Imported from libpng-1.2.20rc4.tar
diff --git a/ANNOUNCE b/ANNOUNCE
index 2c81353..0a7187c 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 @@
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 @@
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 e09eeb3..ffb954e 100644
--- a/CHANGES
+++ b/CHANGES
@@ -1926,6 +1926,9 @@
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 ed8b513..75f58e8 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 @@
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 @@
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 @@
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 @@
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 0245123..b1a946c 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 9f9b122..1501152 100644
--- a/LICENSE
+++ b/LICENSE
@@ -8,7 +8,7 @@
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
@@ -106,4 +106,4 @@
Glenn Randers-Pehrson
glennrp at users.sourceforge.net
-August 30, 2007
+September 1, 2007
diff --git a/Makefile.am b/Makefile.am
index de7dd31..549c2c7 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -33,7 +33,7 @@
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 @@
${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 41913eb..a56378a 100644
--- a/Makefile.in
+++ b/Makefile.in
@@ -87,8 +87,7 @@
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 @@
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 @@
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 @@
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 @@
${srcdir}/contrib/pngsuite/* \
${srcdir}/contrib/visupng/* \
$(TESTS) \
- example.c libpng.txt pngvcrd.c
+ example.c libpng.txt
CLEANFILES = pngout.png libpng12.pc libpng12-config libpng.vers \
libpng.sym
@@ -455,7 +453,6 @@
@AMDEP_TRUE@@am__include@ @am__quote@./$(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@./$(DEPDIR)/libpng12_la-pngpread.Plo@am__quote@
@@ -471,7 +468,6 @@
@AMDEP_TRUE@@am__include@ @am__quote@./$(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 @@
@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 @@
@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 dec2d45..6ae9daf 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 @@
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 @@
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 af33e0e..52cb27b 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 c60ff40..bfaf636 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 <png-mng-implement@lists.sourceforge.net>.
#
@@ -728,8 +728,8 @@
# 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 @@
POW_LIB
LIBOBJS
LIBPNG_DEFINES
-LIBPNG_NO_MMX
HAVE_LD_VERSION_SCRIPT_TRUE
HAVE_LD_VERSION_SCRIPT_FALSE
PNGLIB_VERSION
@@ -1405,7 +1404,7 @@
# 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: $0 [OPTION]... [VAR=VALUE]...
@@ -1475,7 +1474,7 @@
if test -n "$ac_init_help"; then
case $ac_init_help in
- short | recursive ) echo "Configuration of libpng 1.2.20rc3:";;
+ short | recursive ) echo "Configuration of libpng 1.2.20rc4:";;
esac
cat <<\_ACEOF
@@ -1585,7 +1584,7 @@
test -n "$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 @@
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 $@
@@ -2269,7 +2268,7 @@
# Define the identity of the package.
PACKAGE='libpng'
- VERSION='1.2.20rc3'
+ VERSION='1.2.20rc4'
cat >>confdefs.h <<_ACEOF
@@ -2440,7 +2439,7 @@
-PNGLIB_VERSION=1.2.20rc3
+PNGLIB_VERSION=1.2.20rc4
PNGLIB_MAJOR=1
PNGLIB_MINOR=2
PNGLIB_RELEASE=20
@@ -4788,7 +4787,7 @@
;;
*-*-irix6*)
# Find out which ABI we are using.
- echo '#line 4791 "configure"' > conftest.$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 @@
-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 @@
-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 @@
-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 @@
lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
lt_status=$lt_dlunknown
cat > conftest.$ac_ext <<EOF
-#line 10047 "configure"
+#line 10046 "configure"
#include "confdefs.h"
#if HAVE_DLFCN_H
@@ -10144,7 +10143,7 @@
lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
lt_status=$lt_dlunknown
cat > conftest.$ac_ext <<EOF
-#line 10147 "configure"
+#line 10146 "configure"
#include "confdefs.h"
#if HAVE_DLFCN_H
@@ -12564,11 +12563,11 @@
-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:12567: $lt_compile\"" >&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 @@
-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 @@
-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 @@
-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 @@
-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 @@
-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 @@
-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 @@
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 @@
# 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 = $CONFIG_FILES
@@ -21335,7 +21285,7 @@
_ACEOF
cat >>$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 @@
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 @@
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 $CONFIG_STATUS" >&5
diff --git a/configure.ac b/configure.ac
index c7a98e2..244952c 100644
--- a/configure.ac
+++ b/configure.ac
@@ -18,12 +18,12 @@
dnl Version number stuff here:
-AC_INIT([libpng], [1.2.20rc3], [png-mng-implement@lists.sourceforge.net])
+AC_INIT([libpng], [1.2.20rc4], [png-mng-implement@lists.sourceforge.net])
AM_INIT_AUTOMAKE
dnl stop configure from automagically running automake
AM_MAINTAINER_MODE
-PNGLIB_VERSION=1.2.20rc3
+PNGLIB_VERSION=1.2.20rc4
PNGLIB_MAJOR=1
PNGLIB_MINOR=2
PNGLIB_RELEASE=20
@@ -59,18 +59,7 @@
AC_CHECK_LIB(z, zlibVersion, , AC_ERROR([zlib not installed]))
LIBPNG_DEFINES=-DPNG_CONFIGURE_LIBPNG
-AC_MSG_CHECKING(
- [if assembler code in pnggccrd.c can be compiled without PNG_NO_MMX_CODE])
-AC_TRY_COMPILE(
- [#include "$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 93%
rename from libpng-1.2.20rc3.txt
rename to libpng-1.2.20rc4.txt
index ce05124..ba4d9ab 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
<glennrp at users.sourceforge.net>
Copyright (c) 1998-2007 Glenn Randers-Pehrson
@@ -86,9 +86,7 @@
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 @@
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 @@
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 @@
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 d707341..d80f90b 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 <png.h>\fP
@@ -410,7 +410,7 @@
.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
<glennrp at users.sourceforge.net>
Copyright (c) 1998-2007 Glenn Randers-Pehrson
@@ -496,9 +496,7 @@
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 @@
.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 @@
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 @@
.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 @@
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 @@
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 @@
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 @@
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 1aea5a8..06fddfa 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 <png.h>\fP
diff --git a/png.5 b/png.5
index fadb858..317b7ae 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 54d6f7a..e6ffcd4 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 @@
/* 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_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 @@
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 a26a725..a4aa8d0 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_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 @@
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 @@
/* 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 d5ef5e3..01d7578 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 @@
# 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 93ce47b..2d03c56 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 37f3f33..7007ad9 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 75d2ca0..9cf1b84 100644
--- a/pngget.c
+++ b/pngget.c
@@ -841,6 +841,7 @@
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_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 3d910c3..d5952bb 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 @@
/* 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 c26ce61..c914030 100644
--- a/pngread.c
+++ b/pngread.c
@@ -55,12 +55,6 @@
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 4da862e..99bca19 100644
--- a/pngrutil.c
+++ b/pngrutil.c
@@ -2277,1051 +2277,10 @@
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_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 @@
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 @@
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 @@
#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 8d1cbfc..2871140 100644
--- a/pngset.c
+++ b/pngset.c
@@ -1202,51 +1202,13 @@
#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_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..70734a3 100644
--- a/pngtest.c
+++ b/pngtest.c
@@ -1548,4 +1548,4 @@
}
/* 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 edd8ff6..ce4233e 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 8d5b98a..c6df1ef 100644
--- a/pngwrite.c
+++ b/pngwrite.c
@@ -454,14 +454,6 @@
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_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 47c57a2..b948607 100644
--- a/scripts/CMakeLists.txt
+++ b/scripts/CMakeLists.txt
@@ -62,8 +62,6 @@
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 @@
# 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 ${libpng_sources}
- pnggccrd.c
- )
- endif(MSVC)
-else(PNG_MMX)
- add_definitions(-DLIBPNG_NO_MMX)
- add_definitions(-DPNG_NO_MMX_CODE)
-endif(PNG_MMX)
+add_definitions(-DLIBPNG_NO_MMX)
+add_definitions(-DPNG_NO_MMX_CODE)
if(PNG_CONSOLE_IO_SUPPORTED)
add_definitions(-DPNG_CONSOLE_IO_SUPPORTED)
@@ -180,7 +165,7 @@
# SET UP LINKS
set_target_properties(${PNG_LIB_NAME} PROPERTIES
-# VERSION 0.${PNGLIB_RELEASE}.1.2.20rc3
+# VERSION 0.${PNGLIB_RELEASE}.1.2.20rc4
VERSION 0.${PNGLIB_RELEASE}.0
SOVERSION 0
CLEAN_DIRECT_OUTPUT 1)
diff --git a/scripts/libpng-config-head.in b/scripts/libpng-config-head.in
index 0289369..8f8e29b 100755
--- a/scripts/libpng-config-head.in
+++ b/scripts/libpng-config-head.in
@@ -8,7 +8,7 @@
# Modeled after libxml-config.
-version=1.2.20rc3
+version=1.2.20rc4
prefix=""
libdir=""
libs=""
diff --git a/scripts/libpng.pc-configure.in b/scripts/libpng.pc-configure.in
index 03e08e2..ab39e20 100644
--- a/scripts/libpng.pc-configure.in
+++ b/scripts/libpng.pc-configure.in
@@ -5,6 +5,6 @@
Name: libpng
Description: Loads and saves PNG files
-Version: 1.2.20rc3
+Version: 1.2.20rc4
Libs: -L${libdir} -lpng12
Cflags: -I${includedir} @LIBPNG_NO_MMX@
diff --git a/scripts/libpng.pc.in b/scripts/libpng.pc.in
index d3e688d..d83e2d8 100644
--- a/scripts/libpng.pc.in
+++ b/scripts/libpng.pc.in
@@ -5,6 +5,6 @@
Name: libpng
Description: Loads and saves PNG files
-Version: 1.2.20rc3
+Version: 1.2.20rc4
Libs: -L${libdir} -lpng12
Cflags: -I${includedir}
diff --git a/scripts/makefile.32sunu b/scripts/makefile.32sunu
index fb4dc45..1255998 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 0b8564f..aed69ed 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 1d31f0d..f32233f 100644
--- a/scripts/makefile.aix
+++ b/scripts/makefile.aix
@@ -20,7 +20,7 @@
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 d595233..15449c4 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 701e6b1..2812ac2 100644
--- a/scripts/makefile.cygwin
+++ b/scripts/makefile.cygwin
@@ -31,9 +31,6 @@
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 @@
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 @@
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 @@
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 "<variable> 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 "<variable> 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 f84c070..2a27d1f 100644
--- a/scripts/makefile.darwin
+++ b/scripts/makefile.darwin
@@ -19,7 +19,7 @@
# 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.dec b/scripts/makefile.dec
index b18730f..a9d243e 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 d7f49be..4d8e04c 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 a99bef5..1bc5215 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 @@
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 @@
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 6466612..38c896c 100644
--- a/scripts/makefile.hp64
+++ b/scripts/makefile.hp64
@@ -18,7 +18,7 @@
# 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.hpgcc b/scripts/makefile.hpgcc
index 3cfa3e3..34053ce 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 543a14d..d92222d 100644
--- a/scripts/makefile.hpux
+++ b/scripts/makefile.hpux
@@ -18,7 +18,7 @@
# 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.intel b/scripts/makefile.intel
index a155ddd..b0b523a 100644
--- a/scripts/makefile.intel
+++ b/scripts/makefile.intel
@@ -36,7 +36,7 @@
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$(O)
all: test
@@ -61,12 +61,6 @@
pngrutil$(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 9ee81f6..2c83c21 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 09faefd..43963e8 100644
--- a/scripts/makefile.mingw
+++ b/scripts/makefile.mingw
@@ -39,10 +39,6 @@
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 @@
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 @@
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 @@
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 422e516..de9c56a 100644
--- a/scripts/makefile.ne12bsd
+++ b/scripts/makefile.ne12bsd
@@ -14,10 +14,10 @@
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 c1219e2..26be87f 100644
--- a/scripts/makefile.netbsd
+++ b/scripts/makefile.netbsd
@@ -14,10 +14,10 @@
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 a2d840b..e1b8c01 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 9e85b6f..9125962 100644
--- a/scripts/makefile.openbsd
+++ b/scripts/makefile.openbsd
@@ -8,10 +8,10 @@
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 daa5661..3827f51 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 94f428c..58f7ec6 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 371fc32..64a6595 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 4dace29..3b1a286 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 daf96f8..01b10f4 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 7a9fc16..e4e2b74 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 @@
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 = $(OBJS:.o=.pic.o)
@@ -232,7 +232,6 @@
pngmem.o pngmem.pic.o: png.h pngconf.h
pngset.o pngset.pic.o: png.h pngconf.h
pngget.o pngget.pic.o: png.h pngconf.h
-pnggccrd.o pnggccrd.pic.o: png.h pngconf.h
pngread.o pngread.pic.o: png.h pngconf.h
pngrtran.o pngrtran.pic.o: png.h pngconf.h
pngrutil.o pngrutil.pic.o: png.h pngconf.h
diff --git a/scripts/makefile.vcawin32 b/scripts/makefile.vcawin32
index 4c2f3cc..99f5430 100644
--- a/scripts/makefile.vcawin32
+++ b/scripts/makefile.vcawin32
@@ -26,8 +26,7 @@
OBJS1 = png$(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) $(OBJS3)
# Targets
all: libpng.lib
@@ -80,9 +79,6 @@
pngwutil$(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 d6074f3..4a6b3f1 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 05a97cf..b31fa79 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
