl \n\t" // load Prior(x-bpp) into cl - "jmp paeth_dpaeth \n\t" - - "paeth_dabc: \n\t" - // pa <= pc; Raw(x) = Paeth(x) + Raw(x-bpp) - "movb (%2," PBP ",), %%cl \n\t" // load Raw(x-bpp) into cl - - "paeth_dpaeth: \n\t" - "incl %%eax \n\t" - "incl %%ebp \n\t" - // Raw(x) = (Paeth(x) + Paeth_Predictor( a, b, c )) mod 256 - "addb %%cl, -1(%2," PAX ",) \n\t" - "cmpl %%edx, %%eax \n\t" // check against FullLength - "jb paeth_dlp \n\t" - - RESTORE_r11_r12_r13 - RESTORE_GOT_ebx - RESTORE_ebp - - "paeth_dend: \n\t" - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D), - "=a" (dummy_value_a), - "=d" (dummy_value_d) - - : "0" (bpp), // ecx // input regs - "1" (prev_row), // esi/rsi - "2" (row), // edi/rdi - "3" (diff), // eax - "4" (FullLength) // edx - - CLOB_COLON_ebx_ebp_r1X // clobber list - CLOBBER_GOT_ebx - CLOB_COMMA_ebx_ebp - CLOBBER_ebp - CLOB_COMMA_ebX_r1X - CLOBBER_r11_r12_r13 - ); - } - return; // end 1 or 2 bpp (no need to go further with this one) - - case 6: - { -// _ActiveMask2 = 0xffffffff00000000LL; // NOT USED ("_amask_0_4_4") -// _ShiftBpp = 48; // bpp << 3 == bpp * 8 -// _ShiftRem = 16; // 64 - _ShiftBpp - - __asm__ __volatile__ ( -// preload "movl diff, %%ecx \n\t" -// preload "movl row, %1 \n\t" // edi/rdi -// preload "movl prev_row, %0 \n\t" // esi/rsi - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm1 \n\t" - "pxor %%mm0, %%mm0 \n\t" - - "paeth_6lp: \n\t" - // must shift to position Raw(x-bpp) data - "psrlq $16, %%mm1 \n\t" // was _ShiftRem - // do first set of 4 bytes - "movq -8(%0," PCX ",), %%mm3 \n\t" // read c=Prior(x-bpp) bytes - "punpcklbw %%mm0, %%mm1 \n\t" // unpack Low bytes of a - "movq (%0," PCX ",), %%mm2 \n\t" // load b=Prior(x) - "punpcklbw %%mm0, %%mm2 \n\t" // unpack Low bytes of b - // must shift to position Prior(x-bpp) data - "psrlq $16, %%mm3 \n\t" // was _ShiftRem - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - "punpcklbw %%mm0, %%mm3 \n\t" // unpack Low bytes of c - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - "psubw %%mm3, %%mm4 \n\t" - "pxor %%mm7, %%mm7 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "movq %%mm4, %%mm6 \n\t" - "psubw %%mm3, %%mm5 \n\t" - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm4, %%mm0 \n\t" // create mask pav bytes < 0 - "paddw %%mm5, %%mm6 \n\t" - "pand %%mm4, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "pcmpgtw %%mm5, %%mm7 \n\t" // create mask pbv bytes < 0 - "psubw %%mm0, %%mm4 \n\t" - "pand %%mm5, %%mm7 \n\t" // only pbv bytes < 0 in mm0 - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm7, %%mm5 \n\t" - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "pxor %%mm1, %%mm1 \n\t" - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "packuswb %%mm1, %%mm7 \n\t" - "movq -8(%0," PCX ",), %%mm3 \n\t" // load c=Prior(x-bpp) - LOAD_GOT_rbp - "pand " AMASK4_4_0 ", %%mm7 \n\t" // _amask4_4_0 (was _ActiveMask) - RESTORE_rbp - "psrlq $16, %%mm3 \n\t" - "movq (%0," PCX ",), %%mm2 \n\t" // load b=Prior(x) step 1 - "paddb (%1," PCX ",), %%mm7 \n\t" // add Paeth predictor + Raw(x) - "movq %%mm2, %%mm6 \n\t" - "movq %%mm7, (%1," PCX ",) \n\t" // write back updated value - "movq -8(%1," PCX ",), %%mm1 \n\t" - "psllq $48, %%mm6 \n\t" // bpp * 8 = bits per pixel - "movq %%mm7, %%mm5 \n\t" - "psrlq $16, %%mm1 \n\t" // 64 - (bpp * 8) = remainder - "por %%mm6, %%mm3 \n\t" - "psllq $48, %%mm5 \n\t" // was _ShiftBpp - "punpckhbw %%mm0, %%mm3 \n\t" // unpack High bytes of c - "por %%mm5, %%mm1 \n\t" - // do second set of 4 bytes - "punpckhbw %%mm0, %%mm2 \n\t" // unpack High bytes of b - "punpckhbw %%mm0, %%mm1 \n\t" // unpack High bytes of a - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - "psubw %%mm3, %%mm4 \n\t" - "pxor %%mm7, %%mm7 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "movq %%mm4, %%mm6 \n\t" - "psubw %%mm3, %%mm5 \n\t" - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm4, %%mm0 \n\t" // create mask pav bytes < 0 - "paddw %%mm5, %%mm6 \n\t" - "pand %%mm4, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "pcmpgtw %%mm5, %%mm7 \n\t" // create mask pbv bytes < 0 - "psubw %%mm0, %%mm4 \n\t" - "pand %%mm5, %%mm7 \n\t" // only pbv bytes < 0 in mm0 - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm7, %%mm5 \n\t" - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "pxor %%mm1, %%mm1 \n\t" - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "pxor %%mm1, %%mm1 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm0, %%mm0 \n\t" - // step ecx to next set of 8 bytes and repeat loop til done - "addl $8, %%ecx \n\t" - "packuswb %%mm7, %%mm1 \n\t" - "paddb -8(%1," PCX ",), %%mm1 \n\t" // add Paeth predictor + Raw(x) - "cmpl %%eax, %%ecx \n\t" // MMXLength - "movq %%mm1, -8(%1," PCX ",) \n\t" // write back updated value - // mm1 will be used as Raw(x-bpp) next loop - "jb paeth_6lp \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 6 bpp - - case 8: // bpp == 8 - { - __asm__ __volatile__ ( -// preload "movl diff, %%ecx \n\t" -// preload "movl row, %1 \n\t" // edi/rdi -// preload "movl prev_row, %0 \n\t" // esi/rsi - "pxor %%mm0, %%mm0 \n\t" - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PCX ",), %%mm1 \n\t" // only time should need to read - // a=Raw(x-bpp) bytes - "paeth_8lp: \n\t" - // do first set of 4 bytes - "movq -8(%0," PCX ",), %%mm3 \n\t" // read c=Prior(x-bpp) bytes - "punpcklbw %%mm0, %%mm1 \n\t" // unpack Low bytes of a - "movq (%0," PCX ",), %%mm2 \n\t" // load b=Prior(x) - "punpcklbw %%mm0, %%mm2 \n\t" // unpack Low bytes of b - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - "punpcklbw %%mm0, %%mm3 \n\t" // unpack Low bytes of c - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - "psubw %%mm3, %%mm4 \n\t" - "pxor %%mm7, %%mm7 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "movq %%mm4, %%mm6 \n\t" - "psubw %%mm3, %%mm5 \n\t" - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm4, %%mm0 \n\t" // create mask pav bytes < 0 - "paddw %%mm5, %%mm6 \n\t" - "pand %%mm4, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "pcmpgtw %%mm5, %%mm7 \n\t" // create mask pbv bytes < 0 - "psubw %%mm0, %%mm4 \n\t" - "pand %%mm5, %%mm7 \n\t" // only pbv bytes < 0 in mm0 - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm7, %%mm5 \n\t" - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "pxor %%mm1, %%mm1 \n\t" - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "packuswb %%mm1, %%mm7 \n\t" - "movq -8(%0," PCX ",), %%mm3 \n\t" // read c=Prior(x-bpp) bytes - LOAD_GOT_rbp - "pand " AMASK4_4_0 ", %%mm7 \n\t" // _amask4_4_0 (was _ActiveMask) - RESTORE_rbp - "movq (%0," PCX ",), %%mm2 \n\t" // load b=Prior(x) - "paddb (%1," PCX ",), %%mm7 \n\t" // add Paeth predictor + Raw(x) - "punpckhbw %%mm0, %%mm3 \n\t" // unpack High bytes of c - "movq %%mm7, (%1," PCX ",) \n\t" // write back updated value - "movq -8(%1," PCX ",), %%mm1 \n\t" // read a=Raw(x-bpp) bytes - - // do second set of 4 bytes - "punpckhbw %%mm0, %%mm2 \n\t" // unpack High bytes of b - "punpckhbw %%mm0, %%mm1 \n\t" // unpack High bytes of a - // pav = p - a = (a + b - c) - a = b - c - "movq %%mm2, %%mm4 \n\t" - // pbv = p - b = (a + b - c) - b = a - c - "movq %%mm1, %%mm5 \n\t" - "psubw %%mm3, %%mm4 \n\t" - "pxor %%mm7, %%mm7 \n\t" - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - "movq %%mm4, %%mm6 \n\t" - "psubw %%mm3, %%mm5 \n\t" - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - "pcmpgtw %%mm4, %%mm0 \n\t" // create mask pav bytes < 0 - "paddw %%mm5, %%mm6 \n\t" - "pand %%mm4, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "pcmpgtw %%mm5, %%mm7 \n\t" // create mask pbv bytes < 0 - "psubw %%mm0, %%mm4 \n\t" - "pand %%mm5, %%mm7 \n\t" // only pbv bytes < 0 in mm0 - "psubw %%mm0, %%mm4 \n\t" - "psubw %%mm7, %%mm5 \n\t" - "pxor %%mm0, %%mm0 \n\t" - "pcmpgtw %%mm6, %%mm0 \n\t" // create mask pcv bytes < 0 - "pand %%mm6, %%mm0 \n\t" // only pav bytes < 0 in mm7 - "psubw %%mm7, %%mm5 \n\t" - "psubw %%mm0, %%mm6 \n\t" - // test pa <= pb - "movq %%mm4, %%mm7 \n\t" - "psubw %%mm0, %%mm6 \n\t" - "pcmpgtw %%mm5, %%mm7 \n\t" // pa > pb? - "movq %%mm7, %%mm0 \n\t" - // use mm7 mask to merge pa & pb - "pand %%mm7, %%mm5 \n\t" - // use mm0 mask copy to merge a & b - "pand %%mm0, %%mm2 \n\t" - "pandn %%mm4, %%mm7 \n\t" - "pandn %%mm1, %%mm0 \n\t" - "paddw %%mm5, %%mm7 \n\t" - "paddw %%mm2, %%mm0 \n\t" - // test ((pa <= pb)? pa:pb) <= pc - "pcmpgtw %%mm6, %%mm7 \n\t" // pab > pc? - "pxor %%mm1, %%mm1 \n\t" - "pand %%mm7, %%mm3 \n\t" - "pandn %%mm0, %%mm7 \n\t" - "pxor %%mm1, %%mm1 \n\t" - "paddw %%mm3, %%mm7 \n\t" - "pxor %%mm0, %%mm0 \n\t" - // step ecx to next set of 8 bytes and repeat loop til done - "addl $8, %%ecx \n\t" - "packuswb %%mm7, %%mm1 \n\t" - "paddb -8(%1," PCX ",), %%mm1 \n\t" // add Paeth predictor + Raw(x) - "cmpl %%eax, %%ecx \n\t" // MMXLength - "movq %%mm1, -8(%1," PCX ",) \n\t" // write back updated value - // mm1 will be used as Raw(x-bpp) next loop - "jb paeth_8lp \n\t" - - : "=S" (dummy_value_S), // output regs (dummy) - "=D" (dummy_value_D), - "=c" (dummy_value_c), - "=a" (dummy_value_a) - - : "0" (prev_row), // esi/rsi // input regs - "1" (row), // edi/rdi - "2" (diff), // ecx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm2", "%mm3" // clobber list - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 8 bpp - - default: // bpp != 1,2,3,4,6,8: doesn't exist - { - // ERROR: SHOULD NEVER BE REACHED -#if defined(PNG_DEBUG) - png_debug(1, "Internal libpng logic error (GCC " - "png_read_filter_row_mmx_paeth())\n"); -#endif - } - break; - - } // end switch (bpp) - - __asm__ __volatile__ ( - // MMX acceleration complete; now do clean-up - // check if any remaining bytes left to decode -//pre "movl FullLength, %%edx \n\t" -//pre "movl MMXLength, %%eax \n\t" - "cmpl %%edx, %%eax \n\t" - "jnb paeth_end \n\t" - - SAVE_ebp - -//pre "movl row, %2 \n\t" // edi/rdi -//pre "movl prev_row, %1 \n\t" // esi/rsi - // do Paeth decode for remaining bytes - "movl %%eax, %%ebp \n\t" -//pre "subl bpp, %%ebp \n\t" // (bpp is preloaded into ecx) - "subl %%ecx, %%ebp \n\t" // ebp = eax - bpp - "xorl %%ecx, %%ecx \n\t" // zero ecx before using cl & cx below - - SAVE_GOT_ebx - SAVE_r11_r12_r13 - - "paeth_lp2: \n\t" - "xorl %%ebx, %%ebx \n\t" - // pav = p - a = (a + b - c) - a = b - c - "movb (%1," PAX ",), %%bl \n\t" // load Prior(x) into bl - "movb (%1," PBP ",), %%cl \n\t" // load Prior(x-bpp) into cl - "subl %%ecx, %%ebx \n\t" // subtract Prior(x-bpp) - "movl %%ebx, " pa_TEMP " \n\t" // Save pav for later use - "xorl %%ebx, %%ebx \n\t" - // pbv = p - b = (a + b - c) - b = a - c - "movb (%2," PBP ",), %%bl \n\t" // load Raw(x-bpp) into bl - "subl %%ecx, %%ebx \n\t" // subtract Prior(x-bpp) - "movl %%ebx, %%ecx \n\t" - // pcv = p - c = (a + b - c) - c = (a - c) + (b - c) = pav + pbv - "addl " pa_TEMP ", %%ebx \n\t" // pcv = pav + pbv - // pc = abs(pcv) - "testl $0x80000000, %%ebx \n\t" - "jz paeth_pca2 \n\t" - "negl %%ebx \n\t" // reverse sign of neg values - - "paeth_pca2: \n\t" - "movl %%ebx, " pc_TEMP " \n\t" // save pc for later use - // pb = abs(pbv) - "testl $0x80000000, %%ecx \n\t" - "jz paeth_pba2 \n\t" - "negl %%ecx \n\t" // reverse sign of neg values - - "paeth_pba2: \n\t" - "movl %%ecx, " pb_TEMP " \n\t" // save pb for later use - // pa = abs(pav) - "movl " pa_TEMP ", %%ebx \n\t" - "testl $0x80000000, %%ebx \n\t" - "jz paeth_paa2 \n\t" - "negl %%ebx \n\t" // reverse sign of neg values - - "paeth_paa2: \n\t" - "movl %%ebx, " pa_TEMP " \n\t" // save pa for later use - // test if pa <= pb - "cmpl %%ecx, %%ebx \n\t" - "jna paeth_abb2 \n\t" - // pa > pb; now test if pb <= pc - "cmpl " pc_TEMP ", %%ecx \n\t" - "jna paeth_bbc2 \n\t" - // pb > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - "movb (%1," PBP ",), %%cl \n\t" // load Prior(x-bpp) into cl - "jmp paeth_paeth2 \n\t" - - "paeth_bbc2: \n\t" - // pb <= pc; Raw(x) = Paeth(x) + Prior(x) - "movb (%1," PAX ",), %%cl \n\t" // load Prior(x) into cl - "jmp paeth_paeth2 \n\t" - - "paeth_abb2: \n\t" - // pa <= pb; now test if pa <= pc - "cmpl " pc_TEMP ", %%ebx \n\t" - "jna paeth_abc2 \n\t" - // pa > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - "movb (%1," PBP ",), %%cl \n\t" // load Prior(x-bpp) into cl - "jmp paeth_paeth2 \n\t" - - "paeth_abc2: \n\t" - // pa <= pc; Raw(x) = Paeth(x) + Raw(x-bpp) - "movb (%2," PBP ",), %%cl \n\t" // load Raw(x-bpp) into cl - - "paeth_paeth2: \n\t" - "incl %%eax \n\t" - "incl %%ebp \n\t" - // Raw(x) = (Paeth(x) + Paeth_Predictor( a, b, c )) mod 256 - "addb %%cl, -1(%2," PAX ",) \n\t" - "cmpl %%edx, %%eax \n\t" // check against FullLength - "jb paeth_lp2 \n\t" - - RESTORE_r11_r12_r13 - RESTORE_GOT_ebx - RESTORE_ebp - - "paeth_end: \n\t" - "EMMS \n\t" // end MMX; prep for poss. FP instrs. - - : "=c" (dummy_value_c), // output regs (dummy) - "=S" (dummy_value_S), - "=D" (dummy_value_D), - "=a" (dummy_value_a), - "=d" (dummy_value_d) - - : "0" (bpp), // ecx // input regs - "1" (prev_row), // esi/rsi - "2" (row), // edi/rdi - "3" (MMXLength), // eax - "4" (FullLength) // edx - - CLOB_COLON_ebx_ebp_r1X // clobber list - CLOBBER_GOT_ebx - CLOB_COMMA_ebx_ebp - CLOBBER_ebp - CLOB_COMMA_ebX_r1X - CLOBBER_r11_r12_r13 - ); - -} /* end png_read_filter_row_mmx_paeth() */ - -#endif // PNG_x86_64_USE_GOTPCREL || PNG_THREAD_UNSAFE_OK -#endif /* PNG_MMX_READ_FILTER_PAETH_SUPPORTED */ - - - - -#if defined(PNG_MMX_READ_FILTER_SUB_SUPPORTED) - -//===========================================================================// -// // -// P N G _ R E A D _ F I L T E R _ R O W _ M M X _ S U B // -// // -//===========================================================================// - -// Optimized code for PNG Sub filter decoder - -static void /* PRIVATE */ -png_read_filter_row_mmx_sub(png_row_infop row_info, png_bytep row) -{ - unsigned FullLength, MMXLength; // png_uint_32 is actually 64-bit on x86-64 - int bpp; - int dummy_value_a; - int dummy_value_c; - int dummy_value_d; - png_bytep dummy_value_D; - int diff; // __attribute__((used)); - - bpp = (row_info->pixel_depth + 7) >> 3; // calc number of bytes per pixel - FullLength = row_info->rowbytes - bpp; // number of bytes to filter - // (why do we subtract off bpp? not so in avg or paeth...) - - __asm__ __volatile__ ( - SAVE_r15 - SAVE_ebp -//pre "movl row, %1 \n\t" // edi/rdi - "mov %1, " PSI " \n\t" // lp = row -//pre "movl bpp, %%ecx \n\t" - "add " PCX ", %1 \n\t" // rp = row + bpp -//pre "movl FullLength, %%eax \n\t" // bring in via eax... - SAVE_FullLength // ...but store for later use - - "xorl %%eax, %%eax \n\t" - - // get # of bytes to alignment (note: computing _delta_ of two pointers, - // so hereafter %%ebp is sufficient even on 64-bit) - "mov %1, " PBP " \n\t" // take start of row - "add $0xf, " PBP " \n\t" // add 7+8 to incr past alignment bdry -// "andl $0xfffffff8, %%ebp \n\t" // mask to alignment boundary (32-bit!) - CLEAR_BOTTOM_3_BITS PBP "\n\t" // mask to alignment boundary - "sub %1, " PBP " \n\t" // subtract row ptr again => ebp = - "jz sub_go \n\t" // target value of eax at alignment - - "sub_lp1: \n\t" // fix alignment - "movb (" PSI "," PAX ",), %%cl \n\t" - "addb %%cl, (%1," PAX ",) \n\t" - "incl %%eax \n\t" - "cmpl %%ebp, %%eax \n\t" - "jb sub_lp1 \n\t" - - "sub_go: \n\t" - RESTORE_FullLength "%%ecx \n\t" // FullLength -> ecx - "movl %%ecx, %%edx \n\t" - "subl %%eax, %%edx \n\t" // subtract alignment fix - "andl $0x00000007, %%edx \n\t" // calc bytes over mult of 8 - "subl %%edx, %%ecx \n\t" // drop over bytes from length -//out "movl %%ecx, MMXLength \n\t" - "movl %%ebp, %%eax \n\t" // ebp = diff, but no reg constraint(?) - RESTORE_ebp // (could swap ebp and ecx functions, - RESTORE_r15 // but %%cl issues...) - - : "=c" (MMXLength), // 0 // output regs - "=D" (dummy_value_D), // 1 - "=a" (diff) // 2 - - : "0" (bpp), // ecx // input regs - "1" (row), // edi - "2" (FullLength) // eax - - : "%esi", "%edx" // clobber list - _CLOBBER_r15 - _CLOBBER_ebp - ); - - // now do the math for the rest of the row - switch (bpp) - { - case 3: - { -// _ShiftBpp = 24; // == 3 * 8 -// _ShiftRem = 40; // == 64 - 24 - - __asm__ __volatile__ ( -// preload "mov row, %1 \n\t" // edi/rdi - LOAD_GOT_rbp - // load (former) _ActiveMask for 2nd active byte group - "movq " AMASK2_3_3 ", %%mm7 \n\t" // _amask2_3_3 - RESTORE_rbp - -// notused "mov %1, " PSI " \n\t" // lp = row -// preload "movl bpp, %%ecx \n\t" - "add " PCX ", %1 \n\t" // rp = row + bpp - "movq %%mm7, %%mm6 \n\t" -// preload "movl diff, %%edx \n\t" - "psllq $24, %%mm6 \n\t" // move mask in mm6 to cover - // 3rd active byte group - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PDX ",), %%mm1 \n\t" - - "sub_3lp: \n\t" // shift data for adding first - "psrlq $40, %%mm1 \n\t" // bpp bytes (no need for mask; - // shift clears inactive bytes) - // add 1st active group - "movq (%1," PDX ",), %%mm0 \n\t" - "paddb %%mm1, %%mm0 \n\t" - - // add 2nd active group - "movq %%mm0, %%mm1 \n\t" // mov updated Raws to mm1 - "psllq $24, %%mm1 \n\t" // shift data to pos. correctly - "pand %%mm7, %%mm1 \n\t" // mask to use 2nd active group - "paddb %%mm1, %%mm0 \n\t" - - // add 3rd active group - "movq %%mm0, %%mm1 \n\t" // mov updated Raws to mm1 - "psllq $24, %%mm1 \n\t" // shift data to pos. correctly - "pand %%mm6, %%mm1 \n\t" // mask to use 3rd active group - "addl $8, %%edx \n\t" - "paddb %%mm1, %%mm0 \n\t" - - "cmpl %%eax, %%edx \n\t" // MMXLength - "movq %%mm0, -8(%1," PDX ",) \n\t" // write updated Raws to array - "movq %%mm0, %%mm1 \n\t" // prep 1st add at top of loop - "jb sub_3lp \n\t" - - : "=c" (dummy_value_c), // 0 // output regs (dummy) - "=D" (dummy_value_D), // 1 - "=d" (dummy_value_d), // 2 - "=a" (dummy_value_a) // 3 - - : "0" (bpp), // ecx // input regs - "1" (row), // edi - "2" (diff), // edx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm6", "%mm7" // clobber list -#endif - ); - } - break; // end 3 bpp - - case 4: // formerly shared with 6 bpp case via _ShiftBpp and _ShiftRem, - { // but 64-bit PIC/.so problems (could still share, moving vars - // into unused MMX regs via ecx/edx, but kludgy) -// _ShiftBpp = bpp << 3; // 32 (psllq) -// _ShiftRem = 64 - _ShiftBpp; // 32 (psrlq) - - __asm__ __volatile__ ( -// preload "mov row, %1 \n\t" // edi/rdi -// preload "movl diff, %%edx \n\t" -// notused "mov %1, " PSI " \n\t" // lp = row -// preload "movl bpp, %%ecx \n\t" - "add " PCX ", %1 \n\t" // rp = row + bpp - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PDX ",), %%mm1 \n\t" - - "sub_4lp: \n\t" // shift data for adding first - "psrlq $32, %%mm1 \n\t" // bpp bytes (no need for mask; - // shift clears inactive bytes) - "movq (%1," PDX ",), %%mm0 \n\t" - "paddb %%mm1, %%mm0 \n\t" - - // add 2nd active group - "movq %%mm0, %%mm1 \n\t" // mov updated Raws to mm1 - "psllq $32, %%mm1 \n\t" // shift data to pos. correctly - "addl $8, %%edx \n\t" - "paddb %%mm1, %%mm0 \n\t" - - "cmpl %%eax, %%edx \n\t" // MMXLength - "movq %%mm0, -8(%1," PDX ",) \n\t" // write updated Raws to array - "movq %%mm0, %%mm1 \n\t" // prep 1st add at top of loop - "jb sub_4lp \n\t" - - : "=c" (dummy_value_c), // 0 // output regs (dummy) - "=D" (dummy_value_D), // 1 - "=d" (dummy_value_d), // 2 - "=a" (dummy_value_a) // 3 - - : "0" (bpp), // ecx // input regs - "1" (row), // edi - "2" (diff), // edx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - break; // end 4 bpp - - case 1: - { - __asm__ __volatile__ ( -// preload "movl diff, %%edx \n\t" -// preload "mov row, %1 \n\t" // edi/rdi -// preload "cmpl FullLength, %%edx \n\t" - "cmpl %%eax, %%edx \n\t" - "jnb sub_1end \n\t" - "mov %1, " PSI " \n\t" // lp = row -// irrel. "xorl %%ecx, %%ecx \n\t" // (actually bug with preload) -// preload "movl bpp, %%ecx \n\t" - "add " PCX ", %1 \n\t" // rp = row + bpp - - "sub_1lp: \n\t" - "movb (" PSI "," PDX ",), %%cl \n\t" - "addb %%cl, (%1," PDX ",) \n\t" - "incl %%edx \n\t" - "cmpl %%eax, %%edx \n\t" // compare with FullLength - "jb sub_1lp \n\t" - - "sub_1end: \n\t" - - : "=c" (dummy_value_c), // 0 // output regs (dummy) - "=D" (dummy_value_D), // 1 - "=d" (dummy_value_d), // 2 - "=a" (dummy_value_a) // 3 - - : "0" (bpp), // ecx // input regs - "1" (row), // edi - "2" (diff), // edx - "3" (FullLength) // eax - - : "%esi" // clobber list - ); - } - return; // end 1 bpp (bypassing cleanup block!) - - case 2: - { -// _ShiftBpp = 16; // == 2 * 8 -// _ShiftRem = 48; // == 64 - 16 - - __asm__ __volatile__ ( - LOAD_GOT_rbp - // load (former) _ActiveMask for 2nd active byte group - "movq " AMASK4_2_2 ", %%mm7 \n\t" // _amask4_2_2 - RESTORE_rbp -// preload "movl diff, %%edx \n\t" - "movq %%mm7, %%mm6 \n\t" -// preload "mov row, %1 \n\t" // edi/rdi - "psllq $16, %%mm6 \n\t" // move mask in mm6 to cover - // 3rd active byte group -// notused "mov %1, " PSI " \n\t" // lp = row - "movq %%mm6, %%mm5 \n\t" -// preload "movl bpp, %%ecx \n\t" - "add " PCX ", %1 \n\t" // rp = row + bpp - "psllq $16, %%mm5 \n\t" // move mask in mm5 to cover - // 4th active byte group - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PDX ",), %%mm1 \n\t" - - "sub_2lp: \n\t" // shift data for adding first - "psrlq $48, %%mm1 \n\t" // bpp bytes (no need for mask; - // shift clears inactive bytes) - // add 1st active group - "movq (%1," PDX ",), %%mm0 \n\t" - "paddb %%mm1, %%mm0 \n\t" - - // add 2nd active group - "movq %%mm0, %%mm1 \n\t" // mov updated Raws to mm1 - "psllq $16, %%mm1 \n\t" // shift data to pos. correctly - "pand %%mm7, %%mm1 \n\t" // mask to use 2nd active group - "paddb %%mm1, %%mm0 \n\t" - - // add 3rd active group - "movq %%mm0, %%mm1 \n\t" // mov updated Raws to mm1 - "psllq $16, %%mm1 \n\t" // shift data to pos. correctly - "pand %%mm6, %%mm1 \n\t" // mask to use 3rd active group - "paddb %%mm1, %%mm0 \n\t" - - // add 4th active group - "movq %%mm0, %%mm1 \n\t" // mov updated Raws to mm1 - "psllq $16, %%mm1 \n\t" // shift data to pos. correctly - "pand %%mm5, %%mm1 \n\t" // mask to use 4th active group - "addl $8, %%edx \n\t" - "paddb %%mm1, %%mm0 \n\t" - "cmpl %%eax, %%edx \n\t" // MMXLength - "movq %%mm0, -8(%1," PDX ",) \n\t" // write updated Raws to array - "movq %%mm0, %%mm1 \n\t" // prep 1st add at top of loop - "jb sub_2lp \n\t" - - : "=c" (dummy_value_c), // 0 // output regs (dummy) - "=D" (dummy_value_D), // 1 - "=d" (dummy_value_d), // 2 - "=a" (dummy_value_a) // 3 - - : "0" (bpp), // ecx // input regs - "1" (row), // edi - "2" (diff), // edx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1", "%mm5", "%mm6" // clobber list - , "%mm7" -#endif - ); - } - break; // end 2 bpp - - case 6: // formerly shared with 4 bpp case (see comments there) - { -// _ShiftBpp = bpp << 3; // 48 (psllq) -// _ShiftRem = 64 - _ShiftBpp; // 16 (psrlq) - - __asm__ __volatile__ ( -// preload "mov row, %1 \n\t" // edi/rdi -// preload "movl diff, %%edx \n\t" -// notused "mov %1, " PSI " \n\t" // lp = row -// preload "movl bpp, %%ecx \n\t" - "add " PCX ", %1 \n\t" // rp = row + bpp - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PDX ",), %%mm1 \n\t" - - "sub_6lp: \n\t" // shift data for adding first - "psrlq $16, %%mm1 \n\t" // bpp bytes (no need for mask; - // shift clears inactive bytes) - "movq (%1," PDX ",), %%mm0 \n\t" - "paddb %%mm1, %%mm0 \n\t" - - // add 2nd active group - "movq %%mm0, %%mm1 \n\t" // mov updated Raws to mm1 - "psllq $48, %%mm1 \n\t" // shift data to pos. correctly - "addl $8, %%edx \n\t" - "paddb %%mm1, %%mm0 \n\t" - - "cmpl %%eax, %%edx \n\t" // MMXLength - "movq %%mm0, -8(%1," PDX ",) \n\t" // write updated Raws to array - "movq %%mm0, %%mm1 \n\t" // prep 1st add at top of loop - "jb sub_6lp \n\t" - - : "=c" (dummy_value_c), // 0 // output regs (dummy) - "=D" (dummy_value_D), // 1 - "=d" (dummy_value_d), // 2 - "=a" (dummy_value_a) // 3 - - : "0" (bpp), // ecx // input regs - "1" (row), // edi - "2" (diff), // edx - "3" (MMXLength) // eax - -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - : "%mm0", "%mm1" // clobber list -#endif - ); - } - break; // end 6 bpp - - case 8: - { - __asm__ __volatile__ ( -// preload "mov row, %1 \n\t" // edi/rdi -// preload "movl diff, %%edx \n\t" -// notused "mov %1, " PSI " \n\t" // lp = row -// preload "movl bpp, %%ecx \n\t" - "add " PCX ", %1 \n\t" // rp = row + bpp -// preload "movl MMXLength, %%eax \n\t" - - // prime the pump: load the first Raw(x-bpp) data set - "movq -8(%1," PDX ",), %%mm7 \n\t" - "movl %%eax, %%esi \n\t" // copy of MMXLength -> esi - "andl $0x0000003f, %%esi \n\t" // calc bytes over mult of 64 - - "sub_8lp: \n\t" - "movq (%1," PDX ",), %%mm0 \n\t" // load Sub(x) for 1st 8 bytes - "paddb %%mm7, %%mm0 \n\t" - "movq 8(%1," PDX ",), %%mm1 \n\t" // load Sub(x) for 2nd 8 bytes - "movq %%mm0, (%1," PDX ",) \n\t" // write Raw(x) for 1st 8 bytes - - // Now mm0 will be used as Raw(x-bpp) for the 2nd group of 8 bytes. - // This will be repeated for each group of 8 bytes with the 8th - // group being used as the Raw(x-bpp) for the 1st group of the - // next loop. - - "paddb %%mm0, %%mm1 \n\t" - "movq 16(%1," PDX ",), %%mm2 \n\t" // load Sub(x) for 3rd 8 bytes - "movq %%mm1, 8(%1," PDX ",) \n\t" // write Raw(x) for 2nd 8 bytes - "paddb %%mm1, %%mm2 \n\t" - "movq 24(%1," PDX ",), %%mm3 \n\t" // load Sub(x) for 4th 8 bytes - "movq %%mm2, 16(%1," PDX ",) \n\t" // write Raw(x) for 3rd 8 bytes - "paddb %%mm2, %%mm3 \n\t" - "movq 32(%1," PDX ",), %%mm4 \n\t" // load Sub(x) for 5th 8 bytes - "movq %%mm3, 24(%1," PDX ",) \n\t" // write Raw(x) for 4th 8 bytes - "paddb %%mm3, %%mm4 \n\t" - "movq 40(%1," PDX ",), %%mm5 \n\t" // load Sub(x) for 6th 8 bytes - "movq %%mm4, 32(%1," PDX ",) \n\t" // write Raw(x) for 5th 8 bytes - "paddb %%mm4, %%mm5 \n\t" - "movq 48(%1," PDX ",), %%mm6 \n\t" // load Sub(x) for 7th 8 bytes - "movq %%mm5, 40(%1," PDX ",) \n\t" // write Raw(x) for 6th 8 bytes - "paddb %%mm5, %%mm6 \n\t" - "movq 56(%1," PDX ",), %%mm7 \n\t" // load Sub(x) for 8th 8 bytes - "movq %%mm6, 48(%1," PDX ",) \n\t" // write Raw(x) for 7th 8 bytes - "addl $64, %%edx \n\t" - "paddb %%mm6, %%mm7 \n\t" - "cmpl %%esi, %%edx \n\t" // cmp to bytes over mult of 64 - "movq %%mm7, -8(%1," PDX ",) \n\t" // write Raw(x) for 8th 8 bytes - "jb sub_8lp \n\t" - - "cmpl %%eax, %%edx \n\t" // compare to MMXLength - "jnb sub_8lt8 \n\t" - - "sub_8lpA: \n\t" - "movq (%1," PDX ",), %%mm0 \n\t" - "addl $8, %%edx \n\t" - "paddb %%mm7, %%mm0 \n\t" - "cmpl %%eax, %%edx \n\t" // compare to MMXLength - "movq %%mm0, -8(%1," PDX ",) \n\t" // -8 to offset early addl edx - "movq %%mm0, %%mm7 \n\t" // move calculated Raw(x) data - "jb sub_8lpA \n\t" // to mm7 to be new Raw(x-bpp) - // for next loop - "sub_8lt8: \n\t" - - : "=c" (dummy_value_c), // 0 // output regs (dummy) - "=D" (dummy_value_D), // 1 - "=d" (dummy_value_d), // 2 - "=a" (dummy_value_a) // 3 - - : "0" (bpp), // ecx // input regs - "1" (row), // edi - "2" (diff), // edx - "3" (MMXLength) // eax - - : "%esi" // clobber list -#if defined(CLOBBER_MMX_REGS_SUPPORTED) - , "%mm0", "%mm1", "%mm2", "%mm3" - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - } - break; // end 8 bpp - - default: // bpp != 1,2,3,4,6,8: doesn't exist - { - // ERROR: SHOULD NEVER BE REACHED -#if defined(PNG_DEBUG) - png_debug(1, "Internal libpng logic error (GCC " - "png_read_filter_row_mmx_sub())\n"); -#endif - } - break; - - } // end switch (bpp) - - __asm__ __volatile__ ( -//pre "movl MMXLength, %%eax \n\t" -//pre "mov row, %1 \n\t" // edi/rdi -//pre "cmpl FullLength, %%eax \n\t" - "cmpl %%edx, %%eax \n\t" - "jnb sub_end \n\t" - - "mov %1, " PSI " \n\t" // lp = row -//pre "movl bpp, %%ecx \n\t" - "add " PCX ", %1 \n\t" // rp = row + bpp - "xorl %%ecx, %%ecx \n\t" - - "sub_lp2: \n\t" - "movb (" PSI "," PAX ",), %%cl \n\t" - "addb %%cl, (%1," PAX ",) \n\t" - "incl %%eax \n\t" - "cmpl %%edx, %%eax \n\t" // FullLength - "jb sub_lp2 \n\t" - - "sub_end: \n\t" - "EMMS \n\t" // end MMX instructions - - : "=c" (dummy_value_c), // 0 // output regs (dummy) - "=D" (dummy_value_D), // 1 - "=a" (dummy_value_a), // 2 - "=d" (dummy_value_d) // 3 - - : "0" (bpp), // ecx // input regs - "1" (row), // edi - "2" (MMXLength), // eax - "3" (FullLength) // edx - - : "%esi" // clobber list - ); - -} // end of png_read_filter_row_mmx_sub() - -#endif /* PNG_MMX_READ_FILTER_SUB_SUPPORTED */ - - - - -#if defined(PNG_MMX_READ_FILTER_UP_SUPPORTED) - -//===========================================================================// -// // -// P N G _ R E A D _ F I L T E R _ R O W _ M M X _ U P // -// // -//===========================================================================// - -// Optimized code for PNG Up filter decoder - -static void /* PRIVATE */ -png_read_filter_row_mmx_up(png_row_infop row_info, png_bytep row, - png_bytep prev_row) -{ - unsigned len; // png_uint_32 is actually 64-bit on x86-64 - int dummy_value_d; // fix 'forbidden register 3 (dx) was spilled' error - png_bytep dummy_value_S; - png_bytep dummy_value_D; - - len = row_info->rowbytes; // number of bytes to filter - - __asm__ __volatile__ ( - SAVE_GOT_ebx -//pre "mov prev_row, %1 \n\t" // esi/rsi -//pre "movl row, %2 \n\t" // edi/rdi - - "xorl %%ebx, %%ebx \n\t" - "xorl %%eax, %%eax \n\t" - - // get # of bytes to alignment (note: computing _delta_ of two pointers, - // so hereafter %%ecx is sufficient even on 64-bit) - "mov %2, " PCX " \n\t" // take start of row - "add $0x7, " PCX " \n\t" // add 7 to incr past alignment bdry -// "andl $0xfffffff8, %%ecx \n\t" // mask to alignment boundary (32-bit!) - CLEAR_BOTTOM_3_BITS PCX "\n\t" // mask to alignment boundary - "sub %2, " PCX " \n\t" // subtract row ptr again => ebp = - "jz up_go \n\t" // target value of ecx at alignment - - "up_lp1: \n\t" // fix alignment - "movb (%2," PBX ",), %%al \n\t" - "addb (%1," PBX ",), %%al \n\t" - "incl %%ebx \n\t" - "cmpl %%ecx, %%ebx \n\t" - "movb %%al, -1(%2," PBX ",) \n\t" // mov does not affect flags; -1 to - "jb up_lp1 \n\t" // offset incl ebx - - "up_go: \n\t" -//pre "movl len, %%edx \n\t" - "movl %%edx, %%ecx \n\t" - "subl %%ebx, %%edx \n\t" // subtract alignment fix - "andl $0x0000003f, %%edx \n\t" // calc bytes over mult of 64 - "subl %%edx, %%ecx \n\t" // sub over-bytes from original length - - // unrolled loop - use all MMX registers and interleave to reduce - // number of branch instructions (loops) and reduce partial stalls - "up_loop: \n\t" - "movq (%1," PBX ",), %%mm1 \n\t" - "movq (%2," PBX ",), %%mm0 \n\t" - "movq 8(%1," PBX ",), %%mm3 \n\t" - "paddb %%mm1, %%mm0 \n\t" - "movq 8(%2," PBX ",), %%mm2 \n\t" - "movq %%mm0, (%2," PBX ",) \n\t" - "paddb %%mm3, %%mm2 \n\t" - "movq 16(%1," PBX ",), %%mm5 \n\t" - "movq %%mm2, 8(%2," PBX ",) \n\t" - "movq 16(%2," PBX ",), %%mm4 \n\t" - "movq 24(%1," PBX ",), %%mm7 \n\t" - "paddb %%mm5, %%mm4 \n\t" - "movq 24(%2," PBX ",), %%mm6 \n\t" - "movq %%mm4, 16(%2," PBX ",) \n\t" - "paddb %%mm7, %%mm6 \n\t" - "movq 32(%1," PBX ",), %%mm1 \n\t" - "movq %%mm6, 24(%2," PBX ",) \n\t" - "movq 32(%2," PBX ",), %%mm0 \n\t" - "movq 40(%1," PBX ",), %%mm3 \n\t" - "paddb %%mm1, %%mm0 \n\t" - "movq 40(%2," PBX ",), %%mm2 \n\t" - "movq %%mm0, 32(%2," PBX ",) \n\t" - "paddb %%mm3, %%mm2 \n\t" - "movq 48(%1," PBX ",), %%mm5 \n\t" - "movq %%mm2, 40(%2," PBX ",) \n\t" - "movq 48(%2," PBX ",), %%mm4 \n\t" - "movq 56(%1," PBX ",), %%mm7 \n\t" - "paddb %%mm5, %%mm4 \n\t" - "movq 56(%2," PBX ",), %%mm6 \n\t" - "movq %%mm4, 48(%2," PBX ",) \n\t" - "addl $64, %%ebx \n\t" - "paddb %%mm7, %%mm6 \n\t" - "cmpl %%ecx, %%ebx \n\t" - "movq %%mm6, -8(%2," PBX ",) \n\t" // (+56)movq does not affect flags; - "jb up_loop \n\t" // -8 to offset addl ebx - - "cmpl $0, %%edx \n\t" // test for bytes over mult of 64 - "jz up_end \n\t" - - "cmpl $8, %%edx \n\t" // test for less than 8 bytes - "jb up_lt8 \n\t" // [added by lcreeve at netins.net] - - "addl %%edx, %%ecx \n\t" - "andl $0x00000007, %%edx \n\t" // calc bytes over mult of 8 - "subl %%edx, %%ecx \n\t" // drop over-bytes from length - "jz up_lt8 \n\t" - - "up_lpA: \n\t" // use MMX regs to update 8 bytes sim. - "movq (%1," PBX ",), %%mm1 \n\t" - "movq (%2," PBX ",), %%mm0 \n\t" - "addl $8, %%ebx \n\t" - "paddb %%mm1, %%mm0 \n\t" - "cmpl %%ecx, %%ebx \n\t" - "movq %%mm0, -8(%2," PBX ",) \n\t" // movq does not affect flags; -8 to - "jb up_lpA \n\t" // offset add ebx - "cmpl $0, %%edx \n\t" // test for bytes over mult of 8 - "jz up_end \n\t" - - "up_lt8: \n\t" - "xorl %%eax, %%eax \n\t" - "addl %%edx, %%ecx \n\t" // move over byte count into counter - - "up_lp2: \n\t" // use x86 regs for remaining bytes - "movb (%2," PBX ",), %%al \n\t" - "addb (%1," PBX ",), %%al \n\t" - "incl %%ebx \n\t" - "cmpl %%ecx, %%ebx \n\t" - "movb %%al, -1(%2," PBX ",) \n\t" // mov does not affect flags; -1 to - "jb up_lp2 \n\t" // offset inc ebx - - "up_end: \n\t" - "EMMS \n\t" // conversion of filtered row complete - RESTORE_GOT_ebx - - : "=d" (dummy_value_d), // 0 // output regs (dummy) - "=S" (dummy_value_S), // 1 - "=D" (dummy_value_D) // 2 - - : "0" (len), // edx // input regs - "1" (prev_row), // esi - "2" (row) // edi - - : "%eax", "%ecx" // clobber list (no input regs!) - _CLOBBER_GOT_ebx -#if defined(PNG_CLOBBER_MMX_REGS_SUPPORTED) - , "%mm0", "%mm1", "%mm2", "%mm3" - , "%mm4", "%mm5", "%mm6", "%mm7" -#endif - ); - -} // end of png_read_filter_row_mmx_up() - -#endif /* PNG_MMX_READ_FILTER_UP_SUPPORTED */ - - - - -/*===========================================================================*/ -/* */ -/* P N G _ R E A D _ F I L T E R _ R O W */ -/* */ -/*===========================================================================*/ - -/* Optimized png_read_filter_row routines */ - -void /* PRIVATE */ -png_read_filter_row(png_structp png_ptr, png_row_infop row_info, png_bytep - row, png_bytep prev_row, int filter) -{ -#if defined(PNG_DEBUG) - char filtname[10]; -#endif - - if (_mmx_supported == 2) { -#if !defined(PNG_1_0_X) - /* this should have happened in png_init_mmx_flags() already */ - png_warning(png_ptr, "asm_flags may not have been initialized"); -#endif - png_mmx_support(); - } - -#if defined(PNG_DEBUG) - png_debug(1, "in png_read_filter_row (pnggccrd.c)\n"); - switch (filter) - { - case 0: - png_snprintf(filtname, 10, "none"); - break; - - case 1: - png_snprintf(filtname, 10, "sub-%s", -#ifdef PNG_MMX_READ_FILTER_SUB_SUPPORTED -#if !defined(PNG_1_0_X) - ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_SUB) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - _mmx_supported -#endif - ? "MMX" : -#endif - "C"); - break; - - case 2: - png_snprintf(filtname, 10, "up-%s", -#ifdef PNG_MMX_READ_FILTER_UP_SUPPORTED -#if !defined(PNG_1_0_X) - ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_UP) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - _mmx_supported -#endif - ? "MMX" : -#endif - "C"); - break; - - case 3: - png_snprintf(filtname, 10, "avg-%s", -#ifdef PNG_MMX_READ_FILTER_AVG_SUPPORTED -#if !defined(PNG_1_0_X) - ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_AVG) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - _mmx_supported -#endif - ? "MMX" : -#endif - "C"); - break; - - case 4: - png_snprintf(filtname, 10, "paeth-%s", -#ifdef PNG_MMX_READ_FILTER_PAETH_SUPPORTED -#if defined(PNG_x86_64_USE_GOTPCREL) || defined(PNG_THREAD_UNSAFE_OK) -#if !defined(PNG_1_0_X) - ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_PAETH) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - _mmx_supported -#endif - ? "MMX" : -#endif /* PNG_x86_64_USE_GOTPCREL || PNG_THREAD_UNSAFE_OK */ -#endif - "C"); - break; - - default: - png_snprintf(filtname, 10, "unknown"); - break; - } - png_debug2(2, "row_number=%ld, %s, ", png_ptr->row_number, filtname); - //png_debug1(0, "png_ptr=%10p, ", png_ptr); - //png_debug1(0, "asm_flags=0x%08lx, ", png_ptr->asm_flags); - png_debug1(0, "row=%10p, ", row); - png_debug2(0, "pixdepth=%d, bytes=%d, ", (int)row_info->pixel_depth, - (int)((row_info->pixel_depth + 7) >> 3)); - png_debug1(0, "rowbytes=%ld\n", row_info->rowbytes); -#endif /* PNG_DEBUG */ - - switch (filter) - { - case PNG_FILTER_VALUE_NONE: - break; - - case PNG_FILTER_VALUE_SUB: -#ifdef PNG_MMX_READ_FILTER_SUB_SUPPORTED -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_SUB) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - if (_mmx_supported) -#endif - { - png_read_filter_row_mmx_sub(row_info, row); - } - else -#endif - { - png_uint_32 i; - png_uint_32 istop = row_info->rowbytes; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_bytep rp = row + bpp; - png_bytep lp = row; - - for (i = bpp; i < istop; i++) - { - *rp = (png_byte)(((int)(*rp) + (int)(*lp++)) & 0xff); - rp++; - } - } /* end !UseMMX_sub */ - break; - - case PNG_FILTER_VALUE_UP: -#ifdef PNG_MMX_READ_FILTER_UP_SUPPORTED -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_UP) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - if (_mmx_supported) -#endif - { - png_read_filter_row_mmx_up(row_info, row, prev_row); - } - else -#endif - { - png_uint_32 i; - png_uint_32 istop = row_info->rowbytes; - png_bytep rp = row; - png_bytep pp = prev_row; - - for (i = 0; i < istop; ++i) - { - *rp = (png_byte)(((int)(*rp) + (int)(*pp++)) & 0xff); - rp++; - } - } /* end !UseMMX_up */ - break; - - case PNG_FILTER_VALUE_AVG: -#ifdef PNG_MMX_READ_FILTER_AVG_SUPPORTED -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_AVG) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - if (_mmx_supported) -#endif - { - png_read_filter_row_mmx_avg(row_info, row, prev_row); - } - else -#endif - { - png_uint_32 i; - png_bytep rp = row; - png_bytep pp = prev_row; - png_bytep lp = row; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_uint_32 istop = row_info->rowbytes - bpp; - - for (i = 0; i < bpp; i++) - { - *rp = (png_byte)(((int)(*rp) + - ((int)(*pp++) >> 1)) & 0xff); - rp++; - } - - for (i = 0; i < istop; i++) - { - *rp = (png_byte)(((int)(*rp) + - ((int)(*pp++ + *lp++) >> 1)) & 0xff); - rp++; - } - } /* end !UseMMX_avg */ - break; - - case PNG_FILTER_VALUE_PAETH: -#ifdef PNG_MMX_READ_FILTER_PAETH_SUPPORTED -#if defined(PNG_x86_64_USE_GOTPCREL) || defined(PNG_THREAD_UNSAFE_OK) -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_PAETH) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - if (_mmx_supported) -#endif - { - png_read_filter_row_mmx_paeth(row_info, row, prev_row); - } - else -#endif /* PNG_x86_64_USE_GOTPCREL || PNG_THREAD_UNSAFE_OK */ -#endif - { - png_uint_32 i; - png_bytep rp = row; - png_bytep pp = prev_row; - png_bytep lp = row; - png_bytep cp = prev_row; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_uint_32 istop = row_info->rowbytes - bpp; - - for (i = 0; i < bpp; i++) - { - *rp = (png_byte)(((int)(*rp) + (int)(*pp++)) & 0xff); - rp++; - } - - for (i = 0; i < istop; i++) /* use leftover rp,pp */ - { - int a, b, c, pa, pb, pc, p; - - a = *lp++; - b = *pp++; - c = *cp++; - - p = b - c; - pc = a - c; - -#if defined(PNG_USE_ABS) - pa = abs(p); - pb = abs(pc); - pc = abs(p + pc); -#else - pa = p < 0 ? -p : p; - pb = pc < 0 ? -pc : pc; - pc = (p + pc) < 0 ? -(p + pc) : p + pc; -#endif - - /* - if (pa <= pb && pa <= pc) - p = a; - else if (pb <= pc) - p = b; - else - p = c; - */ - - p = (pa <= pb && pa <= pc) ? a : (pb <= pc) ? b : c; - - *rp = (png_byte)(((int)(*rp) + p) & 0xff); - rp++; - } - } /* end !UseMMX_paeth */ - break; - - default: - png_warning(png_ptr, "Ignoring bad row-filter type"); - *row=0; - break; - } -} - -#endif /* PNG_HAVE_MMX_READ_FILTER_ROW */ - - -#endif /* PNG_ASSEMBLER_CODE_SUPPORTED && PNG_USE_PNGGCCRD */ -#endif /* __GNUC__ */ +/* pnggccrd.c was removed from libpng-1.2.20. */ diff --git a/pngget.c b/pngget.c index 75d2ca01..9cf1b845 100644 --- a/pngget.c +++ b/pngget.c @@ -841,6 +841,7 @@ png_get_compression_buffer_size(png_structp png_ptr) png_uint_32 PNGAPI png_get_asm_flags (png_structp png_ptr) { + /* obsolete, to be removed from libpng-1.4.0 */ #ifdef PNG_MMX_CODE_SUPPORTED return (png_uint_32)(png_ptr? png_ptr->asm_flags : 0L); #else @@ -852,94 +853,36 @@ png_get_asm_flags (png_structp png_ptr) png_uint_32 PNGAPI png_get_asm_flagmask (int flag_select) { -#ifdef PNG_MMX_CODE_SUPPORTED - png_uint_32 settable_asm_flags = 0; - - if (flag_select & PNG_SELECT_READ) - settable_asm_flags |= - PNG_ASM_FLAG_MMX_READ_COMBINE_ROW | - PNG_ASM_FLAG_MMX_READ_INTERLACE | - PNG_ASM_FLAG_MMX_READ_FILTER_SUB | - PNG_ASM_FLAG_MMX_READ_FILTER_UP | - PNG_ASM_FLAG_MMX_READ_FILTER_AVG | - PNG_ASM_FLAG_MMX_READ_FILTER_PAETH ; - /* no non-MMX flags yet */ - -#if 0 - /* GRR: no write-flags yet, either, but someday... */ - if (flag_select & PNG_SELECT_WRITE) - settable_asm_flags |= - PNG_ASM_FLAG_MMX_WRITE_ [whatever] ; -#endif /* 0 */ - - return settable_asm_flags; /* _theoretically_ settable capabilities only */ -#else - return (0L); -#endif /* PNG_MMX_CODE_SUPPORTED */ + /* obsolete, to be removed from libpng-1.4.0 */ + flag_select=flag_select; + return 0L; } - /* GRR: could add this: && defined(PNG_MMX_CODE_SUPPORTED) */ /* this function was added to libpng 1.2.0 */ png_uint_32 PNGAPI png_get_mmx_flagmask (int flag_select, int *compilerID) { -#if defined(PNG_MMX_CODE_SUPPORTED) - png_uint_32 settable_mmx_flags = 0; - - if (flag_select & PNG_SELECT_READ) - settable_mmx_flags |= - PNG_ASM_FLAG_MMX_READ_COMBINE_ROW | - PNG_ASM_FLAG_MMX_READ_INTERLACE | - PNG_ASM_FLAG_MMX_READ_FILTER_SUB | - PNG_ASM_FLAG_MMX_READ_FILTER_UP | - PNG_ASM_FLAG_MMX_READ_FILTER_AVG | - PNG_ASM_FLAG_MMX_READ_FILTER_PAETH ; -#if 0 - /* GRR: no MMX write support yet, but someday... */ - if (flag_select & PNG_SELECT_WRITE) - settable_mmx_flags |= - PNG_ASM_FLAG_MMX_WRITE_ [whatever] ; -#endif /* 0 */ - - if (compilerID != NULL) { -#ifdef PNG_USE_PNGVCRD - *compilerID = 1; /* MSVC */ -#else -#ifdef PNG_USE_PNGGCCRD - *compilerID = 2; /* gcc/gas */ -#else - *compilerID = -1; /* unknown (i.e., no asm/MMX code compiled) */ -#endif -#endif - } - - return settable_mmx_flags; /* _theoretically_ settable capabilities only */ -#else - return (0L); -#endif /* ?PNG_MMX_CODE_SUPPORTED */ + /* obsolete, to be removed from libpng-1.4.0 */ + flag_select=flag_select; + *compilerID = -1; /* unknown (i.e., no asm/MMX code compiled) */ + return 0L; } /* this function was added to libpng 1.2.0 */ png_byte PNGAPI png_get_mmx_bitdepth_threshold (png_structp png_ptr) { -#if defined(PNG_MMX_CODE_SUPPORTED) - return (png_byte)(png_ptr? png_ptr->mmx_bitdepth_threshold : 0); -#else + /* obsolete, to be removed from libpng-1.4.0 */ return (png_ptr? 0: 0); -#endif /* ?PNG_MMX_CODE_SUPPORTED */ } /* this function was added to libpng 1.2.0 */ png_uint_32 PNGAPI png_get_mmx_rowbytes_threshold (png_structp png_ptr) { -#if defined(PNG_MMX_CODE_SUPPORTED) - return (png_uint_32)(png_ptr? png_ptr->mmx_rowbytes_threshold : 0L); -#else + /* obsolete, to be removed from libpng-1.4.0 */ return (png_ptr? 0L: 0L); -#endif /* ?PNG_MMX_CODE_SUPPORTED */ } #endif /* ?PNG_1_0_X */ #endif /* ?PNG_ASSEMBLER_CODE_SUPPORTED */ diff --git a/pngpread.c b/pngpread.c index 3d910c30..d5952bbd 100644 --- a/pngpread.c +++ b/pngpread.c @@ -1,7 +1,7 @@ /* pngpread.c - read a png file in push mode * - * Last changed in libpng 1.2.20 August 30, 2007 + * Last changed in libpng 1.2.20 September 1, 2007 * For conditions of distribution and use, see copyright notice in png.h * Copyright (c) 1998-2007 Glenn Randers-Pehrson * (Version 0.96 Copyright (c) 1996, 1997 Andreas Dilger) @@ -1001,11 +1001,6 @@ png_read_push_finish_row(png_structp png_ptr) /* offset to next interlace block in the y direction */ PNG_CONST int FARDATA png_pass_yinc[] = {8, 8, 8, 4, 4, 2, 2}; - /* Width of interlace block. This is not currently used - if you need - * it, uncomment it here and in png.h - PNG_CONST int FARDATA png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; - */ - /* Height of interlace block. This is not currently used - if you need * it, uncomment it here and in png.h PNG_CONST int FARDATA png_pass_height[] = {8, 8, 4, 4, 2, 2, 1}; diff --git a/pngread.c b/pngread.c index c26ce611..c9140309 100644 --- a/pngread.c +++ b/pngread.c @@ -55,12 +55,6 @@ png_create_read_struct_2(png_const_charp user_png_ver, png_voidp error_ptr, if (png_ptr == NULL) return (NULL); -#if !defined(PNG_1_0_X) -#ifdef PNG_MMX_CODE_SUPPORTED - png_init_mmx_flags(png_ptr); /* 1.2.0 addition */ -#endif -#endif /* PNG_1_0_X */ - /* added at libpng-1.2.6 */ #ifdef PNG_SET_USER_LIMITS_SUPPORTED png_ptr->user_width_max=PNG_USER_WIDTH_MAX; diff --git a/pngrutil.c b/pngrutil.c index 4da862e3..99bca193 100644 --- a/pngrutil.c +++ b/pngrutil.c @@ -2277,1051 +2277,10 @@ png_check_chunk_name(png_structp png_ptr, png_bytep chunk_name) to any alpha or transparency value associated with the pixel. If you want all pixels to be combined, pass 0xff (255) in mask. */ -/* Optimized C version of utilities to read a PNG file - * - * Based on code contributed by Nirav Chhatrapati, Intel Corp., 1998. - * Interface to libpng contributed by Gilles Vollant, 1999. - * GNU C port by Greg Roelofs, 1999-2001. - * - */ - -#if defined(PNG_OPTIMIZED_CODE_SUPPORTED) -#if !defined(PNG_HAVE_MMX_COMBINE_ROW) - -/*===========================================================================*/ -/* */ -/* P N G _ C O M B I N E _ R O W */ -/* */ -/*===========================================================================*/ - - -#define BPP2 2 -#define BPP3 3 /* bytes per pixel (a.k.a. pixel_bytes) */ -#define BPP4 4 -#define BPP6 6 /* (defined only to help avoid cut-and-paste errors) */ -#define BPP8 8 - -/* Combines the row recently read in with the previous row. - This routine takes care of alpha and transparency if requested. - This routine also handles the two methods of progressive display - of interlaced images, depending on the mask value. - The mask value describes which pixels are to be combined with - the row. The pattern always repeats every 8 pixels, so just 8 - bits are needed. A one indicates the pixel is to be combined; a - zero indicates the pixel is to be skipped. This is in addition - to any alpha or transparency value associated with the pixel. - If you want all pixels to be combined, pass 0xff (255) in mask. */ - -/* Use this routine for the x86 platform - it uses a faster MMX routine - if the machine supports MMX. */ - void /* PRIVATE */ png_combine_row(png_structp png_ptr, png_bytep row, int mask) { - -#if defined(PNG_USE_LOCAL_ARRAYS) -static PNG_CONST int FARDATA png_pass_start[7] = {0, 4, 0, 2, 0, 1, 0}; -static PNG_CONST int FARDATA png_pass_inc[7] = {8, 8, 4, 4, 2, 2, 1}; -static PNG_CONST int FARDATA png_pass_width[7] = {8, 4, 4, 2, 2, 1, 1}; -#endif - - png_debug(1, "in png_combine_row (pngrutil.c OPTIMIZED)\n"); - - if (mask == 0xff) - { - png_debug(2,"mask == 0xff: doing single png_memcpy()\n"); - png_memcpy(row, png_ptr->row_buf + 1, - (png_size_t)PNG_ROWBYTES(png_ptr->row_info.pixel_depth,png_ptr->width)); - } - else /* (png_combine_row() is never called with mask == 0) */ - { - switch (png_ptr->row_info.pixel_depth) - { - /* most common case: combining 24-bit RGB */ - case 24: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - - { - register png_uint_32 i; - png_uint_32 initial_val = BPP3 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP3 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP3 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP3 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP3; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } /* end of else (_mmx_supported) */ - - break; - } /* end 24 bpp */ - - case 32: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - - { - register png_uint_32 i; - png_uint_32 initial_val = BPP4 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP4 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP4 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP4 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP4; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } - - break; - } /* end 32 bpp */ - - case 8: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - { - register png_uint_32 i; - png_uint_32 initial_val = png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff /* *BPP1 */ ; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } - - break; - } /* end 8 bpp */ - - case 1: /* png_ptr->row_info.pixel_depth */ - { - png_bytep sp; - png_bytep dp; - int s_inc, s_start, s_end; - int m; - int shift; - png_uint_32 i; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 7; - s_inc = 1; - } - else -#endif - { - s_start = 7; - s_end = 0; - s_inc = -1; - } - - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - int value; - - value = (*sp >> shift) & 0x1; - *dp &= (png_byte)((0x7f7f >> (7 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } /* end 1 bpp */ - - case 2: /* png_ptr->row_info.pixel_depth */ - { - png_bytep sp; - png_bytep dp; - int s_start, s_end, s_inc; - int m; - int shift; - png_uint_32 i; - int value; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 6; - s_inc = 2; - } - else -#endif - { - s_start = 6; - s_end = 0; - s_inc = -2; - } - - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - value = (*sp >> shift) & 0x3; - *dp &= (png_byte)((0x3f3f >> (6 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } /* end 2 bpp */ - - case 4: /* png_ptr->row_info.pixel_depth */ - { - png_bytep sp; - png_bytep dp; - int s_start, s_end, s_inc; - int m; - int shift; - png_uint_32 i; - int value; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 4; - s_inc = 4; - } - else -#endif - { - s_start = 4; - s_end = 0; - s_inc = -4; - } - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - value = (*sp >> shift) & 0xf; - *dp &= (png_byte)((0xf0f >> (4 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } /* end 4 bpp */ - - case 16: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - - { - register png_uint_32 i; - png_uint_32 initial_val = BPP2 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP2 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP2 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP2 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP2; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } /* end of else (_mmx_supported) */ - - break; - } /* end 16 bpp */ - - - - case 48: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - { - register png_uint_32 i; - png_uint_32 initial_val = BPP6 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP6 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP6 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP6 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP6; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - } - break; - } /* end 48 bpp */ - - case 64: /* png_ptr->row_info.pixel_depth */ - { - png_bytep srcptr; - png_bytep dstptr; - register png_uint_32 i; - png_uint_32 initial_val = BPP8 * png_pass_start[png_ptr->pass]; - /* png.c: png_pass_start[] = {0, 4, 0, 2, 0, 1, 0}; */ - register int stride = BPP8 * png_pass_inc[png_ptr->pass]; - /* png.c: png_pass_inc[] = {8, 8, 4, 4, 2, 2, 1}; */ - register int rep_bytes = BPP8 * png_pass_width[png_ptr->pass]; - /* png.c: png_pass_width[] = {8, 4, 4, 2, 2, 1, 1}; */ - png_uint_32 len = png_ptr->width &~7; /* reduce to mult of 8 */ - int diff = (int) (png_ptr->width & 7); /* amount lost */ - register png_uint_32 final_val = BPP8 * len; /* GRR bugfix */ - - srcptr = png_ptr->row_buf + 1 + initial_val; - dstptr = row + initial_val; - - for (i = initial_val; i < final_val; i += stride) - { - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - if (diff) /* number of leftover pixels: 3 for pngtest */ - { - final_val += diff*BPP8; - for (; i < final_val; i += stride) - { - if (rep_bytes > (int)(final_val-i)) - rep_bytes = (int)(final_val-i); - png_memcpy(dstptr, srcptr, rep_bytes); - srcptr += stride; - dstptr += stride; - } - } - - break; - } /* end 64 bpp */ - - default: /* png_ptr->row_info.pixel_depth != 1,2,4,8,16,24,32,48,64 */ - { - /* this should never happen */ - png_warning(png_ptr, "Invalid row_info.pixel_depth in pngrutil"); - break; - } - } /* end switch (png_ptr->row_info.pixel_depth) */ - - } /* end if (non-trivial mask) */ - -} /* end png_combine_row() */ -#endif /* PNG_HAVE_MMX_COMBINE_ROW */ - - - -/*===========================================================================*/ -/* */ -/* P N G _ D O _ R E A D _ I N T E R L A C E */ -/* */ -/*===========================================================================*/ - -#if defined(PNG_READ_INTERLACING_SUPPORTED) -#if !defined(PNG_HAVE_MMX_READ_INTERLACE) - -/* png_do_read_interlace() is called after any 16-bit to 8-bit conversion - * has taken place. [GRR: what other steps come before and/or after?] - */ - -void /* PRIVATE */ -png_do_read_interlace(png_structp png_ptr) -{ -#if defined(PNG_USE_LOCAL_ARRAYS) -static PNG_CONST int FARDATA png_pass_inc[7] = {8, 8, 4, 4, 2, 2, 1}; -#endif - png_row_infop row_info = &(png_ptr->row_info); - png_bytep row = png_ptr->row_buf + 1; - int pass = png_ptr->pass; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - png_uint_32 transformations = png_ptr->transformations; -#endif - png_debug(1,"in png_do_read_interlace (pngrutil.c OPTIMIZED)\n"); - - if (row != NULL && row_info != NULL) - { - png_uint_32 final_width; - - final_width = row_info->width * png_pass_inc[pass]; - - switch (row_info->pixel_depth) - { - case 1: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_byte v; - png_uint_32 i; - int j; - - sp = row + (png_size_t)((row_info->width - 1) >> 3); - dp = row + (png_size_t)((final_width - 1) >> 3); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (int)((row_info->width + 7) & 7); - dshift = (int)((final_width + 7) & 7); - s_start = 7; - s_end = 0; - s_inc = -1; - } - else -#endif - { - sshift = 7 - (int)((row_info->width + 7) & 7); - dshift = 7 - (int)((final_width + 7) & 7); - s_start = 0; - s_end = 7; - s_inc = 1; - } - - for (i = row_info->width; i; i--) - { - v = (png_byte)((*sp >> sshift) & 0x1); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0x7f7f >> (7 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - case 2: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_uint_32 i; - - sp = row + (png_size_t)((row_info->width - 1) >> 2); - dp = row + (png_size_t)((final_width - 1) >> 2); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (png_size_t)(((row_info->width + 3) & 3) << 1); - dshift = (png_size_t)(((final_width + 3) & 3) << 1); - s_start = 6; - s_end = 0; - s_inc = -2; - } - else -#endif - { - sshift = (png_size_t)((3 - ((row_info->width + 3) & 3)) << 1); - dshift = (png_size_t)((3 - ((final_width + 3) & 3)) << 1); - s_start = 0; - s_end = 6; - s_inc = 2; - } - - for (i = row_info->width; i; i--) - { - png_byte v; - int j; - - v = (png_byte)((*sp >> sshift) & 0x3); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0x3f3f >> (6 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - case 4: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_uint_32 i; - - sp = row + (png_size_t)((row_info->width - 1) >> 1); - dp = row + (png_size_t)((final_width - 1) >> 1); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (png_size_t)(((row_info->width + 1) & 1) << 2); - dshift = (png_size_t)(((final_width + 1) & 1) << 2); - s_start = 4; - s_end = 0; - s_inc = -4; - } - else -#endif - { - sshift = (png_size_t)((1 - ((row_info->width + 1) & 1)) << 2); - dshift = (png_size_t)((1 - ((final_width + 1) & 1)) << 2); - s_start = 0; - s_end = 4; - s_inc = 4; - } - - for (i = row_info->width; i; i--) - { - png_byte v; - int j; - - v = (png_byte)((*sp >> sshift) & 0xf); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0xf0f >> (4 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - /*====================================================================*/ - - default: /* 8-bit or larger (this is where the routine is modified) */ - { - png_bytep sptr, dp; - png_uint_32 i; - png_size_t pixel_bytes; - int width = (int)row_info->width; - - pixel_bytes = (row_info->pixel_depth >> 3); - - /* point sptr at the last pixel in the pre-expanded row: */ - sptr = row + (width - 1) * pixel_bytes; - - /* point dp at the last pixel position in the expanded row: */ - dp = row + (final_width - 1) * pixel_bytes; - - /* MMX not supported: use modified C code - takes advantage - * of inlining of png_memcpy for a constant */ - /* GRR 19991007: does it? or should pixel_bytes in each - * block be replaced with immediate value (e.g., 1)? */ - /* GRR 19991017: replaced with constants in each case */ - { - if (pixel_bytes == 1) - { - for (i = width; i; i--) - { - int j; - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp-- = *sptr; - } - --sptr; - } - } - else if (pixel_bytes == 3) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, 3); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, 3); - dp -= 3; - } - sptr -= 3; - } - } - else if (pixel_bytes == 2) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, 2); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, 2); - dp -= 2; - } - sptr -= 2; - } - } - else if (pixel_bytes == 4) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, 4); - for (j = 0; j < png_pass_inc[pass]; j++) - { -#if defined(PNG_DEBUG) && defined(PNG_1_0_X) - if (dp < row || dp+3 > row+png_ptr->row_buf_size) - { - printf("dp out of bounds: row=%d, dp=%d, rp=%d\n", - row, dp, row+png_ptr->row_buf_size); - printf("row_buf=%d\n",png_ptr->row_buf_size); - } -#endif - png_memcpy(dp, v, 4); - dp -= 4; - } - sptr -= 4; - } - } - else if (pixel_bytes == 6) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, 6); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, 6); - dp -= 6; - } - sptr -= 6; - } - } - else if (pixel_bytes == 8) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, 8); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, 8); - dp -= 8; - } - sptr -= 8; - } - } - else /* GRR: should never be reached */ - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, pixel_bytes); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, pixel_bytes); - dp -= pixel_bytes; - } - sptr -= pixel_bytes; - } - } - - } - break; - } - } /* end switch (row_info->pixel_depth) */ - - row_info->width = final_width; - - row_info->rowbytes = PNG_ROWBYTES(row_info->pixel_depth,final_width); - } - -} /* end png_do_read_interlace() */ - -#endif /* PNG_HAVE_MMX_READ_INTERLACE */ -#endif /* PNG_READ_INTERLACING_SUPPORTED */ - - - -#if !defined(PNG_HAVE_MMX_READ_FILTER_ROW) -/*===========================================================================*/ -/* */ -/* P N G _ R E A D _ F I L T E R _ R O W */ -/* */ -/*===========================================================================*/ - - -/* Optimized png_read_filter_row routines */ - -void /* PRIVATE */ -png_read_filter_row(png_structp png_ptr, png_row_infop row_info, png_bytep - row, png_bytep prev_row, int filter) -{ -#if defined(PNG_DEBUG) - char filnm[10]; -#endif - - -#if defined(PNG_DEBUG) - png_debug(1, "in png_read_filter_row (pngrutil.c OPTIMIZED)\n"); - switch (filter) - { - case 0: - png_snprintf(filnm, 10, "none"); - break; - - case 1: - png_snprintf(filnm, 10, "sub-%s", - "x86"); - break; - - case 2: - png_snprintf(filnm, 10, "up-%s", - "x86"); - break; - - case 3: - png_snprintf(filnm, 10, "avg-%s", - "x86"); - break; - - case 4: - png_snprintf(filnm, 10, "Paeth-%s", - "x86"); - break; - - default: - png_snprintf(filnm, 10, "unknown"); - break; - } - png_debug2(0, "row_number=%5ld, %10s, ", png_ptr->row_number, filnm); - png_debug1(0, "row=0x%08lx, ", (unsigned long)row); - png_debug2(0, "pixdepth=%2d, bytes=%d, ", (int)row_info->pixel_depth, - (int)((row_info->pixel_depth + 7) >> 3)); - png_debug1(0,"rowbytes=%8ld\n", row_info->rowbytes); -#endif /* PNG_DEBUG */ - - switch (filter) - { - case PNG_FILTER_VALUE_NONE: - break; - - case PNG_FILTER_VALUE_SUB: - { - png_uint_32 i; - png_uint_32 istop = row_info->rowbytes; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_bytep rp = row + bpp; - png_bytep lp = row; - - for (i = bpp; i < istop; i++) - { - *rp = (png_byte)(((int)(*rp) + (int)(*lp++)) & 0xff); - rp++; - } - } - break; - - case PNG_FILTER_VALUE_UP: - { - png_uint_32 i; - png_uint_32 istop = row_info->rowbytes; - png_bytep rp = row; - png_bytep pp = prev_row; - - for (i = 0; i < istop; ++i) - { - *rp = (png_byte)(((int)(*rp) + (int)(*pp++)) & 0xff); - rp++; - } - } - break; - - case PNG_FILTER_VALUE_AVG: - { - png_uint_32 i; - png_bytep rp = row; - png_bytep pp = prev_row; - png_bytep lp = row; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_uint_32 istop = row_info->rowbytes - bpp; - - for (i = 0; i < bpp; i++) - { - *rp = (png_byte)(((int)(*rp) + - ((int)(*pp++) >> 1)) & 0xff); - rp++; - } - - for (i = 0; i < istop; i++) - { - *rp = (png_byte)(((int)(*rp) + - ((int)(*pp++ + *lp++) >> 1)) & 0xff); - rp++; - } - } - break; - - case PNG_FILTER_VALUE_PAETH: - { - png_uint_32 i; - png_bytep rp = row; - png_bytep pp = prev_row; - png_bytep lp = row; - png_bytep cp = prev_row; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_uint_32 istop = row_info->rowbytes - bpp; - - for (i = 0; i < bpp; i++) - { - *rp = (png_byte)(((int)(*rp) + (int)(*pp++)) & 0xff); - rp++; - } - - for (i = 0; i < istop; i++) /* use leftover rp,pp */ - { - int a, b, c, pa, pb, pc, p; - - a = *lp++; - b = *pp++; - c = *cp++; - - p = b - c; - pc = a - c; - -#if defined(PNG_USE_ABS) - pa = abs(p); - pb = abs(pc); - pc = abs(p + pc); -#else - pa = p < 0 ? -p : p; - pb = pc < 0 ? -pc : pc; - pc = (p + pc) < 0 ? -(p + pc) : p + pc; -#endif - - /* - if (pa <= pb && pa <= pc) - p = a; - else if (pb <= pc) - p = b; - else - p = c; - */ - - p = (pa <= pb && pa <= pc) ? a : (pb <= pc) ? b : c; - - *rp = (png_byte)(((int)(*rp) + p) & 0xff); - rp++; - } - } - break; - - default: - png_warning(png_ptr, "Ignoring bad row-filter type"); - *row=0; - break; - } -} - -#endif /* PNG_HAVE_MMX_READ_FILTER_ROW */ -#endif /* PNG_OPTIMIZED_CODE_SUPPORTED */ - -#if !defined(PNG_ASSEMBLER_CODE_SUPPORTED) || \ - !defined(PNG_MMX_CODE_SUPPORTED) || \ - (!defined(PNG_USE_PNGGCCRD) && !defined(PNG_USE_PNGVCRD)) -#if !defined(PNG_OPTIMIZED_CODE_SUPPORTED) -/* Use the unoptimized original C code. This might be removed from a future - * version of libpng if testing proves it to be worthless. */ -void /* PRIVATE */ -png_combine_row(png_structp png_ptr, png_bytep row, int mask) -{ - png_debug(1,"in png_combine_row NOT OPTIMIZED\n"); + png_debug(1,"in png_combine_row\n"); if (mask == 0xff) { png_memcpy(row, png_ptr->row_buf + 1, @@ -3537,7 +2496,7 @@ png_do_read_interlace(png_structp png_ptr) PNG_CONST int png_pass_inc[7] = {8, 8, 4, 4, 2, 2, 1}; #endif - png_debug(1,"in png_do_read_interlace (pngrutil.c NOT OPTIMIZED)\n"); + png_debug(1,"in png_do_read_interlace\n"); if (row != NULL && row_info != NULL) { png_uint_32 final_width; @@ -3750,7 +2709,7 @@ void /* PRIVATE */ png_read_filter_row(png_structp png_ptr, png_row_infop row_info, png_bytep row, png_bytep prev_row, int filter) { - png_debug(1, "in png_read_filter_row (NOT OPTIMIZED)\n"); + png_debug(1, "in png_read_filter_row\n"); png_debug2(2,"row = %lu, filter = %d\n", png_ptr->row_number, filter); switch (filter) { @@ -3868,9 +2827,6 @@ png_read_filter_row(png_structp png_ptr, png_row_infop row_info, png_bytep row, break; } } -#endif /* !PNG_OPTIMIZED_CODE_SUPPORTED */ -#endif /* !PNG_ASSEMBLER_CODE_SUPPORTED || !PNG_MMX_CODE_SUPPORTED || */ - /* (!PNG_USE_PNGGCCRD && !PNG_USE_PNGVCRD) */ void /* PRIVATE */ png_read_finish_row(png_structp png_ptr) @@ -4165,9 +3121,6 @@ defined(PNG_USER_TRANSFORM_PTR_SUPPORTED) #endif png_ptr->big_row_buf = (png_bytep)png_malloc(png_ptr, row_bytes+64); png_ptr->row_buf = png_ptr->big_row_buf+32; -#if defined(PNG_DEBUG) && defined(PNG_USE_PNGGCCRD) && defined(PNG_1_0_X) - png_ptr->row_buf_size = row_bytes; -#endif #ifdef PNG_MAX_MALLOC_64K if ((png_uint_32)png_ptr->rowbytes + 1 > (png_uint_32)65536L) diff --git a/pngset.c b/pngset.c index 8d1cbfcd..2871140d 100644 --- a/pngset.c +++ b/pngset.c @@ -1202,51 +1202,13 @@ png_set_invalid(png_structp png_ptr, png_infop info_ptr, int mask) #ifndef PNG_1_0_X #ifdef PNG_ASSEMBLER_CODE_SUPPORTED -/* this function was added to libpng 1.2.0 and should always exist by default */ +/* function was added to libpng 1.2.0 and should always exist by default */ void PNGAPI png_set_asm_flags (png_structp png_ptr, png_uint_32 asm_flags) { -#ifdef PNG_MMX_CODE_SUPPORTED - png_uint_32 settable_asm_flags; - png_uint_32 settable_mmx_flags; -#endif - if (png_ptr == NULL) - return; -#ifdef PNG_MMX_CODE_SUPPORTED - - settable_mmx_flags = -#ifdef PNG_HAVE_MMX_COMBINE_ROW - PNG_ASM_FLAG_MMX_READ_COMBINE_ROW | -#endif -#ifdef PNG_HAVE_MMX_READ_INTERLACE - PNG_ASM_FLAG_MMX_READ_INTERLACE | -#endif -#ifdef PNG_HAVE_MMX_READ_FILTER_ROW - PNG_ASM_FLAG_MMX_READ_FILTER_SUB | - PNG_ASM_FLAG_MMX_READ_FILTER_UP | - PNG_ASM_FLAG_MMX_READ_FILTER_AVG | - PNG_ASM_FLAG_MMX_READ_FILTER_PAETH | -#endif - 0; - - /* could be some non-MMX ones in the future, but not currently: */ - settable_asm_flags = settable_mmx_flags; - - if (!(png_ptr->asm_flags & PNG_ASM_FLAG_MMX_SUPPORT_COMPILED) || - !(png_ptr->asm_flags & PNG_ASM_FLAG_MMX_SUPPORT_IN_CPU)) - { - /* clear all MMX flags if MMX isn't supported */ - settable_asm_flags &= ~settable_mmx_flags; - png_ptr->asm_flags &= ~settable_mmx_flags; - } - - /* we're replacing the settable bits with those passed in by the user, - * so first zero them out of the master copy, then bitwise-OR in the - * allowed subset that was requested */ - - png_ptr->asm_flags &= ~settable_asm_flags; /* zero them */ - png_ptr->asm_flags |= (asm_flags & settable_asm_flags); /* set them */ -#endif /* ?PNG_MMX_CODE_SUPPORTED */ +/* Obsolete as of libpng-1.2.20 and will be removed from libpng-1.4.0 */ + if (png_ptr != NULL) + png_ptr->asm_flags = 0; } /* this function was added to libpng 1.2.0 */ @@ -1255,12 +1217,9 @@ png_set_mmx_thresholds (png_structp png_ptr, png_byte mmx_bitdepth_threshold, png_uint_32 mmx_rowbytes_threshold) { +/* Obsolete as of libpng-1.2.20 and will be removed from libpng-1.4.0 */ if (png_ptr == NULL) return; -#ifdef PNG_MMX_CODE_SUPPORTED - png_ptr->mmx_bitdepth_threshold = mmx_bitdepth_threshold; - png_ptr->mmx_rowbytes_threshold = mmx_rowbytes_threshold; -#endif /* ?PNG_MMX_CODE_SUPPORTED */ } #endif /* ?PNG_ASSEMBLER_CODE_SUPPORTED */ diff --git a/pngtest.c b/pngtest.c index c46d9177..70734a36 100644 --- a/pngtest.c +++ b/pngtest.c @@ -1548,4 +1548,4 @@ main(int argc, char *argv[]) } /* Generate a compiler error if there is an old png.h in the search path. */ -typedef version_1_2_20rc3 your_png_h_is_not_version_1_2_20rc3; +typedef version_1_2_20rc4 your_png_h_is_not_version_1_2_20rc4; diff --git a/pngvcrd.c b/pngvcrd.c index edd8ff6b..ce4233ef 100644 --- a/pngvcrd.c +++ b/pngvcrd.c @@ -1,3917 +1 @@ - -/* pngvcrd.c - mixed C/assembler version of utilities to read a PNG file - * - * For Intel x86 CPU and Microsoft Visual C++ compiler - * - * Last changed in libpng 1.2.20 August 30, 2007 - * For conditions of distribution and use, see copyright notice in png.h - * Copyright (c) 1998-2007 Glenn Randers-Pehrson - * Copyright (c) 1998, Intel Corporation - * - * Contributed by Nirav Chhatrapati, Intel Corporation, 1998 - * Interface to libpng contributed by Gilles Vollant, 1999 - * - * - * In png_do_read_interlace() in libpng versions 1.0.3a through 1.0.4d, - * a sign error in the post-MMX cleanup code for each pixel_depth resulted - * in bad pixels at the beginning of some rows of some images, and also - * (due to out-of-range memory reads and writes) caused heap corruption - * when compiled with MSVC 6.0. The error was fixed in version 1.0.4e. - * - * [png_read_filter_row_mmx_avg() bpp == 2 bugfix, GRR 20000916] - * - * [runtime MMX configuration, GRR 20010102] - * - * [Copy 6 bytes per pixel, not 4, and use stride of 6, not 4, in the - * second loop of interlace processing of 48-bit pixels, GR-P 20070717] - * - * [move instances of uAll union into local, except for two constant - * instances, GR-P 20070805] - * - * [revised handling of static constants for efficiency, Steve Snyder 20070821] - */ - -#define PNG_INTERNAL -#include "png.h" - -#if defined(PNG_ASSEMBLER_CODE_SUPPORTED) && \ - defined(PNG_USE_PNGVCRD) && defined(PNG_MMX_CODE_SUPPORTED) - -#ifdef PNG_USE_LOCAL_ARRAYS -static PNG_CONST int png_pass_inc[7] = {8, 8, 4, 4, 2, 2, 1}; -#endif - -static int mmx_supported=2; - -int PNGAPI -png_mmx_support(void) -{ - int mmx_supported_local = 0; - _asm { - push ebx //CPUID will trash these - push ecx - push edx - - pushfd //Save Eflag to stack - pop eax //Get Eflag from stack into eax - mov ecx, eax //Make another copy of Eflag in ecx - xor eax, 0x200000 //Toggle ID bit in Eflag [i.e. bit(21)] - push eax //Save modified Eflag back to stack - - popfd //Restored modified value back to Eflag reg - pushfd //Save Eflag to stack - pop eax //Get Eflag from stack - push ecx // save original Eflag to stack - popfd // restore original Eflag - xor eax, ecx //Compare the new Eflag with the original Eflag - jz NOT_SUPPORTED //If the same, CPUID instruction is not supported, - //skip following instructions and jump to - //NOT_SUPPORTED label - - xor eax, eax //Set eax to zero - - _asm _emit 0x0f //CPUID instruction (two bytes opcode) - _asm _emit 0xa2 - - cmp eax, 1 //make sure eax return non-zero value - jl NOT_SUPPORTED //If eax is zero, mmx not supported - - xor eax, eax //set eax to zero - inc eax //Now increment eax to 1. This instruction is - //faster than the instruction "mov eax, 1" - - _asm _emit 0x0f //CPUID instruction - _asm _emit 0xa2 - - and edx, 0x00800000 //mask out all bits but mmx bit(24) - cmp edx, 0 // 0 = mmx not supported - jz NOT_SUPPORTED // non-zero = Yes, mmx IS supported - - mov mmx_supported_local, 1 //set return value to 1 - -NOT_SUPPORTED: - mov eax, mmx_supported_local //move return value to eax - pop edx //CPUID trashed these - pop ecx - pop ebx - } - - //mmx_supported_local=0; // test code for force don't support MMX - //printf("MMX : %u (1=MMX supported)\n",mmx_supported_local); - - mmx_supported = mmx_supported_local; - return mmx_supported_local; -} - -/* Combines the row recently read in with the previous row. - This routine takes care of alpha and transparency if requested. - This routine also handles the two methods of progressive display - of interlaced images, depending on the mask value. - The mask value describes which pixels are to be combined with - the row. The pattern always repeats every 8 pixels, so just 8 - bits are needed. A one indicates the pixel is to be combined; a - zero indicates the pixel is to be skipped. This is in addition - to any alpha or transparency value associated with the pixel. If - you want all pixels to be combined, pass 0xff (255) in mask. */ - -/* Use this routine for x86 platform - uses faster MMX routine if machine - supports MMX */ - -void /* PRIVATE */ -png_combine_row(png_structp png_ptr, png_bytep row, int mask) -{ - static PNG_CONST int offset_table[7] = {0, 4, 0, 2, 0, 1, 0}; - - png_debug(1,"in png_combine_row_asm\n"); - - if (mmx_supported == 2) { -#if !defined(PNG_1_0_X) - /* this should have happened in png_init_mmx_flags() already */ - png_warning(png_ptr, "asm_flags may not have been initialized"); -#endif - png_mmx_support(); - } - - if (mask == 0xff) - { - png_memcpy(row, png_ptr->row_buf + 1, - (png_size_t)PNG_ROWBYTES(png_ptr->row_info.pixel_depth, - png_ptr->width)); - } - /* GRR: add "else if (mask == 0)" case? - * or does png_combine_row() not even get called in that case? */ - else - { - switch (png_ptr->row_info.pixel_depth) - { - case 24: - { - png_bytep srcptr; - png_bytep dstptr; - png_uint_32 len; - int unmask, diff; - - static PNG_CONST __int64 bpp24_mask2=0x0101010202020404, //24bpp - bpp24_mask1=0x0408080810101020, - bpp24_mask0=0x2020404040808080; - - srcptr = png_ptr->row_buf + 1; - dstptr = row; - - unmask = ~mask; - len = (png_ptr->width)&~7; - diff = (png_ptr->width)&7; - -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) - /* && mmx_supported */ ) -#else - if (mmx_supported) -#endif - { - _asm - { - movd mm7, unmask //load bit pattern - psubb mm6,mm6 //zero mm6 - punpcklbw mm7,mm7 - punpcklwd mm7,mm7 - punpckldq mm7,mm7 //fill register with 8 masks - - movq mm0,bpp24_mask0 - movq mm1,bpp24_mask1 - movq mm2,bpp24_mask2 - - pand mm0,mm7 - pand mm1,mm7 - pand mm2,mm7 - - pcmpeqb mm0,mm6 - pcmpeqb mm1,mm6 - pcmpeqb mm2,mm6 - - mov ecx,len //load length of line - mov esi,srcptr //load source - mov ebx,dstptr //load dest - cmp ecx,0 - jz mainloop24end - -mainloop24: - movq mm4,[esi] - pand mm4,mm0 - movq mm6,mm0 - movq mm7,[ebx] - pandn mm6,mm7 - por mm4,mm6 - movq [ebx],mm4 - - - movq mm5,[esi+8] - pand mm5,mm1 - movq mm7,mm1 - movq mm6,[ebx+8] - pandn mm7,mm6 - por mm5,mm7 - movq [ebx+8],mm5 - - movq mm6,[esi+16] - pand mm6,mm2 - movq mm4,mm2 - movq mm7,[ebx+16] - pandn mm4,mm7 - por mm6,mm4 - movq [ebx+16],mm6 - - add esi,24 //inc by 24 bytes processed - add ebx,24 - sub ecx,8 //dec by 8 pixels processed - - ja mainloop24 - -mainloop24end: - mov ecx,diff - cmp ecx,0 - jz end24 - - mov edx,mask - sal edx,24 //make low byte the high byte -secondloop24: - sal edx,1 //move high bit to CF - jnc skip24 //if CF = 0 - mov ax,[esi] - mov [ebx],ax - xor eax,eax - mov al,[esi+2] - mov [ebx+2],al -skip24: - add esi,3 - add ebx,3 - - dec ecx - jnz secondloop24 - -end24: - emms - } - } - else /* mmx not supported - use modified C routine */ - { - register unsigned int incr1, initial_val, final_val; - png_size_t pixel_bytes; - png_uint_32 i; - register int disp = png_pass_inc[png_ptr->pass]; - - pixel_bytes = (png_ptr->row_info.pixel_depth >> 3); - srcptr = png_ptr->row_buf + 1 + offset_table[png_ptr->pass]* - pixel_bytes; - dstptr = row + offset_table[png_ptr->pass]*pixel_bytes; - initial_val = offset_table[png_ptr->pass]*pixel_bytes; - final_val = png_ptr->width*pixel_bytes; - incr1 = (disp)*pixel_bytes; - for (i = initial_val; i < final_val; i += incr1) - { - png_memcpy(dstptr, srcptr, pixel_bytes); - srcptr += incr1; - dstptr += incr1; - } - } /* end of else */ - - break; - } // end 24 bpp - - case 32: - { - png_bytep srcptr; - png_bytep dstptr; - png_uint_32 len; - int unmask, diff; - - static PNG_CONST __int64 bpp32_mask3=0x0101010102020202, //32bpp - bpp32_mask2=0x0404040408080808, - bpp32_mask1=0x1010101020202020, - bpp32_mask0=0x4040404080808080; - - srcptr = png_ptr->row_buf + 1; - dstptr = row; - - unmask = ~mask; - len = (png_ptr->width)&~7; - diff = (png_ptr->width)&7; - -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) - /* && mmx_supported */ ) -#else - if (mmx_supported) -#endif - { - _asm - { - movd mm7, unmask //load bit pattern - psubb mm6,mm6 //zero mm6 - punpcklbw mm7,mm7 - punpcklwd mm7,mm7 - punpckldq mm7,mm7 //fill register with 8 masks - - movq mm0,bpp32_mask0 - movq mm1,bpp32_mask1 - movq mm2,bpp32_mask2 - movq mm3,bpp32_mask3 - - pand mm0,mm7 - pand mm1,mm7 - pand mm2,mm7 - pand mm3,mm7 - - pcmpeqb mm0,mm6 - pcmpeqb mm1,mm6 - pcmpeqb mm2,mm6 - pcmpeqb mm3,mm6 - - mov ecx,len //load length of line - mov esi,srcptr //load source - mov ebx,dstptr //load dest - - cmp ecx,0 //lcr - jz mainloop32end - -mainloop32: - movq mm4,[esi] - pand mm4,mm0 - movq mm6,mm0 - movq mm7,[ebx] - pandn mm6,mm7 - por mm4,mm6 - movq [ebx],mm4 - - movq mm5,[esi+8] - pand mm5,mm1 - movq mm7,mm1 - movq mm6,[ebx+8] - pandn mm7,mm6 - por mm5,mm7 - movq [ebx+8],mm5 - - movq mm6,[esi+16] - pand mm6,mm2 - movq mm4,mm2 - movq mm7,[ebx+16] - pandn mm4,mm7 - por mm6,mm4 - movq [ebx+16],mm6 - - movq mm7,[esi+24] - pand mm7,mm3 - movq mm5,mm3 - movq mm4,[ebx+24] - pandn mm5,mm4 - por mm7,mm5 - movq [ebx+24],mm7 - - add esi,32 //inc by 32 bytes processed - add ebx,32 - sub ecx,8 //dec by 8 pixels processed - - ja mainloop32 - -mainloop32end: - mov ecx,diff - cmp ecx,0 - jz end32 - - mov edx,mask - sal edx,24 //make low byte the high byte -secondloop32: - sal edx,1 //move high bit to CF - jnc skip32 //if CF = 0 - mov eax,[esi] - mov [ebx],eax -skip32: - add esi,4 - add ebx,4 - - dec ecx - jnz secondloop32 - -end32: - emms - } - } - else /* mmx _not supported - Use modified C routine */ - { - register unsigned int incr1, initial_val, final_val; - png_size_t pixel_bytes; - png_uint_32 i; - register int disp = png_pass_inc[png_ptr->pass]; - - pixel_bytes = (png_ptr->row_info.pixel_depth >> 3); - srcptr = png_ptr->row_buf + 1 + offset_table[png_ptr->pass]* - pixel_bytes; - dstptr = row + offset_table[png_ptr->pass]*pixel_bytes; - initial_val = offset_table[png_ptr->pass]*pixel_bytes; - final_val = png_ptr->width*pixel_bytes; - incr1 = (disp)*pixel_bytes; - for (i = initial_val; i < final_val; i += incr1) - { - png_memcpy(dstptr, srcptr, pixel_bytes); - srcptr += incr1; - dstptr += incr1; - } - } /* end of else */ - - break; - } // end 32 bpp - - case 8: - { - png_bytep srcptr; - png_bytep dstptr; - png_uint_32 len; - int m; - int diff, unmask; - - static PNG_CONST __int64 bpp08_mask0=0x0102040810204080; - -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) - /* && mmx_supported */ ) -#else - if (mmx_supported) -#endif - { - srcptr = png_ptr->row_buf + 1; - dstptr = row; - m = 0x80; - unmask = ~mask; - len = png_ptr->width &~7; //reduce to multiple of 8 - diff = png_ptr->width & 7; //amount lost - - _asm - { - movd mm7, unmask //load bit pattern - psubb mm6,mm6 //zero mm6 - punpcklbw mm7,mm7 - punpcklwd mm7,mm7 - punpckldq mm7,mm7 //fill register with 8 masks - - movq mm0,bpp08_mask0 - - pand mm0,mm7 //nonzero if keep byte - pcmpeqb mm0,mm6 //zeros->1s, v versa - - mov ecx,len //load length of line (pixels) - mov esi,srcptr //load source - mov ebx,dstptr //load dest - cmp ecx,0 //lcr - je mainloop8end - -mainloop8: - movq mm4,[esi] - pand mm4,mm0 - movq mm6,mm0 - pandn mm6,[ebx] - por mm4,mm6 - movq [ebx],mm4 - - add esi,8 //inc by 8 bytes processed - add ebx,8 - sub ecx,8 //dec by 8 pixels processed - - ja mainloop8 -mainloop8end: - - mov ecx,diff - cmp ecx,0 - jz end8 - - mov edx,mask - sal edx,24 //make low byte the high byte - -secondloop8: - sal edx,1 //move high bit to CF - jnc skip8 //if CF = 0 - mov al,[esi] - mov [ebx],al -skip8: - inc esi - inc ebx - - dec ecx - jnz secondloop8 -end8: - emms - } - } - else /* mmx not supported - use modified C routine */ - { - register unsigned int incr1, initial_val, final_val; - png_size_t pixel_bytes; - png_uint_32 i; - register int disp = png_pass_inc[png_ptr->pass]; - - pixel_bytes = (png_ptr->row_info.pixel_depth >> 3); - srcptr = png_ptr->row_buf + 1 + offset_table[png_ptr->pass]* - pixel_bytes; - dstptr = row + offset_table[png_ptr->pass]*pixel_bytes; - initial_val = offset_table[png_ptr->pass]*pixel_bytes; - final_val = png_ptr->width*pixel_bytes; - incr1 = (disp)*pixel_bytes; - for (i = initial_val; i < final_val; i += incr1) - { - png_memcpy(dstptr, srcptr, pixel_bytes); - srcptr += incr1; - dstptr += incr1; - } - } /* end of else */ - - break; - } // end 8 bpp - - case 1: - { - png_bytep sp; - png_bytep dp; - int s_inc, s_start, s_end; - int m; - int shift; - png_uint_32 i; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 7; - s_inc = 1; - } - else -#endif - { - s_start = 7; - s_end = 0; - s_inc = -1; - } - - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - int value; - - value = (*sp >> shift) & 0x1; - *dp &= (png_byte)((0x7f7f >> (7 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } - - case 2: - { - png_bytep sp; - png_bytep dp; - int s_start, s_end, s_inc; - int m; - int shift; - png_uint_32 i; - int value; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 6; - s_inc = 2; - } - else -#endif - { - s_start = 6; - s_end = 0; - s_inc = -2; - } - - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - value = (*sp >> shift) & 0x3; - *dp &= (png_byte)((0x3f3f >> (6 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } - - case 4: - { - png_bytep sp; - png_bytep dp; - int s_start, s_end, s_inc; - int m; - int shift; - png_uint_32 i; - int value; - - sp = png_ptr->row_buf + 1; - dp = row; - m = 0x80; -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (png_ptr->transformations & PNG_PACKSWAP) - { - s_start = 0; - s_end = 4; - s_inc = 4; - } - else -#endif - { - s_start = 4; - s_end = 0; - s_inc = -4; - } - shift = s_start; - - for (i = 0; i < png_ptr->width; i++) - { - if (m & mask) - { - value = (*sp >> shift) & 0xf; - *dp &= (png_byte)((0xf0f >> (4 - shift)) & 0xff); - *dp |= (png_byte)(value << shift); - } - - if (shift == s_end) - { - shift = s_start; - sp++; - dp++; - } - else - shift += s_inc; - if (m == 1) - m = 0x80; - else - m >>= 1; - } - break; - } - - case 16: - { - png_bytep srcptr; - png_bytep dstptr; - png_uint_32 len; - int unmask, diff; - static PNG_CONST __int64 bpp16_mask1=0x0101020204040808, - bpp16_mask0=0x1010202040408080; - -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) - /* && mmx_supported */ ) -#else - if (mmx_supported) -#endif - { - srcptr = png_ptr->row_buf + 1; - dstptr = row; - - unmask = ~mask; - len = (png_ptr->width)&~7; - diff = (png_ptr->width)&7; - _asm - { - movd mm7, unmask //load bit pattern - psubb mm6,mm6 //zero mm6 - punpcklbw mm7,mm7 - punpcklwd mm7,mm7 - punpckldq mm7,mm7 //fill register with 8 masks - - movq mm0,bpp16_mask0 - movq mm1,bpp16_mask1 - - pand mm0,mm7 - pand mm1,mm7 - - pcmpeqb mm0,mm6 - pcmpeqb mm1,mm6 - - mov ecx,len //load length of line - mov esi,srcptr //load source - mov ebx,dstptr //load dest - cmp ecx,0 //lcr - jz mainloop16end - -mainloop16: - movq mm4,[esi] - pand mm4,mm0 - movq mm6,mm0 - movq mm7,[ebx] - pandn mm6,mm7 - por mm4,mm6 - movq [ebx],mm4 - - movq mm5,[esi+8] - pand mm5,mm1 - movq mm7,mm1 - movq mm6,[ebx+8] - pandn mm7,mm6 - por mm5,mm7 - movq [ebx+8],mm5 - - add esi,16 //inc by 16 bytes processed - add ebx,16 - sub ecx,8 //dec by 8 pixels processed - - ja mainloop16 - -mainloop16end: - mov ecx,diff - cmp ecx,0 - jz end16 - - mov edx,mask - sal edx,24 //make low byte the high byte -secondloop16: - sal edx,1 //move high bit to CF - jnc skip16 //if CF = 0 - mov ax,[esi] - mov [ebx],ax -skip16: - add esi,2 - add ebx,2 - - dec ecx - jnz secondloop16 -end16: - emms - } - } - else /* mmx not supported - use modified C routine */ - { - register unsigned int incr1, initial_val, final_val; - png_size_t pixel_bytes; - png_uint_32 i; - register int disp = png_pass_inc[png_ptr->pass]; - - pixel_bytes = (png_ptr->row_info.pixel_depth >> 3); - srcptr = png_ptr->row_buf + 1 + offset_table[png_ptr->pass]* - pixel_bytes; - dstptr = row + offset_table[png_ptr->pass]*pixel_bytes; - initial_val = offset_table[png_ptr->pass]*pixel_bytes; - final_val = png_ptr->width*pixel_bytes; - incr1 = (disp)*pixel_bytes; - for (i = initial_val; i < final_val; i += incr1) - { - png_memcpy(dstptr, srcptr, pixel_bytes); - srcptr += incr1; - dstptr += incr1; - } - } /* end of else */ - - break; - } // end 16 bpp - - case 48: - { - png_bytep srcptr; - png_bytep dstptr; - png_uint_32 len; - int unmask, diff; - - static PNG_CONST __int64 bpp48_mask5=0x0101010101010202, - bpp48_mask4=0x0202020204040404, - bpp48_mask3=0x0404080808080808, - bpp48_mask2=0x1010101010102020, - bpp48_mask1=0x2020202040404040, - bpp48_mask0=0x4040808080808080; - -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_COMBINE_ROW) - /* && mmx_supported */ ) -#else - if (mmx_supported) -#endif - { - srcptr = png_ptr->row_buf + 1; - dstptr = row; - - unmask = ~mask; - len = (png_ptr->width)&~7; - diff = (png_ptr->width)&7; - _asm - { - movd mm7, unmask //load bit pattern - psubb mm6,mm6 //zero mm6 - punpcklbw mm7,mm7 - punpcklwd mm7,mm7 - punpckldq mm7,mm7 //fill register with 8 masks - - movq mm0,bpp48_mask0 - movq mm1,bpp48_mask1 - movq mm2,bpp48_mask2 - movq mm3,bpp48_mask3 - movq mm4,bpp48_mask4 - movq mm5,bpp48_mask5 - - pand mm0,mm7 - pand mm1,mm7 - pand mm2,mm7 - pand mm3,mm7 - pand mm4,mm7 - pand mm5,mm7 - - pcmpeqb mm0,mm6 - pcmpeqb mm1,mm6 - pcmpeqb mm2,mm6 - pcmpeqb mm3,mm6 - pcmpeqb mm4,mm6 - pcmpeqb mm5,mm6 - - mov ecx,len //load length of line - mov esi,srcptr //load source - mov ebx,dstptr //load dest - - cmp ecx,0 - jz mainloop48end - -mainloop48: - movq mm7,[esi] - pand mm7,mm0 - movq mm6,mm0 - pandn mm6,[ebx] - por mm7,mm6 - movq [ebx],mm7 - - movq mm6,[esi+8] - pand mm6,mm1 - movq mm7,mm1 - pandn mm7,[ebx+8] - por mm6,mm7 - movq [ebx+8],mm6 - - movq mm6,[esi+16] - pand mm6,mm2 - movq mm7,mm2 - pandn mm7,[ebx+16] - por mm6,mm7 - movq [ebx+16],mm6 - - movq mm7,[esi+24] - pand mm7,mm3 - movq mm6,mm3 - pandn mm6,[ebx+24] - por mm7,mm6 - movq [ebx+24],mm7 - - movq mm6,[esi+32] - pand mm6,mm4 - movq mm7,mm4 - pandn mm7,[ebx+32] - por mm6,mm7 - movq [ebx+32],mm6 - - movq mm7,[esi+40] - pand mm7,mm5 - movq mm6,mm5 - pandn mm6,[ebx+40] - por mm7,mm6 - movq [ebx+40],mm7 - - add esi,48 //inc by 32 bytes processed - add ebx,48 - sub ecx,8 //dec by 8 pixels processed - - ja mainloop48 -mainloop48end: - - mov ecx,diff - cmp ecx,0 - jz end48 - - mov edx,mask - sal edx,24 //make low byte the high byte - -secondloop48: - sal edx,1 //move high bit to CF - jnc skip48 //if CF = 0 - mov eax,[esi] - mov [ebx],eax - mov ax,[esi+4] // These 2 lines added 20070717 - mov [ebx+4],ax // Glenn R-P -skip48: - add esi,6 // Changed 4 to 6 on these 2 - add ebx,6 // lines. Glenn R-P 20070717 - - dec ecx - jnz secondloop48 - -end48: - emms - } - } - else /* mmx _not supported - Use modified C routine */ - { - register unsigned int incr1, initial_val, final_val; - png_size_t pixel_bytes; - png_uint_32 i; - register int disp = png_pass_inc[png_ptr->pass]; - - pixel_bytes = (png_ptr->row_info.pixel_depth >> 3); - srcptr = png_ptr->row_buf + 1 + offset_table[png_ptr->pass]* - pixel_bytes; - dstptr = row + offset_table[png_ptr->pass]*pixel_bytes; - initial_val = offset_table[png_ptr->pass]*pixel_bytes; - final_val = png_ptr->width*pixel_bytes; - incr1 = (disp)*pixel_bytes; - for (i = initial_val; i < final_val; i += incr1) - { - png_memcpy(dstptr, srcptr, pixel_bytes); - srcptr += incr1; - dstptr += incr1; - } - } /* end of else */ - - break; - } // end 48 bpp - - default: - { - png_bytep sptr; - png_bytep dp; - png_size_t pixel_bytes; - unsigned int i; - register int disp = png_pass_inc[png_ptr->pass]; // get the offset - register unsigned int incr1, initial_val, final_val; - - pixel_bytes = (png_ptr->row_info.pixel_depth >> 3); - sptr = png_ptr->row_buf + 1 + offset_table[png_ptr->pass]* - pixel_bytes; - dp = row + offset_table[png_ptr->pass]*pixel_bytes; - initial_val = offset_table[png_ptr->pass]*pixel_bytes; - final_val = png_ptr->width*pixel_bytes; - incr1 = (disp)*pixel_bytes; - for (i = initial_val; i < final_val; i += incr1) - { - png_memcpy(dp, sptr, pixel_bytes); - sptr += incr1; - dp += incr1; - } - break; - } - } /* end switch (png_ptr->row_info.pixel_depth) */ - } /* end if (non-trivial mask) */ - -} /* end png_combine_row() */ - - -#if defined(PNG_READ_INTERLACING_SUPPORTED) - -void /* PRIVATE */ -png_do_read_interlace(png_structp png_ptr) -{ - png_row_infop row_info = &(png_ptr->row_info); - png_bytep row = png_ptr->row_buf + 1; - int pass = png_ptr->pass; - png_uint_32 transformations = png_ptr->transformations; - - png_debug(1,"in png_do_read_interlace\n"); - - if (mmx_supported == 2) { -#if !defined(PNG_1_0_X) - /* this should have happened in png_init_mmx_flags() already */ - png_warning(png_ptr, "asm_flags may not have been initialized"); -#endif - png_mmx_support(); - } - - if (row != NULL && row_info != NULL) - { - png_uint_32 final_width; - - final_width = row_info->width * png_pass_inc[pass]; - - switch (row_info->pixel_depth) - { - case 1: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_byte v; - png_uint_32 i; - int j; - - sp = row + (png_size_t)((row_info->width - 1) >> 3); - dp = row + (png_size_t)((final_width - 1) >> 3); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (int)((row_info->width + 7) & 7); - dshift = (int)((final_width + 7) & 7); - s_start = 7; - s_end = 0; - s_inc = -1; - } - else -#endif - { - sshift = 7 - (int)((row_info->width + 7) & 7); - dshift = 7 - (int)((final_width + 7) & 7); - s_start = 0; - s_end = 7; - s_inc = 1; - } - - for (i = row_info->width; i; i--) - { - v = (png_byte)((*sp >> sshift) & 0x1); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0x7f7f >> (7 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - case 2: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_uint_32 i; - - sp = row + (png_size_t)((row_info->width - 1) >> 2); - dp = row + (png_size_t)((final_width - 1) >> 2); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (png_size_t)(((row_info->width + 3) & 3) << 1); - dshift = (png_size_t)(((final_width + 3) & 3) << 1); - s_start = 6; - s_end = 0; - s_inc = -2; - } - else -#endif - { - sshift = (png_size_t)((3 - ((row_info->width + 3) & 3)) << 1); - dshift = (png_size_t)((3 - ((final_width + 3) & 3)) << 1); - s_start = 0; - s_end = 6; - s_inc = 2; - } - - for (i = row_info->width; i; i--) - { - png_byte v; - int j; - - v = (png_byte)((*sp >> sshift) & 0x3); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0x3f3f >> (6 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - case 4: - { - png_bytep sp, dp; - int sshift, dshift; - int s_start, s_end, s_inc; - png_uint_32 i; - - sp = row + (png_size_t)((row_info->width - 1) >> 1); - dp = row + (png_size_t)((final_width - 1) >> 1); -#if defined(PNG_READ_PACKSWAP_SUPPORTED) - if (transformations & PNG_PACKSWAP) - { - sshift = (png_size_t)(((row_info->width + 1) & 1) << 2); - dshift = (png_size_t)(((final_width + 1) & 1) << 2); - s_start = 4; - s_end = 0; - s_inc = -4; - } - else -#endif - { - sshift = (png_size_t)((1 - ((row_info->width + 1) & 1)) << 2); - dshift = (png_size_t)((1 - ((final_width + 1) & 1)) << 2); - s_start = 0; - s_end = 4; - s_inc = 4; - } - - for (i = row_info->width; i; i--) - { - png_byte v; - int j; - - v = (png_byte)((*sp >> sshift) & 0xf); - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp &= (png_byte)((0xf0f >> (4 - dshift)) & 0xff); - *dp |= (png_byte)(v << dshift); - if (dshift == s_end) - { - dshift = s_start; - dp--; - } - else - dshift += s_inc; - } - if (sshift == s_end) - { - sshift = s_start; - sp--; - } - else - sshift += s_inc; - } - break; - } - - default: // This is the place where the routine is modified - { - static PNG_CONST __int64 const4 = 0x0000000000FFFFFF; - // static PNG_CONST __int64 const5 = 0x000000FFFFFF0000; // unused... - static PNG_CONST __int64 const6 = 0x00000000000000FF; - png_bytep sptr, dp; - png_uint_32 i; - png_size_t pixel_bytes; - int width = row_info->width; - - pixel_bytes = (row_info->pixel_depth >> 3); - - sptr = row + (width - 1) * pixel_bytes; - dp = row + (final_width - 1) * pixel_bytes; - // New code by Nirav Chhatrapati - Intel Corporation - // sign fix by GRR - // NOTE: there is NO MMX code for 48-bit and 64-bit images - - // use MMX routine if machine supports it -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_INTERLACE) - /* && mmx_supported */ ) -#else - if (mmx_supported) -#endif - { - if (pixel_bytes == 3) - { - if (((pass == 4) || (pass == 5)) && width) - { - int width_mmx = ((width >> 1) << 1) - 8; - if (width_mmx < 0) - width_mmx = 0; - width -= width_mmx; // 8 or 9 pix, 24 or 27 bytes - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub esi, 3 - sub edi, 9 -loop_pass4: - movq mm0, [esi] ; X X v2 v1 v0 v5 v4 v3 - movq mm7, mm0 ; X X v2 v1 v0 v5 v4 v3 - movq mm6, mm0 ; X X v2 v1 v0 v5 v4 v3 - psllq mm0, 24 ; v1 v0 v5 v4 v3 0 0 0 - pand mm7, const4 ; 0 0 0 0 0 v5 v4 v3 - psrlq mm6, 24 ; 0 0 0 X X v2 v1 v0 - por mm0, mm7 ; v1 v0 v5 v4 v3 v5 v4 v3 - movq mm5, mm6 ; 0 0 0 X X v2 v1 v0 - psllq mm6, 8 ; 0 0 X X v2 v1 v0 0 - movq [edi], mm0 ; move quad to memory - psrlq mm5, 16 ; 0 0 0 0 0 X X v2 - pand mm5, const6 ; 0 0 0 0 0 0 0 v2 - por mm6, mm5 ; 0 0 X X v2 v1 v0 v2 - movd [edi+8], mm6 ; move double to memory - sub esi, 6 - sub edi, 12 - sub ecx, 2 - jnz loop_pass4 - EMMS - } - } - - sptr -= width_mmx*3; - dp -= width_mmx*6; - for (i = width; i; i--) - { - png_byte v[8]; - int j; - - png_memcpy(v, sptr, 3); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, 3); - dp -= 3; - } - sptr -= 3; - } - } - else if (((pass == 2) || (pass == 3)) && width) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width - sub edi, 9 // (png_pass_inc[pass] - 1)*pixel_bytes -loop_pass2: - movd mm0, [esi] ; X X X X X v2 v1 v0 - pand mm0, const4 ; 0 0 0 0 0 v2 v1 v0 - movq mm1, mm0 ; 0 0 0 0 0 v2 v1 v0 - psllq mm0, 16 ; 0 0 0 v2 v1 v0 0 0 - movq mm2, mm0 ; 0 0 0 v2 v1 v0 0 0 - psllq mm0, 24 ; v2 v1 v0 0 0 0 0 0 - psrlq mm1, 8 ; 0 0 0 0 0 0 v2 v1 - por mm0, mm2 ; v2 v1 v0 v2 v1 v0 0 0 - por mm0, mm1 ; v2 v1 v0 v2 v1 v0 v2 v1 - movq [edi+4], mm0 ; move to memory - psrlq mm0, 16 ; 0 0 v2 v1 v0 v2 v1 v0 - movd [edi], mm0 ; move to memory - sub esi, 3 - sub edi, 12 - dec ecx - jnz loop_pass2 - EMMS - } - } - else if (width) /* && ((pass == 0) || (pass == 1))) */ - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width - sub edi, 21 // (png_pass_inc[pass] - 1)*pixel_bytes -loop_pass0: - movd mm0, [esi] ; X X X X X v2 v1 v0 - pand mm0, const4 ; 0 0 0 0 0 v2 v1 v0 - movq mm1, mm0 ; 0 0 0 0 0 v2 v1 v0 - psllq mm0, 16 ; 0 0 0 v2 v1 v0 0 0 - movq mm2, mm0 ; 0 0 0 v2 v1 v0 0 0 - psllq mm0, 24 ; v2 v1 v0 0 0 0 0 0 - psrlq mm1, 8 ; 0 0 0 0 0 0 v2 v1 - por mm0, mm2 ; v2 v1 v0 v2 v1 v0 0 0 - por mm0, mm1 ; v2 v1 v0 v2 v1 v0 v2 v1 - movq mm3, mm0 ; v2 v1 v0 v2 v1 v0 v2 v1 - psllq mm0, 16 ; v0 v2 v1 v0 v2 v1 0 0 - movq mm4, mm3 ; v2 v1 v0 v2 v1 v0 v2 v1 - punpckhdq mm3, mm0 ; v0 v2 v1 v0 v2 v1 v0 v2 - movq [edi+16] , mm4 - psrlq mm0, 32 ; 0 0 0 0 v0 v2 v1 v0 - movq [edi+8] , mm3 - punpckldq mm0, mm4 ; v1 v0 v2 v1 v0 v2 v1 v0 - sub esi, 3 - movq [edi], mm0 - sub edi, 24 - //sub esi, 3 - dec ecx - jnz loop_pass0 - EMMS - } - } - } /* end of pixel_bytes == 3 */ - - else if (pixel_bytes == 1) - { - if (((pass == 4) || (pass == 5)) && width) - { - int width_mmx = ((width >> 3) << 3); - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub edi, 15 - sub esi, 7 -loop1_pass4: - movq mm0, [esi] ; v0 v1 v2 v3 v4 v5 v6 v7 - movq mm1, mm0 ; v0 v1 v2 v3 v4 v5 v6 v7 - punpcklbw mm0, mm0 ; v4 v4 v5 v5 v6 v6 v7 v7 - //movq mm1, mm0 ; v0 v0 v1 v1 v2 v2 v3 v3 - punpckhbw mm1, mm1 ;v0 v0 v1 v1 v2 v2 v3 v3 - movq [edi+8], mm1 ; move to memory v0 v1 v2 and v3 - sub esi, 8 - movq [edi], mm0 ; move to memory v4 v5 v6 and v7 - //sub esi, 4 - sub edi, 16 - sub ecx, 8 - jnz loop1_pass4 - EMMS - } - } - - sptr -= width_mmx; - dp -= width_mmx*2; - for (i = width; i; i--) - { - int j; - - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp-- = *sptr; - } - sptr --; - } - } - else if (((pass == 2) || (pass == 3)) && width) - { - int width_mmx = ((width >> 2) << 2); - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub edi, 15 - sub esi, 3 -loop1_pass2: - movd mm0, [esi] ; X X X X v0 v1 v2 v3 - punpcklbw mm0, mm0 ; v0 v0 v1 v1 v2 v2 v3 v3 - movq mm1, mm0 ; v0 v0 v1 v1 v2 v2 v3 v3 - punpcklwd mm0, mm0 ; v2 v2 v2 v2 v3 v3 v3 v3 - punpckhwd mm1, mm1 ; v0 v0 v0 v0 v1 v1 v1 v1 - movq [edi], mm0 ; move to memory v2 and v3 - sub esi, 4 - movq [edi+8], mm1 ; move to memory v1 and v0 - sub edi, 16 - sub ecx, 4 - jnz loop1_pass2 - EMMS - } - } - - sptr -= width_mmx; - dp -= width_mmx*4; - for (i = width; i; i--) - { - int j; - - for (j = 0; j < png_pass_inc[pass]; j++) - { - *dp-- = *sptr; - } - sptr --; - } - } - else if (width) /* && ((pass == 0) || (pass == 1))) */ - { - int width_mmx = ((width >> 2) << 2); - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub edi, 31 - sub esi, 3 -loop1_pass0: - movd mm0, [esi] ; X X X X v0 v1 v2 v3 - movq mm1, mm0 ; X X X X v0 v1 v2 v3 - punpcklbw mm0, mm0 ; v0 v0 v1 v1 v2 v2 v3 v3 - movq mm2, mm0 ; v0 v0 v1 v1 v2 v2 v3 v3 - punpcklwd mm0, mm0 ; v2 v2 v2 v2 v3 v3 v3 v3 - movq mm3, mm0 ; v2 v2 v2 v2 v3 v3 v3 v3 - punpckldq mm0, mm0 ; v3 v3 v3 v3 v3 v3 v3 v3 - punpckhdq mm3, mm3 ; v2 v2 v2 v2 v2 v2 v2 v2 - movq [edi], mm0 ; move to memory v3 - punpckhwd mm2, mm2 ; v0 v0 v0 v0 v1 v1 v1 v1 - movq [edi+8], mm3 ; move to memory v2 - movq mm4, mm2 ; v0 v0 v0 v0 v1 v1 v1 v1 - punpckldq mm2, mm2 ; v1 v1 v1 v1 v1 v1 v1 v1 - punpckhdq mm4, mm4 ; v0 v0 v0 v0 v0 v0 v0 v0 - movq [edi+16], mm2 ; move to memory v1 - movq [edi+24], mm4 ; move to memory v0 - sub esi, 4 - sub edi, 32 - sub ecx, 4 - jnz loop1_pass0 - EMMS - } - } - - sptr -= width_mmx; - dp -= width_mmx*8; - for (i = width; i; i--) - { - int j; - - /* I simplified this part in version 1.0.4e - * here and in several other instances where - * pixel_bytes == 1 -- GR-P - * - * Original code: - * - * png_byte v[8]; - * png_memcpy(v, sptr, pixel_bytes); - * for (j = 0; j < png_pass_inc[pass]; j++) - * { - * png_memcpy(dp, v, pixel_bytes); - * dp -= pixel_bytes; - * } - * sptr -= pixel_bytes; - * - * Replacement code is in the next three lines: - */ - - for (j = 0; j < png_pass_inc[pass]; j++) - *dp-- = *sptr; - sptr--; - } - } - } /* end of pixel_bytes == 1 */ - - else if (pixel_bytes == 2) - { - if (((pass == 4) || (pass == 5)) && width) - { - int width_mmx = ((width >> 1) << 1) ; - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub esi, 2 - sub edi, 6 -loop2_pass4: - movd mm0, [esi] ; X X X X v1 v0 v3 v2 - punpcklwd mm0, mm0 ; v1 v0 v1 v0 v3 v2 v3 v2 - sub esi, 4 - movq [edi], mm0 - sub edi, 8 - sub ecx, 2 - jnz loop2_pass4 - EMMS - } - } - - sptr -= (width_mmx*2 - 2); // sign fixed - dp -= (width_mmx*4 - 2); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 2; - png_memcpy(v, sptr, 2); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 2; - png_memcpy(dp, v, 2); - } - } - } - else if (((pass == 2) || (pass == 3)) && width) - { - int width_mmx = ((width >> 1) << 1) ; - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub esi, 2 - sub edi, 14 -loop2_pass2: - movd mm0, [esi] ; X X X X v1 v0 v3 v2 - punpcklwd mm0, mm0 ; v1 v0 v1 v0 v3 v2 v3 v2 - movq mm1, mm0 ; v1 v0 v1 v0 v3 v2 v3 v2 - punpckldq mm0, mm0 ; v3 v2 v3 v2 v3 v2 v3 v2 - punpckhdq mm1, mm1 ; v1 v0 v1 v0 v1 v0 v1 v0 - movq [edi], mm0 - sub esi, 4 - movq [edi + 8], mm1 - //sub esi, 4 - sub edi, 16 - sub ecx, 2 - jnz loop2_pass2 - EMMS - } - } - - sptr -= (width_mmx*2 - 2); // sign fixed - dp -= (width_mmx*8 - 2); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 2; - png_memcpy(v, sptr, 2); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 2; - png_memcpy(dp, v, 2); - } - } - } - else if (width) /* && ((pass == 0) || (pass == 1))) */ - { - int width_mmx = ((width >> 1) << 1); - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub esi, 2 - sub edi, 30 -loop2_pass0: - movd mm0, [esi] ; X X X X v1 v0 v3 v2 - punpcklwd mm0, mm0 ; v1 v0 v1 v0 v3 v2 v3 v2 - movq mm1, mm0 ; v1 v0 v1 v0 v3 v2 v3 v2 - punpckldq mm0, mm0 ; v3 v2 v3 v2 v3 v2 v3 v2 - punpckhdq mm1, mm1 ; v1 v0 v1 v0 v1 v0 v1 v0 - movq [edi], mm0 - movq [edi + 8], mm0 - movq [edi + 16], mm1 - movq [edi + 24], mm1 - sub esi, 4 - sub edi, 32 - sub ecx, 2 - jnz loop2_pass0 - EMMS - } - } - - sptr -= (width_mmx*2 - 2); // sign fixed - dp -= (width_mmx*16 - 2); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 2; - png_memcpy(v, sptr, 2); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 2; - png_memcpy(dp, v, 2); - } - } - } - } /* end of pixel_bytes == 2 */ - - else if (pixel_bytes == 4) - { - if (((pass == 4) || (pass == 5)) && width) - { - int width_mmx = ((width >> 1) << 1) ; - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub esi, 4 - sub edi, 12 -loop4_pass4: - movq mm0, [esi] ; v3 v2 v1 v0 v7 v6 v5 v4 - movq mm1, mm0 ; v3 v2 v1 v0 v7 v6 v5 v4 - punpckldq mm0, mm0 ; v7 v6 v5 v4 v7 v6 v5 v4 - punpckhdq mm1, mm1 ; v3 v2 v1 v0 v3 v2 v1 v0 - movq [edi], mm0 - sub esi, 8 - movq [edi + 8], mm1 - sub edi, 16 - sub ecx, 2 - jnz loop4_pass4 - EMMS - } - } - - sptr -= (width_mmx*4 - 4); // sign fixed - dp -= (width_mmx*8 - 4); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 4; - png_memcpy(v, sptr, 4); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 4; - png_memcpy(dp, v, 4); - } - } - } - else if (((pass == 2) || (pass == 3)) && width) - { - int width_mmx = ((width >> 1) << 1) ; - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub esi, 4 - sub edi, 28 -loop4_pass2: - movq mm0, [esi] ; v3 v2 v1 v0 v7 v6 v5 v4 - movq mm1, mm0 ; v3 v2 v1 v0 v7 v6 v5 v4 - punpckldq mm0, mm0 ; v7 v6 v5 v4 v7 v6 v5 v4 - punpckhdq mm1, mm1 ; v3 v2 v1 v0 v3 v2 v1 v0 - movq [edi], mm0 - movq [edi + 8], mm0 - movq [edi+16], mm1 - movq [edi + 24], mm1 - sub esi, 8 - sub edi, 32 - sub ecx, 2 - jnz loop4_pass2 - EMMS - } - } - - sptr -= (width_mmx*4 - 4); // sign fixed - dp -= (width_mmx*16 - 4); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 4; - png_memcpy(v, sptr, 4); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 4; - png_memcpy(dp, v, 4); - } - } - } - else if (width) /* && ((pass == 0) || (pass == 1))) */ - { - int width_mmx = ((width >> 1) << 1) ; - width -= width_mmx; - if (width_mmx) - { - _asm - { - mov esi, sptr - mov edi, dp - mov ecx, width_mmx - sub esi, 4 - sub edi, 60 -loop4_pass0: - movq mm0, [esi] ; v3 v2 v1 v0 v7 v6 v5 v4 - movq mm1, mm0 ; v3 v2 v1 v0 v7 v6 v5 v4 - punpckldq mm0, mm0 ; v7 v6 v5 v4 v7 v6 v5 v4 - punpckhdq mm1, mm1 ; v3 v2 v1 v0 v3 v2 v1 v0 - movq [edi], mm0 - movq [edi + 8], mm0 - movq [edi + 16], mm0 - movq [edi + 24], mm0 - movq [edi+32], mm1 - movq [edi + 40], mm1 - movq [edi+ 48], mm1 - sub esi, 8 - movq [edi + 56], mm1 - sub edi, 64 - sub ecx, 2 - jnz loop4_pass0 - EMMS - } - } - - sptr -= (width_mmx*4 - 4); // sign fixed - dp -= (width_mmx*32 - 4); // sign fixed - for (i = width; i; i--) - { - png_byte v[8]; - int j; - sptr -= 4; - png_memcpy(v, sptr, 4); - for (j = 0; j < png_pass_inc[pass]; j++) - { - dp -= 4; - png_memcpy(dp, v, 4); - } - } - } - - } /* end of pixel_bytes == 4 */ - - else if (pixel_bytes == 6) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, 6); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, 6); - dp -= 6; - } - sptr -= 6; - } - } /* end of pixel_bytes == 6 */ - - else - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, pixel_bytes); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, pixel_bytes); - dp -= pixel_bytes; - } - sptr-= pixel_bytes; - } - } - } /* end of mmx_supported */ - - else /* MMX not supported: use modified C code - takes advantage - * of inlining of memcpy for a constant */ - { - if (pixel_bytes == 1) - { - for (i = width; i; i--) - { - int j; - for (j = 0; j < png_pass_inc[pass]; j++) - *dp-- = *sptr; - sptr--; - } - } - else if (pixel_bytes == 3) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, pixel_bytes); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, pixel_bytes); - dp -= pixel_bytes; - } - sptr -= pixel_bytes; - } - } - else if (pixel_bytes == 2) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, pixel_bytes); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, pixel_bytes); - dp -= pixel_bytes; - } - sptr -= pixel_bytes; - } - } - else if (pixel_bytes == 4) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, pixel_bytes); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, pixel_bytes); - dp -= pixel_bytes; - } - sptr -= pixel_bytes; - } - } - else if (pixel_bytes == 6) - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, pixel_bytes); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, pixel_bytes); - dp -= pixel_bytes; - } - sptr -= pixel_bytes; - } - } - else - { - for (i = width; i; i--) - { - png_byte v[8]; - int j; - png_memcpy(v, sptr, pixel_bytes); - for (j = 0; j < png_pass_inc[pass]; j++) - { - png_memcpy(dp, v, pixel_bytes); - dp -= pixel_bytes; - } - sptr -= pixel_bytes; - } - } - - } /* end of MMX not supported */ - break; - } - } /* end switch (row_info->pixel_depth) */ - - row_info->width = final_width; - - row_info->rowbytes = PNG_ROWBYTES(row_info->pixel_depth,final_width); - } - -} - -#endif /* PNG_READ_INTERLACING_SUPPORTED */ - - -// These global constants are declared -// here to ensure alignment on 8-byte boundaries. - union uAll { - __int64 use; - double double_align; - long long long_long_align; - } ; - static PNG_CONST union uAll LBCarryMask = {0x0101010101010101}, - HBClearMask = {0x7f7f7f7f7f7f7f7f}; - -// Optimized code for PNG Average filter decoder -void /* PRIVATE */ -png_read_filter_row_mmx_avg(png_row_infop row_info, png_bytep row - , png_bytep prev_row) -{ - // These variables are declared - // here to ensure alignment on 8-byte boundaries. - union uAll ActiveMask, ShiftBpp, ShiftRem; - - int bpp; - png_uint_32 FullLength; - png_uint_32 MMXLength; - //png_uint_32 len; - int diff; - - bpp = (row_info->pixel_depth + 7) >> 3; // Get # bytes per pixel - FullLength = row_info->rowbytes; // # of bytes to filter - _asm { - // Init address pointers and offset - mov edi, row // edi ==> Avg(x) - xor ebx, ebx // ebx ==> x - mov edx, edi - mov esi, prev_row // esi ==> Prior(x) - sub edx, bpp // edx ==> Raw(x-bpp) - - xor eax, eax - // Compute the Raw value for the first bpp bytes - // Raw(x) = Avg(x) + (Prior(x)/2) -davgrlp: - mov al, [esi + ebx] // Load al with Prior(x) - inc ebx - shr al, 1 // divide by 2 - add al, [edi+ebx-1] // Add Avg(x); -1 to offset inc ebx - cmp ebx, bpp - mov [edi+ebx-1], al // Write back Raw(x); - // mov does not affect flags; -1 to offset inc ebx - jb davgrlp - // get # of bytes to alignment - mov diff, edi // take start of row - add diff, ebx // add bpp - add diff, 0xf // add 7 + 8 to incr past alignment boundary - and diff, 0xfffffff8 // mask to alignment boundary - sub diff, edi // subtract from start ==> value ebx at alignment - jz davggo - // fix alignment - // Compute the Raw value for the bytes upto the alignment boundary - // Raw(x) = Avg(x) + ((Raw(x-bpp) + Prior(x))/2) - xor ecx, ecx -davglp1: - xor eax, eax - mov cl, [esi + ebx] // load cl with Prior(x) - mov al, [edx + ebx] // load al with Raw(x-bpp) - add ax, cx - inc ebx - shr ax, 1 // divide by 2 - add al, [edi+ebx-1] // Add Avg(x); -1 to offset inc ebx - cmp ebx, diff // Check if at alignment boundary - mov [edi+ebx-1], al // Write back Raw(x); - // mov does not affect flags; -1 to offset inc ebx - jb davglp1 // Repeat until at alignment boundary -davggo: - mov eax, FullLength - mov ecx, eax - sub eax, ebx // subtract alignment fix - and eax, 0x00000007 // calc bytes over mult of 8 - sub ecx, eax // drop over bytes from original length - mov MMXLength, ecx - } // end _asm block - // Now do the math for the rest of the row - switch ( bpp ) - { - case 3: - { - ActiveMask.use = 0x0000000000ffffff; - ShiftBpp.use = 24; // == 3 * 8 - ShiftRem.use = 40; // == 64 - 24 - _asm { - // Re-init address pointers and offset - movq mm7, ActiveMask - mov ebx, diff // ebx ==> x = offset to alignment boundary - movq mm5, LBCarryMask - mov edi, row // edi ==> Avg(x) - movq mm4, HBClearMask - mov esi, prev_row // esi ==> Prior(x) - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm2, [edi + ebx - 8] // Load previous aligned 8 bytes - // (we correct position in loop below) -davg3lp: - movq mm0, [edi + ebx] // Load mm0 with Avg(x) - // Add (Prev_row/2) to Average - movq mm3, mm5 - psrlq mm2, ShiftRem // Correct position Raw(x-bpp) data - movq mm1, [esi + ebx] // Load mm1 with Prior(x) - movq mm6, mm7 - pand mm3, mm1 // get lsb for each prev_row byte - psrlq mm1, 1 // divide prev_row bytes by 2 - pand mm1, mm4 // clear invalid bit 7 of each byte - paddb mm0, mm1 // add (Prev_row/2) to Avg for each byte - // Add 1st active group (Raw(x-bpp)/2) to Average with LBCarry - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm6 // Leave only Active Group 1 bytes to add to Avg - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active - // byte - // Add 2nd active group (Raw(x-bpp)/2) to Average with LBCarry - psllq mm6, ShiftBpp // shift the mm6 mask to cover bytes 3-5 - movq mm2, mm0 // mov updated Raws to mm2 - psllq mm2, ShiftBpp // shift data to position correctly - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm6 // Leave only Active Group 2 bytes to add to Avg - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active - // byte - - // Add 3rd active group (Raw(x-bpp)/2) to Average with LBCarry - psllq mm6, ShiftBpp // shift the mm6 mask to cover the last two - // bytes - movq mm2, mm0 // mov updated Raws to mm2 - psllq mm2, ShiftBpp // shift data to position correctly - // Data only needs to be shifted once here to - // get the correct x-bpp offset. - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm6 // Leave only Active Group 2 bytes to add to Avg - add ebx, 8 - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active - // byte - - // Now ready to write back to memory - movq [edi + ebx - 8], mm0 - // Move updated Raw(x) to use as Raw(x-bpp) for next loop - cmp ebx, MMXLength - movq mm2, mm0 // mov updated Raw(x) to mm2 - jb davg3lp - } // end _asm block - } - break; - - case 6: - case 4: - case 7: - case 5: - { - ActiveMask.use = 0xffffffffffffffff; // use shift below to clear - // appropriate inactive bytes - ShiftBpp.use = bpp << 3; - ShiftRem.use = 64 - ShiftBpp.use; - _asm { - movq mm4, HBClearMask - // Re-init address pointers and offset - mov ebx, diff // ebx ==> x = offset to alignment boundary - // Load ActiveMask and clear all bytes except for 1st active group - movq mm7, ActiveMask - mov edi, row // edi ==> Avg(x) - psrlq mm7, ShiftRem - mov esi, prev_row // esi ==> Prior(x) - movq mm6, mm7 - movq mm5, LBCarryMask - psllq mm6, ShiftBpp // Create mask for 2nd active group - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm2, [edi + ebx - 8] // Load previous aligned 8 bytes - // (we correct position in loop below) -davg4lp: - movq mm0, [edi + ebx] - psrlq mm2, ShiftRem // shift data to position correctly - movq mm1, [esi + ebx] - // Add (Prev_row/2) to Average - movq mm3, mm5 - pand mm3, mm1 // get lsb for each prev_row byte - psrlq mm1, 1 // divide prev_row bytes by 2 - pand mm1, mm4 // clear invalid bit 7 of each byte - paddb mm0, mm1 // add (Prev_row/2) to Avg for each byte - // Add 1st active group (Raw(x-bpp)/2) to Average with LBCarry - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm7 // Leave only Active Group 1 bytes to add to Avg - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active - // byte - // Add 2nd active group (Raw(x-bpp)/2) to Average with LBCarry - movq mm2, mm0 // mov updated Raws to mm2 - psllq mm2, ShiftBpp // shift data to position correctly - add ebx, 8 - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm6 // Leave only Active Group 2 bytes to add to Avg - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active - // byte - cmp ebx, MMXLength - // Now ready to write back to memory - movq [edi + ebx - 8], mm0 - // Prep Raw(x-bpp) for next loop - movq mm2, mm0 // mov updated Raws to mm2 - jb davg4lp - } // end _asm block - } - break; - case 2: - { - ActiveMask.use = 0x000000000000ffff; - ShiftBpp.use = 16; // == 2 * 8 [BUGFIX] - ShiftRem.use = 48; // == 64 - 16 [BUGFIX] - _asm { - // Load ActiveMask - movq mm7, ActiveMask - // Re-init address pointers and offset - mov ebx, diff // ebx ==> x = offset to alignment boundary - movq mm5, LBCarryMask - mov edi, row // edi ==> Avg(x) - movq mm4, HBClearMask - mov esi, prev_row // esi ==> Prior(x) - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm2, [edi + ebx - 8] // Load previous aligned 8 bytes - // (we correct position in loop below) -davg2lp: - movq mm0, [edi + ebx] - psrlq mm2, ShiftRem // shift data to position correctly [BUGFIX] - movq mm1, [esi + ebx] - // Add (Prev_row/2) to Average - movq mm3, mm5 - pand mm3, mm1 // get lsb for each prev_row byte - psrlq mm1, 1 // divide prev_row bytes by 2 - pand mm1, mm4 // clear invalid bit 7 of each byte - movq mm6, mm7 - paddb mm0, mm1 // add (Prev_row/2) to Avg for each byte - // Add 1st active group (Raw(x-bpp)/2) to Average with LBCarry - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm6 // Leave only Active Group 1 bytes to add to Avg - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active byte - // Add 2nd active group (Raw(x-bpp)/2) to Average with LBCarry - psllq mm6, ShiftBpp // shift the mm6 mask to cover bytes 2 & 3 - movq mm2, mm0 // mov updated Raws to mm2 - psllq mm2, ShiftBpp // shift data to position correctly - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm6 // Leave only Active Group 2 bytes to add to Avg - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active byte - - // Add rdd active group (Raw(x-bpp)/2) to Average with LBCarry - psllq mm6, ShiftBpp // shift the mm6 mask to cover bytes 4 & 5 - movq mm2, mm0 // mov updated Raws to mm2 - psllq mm2, ShiftBpp // shift data to position correctly - // Data only needs to be shifted once here to - // get the correct x-bpp offset. - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm6 // Leave only Active Group 2 bytes to add to Avg - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active byte - - // Add 4th active group (Raw(x-bpp)/2) to Average with LBCarry - psllq mm6, ShiftBpp // shift the mm6 mask to cover bytes 6 & 7 - movq mm2, mm0 // mov updated Raws to mm2 - psllq mm2, ShiftBpp // shift data to position correctly - // Data only needs to be shifted once here to - // get the correct x-bpp offset. - add ebx, 8 - movq mm1, mm3 // now use mm1 for getting LBCarrys - pand mm1, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 (Only valid for active group) - psrlq mm2, 1 // divide raw bytes by 2 - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm2, mm1 // add LBCarrys to (Raw(x-bpp)/2) for each byte - pand mm2, mm6 // Leave only Active Group 2 bytes to add to Avg - paddb mm0, mm2 // add (Raw/2) + LBCarrys to Avg for each Active byte - - cmp ebx, MMXLength - // Now ready to write back to memory - movq [edi + ebx - 8], mm0 - // Prep Raw(x-bpp) for next loop - movq mm2, mm0 // mov updated Raws to mm2 - jb davg2lp - } // end _asm block - } - break; - - case 1: // bpp == 1 - { - _asm { - // Re-init address pointers and offset - mov ebx, diff // ebx ==> x = offset to alignment boundary - mov edi, row // edi ==> Avg(x) - cmp ebx, FullLength // Test if offset at end of array - jnb davg1end - // Do Paeth decode for remaining bytes - mov esi, prev_row // esi ==> Prior(x) - mov edx, edi - xor ecx, ecx // zero ecx before using cl & cx in loop below - sub edx, bpp // edx ==> Raw(x-bpp) -davg1lp: - // Raw(x) = Avg(x) + ((Raw(x-bpp) + Prior(x))/2) - xor eax, eax - mov cl, [esi + ebx] // load cl with Prior(x) - mov al, [edx + ebx] // load al with Raw(x-bpp) - add ax, cx - inc ebx - shr ax, 1 // divide by 2 - add al, [edi+ebx-1] // Add Avg(x); -1 to offset inc ebx - cmp ebx, FullLength // Check if at end of array - mov [edi+ebx-1], al // Write back Raw(x); - // mov does not affect flags; -1 to offset inc ebx - jb davg1lp -davg1end: - } // end _asm block - } - return; - - case 8: // bpp == 8 - { - _asm { - // Re-init address pointers and offset - mov ebx, diff // ebx ==> x = offset to alignment boundary - movq mm5, LBCarryMask - mov edi, row // edi ==> Avg(x) - movq mm4, HBClearMask - mov esi, prev_row // esi ==> Prior(x) - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm2, [edi + ebx - 8] // Load previous aligned 8 bytes - // (NO NEED to correct position in loop below) -davg8lp: - movq mm0, [edi + ebx] - movq mm3, mm5 - movq mm1, [esi + ebx] - add ebx, 8 - pand mm3, mm1 // get lsb for each prev_row byte - psrlq mm1, 1 // divide prev_row bytes by 2 - pand mm3, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 - psrlq mm2, 1 // divide raw bytes by 2 - pand mm1, mm4 // clear invalid bit 7 of each byte - paddb mm0, mm3 // add LBCarrys to Avg for each byte - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm0, mm1 // add (Prev_row/2) to Avg for each byte - paddb mm0, mm2 // add (Raw/2) to Avg for each byte - cmp ebx, MMXLength - movq [edi + ebx - 8], mm0 - movq mm2, mm0 // reuse as Raw(x-bpp) - jb davg8lp - } // end _asm block - } - break; - default: // bpp greater than 8 - { - _asm { - movq mm5, LBCarryMask - // Re-init address pointers and offset - mov ebx, diff // ebx ==> x = offset to alignment boundary - mov edi, row // edi ==> Avg(x) - movq mm4, HBClearMask - mov edx, edi - mov esi, prev_row // esi ==> Prior(x) - sub edx, bpp // edx ==> Raw(x-bpp) -davgAlp: - movq mm0, [edi + ebx] - movq mm3, mm5 - movq mm1, [esi + ebx] - pand mm3, mm1 // get lsb for each prev_row byte - movq mm2, [edx + ebx] - psrlq mm1, 1 // divide prev_row bytes by 2 - pand mm3, mm2 // get LBCarrys for each byte where both - // lsb's were == 1 - psrlq mm2, 1 // divide raw bytes by 2 - pand mm1, mm4 // clear invalid bit 7 of each byte - paddb mm0, mm3 // add LBCarrys to Avg for each byte - pand mm2, mm4 // clear invalid bit 7 of each byte - paddb mm0, mm1 // add (Prev_row/2) to Avg for each byte - add ebx, 8 - paddb mm0, mm2 // add (Raw/2) to Avg for each byte - cmp ebx, MMXLength - movq [edi + ebx - 8], mm0 - jb davgAlp - } // end _asm block - } - break; - } // end switch ( bpp ) - - _asm { - // MMX acceleration complete now do clean-up - // Check if any remaining bytes left to decode - mov ebx, MMXLength // ebx ==> x = offset bytes remaining after MMX - mov edi, row // edi ==> Avg(x) - cmp ebx, FullLength // Test if offset at end of array - jnb davgend - // Do Paeth decode for remaining bytes - mov esi, prev_row // esi ==> Prior(x) - mov edx, edi - xor ecx, ecx // zero ecx before using cl & cx in loop below - sub edx, bpp // edx ==> Raw(x-bpp) -davglp2: - // Raw(x) = Avg(x) + ((Raw(x-bpp) + Prior(x))/2) - xor eax, eax - mov cl, [esi + ebx] // load cl with Prior(x) - mov al, [edx + ebx] // load al with Raw(x-bpp) - add ax, cx - inc ebx - shr ax, 1 // divide by 2 - add al, [edi+ebx-1] // Add Avg(x); -1 to offset inc ebx - cmp ebx, FullLength // Check if at end of array - mov [edi+ebx-1], al // Write back Raw(x); - // mov does not affect flags; -1 to offset inc ebx - jb davglp2 -davgend: - emms // End MMX instructions; prep for possible FP instrs. - } // end _asm block -} - -// Optimized code for PNG Paeth filter decoder -void /* PRIVATE */ -png_read_filter_row_mmx_paeth(png_row_infop row_info, png_bytep row, - png_bytep prev_row) -{ - // These variables are declared - // here to ensure alignment on 8-byte boundaries. - union uAll ActiveMask, ActiveMask2, ActiveMaskEnd, ShiftBpp, ShiftRem; - - png_uint_32 FullLength; - png_uint_32 MMXLength; - //png_uint_32 len; - int bpp; - int diff; - //int ptemp; - int patemp, pbtemp, pctemp; - - bpp = (row_info->pixel_depth + 7) >> 3; // Get # bytes per pixel - FullLength = row_info->rowbytes; // # of bytes to filter - _asm - { - xor ebx, ebx // ebx ==> x offset - mov edi, row - xor edx, edx // edx ==> x-bpp offset - mov esi, prev_row - xor eax, eax - - // Compute the Raw value for the first bpp bytes - // Note: the formula works out to be always - // Paeth(x) = Raw(x) + Prior(x) where x < bpp -dpthrlp: - mov al, [edi + ebx] - add al, [esi + ebx] - inc ebx - cmp ebx, bpp - mov [edi + ebx - 1], al - jb dpthrlp - // get # of bytes to alignment - mov diff, edi // take start of row - add diff, ebx // add bpp - xor ecx, ecx - add diff, 0xf // add 7 + 8 to incr past alignment boundary - and diff, 0xfffffff8 // mask to alignment boundary - sub diff, edi // subtract from start ==> value ebx at alignment - jz dpthgo - // fix alignment -dpthlp1: - xor eax, eax - // pav = p - a = (a + b - c) - a = b - c - mov al, [esi + ebx] // load Prior(x) into al - mov cl, [esi + edx] // load Prior(x-bpp) into cl - sub eax, ecx // subtract Prior(x-bpp) - mov patemp, eax // Save pav for later use - xor eax, eax - // pbv = p - b = (a + b - c) - b = a - c - mov al, [edi + edx] // load Raw(x-bpp) into al - sub eax, ecx // subtract Prior(x-bpp) - mov ecx, eax - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - add eax, patemp // pcv = pav + pbv - // pc = abs(pcv) - test eax, 0x80000000 - jz dpthpca - neg eax // reverse sign of neg values -dpthpca: - mov pctemp, eax // save pc for later use - // pb = abs(pbv) - test ecx, 0x80000000 - jz dpthpba - neg ecx // reverse sign of neg values -dpthpba: - mov pbtemp, ecx // save pb for later use - // pa = abs(pav) - mov eax, patemp - test eax, 0x80000000 - jz dpthpaa - neg eax // reverse sign of neg values -dpthpaa: - mov patemp, eax // save pa for later use - // test if pa <= pb - cmp eax, ecx - jna dpthabb - // pa > pb; now test if pb <= pc - cmp ecx, pctemp - jna dpthbbc - // pb > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - mov cl, [esi + edx] // load Prior(x-bpp) into cl - jmp dpthpaeth -dpthbbc: - // pb <= pc; Raw(x) = Paeth(x) + Prior(x) - mov cl, [esi + ebx] // load Prior(x) into cl - jmp dpthpaeth -dpthabb: - // pa <= pb; now test if pa <= pc - cmp eax, pctemp - jna dpthabc - // pa > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - mov cl, [esi + edx] // load Prior(x-bpp) into cl - jmp dpthpaeth -dpthabc: - // pa <= pc; Raw(x) = Paeth(x) + Raw(x-bpp) - mov cl, [edi + edx] // load Raw(x-bpp) into cl -dpthpaeth: - inc ebx - inc edx - // Raw(x) = (Paeth(x) + Paeth_Predictor( a, b, c )) mod 256 - add [edi + ebx - 1], cl - cmp ebx, diff - jb dpthlp1 -dpthgo: - mov ecx, FullLength - mov eax, ecx - sub eax, ebx // subtract alignment fix - and eax, 0x00000007 // calc bytes over mult of 8 - sub ecx, eax // drop over bytes from original length - mov MMXLength, ecx - } // end _asm block - // Now do the math for the rest of the row - switch ( bpp ) - { - case 3: - { - ActiveMask.use = 0x0000000000ffffff; - ActiveMaskEnd.use = 0xffff000000000000; - ShiftBpp.use = 24; // == bpp(3) * 8 - ShiftRem.use = 40; // == 64 - 24 - _asm - { - mov ebx, diff - mov edi, row - mov esi, prev_row - pxor mm0, mm0 - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm1, [edi+ebx-8] -dpth3lp: - psrlq mm1, ShiftRem // shift last 3 bytes to 1st 3 bytes - movq mm2, [esi + ebx] // load b=Prior(x) - punpcklbw mm1, mm0 // Unpack High bytes of a - movq mm3, [esi+ebx-8] // Prep c=Prior(x-bpp) bytes - punpcklbw mm2, mm0 // Unpack High bytes of b - psrlq mm3, ShiftRem // shift last 3 bytes to 1st 3 bytes - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - punpcklbw mm3, mm0 // Unpack High bytes of c - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - psubw mm4, mm3 - pxor mm7, mm7 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - movq mm6, mm4 - psubw mm5, mm3 - - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm4 // Create mask pav bytes < 0 - paddw mm6, mm5 - pand mm0, mm4 // Only pav bytes < 0 in mm7 - pcmpgtw mm7, mm5 // Create mask pbv bytes < 0 - psubw mm4, mm0 - pand mm7, mm5 // Only pbv bytes < 0 in mm0 - psubw mm4, mm0 - psubw mm5, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm5, mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - pandn mm7, mm4 - pandn mm0, mm1 - paddw mm7, mm5 - paddw mm0, mm2 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - pxor mm1, mm1 - pand mm3, mm7 - pandn mm7, mm0 - paddw mm7, mm3 - pxor mm0, mm0 - packuswb mm7, mm1 - movq mm3, [esi + ebx] // load c=Prior(x-bpp) - pand mm7, ActiveMask - movq mm2, mm3 // load b=Prior(x) step 1 - paddb mm7, [edi + ebx] // add Paeth predictor with Raw(x) - punpcklbw mm3, mm0 // Unpack High bytes of c - movq [edi + ebx], mm7 // write back updated value - movq mm1, mm7 // Now mm1 will be used as Raw(x-bpp) - // Now do Paeth for 2nd set of bytes (3-5) - psrlq mm2, ShiftBpp // load b=Prior(x) step 2 - punpcklbw mm1, mm0 // Unpack High bytes of a - pxor mm7, mm7 - punpcklbw mm2, mm0 // Unpack High bytes of b - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - psubw mm5, mm3 - psubw mm4, mm3 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = - // pav + pbv = pbv + pav - movq mm6, mm5 - paddw mm6, mm4 - - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm5 // Create mask pbv bytes < 0 - pcmpgtw mm7, mm4 // Create mask pav bytes < 0 - pand mm0, mm5 // Only pbv bytes < 0 in mm0 - pand mm7, mm4 // Only pav bytes < 0 in mm7 - psubw mm5, mm0 - psubw mm4, mm7 - psubw mm5, mm0 - psubw mm4, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - pandn mm7, mm4 - pandn mm0, mm1 - paddw mm7, mm5 - paddw mm0, mm2 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - movq mm2, [esi + ebx] // load b=Prior(x) - pand mm3, mm7 - pandn mm7, mm0 - pxor mm1, mm1 - paddw mm7, mm3 - pxor mm0, mm0 - packuswb mm7, mm1 - movq mm3, mm2 // load c=Prior(x-bpp) step 1 - pand mm7, ActiveMask - punpckhbw mm2, mm0 // Unpack High bytes of b - psllq mm7, ShiftBpp // Shift bytes to 2nd group of 3 bytes - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - paddb mm7, [edi + ebx] // add Paeth predictor with Raw(x) - psllq mm3, ShiftBpp // load c=Prior(x-bpp) step 2 - movq [edi + ebx], mm7 // write back updated value - movq mm1, mm7 - punpckhbw mm3, mm0 // Unpack High bytes of c - psllq mm1, ShiftBpp // Shift bytes - // Now mm1 will be used as Raw(x-bpp) - // Now do Paeth for 3rd, and final, set of bytes (6-7) - pxor mm7, mm7 - punpckhbw mm1, mm0 // Unpack High bytes of a - psubw mm4, mm3 - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - movq mm6, mm4 - psubw mm5, mm3 - pxor mm0, mm0 - paddw mm6, mm5 - - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm4 // Create mask pav bytes < 0 - pcmpgtw mm7, mm5 // Create mask pbv bytes < 0 - pand mm0, mm4 // Only pav bytes < 0 in mm7 - pand mm7, mm5 // Only pbv bytes < 0 in mm0 - psubw mm4, mm0 - psubw mm5, mm7 - psubw mm4, mm0 - psubw mm5, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - pandn mm0, mm1 - pandn mm7, mm4 - paddw mm0, mm2 - paddw mm7, mm5 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - pand mm3, mm7 - pandn mm7, mm0 - paddw mm7, mm3 - pxor mm1, mm1 - packuswb mm1, mm7 - // Step ebx to next set of 8 bytes and repeat loop til done - add ebx, 8 - pand mm1, ActiveMaskEnd - paddb mm1, [edi + ebx - 8] // add Paeth predictor with Raw(x) - - cmp ebx, MMXLength - pxor mm0, mm0 // pxor does not affect flags - movq [edi + ebx - 8], mm1 // write back updated value - // mm1 will be used as Raw(x-bpp) next loop - // mm3 ready to be used as Prior(x-bpp) next loop - jb dpth3lp - } // end _asm block - } - break; - - case 6: - case 7: - case 5: - { - ActiveMask.use = 0x00000000ffffffff; - ActiveMask2.use = 0xffffffff00000000; - ShiftBpp.use = bpp << 3; // == bpp * 8 - ShiftRem.use = 64 - ShiftBpp.use; - _asm - { - mov ebx, diff - mov edi, row - mov esi, prev_row - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm1, [edi+ebx-8] - pxor mm0, mm0 -dpth6lp: - // Must shift to position Raw(x-bpp) data - psrlq mm1, ShiftRem - // Do first set of 4 bytes - movq mm3, [esi+ebx-8] // read c=Prior(x-bpp) bytes - punpcklbw mm1, mm0 // Unpack Low bytes of a - movq mm2, [esi + ebx] // load b=Prior(x) - punpcklbw mm2, mm0 // Unpack Low bytes of b - // Must shift to position Prior(x-bpp) data - psrlq mm3, ShiftRem - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - punpcklbw mm3, mm0 // Unpack Low bytes of c - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - psubw mm4, mm3 - pxor mm7, mm7 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - movq mm6, mm4 - psubw mm5, mm3 - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm4 // Create mask pav bytes < 0 - paddw mm6, mm5 - pand mm0, mm4 // Only pav bytes < 0 in mm7 - pcmpgtw mm7, mm5 // Create mask pbv bytes < 0 - psubw mm4, mm0 - pand mm7, mm5 // Only pbv bytes < 0 in mm0 - psubw mm4, mm0 - psubw mm5, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm5, mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - pandn mm7, mm4 - pandn mm0, mm1 - paddw mm7, mm5 - paddw mm0, mm2 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - pxor mm1, mm1 - pand mm3, mm7 - pandn mm7, mm0 - paddw mm7, mm3 - pxor mm0, mm0 - packuswb mm7, mm1 - movq mm3, [esi + ebx - 8] // load c=Prior(x-bpp) - pand mm7, ActiveMask - psrlq mm3, ShiftRem - movq mm2, [esi + ebx] // load b=Prior(x) step 1 - paddb mm7, [edi + ebx] // add Paeth predictor with Raw(x) - movq mm6, mm2 - movq [edi + ebx], mm7 // write back updated value - movq mm1, [edi+ebx-8] - psllq mm6, ShiftBpp - movq mm5, mm7 - psrlq mm1, ShiftRem - por mm3, mm6 - psllq mm5, ShiftBpp - punpckhbw mm3, mm0 // Unpack High bytes of c - por mm1, mm5 - // Do second set of 4 bytes - punpckhbw mm2, mm0 // Unpack High bytes of b - punpckhbw mm1, mm0 // Unpack High bytes of a - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - psubw mm4, mm3 - pxor mm7, mm7 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - movq mm6, mm4 - psubw mm5, mm3 - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm4 // Create mask pav bytes < 0 - paddw mm6, mm5 - pand mm0, mm4 // Only pav bytes < 0 in mm7 - pcmpgtw mm7, mm5 // Create mask pbv bytes < 0 - psubw mm4, mm0 - pand mm7, mm5 // Only pbv bytes < 0 in mm0 - psubw mm4, mm0 - psubw mm5, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm5, mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - pandn mm7, mm4 - pandn mm0, mm1 - paddw mm7, mm5 - paddw mm0, mm2 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - pxor mm1, mm1 - pand mm3, mm7 - pandn mm7, mm0 - pxor mm1, mm1 - paddw mm7, mm3 - pxor mm0, mm0 - // Step ex to next set of 8 bytes and repeat loop til done - add ebx, 8 - packuswb mm1, mm7 - paddb mm1, [edi + ebx - 8] // add Paeth predictor with Raw(x) - cmp ebx, MMXLength - movq [edi + ebx - 8], mm1 // write back updated value - // mm1 will be used as Raw(x-bpp) next loop - jb dpth6lp - } // end _asm block - } - break; - - case 4: - { - ActiveMask.use = 0x00000000ffffffff; - _asm { - mov ebx, diff - mov edi, row - mov esi, prev_row - pxor mm0, mm0 - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm1, [edi+ebx-8] // Only time should need to read - // a=Raw(x-bpp) bytes -dpth4lp: - // Do first set of 4 bytes - movq mm3, [esi+ebx-8] // read c=Prior(x-bpp) bytes - punpckhbw mm1, mm0 // Unpack Low bytes of a - movq mm2, [esi + ebx] // load b=Prior(x) - punpcklbw mm2, mm0 // Unpack High bytes of b - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - punpckhbw mm3, mm0 // Unpack High bytes of c - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - psubw mm4, mm3 - pxor mm7, mm7 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - movq mm6, mm4 - psubw mm5, mm3 - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm4 // Create mask pav bytes < 0 - paddw mm6, mm5 - pand mm0, mm4 // Only pav bytes < 0 in mm7 - pcmpgtw mm7, mm5 // Create mask pbv bytes < 0 - psubw mm4, mm0 - pand mm7, mm5 // Only pbv bytes < 0 in mm0 - psubw mm4, mm0 - psubw mm5, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm5, mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - pandn mm7, mm4 - pandn mm0, mm1 - paddw mm7, mm5 - paddw mm0, mm2 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - pxor mm1, mm1 - pand mm3, mm7 - pandn mm7, mm0 - paddw mm7, mm3 - pxor mm0, mm0 - packuswb mm7, mm1 - movq mm3, [esi + ebx] // load c=Prior(x-bpp) - pand mm7, ActiveMask - movq mm2, mm3 // load b=Prior(x) step 1 - paddb mm7, [edi + ebx] // add Paeth predictor with Raw(x) - punpcklbw mm3, mm0 // Unpack High bytes of c - movq [edi + ebx], mm7 // write back updated value - movq mm1, mm7 // Now mm1 will be used as Raw(x-bpp) - // Do second set of 4 bytes - punpckhbw mm2, mm0 // Unpack Low bytes of b - punpcklbw mm1, mm0 // Unpack Low bytes of a - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - psubw mm4, mm3 - pxor mm7, mm7 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - movq mm6, mm4 - psubw mm5, mm3 - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm4 // Create mask pav bytes < 0 - paddw mm6, mm5 - pand mm0, mm4 // Only pav bytes < 0 in mm7 - pcmpgtw mm7, mm5 // Create mask pbv bytes < 0 - psubw mm4, mm0 - pand mm7, mm5 // Only pbv bytes < 0 in mm0 - psubw mm4, mm0 - psubw mm5, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm5, mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - pandn mm7, mm4 - pandn mm0, mm1 - paddw mm7, mm5 - paddw mm0, mm2 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - pxor mm1, mm1 - pand mm3, mm7 - pandn mm7, mm0 - pxor mm1, mm1 - paddw mm7, mm3 - pxor mm0, mm0 - // Step ex to next set of 8 bytes and repeat loop til done - add ebx, 8 - packuswb mm1, mm7 - paddb mm1, [edi + ebx - 8] // add Paeth predictor with Raw(x) - cmp ebx, MMXLength - movq [edi + ebx - 8], mm1 // write back updated value - // mm1 will be used as Raw(x-bpp) next loop - jb dpth4lp - } // end _asm block - } - break; - case 8: // bpp == 8 - { - ActiveMask.use = 0x00000000ffffffff; - _asm { - mov ebx, diff - mov edi, row - mov esi, prev_row - pxor mm0, mm0 - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm1, [edi+ebx-8] // Only time should need to read - // a=Raw(x-bpp) bytes -dpth8lp: - // Do first set of 4 bytes - movq mm3, [esi+ebx-8] // read c=Prior(x-bpp) bytes - punpcklbw mm1, mm0 // Unpack Low bytes of a - movq mm2, [esi + ebx] // load b=Prior(x) - punpcklbw mm2, mm0 // Unpack Low bytes of b - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - punpcklbw mm3, mm0 // Unpack Low bytes of c - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - psubw mm4, mm3 - pxor mm7, mm7 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - movq mm6, mm4 - psubw mm5, mm3 - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm4 // Create mask pav bytes < 0 - paddw mm6, mm5 - pand mm0, mm4 // Only pav bytes < 0 in mm7 - pcmpgtw mm7, mm5 // Create mask pbv bytes < 0 - psubw mm4, mm0 - pand mm7, mm5 // Only pbv bytes < 0 in mm0 - psubw mm4, mm0 - psubw mm5, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm5, mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - pandn mm7, mm4 - pandn mm0, mm1 - paddw mm7, mm5 - paddw mm0, mm2 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - pxor mm1, mm1 - pand mm3, mm7 - pandn mm7, mm0 - paddw mm7, mm3 - pxor mm0, mm0 - packuswb mm7, mm1 - movq mm3, [esi+ebx-8] // read c=Prior(x-bpp) bytes - pand mm7, ActiveMask - movq mm2, [esi + ebx] // load b=Prior(x) - paddb mm7, [edi + ebx] // add Paeth predictor with Raw(x) - punpckhbw mm3, mm0 // Unpack High bytes of c - movq [edi + ebx], mm7 // write back updated value - movq mm1, [edi+ebx-8] // read a=Raw(x-bpp) bytes - - // Do second set of 4 bytes - punpckhbw mm2, mm0 // Unpack High bytes of b - punpckhbw mm1, mm0 // Unpack High bytes of a - // pav = p - a = (a + b - c) - a = b - c - movq mm4, mm2 - // pbv = p - b = (a + b - c) - b = a - c - movq mm5, mm1 - psubw mm4, mm3 - pxor mm7, mm7 - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - movq mm6, mm4 - psubw mm5, mm3 - // pa = abs(p-a) = abs(pav) - // pb = abs(p-b) = abs(pbv) - // pc = abs(p-c) = abs(pcv) - pcmpgtw mm0, mm4 // Create mask pav bytes < 0 - paddw mm6, mm5 - pand mm0, mm4 // Only pav bytes < 0 in mm7 - pcmpgtw mm7, mm5 // Create mask pbv bytes < 0 - psubw mm4, mm0 - pand mm7, mm5 // Only pbv bytes < 0 in mm0 - psubw mm4, mm0 - psubw mm5, mm7 - pxor mm0, mm0 - pcmpgtw mm0, mm6 // Create mask pcv bytes < 0 - pand mm0, mm6 // Only pav bytes < 0 in mm7 - psubw mm5, mm7 - psubw mm6, mm0 - // test pa <= pb - movq mm7, mm4 - psubw mm6, mm0 - pcmpgtw mm7, mm5 // pa > pb? - movq mm0, mm7 - // use mm7 mask to merge pa & pb - pand mm5, mm7 - // use mm0 mask copy to merge a & b - pand mm2, mm0 - pandn mm7, mm4 - pandn mm0, mm1 - paddw mm7, mm5 - paddw mm0, mm2 - // test ((pa <= pb)? pa:pb) <= pc - pcmpgtw mm7, mm6 // pab > pc? - pxor mm1, mm1 - pand mm3, mm7 - pandn mm7, mm0 - pxor mm1, mm1 - paddw mm7, mm3 - pxor mm0, mm0 - // Step ex to next set of 8 bytes and repeat loop til done - add ebx, 8 - packuswb mm1, mm7 - paddb mm1, [edi + ebx - 8] // add Paeth predictor with Raw(x) - cmp ebx, MMXLength - movq [edi + ebx - 8], mm1 // write back updated value - // mm1 will be used as Raw(x-bpp) next loop - jb dpth8lp - } // end _asm block - } - break; - - case 1: // bpp = 1 - case 2: // bpp = 2 - default: // bpp > 8 - { - _asm { - mov ebx, diff - cmp ebx, FullLength - jnb dpthdend - mov edi, row - mov esi, prev_row - // Do Paeth decode for remaining bytes - mov edx, ebx - xor ecx, ecx // zero ecx before using cl & cx in loop below - sub edx, bpp // Set edx = ebx - bpp -dpthdlp: - xor eax, eax - // pav = p - a = (a + b - c) - a = b - c - mov al, [esi + ebx] // load Prior(x) into al - mov cl, [esi + edx] // load Prior(x-bpp) into cl - sub eax, ecx // subtract Prior(x-bpp) - mov patemp, eax // Save pav for later use - xor eax, eax - // pbv = p - b = (a + b - c) - b = a - c - mov al, [edi + edx] // load Raw(x-bpp) into al - sub eax, ecx // subtract Prior(x-bpp) - mov ecx, eax - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - add eax, patemp // pcv = pav + pbv - // pc = abs(pcv) - test eax, 0x80000000 - jz dpthdpca - neg eax // reverse sign of neg values -dpthdpca: - mov pctemp, eax // save pc for later use - // pb = abs(pbv) - test ecx, 0x80000000 - jz dpthdpba - neg ecx // reverse sign of neg values -dpthdpba: - mov pbtemp, ecx // save pb for later use - // pa = abs(pav) - mov eax, patemp - test eax, 0x80000000 - jz dpthdpaa - neg eax // reverse sign of neg values -dpthdpaa: - mov patemp, eax // save pa for later use - // test if pa <= pb - cmp eax, ecx - jna dpthdabb - // pa > pb; now test if pb <= pc - cmp ecx, pctemp - jna dpthdbbc - // pb > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - mov cl, [esi + edx] // load Prior(x-bpp) into cl - jmp dpthdpaeth -dpthdbbc: - // pb <= pc; Raw(x) = Paeth(x) + Prior(x) - mov cl, [esi + ebx] // load Prior(x) into cl - jmp dpthdpaeth -dpthdabb: - // pa <= pb; now test if pa <= pc - cmp eax, pctemp - jna dpthdabc - // pa > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - mov cl, [esi + edx] // load Prior(x-bpp) into cl - jmp dpthdpaeth -dpthdabc: - // pa <= pc; Raw(x) = Paeth(x) + Raw(x-bpp) - mov cl, [edi + edx] // load Raw(x-bpp) into cl -dpthdpaeth: - inc ebx - inc edx - // Raw(x) = (Paeth(x) + Paeth_Predictor( a, b, c )) mod 256 - add [edi + ebx - 1], cl - cmp ebx, FullLength - jb dpthdlp -dpthdend: - } // end _asm block - } - return; // No need to go further with this one - } // end switch ( bpp ) - _asm - { - // MMX acceleration complete now do clean-up - // Check if any remaining bytes left to decode - mov ebx, MMXLength - cmp ebx, FullLength - jnb dpthend - mov edi, row - mov esi, prev_row - // Do Paeth decode for remaining bytes - mov edx, ebx - xor ecx, ecx // zero ecx before using cl & cx in loop below - sub edx, bpp // Set edx = ebx - bpp -dpthlp2: - xor eax, eax - // pav = p - a = (a + b - c) - a = b - c - mov al, [esi + ebx] // load Prior(x) into al - mov cl, [esi + edx] // load Prior(x-bpp) into cl - sub eax, ecx // subtract Prior(x-bpp) - mov patemp, eax // Save pav for later use - xor eax, eax - // pbv = p - b = (a + b - c) - b = a - c - mov al, [edi + edx] // load Raw(x-bpp) into al - sub eax, ecx // subtract Prior(x-bpp) - mov ecx, eax - // pcv = p - c = (a + b - c) -c = (a - c) + (b - c) = pav + pbv - add eax, patemp // pcv = pav + pbv - // pc = abs(pcv) - test eax, 0x80000000 - jz dpthpca2 - neg eax // reverse sign of neg values -dpthpca2: - mov pctemp, eax // save pc for later use - // pb = abs(pbv) - test ecx, 0x80000000 - jz dpthpba2 - neg ecx // reverse sign of neg values -dpthpba2: - mov pbtemp, ecx // save pb for later use - // pa = abs(pav) - mov eax, patemp - test eax, 0x80000000 - jz dpthpaa2 - neg eax // reverse sign of neg values -dpthpaa2: - mov patemp, eax // save pa for later use - // test if pa <= pb - cmp eax, ecx - jna dpthabb2 - // pa > pb; now test if pb <= pc - cmp ecx, pctemp - jna dpthbbc2 - // pb > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - mov cl, [esi + edx] // load Prior(x-bpp) into cl - jmp dpthpaeth2 -dpthbbc2: - // pb <= pc; Raw(x) = Paeth(x) + Prior(x) - mov cl, [esi + ebx] // load Prior(x) into cl - jmp dpthpaeth2 -dpthabb2: - // pa <= pb; now test if pa <= pc - cmp eax, pctemp - jna dpthabc2 - // pa > pc; Raw(x) = Paeth(x) + Prior(x-bpp) - mov cl, [esi + edx] // load Prior(x-bpp) into cl - jmp dpthpaeth2 -dpthabc2: - // pa <= pc; Raw(x) = Paeth(x) + Raw(x-bpp) - mov cl, [edi + edx] // load Raw(x-bpp) into cl -dpthpaeth2: - inc ebx - inc edx - // Raw(x) = (Paeth(x) + Paeth_Predictor( a, b, c )) mod 256 - add [edi + ebx - 1], cl - cmp ebx, FullLength - jb dpthlp2 -dpthend: - emms // End MMX instructions; prep for possible FP instrs. - } // end _asm block -} - -// Optimized code for PNG Sub filter decoder -void /* PRIVATE */ -png_read_filter_row_mmx_sub(png_row_infop row_info, png_bytep row) -{ - // These variables are declared - // here to ensure alignment on 8-byte boundaries. - union uAll ActiveMask, ShiftBpp, ShiftRem; - - //int test; - int bpp; - png_uint_32 FullLength; - png_uint_32 MMXLength; - int diff; - - bpp = (row_info->pixel_depth + 7) >> 3; // Get # bytes per pixel - FullLength = row_info->rowbytes - bpp; // # of bytes to filter - _asm { - mov edi, row - mov esi, edi // lp = row - add edi, bpp // rp = row + bpp - xor eax, eax - // get # of bytes to alignment - mov diff, edi // take start of row - add diff, 0xf // add 7 + 8 to incr past - // alignment boundary - xor ebx, ebx - and diff, 0xfffffff8 // mask to alignment boundary - sub diff, edi // subtract from start ==> value - // ebx at alignment - jz dsubgo - // fix alignment -dsublp1: - mov al, [esi+ebx] - add [edi+ebx], al - inc ebx - cmp ebx, diff - jb dsublp1 -dsubgo: - mov ecx, FullLength - mov edx, ecx - sub edx, ebx // subtract alignment fix - and edx, 0x00000007 // calc bytes over mult of 8 - sub ecx, edx // drop over bytes from length - mov MMXLength, ecx - } // end _asm block - - // Now do the math for the rest of the row - switch ( bpp ) - { - case 3: - { - ActiveMask.use = 0x0000ffffff000000; - ShiftBpp.use = 24; // == 3 * 8 - ShiftRem.use = 40; // == 64 - 24 - _asm { - mov edi, row - movq mm7, ActiveMask // Load ActiveMask for 2nd active byte group - mov esi, edi // lp = row - add edi, bpp // rp = row + bpp - movq mm6, mm7 - mov ebx, diff - psllq mm6, ShiftBpp // Move mask in mm6 to cover 3rd active - // byte group - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm1, [edi+ebx-8] -dsub3lp: - psrlq mm1, ShiftRem // Shift data for adding 1st bpp bytes - // no need for mask; shift clears inactive bytes - // Add 1st active group - movq mm0, [edi+ebx] - paddb mm0, mm1 - // Add 2nd active group - movq mm1, mm0 // mov updated Raws to mm1 - psllq mm1, ShiftBpp // shift data to position correctly - pand mm1, mm7 // mask to use only 2nd active group - paddb mm0, mm1 - // Add 3rd active group - movq mm1, mm0 // mov updated Raws to mm1 - psllq mm1, ShiftBpp // shift data to position correctly - pand mm1, mm6 // mask to use only 3rd active group - add ebx, 8 - paddb mm0, mm1 - cmp ebx, MMXLength - movq [edi+ebx-8], mm0 // Write updated Raws back to array - // Prep for doing 1st add at top of loop - movq mm1, mm0 - jb dsub3lp - } // end _asm block - } - break; - - case 1: - { - // Placed here just in case this is a duplicate of the - // non-MMX code for the SUB filter in png_read_filter_row below - // - // png_bytep rp; - // png_bytep lp; - // png_uint_32 i; - // bpp = (row_info->pixel_depth + 7) >> 3; - // for (i = (png_uint_32)bpp, rp = row + bpp, lp = row; - // i < row_info->rowbytes; i++, rp++, lp++) - // { - // *rp = (png_byte)(((int)(*rp) + (int)(*lp)) & 0xff); - // } - _asm { - mov ebx, diff - mov edi, row - cmp ebx, FullLength - jnb dsub1end - mov esi, edi // lp = row - xor eax, eax - add edi, bpp // rp = row + bpp -dsub1lp: - mov al, [esi+ebx] - add [edi+ebx], al - inc ebx - cmp ebx, FullLength - jb dsub1lp -dsub1end: - } // end _asm block - } - return; - - case 6: - case 7: - case 4: - case 5: - { - ShiftBpp.use = bpp << 3; - ShiftRem.use = 64 - ShiftBpp.use; - _asm { - mov edi, row - mov ebx, diff - mov esi, edi // lp = row - add edi, bpp // rp = row + bpp - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm1, [edi+ebx-8] -dsub4lp: - psrlq mm1, ShiftRem // Shift data for adding 1st bpp bytes - // no need for mask; shift clears inactive bytes - movq mm0, [edi+ebx] - paddb mm0, mm1 - // Add 2nd active group - movq mm1, mm0 // mov updated Raws to mm1 - psllq mm1, ShiftBpp // shift data to position correctly - // there is no need for any mask - // since shift clears inactive bits/bytes - add ebx, 8 - paddb mm0, mm1 - cmp ebx, MMXLength - movq [edi+ebx-8], mm0 - movq mm1, mm0 // Prep for doing 1st add at top of loop - jb dsub4lp - } // end _asm block - } - break; - - case 2: - { - ActiveMask.use = 0x00000000ffff0000; - ShiftBpp.use = 16; // == 2 * 8 - ShiftRem.use = 48; // == 64 - 16 - _asm { - movq mm7, ActiveMask // Load ActiveMask for 2nd active byte group - mov ebx, diff - movq mm6, mm7 - mov edi, row - psllq mm6, ShiftBpp // Move mask in mm6 to cover 3rd active - // byte group - mov esi, edi // lp = row - movq mm5, mm6 - add edi, bpp // rp = row + bpp - psllq mm5, ShiftBpp // Move mask in mm5 to cover 4th active - // byte group - // PRIME the pump (load the first Raw(x-bpp) data set - movq mm1, [edi+ebx-8] -dsub2lp: - // Add 1st active group - psrlq mm1, ShiftRem // Shift data for adding 1st bpp bytes - // no need for mask; shift clears inactive - // bytes - movq mm0, [edi+ebx] - paddb mm0, mm1 - // Add 2nd active group - movq mm1, mm0 // mov updated Raws to mm1 - psllq mm1, ShiftBpp // shift data to position correctly - pand mm1, mm7 // mask to use only 2nd active group - paddb mm0, mm1 - // Add 3rd active group - movq mm1, mm0 // mov updated Raws to mm1 - psllq mm1, ShiftBpp // shift data to position correctly - pand mm1, mm6 // mask to use only 3rd active group - paddb mm0, mm1 - // Add 4th active group - movq mm1, mm0 // mov updated Raws to mm1 - psllq mm1, ShiftBpp // shift data to position correctly - pand mm1, mm5 // mask to use only 4th active group - add ebx, 8 - paddb mm0, mm1 - cmp ebx, MMXLength - movq [edi+ebx-8], mm0 // Write updated Raws back to array - movq mm1, mm0 // Prep for doing 1st add at top of loop - jb dsub2lp - } // end _asm block - } - break; - case 8: - { - _asm { - mov edi, row - mov ebx, diff - mov esi, edi // lp = row - add edi, bpp // rp = row + bpp - mov ecx, MMXLength - movq mm7, [edi+ebx-8] // PRIME the pump (load the first - // Raw(x-bpp) data set - and ecx, 0x0000003f // calc bytes over mult of 64 -dsub8lp: - movq mm0, [edi+ebx] // Load Sub(x) for 1st 8 bytes - paddb mm0, mm7 - movq mm1, [edi+ebx+8] // Load Sub(x) for 2nd 8 bytes - movq [edi+ebx], mm0 // Write Raw(x) for 1st 8 bytes - // Now mm0 will be used as Raw(x-bpp) for - // the 2nd group of 8 bytes. This will be - // repeated for each group of 8 bytes with - // the 8th group being used as the Raw(x-bpp) - // for the 1st group of the next loop. - paddb mm1, mm0 - movq mm2, [edi+ebx+16] // Load Sub(x) for 3rd 8 bytes - movq [edi+ebx+8], mm1 // Write Raw(x) for 2nd 8 bytes - paddb mm2, mm1 - movq mm3, [edi+ebx+24] // Load Sub(x) for 4th 8 bytes - movq [edi+ebx+16], mm2 // Write Raw(x) for 3rd 8 bytes - paddb mm3, mm2 - movq mm4, [edi+ebx+32] // Load Sub(x) for 5th 8 bytes - movq [edi+ebx+24], mm3 // Write Raw(x) for 4th 8 bytes - paddb mm4, mm3 - movq mm5, [edi+ebx+40] // Load Sub(x) for 6th 8 bytes - movq [edi+ebx+32], mm4 // Write Raw(x) for 5th 8 bytes - paddb mm5, mm4 - movq mm6, [edi+ebx+48] // Load Sub(x) for 7th 8 bytes - movq [edi+ebx+40], mm5 // Write Raw(x) for 6th 8 bytes - paddb mm6, mm5 - movq mm7, [edi+ebx+56] // Load Sub(x) for 8th 8 bytes - movq [edi+ebx+48], mm6 // Write Raw(x) for 7th 8 bytes - add ebx, 64 - paddb mm7, mm6 - cmp ebx, ecx - movq [edi+ebx-8], mm7 // Write Raw(x) for 8th 8 bytes - jb dsub8lp - cmp ebx, MMXLength - jnb dsub8lt8 -dsub8lpA: - movq mm0, [edi+ebx] - add ebx, 8 - paddb mm0, mm7 - cmp ebx, MMXLength - movq [edi+ebx-8], mm0 // use -8 to offset early add to ebx - movq mm7, mm0 // Move calculated Raw(x) data to mm1 to - // be the new Raw(x-bpp) for the next loop - jb dsub8lpA -dsub8lt8: - } // end _asm block - } - break; - - default: // bpp greater than 8 bytes - { - _asm { - mov ebx, diff - mov edi, row - mov esi, edi // lp = row - add edi, bpp // rp = row + bpp -dsubAlp: - movq mm0, [edi+ebx] - movq mm1, [esi+ebx] - add ebx, 8 - paddb mm0, mm1 - cmp ebx, MMXLength - movq [edi+ebx-8], mm0 // mov does not affect flags; -8 to offset - // add ebx - jb dsubAlp - } // end _asm block - } - break; - - } // end switch ( bpp ) - - _asm { - mov ebx, MMXLength - mov edi, row - cmp ebx, FullLength - jnb dsubend - mov esi, edi // lp = row - xor eax, eax - add edi, bpp // rp = row + bpp -dsublp2: - mov al, [esi+ebx] - add [edi+ebx], al - inc ebx - cmp ebx, FullLength - jb dsublp2 -dsubend: - emms // End MMX instructions; prep for possible FP instrs. - } // end _asm block -} - -// Optimized code for PNG Up filter decoder -void /* PRIVATE */ -png_read_filter_row_mmx_up(png_row_infop row_info, png_bytep row, - png_bytep prev_row) -{ - png_uint_32 len; - len = row_info->rowbytes; // # of bytes to filter - _asm { - mov edi, row - // get # of bytes to alignment - mov ecx, edi - xor ebx, ebx - add ecx, 0x7 - xor eax, eax - and ecx, 0xfffffff8 - mov esi, prev_row - sub ecx, edi - jz dupgo - // fix alignment -duplp1: - mov al, [edi+ebx] - add al, [esi+ebx] - inc ebx - cmp ebx, ecx - mov [edi + ebx-1], al // mov does not affect flags; -1 to offset inc ebx - jb duplp1 -dupgo: - mov ecx, len - mov edx, ecx - sub edx, ebx // subtract alignment fix - and edx, 0x0000003f // calc bytes over mult of 64 - sub ecx, edx // drop over bytes from length - // Unrolled loop - use all MMX registers and interleave to reduce - // number of branch instructions (loops) and reduce partial stalls -duploop: - movq mm1, [esi+ebx] - movq mm0, [edi+ebx] - movq mm3, [esi+ebx+8] - paddb mm0, mm1 - movq mm2, [edi+ebx+8] - movq [edi+ebx], mm0 - paddb mm2, mm3 - movq mm5, [esi+ebx+16] - movq [edi+ebx+8], mm2 - movq mm4, [edi+ebx+16] - movq mm7, [esi+ebx+24] - paddb mm4, mm5 - movq mm6, [edi+ebx+24] - movq [edi+ebx+16], mm4 - paddb mm6, mm7 - movq mm1, [esi+ebx+32] - movq [edi+ebx+24], mm6 - movq mm0, [edi+ebx+32] - movq mm3, [esi+ebx+40] - paddb mm0, mm1 - movq mm2, [edi+ebx+40] - movq [edi+ebx+32], mm0 - paddb mm2, mm3 - movq mm5, [esi+ebx+48] - movq [edi+ebx+40], mm2 - movq mm4, [edi+ebx+48] - movq mm7, [esi+ebx+56] - paddb mm4, mm5 - movq mm6, [edi+ebx+56] - movq [edi+ebx+48], mm4 - add ebx, 64 - paddb mm6, mm7 - cmp ebx, ecx - movq [edi+ebx-8], mm6 // (+56)movq does not affect flags; - // -8 to offset add ebx - jb duploop - - cmp edx, 0 // Test for bytes over mult of 64 - jz dupend - - - // 2 lines added by lcreeve at netins.net - // (mail 11 Jul 98 in png-implement list) - cmp edx, 8 //test for less than 8 bytes - jb duplt8 - - - add ecx, edx - and edx, 0x00000007 // calc bytes over mult of 8 - sub ecx, edx // drop over bytes from length - jz duplt8 - // Loop using MMX registers mm0 & mm1 to update 8 bytes simultaneously -duplpA: - movq mm1, [esi+ebx] - movq mm0, [edi+ebx] - add ebx, 8 - paddb mm0, mm1 - cmp ebx, ecx - movq [edi+ebx-8], mm0 // movq does not affect flags; -8 to offset add ebx - jb duplpA - cmp edx, 0 // Test for bytes over mult of 8 - jz dupend -duplt8: - xor eax, eax - add ecx, edx // move over byte count into counter - // Loop using x86 registers to update remaining bytes -duplp2: - mov al, [edi + ebx] - add al, [esi + ebx] - inc ebx - cmp ebx, ecx - mov [edi + ebx-1], al // mov does not affect flags; -1 to offset inc ebx - jb duplp2 -dupend: - // Conversion of filtered row completed - emms // End MMX instructions; prep for possible FP instrs. - } // end _asm block -} - - -// Optimized png_read_filter_row routines -void /* PRIVATE */ -png_read_filter_row(png_structp png_ptr, png_row_infop row_info, png_bytep - row, png_bytep prev_row, int filter) -{ -#ifdef PNG_DEBUG - char filnm[10]; -#endif - - if (mmx_supported == 2) { -#if !defined(PNG_1_0_X) - /* this should have happened in png_init_mmx_flags() already */ - png_warning(png_ptr, "asm_flags may not have been initialized"); -#endif - png_mmx_support(); - } - -#ifdef PNG_DEBUG - png_debug(1, "in png_read_filter_row\n"); - switch (filter) - { - case 0: png_snprintf(filnm, 10, "none"); - break; -#if !defined(PNG_1_0_X) - case 1: png_snprintf(filnm, 10, "sub-%s", - (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_SUB)? "MMX" : "x86"); - break; - case 2: png_snprintf(filnm, 10, "up-%s", - (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_UP)? "MMX" : "x86"); - break; - case 3: png_snprintf(filnm, 10, "avg-%s", - (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_AVG)? "MMX" : "x86"); - break; - case 4: png_snprintf(filnm, 10, "Paeth-%s", - (png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_PAETH)? "MMX":"x86"); - break; -#else - case 1: png_snprintf(filnm, 10, "sub"); - break; - case 2: png_snprintf(filnm, 10, "up"); - break; - case 3: png_snprintf(filnm, 10, "avg"); - break; - case 4: png_snprintf(filnm, 10, "Paeth"); - break; -#endif - default: png_snprintf(filnm, 10, "unknw"); - break; - } - png_debug2(0,"row=%5d, %s, ", png_ptr->row_number, filnm); - png_debug2(0, "pd=%2d, b=%d, ", (int)row_info->pixel_depth, - (int)((row_info->pixel_depth + 7) >> 3)); - png_debug1(0,"len=%8d, ", row_info->rowbytes); -#endif /* PNG_DEBUG */ - - switch (filter) - { - case PNG_FILTER_VALUE_NONE: - break; - - case PNG_FILTER_VALUE_SUB: - { -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_SUB) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - if (mmx_supported) -#endif - { - png_read_filter_row_mmx_sub(row_info, row); - } - else - { - png_uint_32 i; - png_uint_32 istop = row_info->rowbytes; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_bytep rp = row + bpp; - png_bytep lp = row; - - for (i = bpp; i < istop; i++) - { - *rp = (png_byte)(((int)(*rp) + (int)(*lp++)) & 0xff); - rp++; - } - } - break; - } - - case PNG_FILTER_VALUE_UP: - { -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_UP) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - if (mmx_supported) -#endif - { - png_read_filter_row_mmx_up(row_info, row, prev_row); - } - else - { - png_uint_32 i; - png_uint_32 istop = row_info->rowbytes; - png_bytep rp = row; - png_bytep pp = prev_row; - - for (i = 0; i < istop; ++i) - { - *rp = (png_byte)(((int)(*rp) + (int)(*pp++)) & 0xff); - rp++; - } - } - break; - } - - case PNG_FILTER_VALUE_AVG: - { -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_AVG) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - if (mmx_supported) -#endif - { - png_read_filter_row_mmx_avg(row_info, row, prev_row); - } - else - { - png_uint_32 i; - png_bytep rp = row; - png_bytep pp = prev_row; - png_bytep lp = row; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_uint_32 istop = row_info->rowbytes - bpp; - - for (i = 0; i < bpp; i++) - { - *rp = (png_byte)(((int)(*rp) + - ((int)(*pp++) >> 1)) & 0xff); - rp++; - } - - for (i = 0; i < istop; i++) - { - *rp = (png_byte)(((int)(*rp) + - ((int)(*pp++ + *lp++) >> 1)) & 0xff); - rp++; - } - } - break; - } - - case PNG_FILTER_VALUE_PAETH: - { -#if !defined(PNG_1_0_X) - if ((png_ptr->asm_flags & PNG_ASM_FLAG_MMX_READ_FILTER_PAETH) && - (row_info->pixel_depth >= png_ptr->mmx_bitdepth_threshold) && - (row_info->rowbytes >= png_ptr->mmx_rowbytes_threshold)) -#else - if (mmx_supported) -#endif - { - png_read_filter_row_mmx_paeth(row_info, row, prev_row); - } - else - { - png_uint_32 i; - png_bytep rp = row; - png_bytep pp = prev_row; - png_bytep lp = row; - png_bytep cp = prev_row; - png_uint_32 bpp = (row_info->pixel_depth + 7) >> 3; - png_uint_32 istop=row_info->rowbytes - bpp; - - for (i = 0; i < bpp; i++) - { - *rp = (png_byte)(((int)(*rp) + (int)(*pp++)) & 0xff); - rp++; - } - - for (i = 0; i < istop; i++) // use leftover rp,pp - { - int a, b, c, pa, pb, pc, p; - - a = *lp++; - b = *pp++; - c = *cp++; - - p = b - c; - pc = a - c; - -#ifdef PNG_USE_ABS - pa = abs(p); - pb = abs(pc); - pc = abs(p + pc); -#else - pa = p < 0 ? -p : p; - pb = pc < 0 ? -pc : pc; - pc = (p + pc) < 0 ? -(p + pc) : p + pc; -#endif - - /* - if (pa <= pb && pa <= pc) - p = a; - else if (pb <= pc) - p = b; - else - p = c; - */ - - p = (pa <= pb && pa <=pc) ? a : (pb <= pc) ? b : c; - - *rp = (png_byte)(((int)(*rp) + p) & 0xff); - rp++; - } - } - break; - } - - default: - png_warning(png_ptr, "Ignoring bad row filter type"); - *row=0; - break; - } -} - -#endif /* PNG_MMX_CODE_SUPPORTED && PNG_USE_PNGVCRD */ +/* pnggvrd.c was removed from libpng-1.2.20. */ diff --git a/pngwrite.c b/pngwrite.c index 8d5b98ac..c6df1ef3 100644 --- a/pngwrite.c +++ b/pngwrite.c @@ -454,14 +454,6 @@ png_create_write_struct_2(png_const_charp user_png_ver, png_voidp error_ptr, if (png_ptr == NULL) return (NULL); -#if !defined(PNG_1_0_X) -#ifdef PNG_ASSEMBLER_CODE_SUPPORTED -#ifdef PNG_MMX_CODE_SUPPORTED - png_init_mmx_flags(png_ptr); /* 1.2.0 addition */ -#endif -#endif -#endif /* PNG_1_0_X */ - /* added at libpng-1.2.6 */ #ifdef PNG_SET_USER_LIMITS_SUPPORTED png_ptr->user_width_max=PNG_USER_WIDTH_MAX; @@ -670,14 +662,6 @@ png_write_init_3(png_structpp ptr_ptr, png_const_charp user_png_ver, png_ptr->user_height_max=PNG_USER_HEIGHT_MAX; #endif -#if !defined(PNG_1_0_X) -#ifdef PNG_ASSEMBLER_CODE_SUPPORTED -#ifdef PNG_MMX_CODE_SUPPORTED - png_init_mmx_flags(png_ptr); /* 1.2.0 addition */ -#endif -#endif -#endif /* PNG_1_0_X */ - #ifdef PNG_SETJMP_SUPPORTED /* restore jump buffer */ png_memcpy(png_ptr->jmpbuf, tmp_jmp, png_sizeof (jmp_buf)); diff --git a/scripts/CMakeLists.txt b/scripts/CMakeLists.txt index 47c57a26..b948607e 100644 --- a/scripts/CMakeLists.txt +++ b/scripts/CMakeLists.txt @@ -62,8 +62,6 @@ else(NOT WIN32) endif("$ENV{${TEXT}}" STREQUAL "") endif(NOT WIN32) -option(PNG_MMX "Use MMX assembler code (x86 only)" ${png_asm_tmp}) - # SET LIBNAME # msvc does not append 'lib' - do it here to have consistent name if(MSVC) @@ -101,21 +99,8 @@ set(pngtest_sources # SOME NEEDED DEFINITIONS add_definitions(-DZLIB_DLL) -if(PNG_MMX) - if(MSVC) - add_definitions(-DPNG_NO_MODULEDEF -D_CRT_SECURE_NO_DEPRECATE) - set(libpng_sources ${libpng_sources} - pngvcrd.c - ) - else(MSVC) - set(libpng_sources ${libpng_sources} - pnggccrd.c - ) - endif(MSVC) -else(PNG_MMX) - add_definitions(-DLIBPNG_NO_MMX) - add_definitions(-DPNG_NO_MMX_CODE) -endif(PNG_MMX) +add_definitions(-DLIBPNG_NO_MMX) +add_definitions(-DPNG_NO_MMX_CODE) if(PNG_CONSOLE_IO_SUPPORTED) add_definitions(-DPNG_CONSOLE_IO_SUPPORTED) @@ -180,7 +165,7 @@ configure_file(${PNG_SOURCE_DIR}/scripts/libpng-config.in # SET UP LINKS set_target_properties(${PNG_LIB_NAME} PROPERTIES -# VERSION 0.${PNGLIB_RELEASE}.1.2.20rc3 +# VERSION 0.${PNGLIB_RELEASE}.1.2.20rc4 VERSION 0.${PNGLIB_RELEASE}.0 SOVERSION 0 CLEAN_DIRECT_OUTPUT 1) diff --git a/scripts/libpng-config-head.in b/scripts/libpng-config-head.in index 0289369f..8f8e29b5 100755 --- a/scripts/libpng-config-head.in +++ b/scripts/libpng-config-head.in @@ -8,7 +8,7 @@ # Modeled after libxml-config. -version=1.2.20rc3 +version=1.2.20rc4 prefix="" libdir="" libs="" diff --git a/scripts/libpng.pc-configure.in b/scripts/libpng.pc-configure.in index 03e08e20..ab39e205 100644 --- a/scripts/libpng.pc-configure.in +++ b/scripts/libpng.pc-configure.in @@ -5,6 +5,6 @@ includedir=@includedir@/libpng12 Name: libpng Description: Loads and saves PNG files -Version: 1.2.20rc3 +Version: 1.2.20rc4 Libs: -L${libdir} -lpng12 Cflags: -I${includedir} @LIBPNG_NO_MMX@ diff --git a/scripts/libpng.pc.in b/scripts/libpng.pc.in index d3e688d9..d83e2d8b 100644 --- a/scripts/libpng.pc.in +++ b/scripts/libpng.pc.in @@ -5,6 +5,6 @@ includedir=@includedir@/libpng12 Name: libpng Description: Loads and saves PNG files -Version: 1.2.20rc3 +Version: 1.2.20rc4 Libs: -L${libdir} -lpng12 Cflags: -I${includedir} diff --git a/scripts/makefile.32sunu b/scripts/makefile.32sunu index fb4dc459..12559988 100644 --- a/scripts/makefile.32sunu +++ b/scripts/makefile.32sunu @@ -8,7 +8,7 @@ # Library name: LIBNAME=libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.64sunu b/scripts/makefile.64sunu index 0b8564f7..aed69ed4 100644 --- a/scripts/makefile.64sunu +++ b/scripts/makefile.64sunu @@ -8,7 +8,7 @@ # Library name: LIBNAME=libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.aix b/scripts/makefile.aix index 1d31f0d0..f32233f2 100644 --- a/scripts/makefile.aix +++ b/scripts/makefile.aix @@ -20,7 +20,7 @@ LN_SF = ln -f -s LIBNAME=libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) prefix=/usr/local diff --git a/scripts/makefile.beos b/scripts/makefile.beos index d5952337..15449c43 100644 --- a/scripts/makefile.beos +++ b/scripts/makefile.beos @@ -8,7 +8,7 @@ # Library name: LIBNAME=libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.cygwin b/scripts/makefile.cygwin index 701e6b12..2812ac25 100644 --- a/scripts/makefile.cygwin +++ b/scripts/makefile.cygwin @@ -31,9 +31,6 @@ endif DESTDIR= -# To disable assembler optimizations, add '-DPNG_NO_MMX_CODE' to -# $CFLAGS. To enable, add pnggccrd.o to the dependencies. - CC=gcc ifdef MINGW MINGW_CCFLAGS=-mno-cygwin -I/usr/include/mingw @@ -77,7 +74,7 @@ CFLAGS= $(strip $(MINGW_CCFLAGS) $(addprefix -I,$(ZLIBINC)) \ LIBNAME = libpng12 PNGMAJ = 0 CYGDLL = 12 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) SHAREDLIB=cygpng$(CYGDLL).dll @@ -112,7 +109,7 @@ DL =$(D)$(LIBPATH) OBJS = png.o pngset.o pngget.o pngrutil.o pngtrans.o pngwutil.o \ pngread.o pngrio.o pngwio.o pngwrite.o pngrtran.o \ - pngwtran.o pngmem.o pngerror.o pngpread.o # pnggccrd.o + pngwtran.o pngmem.o pngerror.o pngpread.o OBJSDLL = $(OBJS:.o=.pic.o) @@ -156,20 +153,6 @@ shared: all-shared all-static: $(STATLIB) pngtest-stat$(EXE) all-shared: $(SHAREDLIB) pngtest$(EXE) -# pnggccrd.o: pnggccrd.c png.h pngconf.h -# @echo "" -# @echo ' You can ignore the "control reaches end of non-void function"' -# @echo ' warning and " defined but not used" warnings:' -# @echo "" -# $(CC) -c $(CFLAGS) -o $@ $< - -# pnggccrd.pic.o: pnggccrd.c png.h pngconf.h -# @echo "" -# @echo ' You can ignore the "control reaches end of non-void function"' -# @echo ' warning and " defined but not used" warnings:' -# @echo "" -# $(CC) -c $(CFLAGS) -DPNG_BUILD_DLL -o $@ $< - $(STATLIB): $(OBJS) ar rc $@ $(OBJS) $(RANLIB) $@ diff --git a/scripts/makefile.darwin b/scripts/makefile.darwin index f84c0707..2a27d1f7 100644 --- a/scripts/makefile.darwin +++ b/scripts/makefile.darwin @@ -19,7 +19,7 @@ ZLIBINC=../zlib # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.dec b/scripts/makefile.dec index b18730fa..a9d243e1 100644 --- a/scripts/makefile.dec +++ b/scripts/makefile.dec @@ -5,7 +5,7 @@ # Library name: PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) LIBNAME = libpng12 diff --git a/scripts/makefile.elf b/scripts/makefile.elf index d7f49bed..4d8e04c4 100644 --- a/scripts/makefile.elf +++ b/scripts/makefile.elf @@ -12,7 +12,7 @@ # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.gcmmx b/scripts/makefile.gcmmx index a99bef58..1bc5215b 100644 --- a/scripts/makefile.gcmmx +++ b/scripts/makefile.gcmmx @@ -7,8 +7,6 @@ # CAUTION: Do not use this makefile with gcc versions 2.7.2.2 and earlier. -# WARNING: The assembler code in pnggccrd.c may not be thread safe. - # NOTE: When testing MMX performance on a multitasking system, make sure # there are no floating-point programs (e.g., SETI@Home) running in # the background! Context switches between MMX and FPU are expensive. @@ -16,7 +14,7 @@ # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: @@ -98,7 +96,7 @@ DM=$(DESTDIR)$(MANPATH) OBJS = png.o pngset.o pngget.o pngrutil.o pngtrans.o pngwutil.o \ pngread.o pngrio.o pngwio.o pngwrite.o pngrtran.o \ - pngwtran.o pngmem.o pngerror.o pngpread.o pnggccrd.o + pngwtran.o pngmem.o pngerror.o pngpread.o OBJSDLL = $(OBJS:.o=.pic.o) @@ -127,12 +125,6 @@ libpng-config: cat scripts/libpng-config-body.in ) > libpng-config chmod +x libpng-config -pnggccrd.o: pnggccrd.c png.h pngconf.h - $(CC) -c $(CFLAGS) -o $@ $*.c - -pnggccrd.pic.o: pnggccrd.c png.h pngconf.h - $(CC) -c $(CFLAGS) -fPIC -o $@ pnggccrd.c - $(LIBSO): $(LIBSOMAJ) $(LN_SF) $(LIBSOMAJ) $(LIBSO) diff --git a/scripts/makefile.hp64 b/scripts/makefile.hp64 index 64666121..38c896cd 100644 --- a/scripts/makefile.hp64 +++ b/scripts/makefile.hp64 @@ -18,7 +18,7 @@ ZLIBINC=/opt/zlib/include # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.hpgcc b/scripts/makefile.hpgcc index 3cfa3e3f..34053ce7 100644 --- a/scripts/makefile.hpgcc +++ b/scripts/makefile.hpgcc @@ -8,7 +8,7 @@ # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.hpux b/scripts/makefile.hpux index 543a14da..d92222d5 100644 --- a/scripts/makefile.hpux +++ b/scripts/makefile.hpux @@ -18,7 +18,7 @@ ZLIBINC=/opt/zlib/include # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.intel b/scripts/makefile.intel index a155ddd5..b0b523af 100644 --- a/scripts/makefile.intel +++ b/scripts/makefile.intel @@ -36,7 +36,7 @@ O=.obj OBJS=png$(O) pngset$(O) pngget$(O) pngrutil$(O) pngtrans$(O) pngwutil$(O) \ pngmem$(O) pngpread$(O) pngread$(O) pngerror$(O) pngwrite$(O) \ -pngrtran$(O) pngwtran$(O) pngrio$(O) pngwio$(O) pngvcrd$(O) pnggccrd$(O) +pngrtran$(O) pngwtran$(O) pngrio$(O) pngwio$(O) all: test @@ -61,12 +61,6 @@ pngrtran$(O): png.h pngconf.h pngrutil$(O): png.h pngconf.h $(CC) $(CFLAGS) $*.c $(ERRFILE) -pnggccrd$(O): png.h pngconf.h - $(CC) $(CFLAGS) $*.c $(ERRFILE) - -pngvcrd$(O): png.h pngconf.h - $(CC) $(CFLAGS) $*.c $(ERRFILE) - pngerror$(O): png.h pngconf.h $(CC) $(CFLAGS) $*.c $(ERRFILE) diff --git a/scripts/makefile.linux b/scripts/makefile.linux index 9ee81f66..2c83c21f 100644 --- a/scripts/makefile.linux +++ b/scripts/makefile.linux @@ -6,7 +6,7 @@ # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.mingw b/scripts/makefile.mingw index 09faefd9..43963e85 100644 --- a/scripts/makefile.mingw +++ b/scripts/makefile.mingw @@ -39,10 +39,6 @@ RANLIB=ranlib MKDIR_P=/bin/mkdir -pv - -# To disable assembler optimizations, add '-DPNG_NO_MMX_CODE' to -# $CFLAGS. To enable them, add pnggccrd.o to the dependencies. - # Where "make install" puts libpng*.a, *png*.dll, png.h, and pngconf.h ifndef prefix prefix=/usr @@ -78,7 +74,7 @@ CFLAGS= $(strip $(MINGW_CCFLAGS) $(addprefix -I,$(ZLIBINC)) \ LIBNAME = libpng12 PNGMAJ = 0 MINGDLL = 12 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) SHAREDLIB=libpng$(MINGDLL).dll @@ -109,7 +105,7 @@ DL =$(D)$(LIBPATH) OBJS = png.o pngset.o pngget.o pngrutil.o pngtrans.o pngwutil.o \ pngread.o pngrio.o pngwio.o pngwrite.o pngrtran.o \ - pngwtran.o pngmem.o pngerror.o pngpread.o # pnggccrd.o + pngwtran.o pngmem.o pngerror.o pngpread.o OBJSDLL = $(OBJS:.o=.pic.o) @@ -282,7 +278,6 @@ pngwrite.o pngwrite.pic.o: png.h pngconf.h pngwrite.c pngwtran.o pngwtran.pic.o: png.h pngconf.h pngwtran.c pngwutil.o pngwutil.pic.o: png.h pngconf.h pngwutil.c pngpread.o pngpread.pic.o: png.h pngconf.h pngpread.c -# pnggccrd.o pnggccrd.pic.o: png.h pngconf.h pnggccrd.c pngtest.o pngtest.pic.o: png.h pngconf.h pngtest.c diff --git a/scripts/makefile.ne12bsd b/scripts/makefile.ne12bsd index 422e5167..de9c56a8 100644 --- a/scripts/makefile.ne12bsd +++ b/scripts/makefile.ne12bsd @@ -14,10 +14,10 @@ INCSDIR=${LOCALBASE}/include/libpng12 LIB= png12 SHLIB_MAJOR= 0 -SHLIB_MINOR= 1.2.20rc3 -SRCS= pnggccrd.c png.c pngset.c pngget.c pngrutil.c pngtrans.c pngwutil.c \ - pngread.c pngrio.c pngwio.c pngwrite.c pngrtran.c \ - pngwtran.c pngmem.c pngerror.c pngpread.c +SHLIB_MINOR= 1.2.20rc4 +SRCS= png.c pngset.c pngget.c pngrutil.c pngtrans.c pngwutil.c \ + pngread.c pngrio.c pngwio.c pngwrite.c pngrtran.c \ + pngwtran.c pngmem.c pngerror.c pngpread.c INCS= png.h pngconf.h MAN= libpng.3 libpngpf.3 png.5 diff --git a/scripts/makefile.netbsd b/scripts/makefile.netbsd index c1219e2a..26be87f1 100644 --- a/scripts/makefile.netbsd +++ b/scripts/makefile.netbsd @@ -14,10 +14,10 @@ INCSDIR=${LOCALBASE}/include/libpng LIB= png SHLIB_MAJOR= 3 -SHLIB_MINOR= 1.2.20rc3 -SRCS= pnggccrd.c png.c pngset.c pngget.c pngrutil.c pngtrans.c pngwutil.c \ - pngread.c pngrio.c pngwio.c pngwrite.c pngrtran.c \ - pngwtran.c pngmem.c pngerror.c pngpread.c +SHLIB_MINOR= 1.2.20rc4 +SRCS= png.c pngset.c pngget.c pngrutil.c pngtrans.c pngwutil.c \ + pngread.c pngrio.c pngwio.c pngwrite.c pngrtran.c \ + pngwtran.c pngmem.c pngerror.c pngpread.c INCS= png.h pngconf.h MAN= libpng.3 libpngpf.3 png.5 diff --git a/scripts/makefile.nommx b/scripts/makefile.nommx index a2d840bf..e1b8c01d 100644 --- a/scripts/makefile.nommx +++ b/scripts/makefile.nommx @@ -7,7 +7,7 @@ # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.openbsd b/scripts/makefile.openbsd index 9e85b6ff..91259624 100644 --- a/scripts/makefile.openbsd +++ b/scripts/makefile.openbsd @@ -8,10 +8,10 @@ LIBDIR= ${PREFIX}/lib MANDIR= ${PREFIX}/man/cat SHLIB_MAJOR= 0 -SHLIB_MINOR= 1.2.20rc3 +SHLIB_MINOR= 1.2.20rc4 LIB= png -SRCS= png.c pngerror.c pnggccrd.c pngget.c pngmem.c pngpread.c \ +SRCS= png.c pngerror.c pngget.c pngmem.c pngpread.c \ pngread.c pngrio.c pngrtran.c pngrutil.c pngset.c pngtrans.c \ pngwio.c pngwrite.c pngwtran.c pngwutil.c diff --git a/scripts/makefile.sco b/scripts/makefile.sco index daa5661d..3827f515 100644 --- a/scripts/makefile.sco +++ b/scripts/makefile.sco @@ -9,7 +9,7 @@ # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.sggcc b/scripts/makefile.sggcc index 94f428cd..58f7ec6f 100644 --- a/scripts/makefile.sggcc +++ b/scripts/makefile.sggcc @@ -6,7 +6,7 @@ # Library name: LIBNAME=libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.sgi b/scripts/makefile.sgi index 371fc32e..64a65950 100644 --- a/scripts/makefile.sgi +++ b/scripts/makefile.sgi @@ -6,7 +6,7 @@ # Library name: LIBNAME=libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.so9 b/scripts/makefile.so9 index 4dace290..3b1a2864 100644 --- a/scripts/makefile.so9 +++ b/scripts/makefile.so9 @@ -8,7 +8,7 @@ # Library name: PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) LIBNAME = libpng12 diff --git a/scripts/makefile.solaris b/scripts/makefile.solaris index daf96f87..01b10f49 100644 --- a/scripts/makefile.solaris +++ b/scripts/makefile.solaris @@ -8,7 +8,7 @@ # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: diff --git a/scripts/makefile.solaris-x86 b/scripts/makefile.solaris-x86 index 7a9fc164..e4e2b74d 100644 --- a/scripts/makefile.solaris-x86 +++ b/scripts/makefile.solaris-x86 @@ -8,7 +8,7 @@ # Library name: LIBNAME = libpng12 PNGMAJ = 0 -PNGMIN = 1.2.20rc3 +PNGMIN = 1.2.20rc4 PNGVER = $(PNGMAJ).$(PNGMIN) # Shared library names: @@ -69,7 +69,7 @@ DM=$(DESTDIR)$(MANPATH) OBJS = png.o pngset.o pngget.o pngrutil.o pngtrans.o pngwutil.o \ pngread.o pngrio.o pngwio.o pngwrite.o pngrtran.o \ - pngwtran.o pngmem.o pngerror.o pngpread.o pnggccrd.o + pngwtran.o pngmem.o pngerror.o pngpread.o OBJSDLL = $(OBJS:.o=.pic.o) @@ -232,7 +232,6 @@ pngwio.o pngwio.pic.o: png.h pngconf.h pngmem.o pngmem.pic.o: png.h pngconf.h pngset.o pngset.pic.o: png.h pngconf.h pngget.o pngget.pic.o: png.h pngconf.h -pnggccrd.o pnggccrd.pic.o: png.h pngconf.h pngread.o pngread.pic.o: png.h pngconf.h pngrtran.o pngrtran.pic.o: png.h pngconf.h pngrutil.o pngrutil.pic.o: png.h pngconf.h diff --git a/scripts/makefile.vcawin32 b/scripts/makefile.vcawin32 index 4c2f3cc0..99f54305 100644 --- a/scripts/makefile.vcawin32 +++ b/scripts/makefile.vcawin32 @@ -26,8 +26,7 @@ O=.obj OBJS1 = png$(O) pngerror$(O) pngget$(O) pngmem$(O) pngpread$(O) OBJS2 = pngread$(O) pngrio$(O) pngrtran$(O) pngrutil$(O) pngset$(O) OBJS3 = pngtrans$(O) pngwio$(O) pngwrite$(O) pngwtran$(O) pngwutil$(O) -OBJS4 = pngvcrd$(O) -OBJS = $(OBJS1) $(OBJS2) $(OBJS3) $(OBJS4) +OBJS = $(OBJS1) $(OBJS2) $(OBJS3) # Targets all: libpng.lib @@ -80,9 +79,6 @@ pngwtran$(O): png.h pngconf.h pngwutil$(O): png.h pngconf.h $(CC) -c $(CFLAGS) $*.c $(ERRFILE) -pngvcrd$(O): png.h pngconf.h - $(CC) -c $(CFLAGS) $*.c $(ERRFILE) - libpng.lib: $(OBJS) -$(RM) $@ $(AR) $(ARFLAGS) -out:$@ $(OBJS) $(ERRFILE) diff --git a/scripts/pngos2.def b/scripts/pngos2.def index d6074f36..4a6b3f1d 100644 --- a/scripts/pngos2.def +++ b/scripts/pngos2.def @@ -2,7 +2,7 @@ ; PNG.LIB module definition file for OS/2 ;---------------------------------------- -; Version 1.2.20rc3 +; Version 1.2.20rc4 LIBRARY PNG DESCRIPTION "PNG image compression library for OS/2" diff --git a/scripts/pngw32.def b/scripts/pngw32.def index 05a97cf9..b31fa79e 100644 --- a/scripts/pngw32.def +++ b/scripts/pngw32.def @@ -5,7 +5,7 @@ LIBRARY EXPORTS -;Version 1.2.20rc3 +;Version 1.2.20rc4 png_build_grayscale_palette @1 png_check_sig @2 png_chunk_error @3