svn merge ^/branches/released

author: marha <marha@users.sourceforge.net> 2009-11-20 14:53:20 +0000
committer: marha <marha@users.sourceforge.net> 2009-11-20 14:53:20 +0000
commit: 9e11b82359e90c0963f56c4695a4dcc8731e45c3 (patch)
tree: ffa3a53246352ca44ccfc4d90e7cca993ad0b02e /pixman/pixman
parent: 607c11e20aca5c9f13d5e6ab80153b10c249162b (diff)
parent: 4ac4a5b7ce8cc8f195d69a42da10d386eaa5c056 (diff)
download: vcxsrv-9e11b82359e90c0963f56c4695a4dcc8731e45c3.tar.gz
vcxsrv-9e11b82359e90c0963f56c4695a4dcc8731e45c3.tar.bz2
vcxsrv-9e11b82359e90c0963f56c4695a4dcc8731e45c3.zip
20 files changed, 6927 insertions, 6822 deletions
diff --git a/pixman/pixman/Makefile.am b/pixman/pixman/Makefile.am
index e19fa6e7f..cd0102313 100644
--- a/pixman/pixman/Makefile.am
+++ b/pixman/pixman/Makefile.am
@@ -1,7 +1,6 @@
 lib_LTLIBRARIES = libpixman-1.la
 libpixman_1_la_LDFLAGS = -version-info $(LT_VERSION_INFO) -no-undefined
 libpixman_1_la_LIBADD = @DEP_LIBS@ -lm
-libpixman_1_la_CFLAGS = -DPIXMAN_DISABLE_DEPRECATED
 libpixman_1_la_SOURCES =			\
 	pixman.h				\
 	pixman-accessor.h			\
@@ -64,6 +63,8 @@ libpixman_mmx_la_CFLAGS = $(DEP_CFLAGS) $(MMX_CFLAGS)
 libpixman_mmx_la_LIBADD = $(DEP_LIBS)
 libpixman_1_la_LDFLAGS += $(MMX_LDFLAGS)
 libpixman_1_la_LIBADD += libpixman-mmx.la
+
+$(libpixman_mmx_la_SOURCES:.c=.s) : ASM_CFLAGS=$(MMX_CFLAGS)
 endif
 
 # vmx code
@@ -75,6 +76,8 @@ libpixman_vmx_la_SOURCES = \
 libpixman_vmx_la_CFLAGS = $(DEP_CFLAGS) $(VMX_CFLAGS)
 libpixman_vmx_la_LIBADD = $(DEP_LIBS)
 libpixman_1_la_LIBADD += libpixman-vmx.la
+
+$(libpixman_vmx_la_SOURCES:.c=.s) : ASM_CFLAGS=$(VMX_CFLAGS)
 endif
 
 # sse2 code
@@ -86,6 +89,8 @@ libpixman_sse2_la_CFLAGS = $(DEP_CFLAGS) $(SSE2_CFLAGS)
 libpixman_sse2_la_LIBADD = $(DEP_LIBS)
 libpixman_1_la_LDFLAGS += $(SSE2_LDFLAGS)
 libpixman_1_la_LIBADD += libpixman-sse2.la
+
+$(libpixman_sse2_la_SOURCES:.c=.s) : ASM_CFLAGS=$(SSE2_CFLAGS)
 endif
 
 # arm simd code
@@ -96,16 +101,23 @@ libpixman_arm_simd_la_SOURCES = \
 libpixman_arm_simd_la_CFLAGS = $(DEP_CFLAGS) $(ARM_SIMD_CFLAGS)
 libpixman_arm_simd_la_LIBADD = $(DEP_LIBS)
 libpixman_1_la_LIBADD += libpixman-arm-simd.la
+
+$(libpixman_arm_simd_la_SOURCES:.c=.s) : ASM_CFLAGS=$(ARM_SIMD_CFLAGS)
 endif
 
 # arm neon code
 if USE_ARM_NEON
 noinst_LTLIBRARIES += libpixman-arm-neon.la
 libpixman_arm_neon_la_SOURCES = \
-        pixman-arm-neon.c
-libpixman_arm_neon_la_CFLAGS = $(DEP_CFLAGS) $(ARM_NEON_CFLAGS)
+        pixman-arm-neon.c	\
+        pixman-arm-neon-asm.S	\
+        pixman-arm-neon-asm.h
+libpixman_arm_neon_la_CFLAGS = $(DEP_CFLAGS)
 libpixman_arm_neon_la_LIBADD = $(DEP_LIBS)
 libpixman_1_la_LIBADD += libpixman-arm-neon.la
-endif
 
+$(libpixman_arm_neon_la_SOURCES:.c=.s) : ASM_CFLAGS=
+endif
 
+.c.s : $(libpixmaninclude_HEADERS) $(BUILT_SOURCES)
+	$(CC) $(CFLAGS) $(ASM_CFLAGS) -DHAVE_CONFIG_H -I$(srcdir) -I$(builddir) -I$(top_builddir) -S -o $@ $<
diff --git a/pixman/pixman/Makefile.in b/pixman/pixman/Makefile.in
index 1ca63f731..a0e4f9e7b 100644
--- a/pixman/pixman/Makefile.in
+++ b/pixman/pixman/Makefile.in
@@ -1,4 +1,4 @@
-# Makefile.in generated by automake 1.10.2 from Makefile.am.
+# Makefile.in generated by automake 1.10.1 from Makefile.am.
 # @configure_input@
 
 # Copyright (C) 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002,
@@ -77,37 +77,28 @@ libLTLIBRARIES_INSTALL = $(INSTALL)
 LTLIBRARIES = $(lib_LTLIBRARIES) $(noinst_LTLIBRARIES)
 libpixman_1_la_DEPENDENCIES = $(am__append_3) $(am__append_5) \
 	$(am__append_8) $(am__append_10) $(am__append_12)
-am_libpixman_1_la_OBJECTS = libpixman_1_la-pixman-access.lo \
-	libpixman_1_la-pixman-access-accessors.lo \
-	libpixman_1_la-pixman-cpu.lo \
-	libpixman_1_la-pixman-gradient-walker.lo \
-	libpixman_1_la-pixman-region16.lo \
-	libpixman_1_la-pixman-region32.lo \
-	libpixman_1_la-pixman-image.lo \
-	libpixman_1_la-pixman-implementation.lo \
-	libpixman_1_la-pixman-combine32.lo \
-	libpixman_1_la-pixman-combine64.lo \
-	libpixman_1_la-pixman-general.lo libpixman_1_la-pixman.lo \
-	libpixman_1_la-pixman-fast-path.lo \
-	libpixman_1_la-pixman-solid-fill.lo \
-	libpixman_1_la-pixman-conical-gradient.lo \
-	libpixman_1_la-pixman-linear-gradient.lo \
-	libpixman_1_la-pixman-radial-gradient.lo \
-	libpixman_1_la-pixman-bits-image.lo \
-	libpixman_1_la-pixman-utils.lo libpixman_1_la-pixman-edge.lo \
-	libpixman_1_la-pixman-edge-accessors.lo \
-	libpixman_1_la-pixman-trap.lo libpixman_1_la-pixman-timer.lo \
-	libpixman_1_la-pixman-matrix.lo
+am_libpixman_1_la_OBJECTS = pixman-access.lo \
+	pixman-access-accessors.lo pixman-cpu.lo \
+	pixman-gradient-walker.lo pixman-region16.lo \
+	pixman-region32.lo pixman-image.lo pixman-implementation.lo \
+	pixman-combine32.lo pixman-combine64.lo pixman-general.lo \
+	pixman.lo pixman-fast-path.lo pixman-solid-fill.lo \
+	pixman-conical-gradient.lo pixman-linear-gradient.lo \
+	pixman-radial-gradient.lo pixman-bits-image.lo pixman-utils.lo \
+	pixman-edge.lo pixman-edge-accessors.lo pixman-trap.lo \
+	pixman-timer.lo pixman-matrix.lo
 libpixman_1_la_OBJECTS = $(am_libpixman_1_la_OBJECTS)
 libpixman_1_la_LINK = $(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) \
-	$(LIBTOOLFLAGS) --mode=link $(CCLD) $(libpixman_1_la_CFLAGS) \
-	$(CFLAGS) $(libpixman_1_la_LDFLAGS) $(LDFLAGS) -o $@
+	$(LIBTOOLFLAGS) --mode=link $(CCLD) $(AM_CFLAGS) $(CFLAGS) \
+	$(libpixman_1_la_LDFLAGS) $(LDFLAGS) -o $@
 am__DEPENDENCIES_1 =
 @USE_ARM_NEON_TRUE@libpixman_arm_neon_la_DEPENDENCIES =  \
 @USE_ARM_NEON_TRUE@	$(am__DEPENDENCIES_1)
-am__libpixman_arm_neon_la_SOURCES_DIST = pixman-arm-neon.c
+am__libpixman_arm_neon_la_SOURCES_DIST = pixman-arm-neon.c \
+	pixman-arm-neon-asm.S pixman-arm-neon-asm.h
 @USE_ARM_NEON_TRUE@am_libpixman_arm_neon_la_OBJECTS =  \
-@USE_ARM_NEON_TRUE@	libpixman_arm_neon_la-pixman-arm-neon.lo
+@USE_ARM_NEON_TRUE@	libpixman_arm_neon_la-pixman-arm-neon.lo \
+@USE_ARM_NEON_TRUE@	pixman-arm-neon-asm.lo
 libpixman_arm_neon_la_OBJECTS = $(am_libpixman_arm_neon_la_OBJECTS)
 libpixman_arm_neon_la_LINK = $(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) \
 	$(LIBTOOLFLAGS) --mode=link $(CCLD) \
@@ -156,6 +147,11 @@ libpixman_vmx_la_LINK = $(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) \
 DEFAULT_INCLUDES = -I.@am__isrc@ -I$(top_builddir)
 depcomp = $(SHELL) $(top_srcdir)/depcomp
 am__depfiles_maybe = depfiles
+CPPASCOMPILE = $(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) \
+	$(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CCASFLAGS) $(CCASFLAGS)
+LTCPPASCOMPILE = $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) \
+	--mode=compile $(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) \
+	$(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CCASFLAGS) $(CCASFLAGS)
 COMPILE = $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) \
 	$(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS)
 LTCOMPILE = $(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) \
@@ -182,30 +178,37 @@ DISTFILES = $(DIST_COMMON) $(DIST_SOURCES) $(TEXINFOS) $(EXTRA_DIST)
 ACLOCAL = @ACLOCAL@
 AMTAR = @AMTAR@
 AR = @AR@
-ARM_NEON_CFLAGS = @ARM_NEON_CFLAGS@
 ARM_SIMD_CFLAGS = @ARM_SIMD_CFLAGS@
 AUTOCONF = @AUTOCONF@
 AUTOHEADER = @AUTOHEADER@
 AUTOMAKE = @AUTOMAKE@
 AWK = @AWK@
 CC = @CC@
+CCAS = @CCAS@
+CCASDEPMODE = @CCASDEPMODE@
+CCASFLAGS = @CCASFLAGS@
 CCDEPMODE = @CCDEPMODE@
 CFLAGS = @CFLAGS@
 CPP = @CPP@
 CPPFLAGS = @CPPFLAGS@
+CXX = @CXX@
+CXXCPP = @CXXCPP@
+CXXDEPMODE = @CXXDEPMODE@
+CXXFLAGS = @CXXFLAGS@
 CYGPATH_W = @CYGPATH_W@
 DEFS = @DEFS@
 DEPDIR = @DEPDIR@
 DEP_CFLAGS = @DEP_CFLAGS@
 DEP_LIBS = @DEP_LIBS@
 DSYMUTIL = @DSYMUTIL@
-DUMPBIN = @DUMPBIN@
+ECHO = @ECHO@
 ECHO_C = @ECHO_C@
 ECHO_N = @ECHO_N@
 ECHO_T = @ECHO_T@
 EGREP = @EGREP@
 EXEEXT = @EXEEXT@
-FGREP = @FGREP@
+F77 = @F77@
+FFLAGS = @FFLAGS@
 GREP = @GREP@
 GTK_CFLAGS = @GTK_CFLAGS@
 GTK_LIBS = @GTK_LIBS@
@@ -214,12 +217,10 @@ INSTALL_DATA = @INSTALL_DATA@
 INSTALL_PROGRAM = @INSTALL_PROGRAM@
 INSTALL_SCRIPT = @INSTALL_SCRIPT@
 INSTALL_STRIP_PROGRAM = @INSTALL_STRIP_PROGRAM@
-LD = @LD@
 LDFLAGS = @LDFLAGS@
 LIBOBJS = @LIBOBJS@
 LIBS = @LIBS@
 LIBTOOL = @LIBTOOL@
-LIPO = @LIPO@
 LN_S = @LN_S@
 LTLIBOBJS = @LTLIBOBJS@
 LT_VERSION_INFO = @LT_VERSION_INFO@
@@ -227,12 +228,8 @@ MAKEINFO = @MAKEINFO@
 MKDIR_P = @MKDIR_P@
 MMX_CFLAGS = @MMX_CFLAGS@
 MMX_LDFLAGS = @MMX_LDFLAGS@
-NM = @NM@
 NMEDIT = @NMEDIT@
-OBJDUMP = @OBJDUMP@
 OBJEXT = @OBJEXT@
-OTOOL = @OTOOL@
-OTOOL64 = @OTOOL64@
 PACKAGE = @PACKAGE@
 PACKAGE_BUGREPORT = @PACKAGE_BUGREPORT@
 PACKAGE_NAME = @PACKAGE_NAME@
@@ -260,7 +257,8 @@ abs_srcdir = @abs_srcdir@
 abs_top_builddir = @abs_top_builddir@
 abs_top_srcdir = @abs_top_srcdir@
 ac_ct_CC = @ac_ct_CC@
-ac_ct_DUMPBIN = @ac_ct_DUMPBIN@
+ac_ct_CXX = @ac_ct_CXX@
+ac_ct_F77 = @ac_ct_F77@
 am__include = @am__include@
 am__leading_dot = @am__leading_dot@
 am__quote = @am__quote@
@@ -291,7 +289,6 @@ libdir = @libdir@
 libexecdir = @libexecdir@
 localedir = @localedir@
 localstatedir = @localstatedir@
-lt_ECHO = @lt_ECHO@
 mandir = @mandir@
 mkdir_p = @mkdir_p@
 oldincludedir = @oldincludedir@
@@ -312,7 +309,6 @@ libpixman_1_la_LDFLAGS = -version-info $(LT_VERSION_INFO) \
 	-no-undefined $(am__append_2) $(am__append_7)
 libpixman_1_la_LIBADD = @DEP_LIBS@ -lm $(am__append_3) $(am__append_5) \
 	$(am__append_8) $(am__append_10) $(am__append_12)
-libpixman_1_la_CFLAGS = -DPIXMAN_DISABLE_DEPRECATED
 libpixman_1_la_SOURCES = \
 	pixman.h				\
 	pixman-accessor.h			\
@@ -377,27 +373,29 @@ CLEANFILES = pixman-combine32.c pixman-combine64.c pixman-combine32.h pixman-com
 @USE_ARM_SIMD_TRUE@libpixman_arm_simd_la_CFLAGS = $(DEP_CFLAGS) $(ARM_SIMD_CFLAGS)
 @USE_ARM_SIMD_TRUE@libpixman_arm_simd_la_LIBADD = $(DEP_LIBS)
 @USE_ARM_NEON_TRUE@libpixman_arm_neon_la_SOURCES = \
-@USE_ARM_NEON_TRUE@        pixman-arm-neon.c
+@USE_ARM_NEON_TRUE@        pixman-arm-neon.c	\
+@USE_ARM_NEON_TRUE@        pixman-arm-neon-asm.S	\
+@USE_ARM_NEON_TRUE@        pixman-arm-neon-asm.h
 
-@USE_ARM_NEON_TRUE@libpixman_arm_neon_la_CFLAGS = $(DEP_CFLAGS) $(ARM_NEON_CFLAGS)
+@USE_ARM_NEON_TRUE@libpixman_arm_neon_la_CFLAGS = $(DEP_CFLAGS)
 @USE_ARM_NEON_TRUE@libpixman_arm_neon_la_LIBADD = $(DEP_LIBS)
 all: $(BUILT_SOURCES)
 	$(MAKE) $(AM_MAKEFLAGS) all-am
 
 .SUFFIXES:
-.SUFFIXES: .c .lo .o .obj
+.SUFFIXES: .S .c .lo .o .obj .s
 $(srcdir)/Makefile.in:  $(srcdir)/Makefile.am  $(am__configure_deps)
 	@for dep in $?; do \
 	  case '$(am__configure_deps)' in \
 	    *$$dep*) \
-	      ( cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh ) \
-	        && { if test -f $@; then exit 0; else break; fi; }; \
+	      cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh \
+		&& exit 0; \
 	      exit 1;; \
 	  esac; \
 	done; \
-	echo ' cd $(top_srcdir) && $(AUTOMAKE) --gnu  pixman/Makefile'; \
+	echo ' cd $(top_srcdir) && $(AUTOMAKE) --foreign  pixman/Makefile'; \
 	cd $(top_srcdir) && \
-	  $(AUTOMAKE) --gnu  pixman/Makefile
+	  $(AUTOMAKE) --foreign  pixman/Makefile
 .PRECIOUS: Makefile
 Makefile: $(srcdir)/Makefile.in $(top_builddir)/config.status
 	@case '$?' in \
@@ -472,35 +470,57 @@ mostlyclean-compile:
 distclean-compile:
 	-rm -f *.tab.c
 
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-access-accessors.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-access.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-bits-image.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-combine32.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-combine64.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-conical-gradient.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-cpu.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-edge-accessors.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-edge.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-fast-path.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-general.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-gradient-walker.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-image.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-implementation.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-linear-gradient.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-matrix.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-radial-gradient.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-region16.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-region32.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-solid-fill.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-timer.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-trap.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman-utils.Plo@am__quote@
-@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_1_la-pixman.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_arm_neon_la-pixman-arm-neon.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_arm_simd_la-pixman-arm-simd.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_mmx_la-pixman-mmx.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_sse2_la-pixman-sse2.Plo@am__quote@
 @AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/libpixman_vmx_la-pixman-vmx.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-access-accessors.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-access.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-arm-neon-asm.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-bits-image.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-combine32.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-combine64.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-conical-gradient.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-cpu.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-edge-accessors.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-edge.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-fast-path.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-general.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-gradient-walker.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-image.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-implementation.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-linear-gradient.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-matrix.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-radial-gradient.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-region16.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-region32.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-solid-fill.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-timer.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-trap.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman-utils.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/pixman.Plo@am__quote@
+
+.S.o:
+@am__fastdepCCAS_TRUE@	$(CPPASCOMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
+@am__fastdepCCAS_TRUE@	mv -f $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCCAS_FALSE@	source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCCAS_FALSE@	DEPDIR=$(DEPDIR) $(CCASDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCCAS_FALSE@	$(CPPASCOMPILE) -c -o $@ $<
+
+.S.obj:
+@am__fastdepCCAS_TRUE@	$(CPPASCOMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ `$(CYGPATH_W) '$<'`
+@am__fastdepCCAS_TRUE@	mv -f $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCCAS_FALSE@	source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCCAS_FALSE@	DEPDIR=$(DEPDIR) $(CCASDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCCAS_FALSE@	$(CPPASCOMPILE) -c -o $@ `$(CYGPATH_W) '$<'`
+
+.S.lo:
+@am__fastdepCCAS_TRUE@	$(LTCPPASCOMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
+@am__fastdepCCAS_TRUE@	mv -f $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Plo
+@AMDEP_TRUE@@am__fastdepCCAS_FALSE@	source='$<' object='$@' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCCAS_FALSE@	DEPDIR=$(DEPDIR) $(CCASDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCCAS_FALSE@	$(LTCPPASCOMPILE) -c -o $@ $<
 
 .c.o:
 @am__fastdepCC_TRUE@	$(COMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
@@ -523,174 +543,6 @@ distclean-compile:
 @AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
 @am__fastdepCC_FALSE@	$(LTCOMPILE) -c -o $@ $<
 
-libpixman_1_la-pixman-access.lo: pixman-access.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-access.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-access.Tpo -c -o libpixman_1_la-pixman-access.lo `test -f 'pixman-access.c' || echo '$(srcdir)/'`pixman-access.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-access.Tpo $(DEPDIR)/libpixman_1_la-pixman-access.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-access.c' object='libpixman_1_la-pixman-access.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-access.lo `test -f 'pixman-access.c' || echo '$(srcdir)/'`pixman-access.c
-
-libpixman_1_la-pixman-access-accessors.lo: pixman-access-accessors.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-access-accessors.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-access-accessors.Tpo -c -o libpixman_1_la-pixman-access-accessors.lo `test -f 'pixman-access-accessors.c' || echo '$(srcdir)/'`pixman-access-accessors.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-access-accessors.Tpo $(DEPDIR)/libpixman_1_la-pixman-access-accessors.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-access-accessors.c' object='libpixman_1_la-pixman-access-accessors.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-access-accessors.lo `test -f 'pixman-access-accessors.c' || echo '$(srcdir)/'`pixman-access-accessors.c
-
-libpixman_1_la-pixman-cpu.lo: pixman-cpu.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-cpu.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-cpu.Tpo -c -o libpixman_1_la-pixman-cpu.lo `test -f 'pixman-cpu.c' || echo '$(srcdir)/'`pixman-cpu.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-cpu.Tpo $(DEPDIR)/libpixman_1_la-pixman-cpu.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-cpu.c' object='libpixman_1_la-pixman-cpu.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-cpu.lo `test -f 'pixman-cpu.c' || echo '$(srcdir)/'`pixman-cpu.c
-
-libpixman_1_la-pixman-gradient-walker.lo: pixman-gradient-walker.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-gradient-walker.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-gradient-walker.Tpo -c -o libpixman_1_la-pixman-gradient-walker.lo `test -f 'pixman-gradient-walker.c' || echo '$(srcdir)/'`pixman-gradient-walker.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-gradient-walker.Tpo $(DEPDIR)/libpixman_1_la-pixman-gradient-walker.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-gradient-walker.c' object='libpixman_1_la-pixman-gradient-walker.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-gradient-walker.lo `test -f 'pixman-gradient-walker.c' || echo '$(srcdir)/'`pixman-gradient-walker.c
-
-libpixman_1_la-pixman-region16.lo: pixman-region16.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-region16.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-region16.Tpo -c -o libpixman_1_la-pixman-region16.lo `test -f 'pixman-region16.c' || echo '$(srcdir)/'`pixman-region16.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-region16.Tpo $(DEPDIR)/libpixman_1_la-pixman-region16.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-region16.c' object='libpixman_1_la-pixman-region16.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-region16.lo `test -f 'pixman-region16.c' || echo '$(srcdir)/'`pixman-region16.c
-
-libpixman_1_la-pixman-region32.lo: pixman-region32.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-region32.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-region32.Tpo -c -o libpixman_1_la-pixman-region32.lo `test -f 'pixman-region32.c' || echo '$(srcdir)/'`pixman-region32.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-region32.Tpo $(DEPDIR)/libpixman_1_la-pixman-region32.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-region32.c' object='libpixman_1_la-pixman-region32.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-region32.lo `test -f 'pixman-region32.c' || echo '$(srcdir)/'`pixman-region32.c
-
-libpixman_1_la-pixman-image.lo: pixman-image.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-image.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-image.Tpo -c -o libpixman_1_la-pixman-image.lo `test -f 'pixman-image.c' || echo '$(srcdir)/'`pixman-image.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-image.Tpo $(DEPDIR)/libpixman_1_la-pixman-image.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-image.c' object='libpixman_1_la-pixman-image.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-image.lo `test -f 'pixman-image.c' || echo '$(srcdir)/'`pixman-image.c
-
-libpixman_1_la-pixman-implementation.lo: pixman-implementation.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-implementation.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-implementation.Tpo -c -o libpixman_1_la-pixman-implementation.lo `test -f 'pixman-implementation.c' || echo '$(srcdir)/'`pixman-implementation.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-implementation.Tpo $(DEPDIR)/libpixman_1_la-pixman-implementation.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-implementation.c' object='libpixman_1_la-pixman-implementation.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-implementation.lo `test -f 'pixman-implementation.c' || echo '$(srcdir)/'`pixman-implementation.c
-
-libpixman_1_la-pixman-combine32.lo: pixman-combine32.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-combine32.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-combine32.Tpo -c -o libpixman_1_la-pixman-combine32.lo `test -f 'pixman-combine32.c' || echo '$(srcdir)/'`pixman-combine32.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-combine32.Tpo $(DEPDIR)/libpixman_1_la-pixman-combine32.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-combine32.c' object='libpixman_1_la-pixman-combine32.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-combine32.lo `test -f 'pixman-combine32.c' || echo '$(srcdir)/'`pixman-combine32.c
-
-libpixman_1_la-pixman-combine64.lo: pixman-combine64.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-combine64.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-combine64.Tpo -c -o libpixman_1_la-pixman-combine64.lo `test -f 'pixman-combine64.c' || echo '$(srcdir)/'`pixman-combine64.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-combine64.Tpo $(DEPDIR)/libpixman_1_la-pixman-combine64.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-combine64.c' object='libpixman_1_la-pixman-combine64.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-combine64.lo `test -f 'pixman-combine64.c' || echo '$(srcdir)/'`pixman-combine64.c
-
-libpixman_1_la-pixman-general.lo: pixman-general.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-general.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-general.Tpo -c -o libpixman_1_la-pixman-general.lo `test -f 'pixman-general.c' || echo '$(srcdir)/'`pixman-general.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-general.Tpo $(DEPDIR)/libpixman_1_la-pixman-general.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-general.c' object='libpixman_1_la-pixman-general.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-general.lo `test -f 'pixman-general.c' || echo '$(srcdir)/'`pixman-general.c
-
-libpixman_1_la-pixman.lo: pixman.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman.Tpo -c -o libpixman_1_la-pixman.lo `test -f 'pixman.c' || echo '$(srcdir)/'`pixman.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman.Tpo $(DEPDIR)/libpixman_1_la-pixman.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman.c' object='libpixman_1_la-pixman.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman.lo `test -f 'pixman.c' || echo '$(srcdir)/'`pixman.c
-
-libpixman_1_la-pixman-fast-path.lo: pixman-fast-path.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-fast-path.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-fast-path.Tpo -c -o libpixman_1_la-pixman-fast-path.lo `test -f 'pixman-fast-path.c' || echo '$(srcdir)/'`pixman-fast-path.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-fast-path.Tpo $(DEPDIR)/libpixman_1_la-pixman-fast-path.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-fast-path.c' object='libpixman_1_la-pixman-fast-path.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-fast-path.lo `test -f 'pixman-fast-path.c' || echo '$(srcdir)/'`pixman-fast-path.c
-
-libpixman_1_la-pixman-solid-fill.lo: pixman-solid-fill.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-solid-fill.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-solid-fill.Tpo -c -o libpixman_1_la-pixman-solid-fill.lo `test -f 'pixman-solid-fill.c' || echo '$(srcdir)/'`pixman-solid-fill.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-solid-fill.Tpo $(DEPDIR)/libpixman_1_la-pixman-solid-fill.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-solid-fill.c' object='libpixman_1_la-pixman-solid-fill.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-solid-fill.lo `test -f 'pixman-solid-fill.c' || echo '$(srcdir)/'`pixman-solid-fill.c
-
-libpixman_1_la-pixman-conical-gradient.lo: pixman-conical-gradient.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-conical-gradient.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-conical-gradient.Tpo -c -o libpixman_1_la-pixman-conical-gradient.lo `test -f 'pixman-conical-gradient.c' || echo '$(srcdir)/'`pixman-conical-gradient.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-conical-gradient.Tpo $(DEPDIR)/libpixman_1_la-pixman-conical-gradient.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-conical-gradient.c' object='libpixman_1_la-pixman-conical-gradient.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-conical-gradient.lo `test -f 'pixman-conical-gradient.c' || echo '$(srcdir)/'`pixman-conical-gradient.c
-
-libpixman_1_la-pixman-linear-gradient.lo: pixman-linear-gradient.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-linear-gradient.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-linear-gradient.Tpo -c -o libpixman_1_la-pixman-linear-gradient.lo `test -f 'pixman-linear-gradient.c' || echo '$(srcdir)/'`pixman-linear-gradient.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-linear-gradient.Tpo $(DEPDIR)/libpixman_1_la-pixman-linear-gradient.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-linear-gradient.c' object='libpixman_1_la-pixman-linear-gradient.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-linear-gradient.lo `test -f 'pixman-linear-gradient.c' || echo '$(srcdir)/'`pixman-linear-gradient.c
-
-libpixman_1_la-pixman-radial-gradient.lo: pixman-radial-gradient.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-radial-gradient.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-radial-gradient.Tpo -c -o libpixman_1_la-pixman-radial-gradient.lo `test -f 'pixman-radial-gradient.c' || echo '$(srcdir)/'`pixman-radial-gradient.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-radial-gradient.Tpo $(DEPDIR)/libpixman_1_la-pixman-radial-gradient.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-radial-gradient.c' object='libpixman_1_la-pixman-radial-gradient.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-radial-gradient.lo `test -f 'pixman-radial-gradient.c' || echo '$(srcdir)/'`pixman-radial-gradient.c
-
-libpixman_1_la-pixman-bits-image.lo: pixman-bits-image.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-bits-image.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-bits-image.Tpo -c -o libpixman_1_la-pixman-bits-image.lo `test -f 'pixman-bits-image.c' || echo '$(srcdir)/'`pixman-bits-image.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-bits-image.Tpo $(DEPDIR)/libpixman_1_la-pixman-bits-image.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-bits-image.c' object='libpixman_1_la-pixman-bits-image.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-bits-image.lo `test -f 'pixman-bits-image.c' || echo '$(srcdir)/'`pixman-bits-image.c
-
-libpixman_1_la-pixman-utils.lo: pixman-utils.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-utils.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-utils.Tpo -c -o libpixman_1_la-pixman-utils.lo `test -f 'pixman-utils.c' || echo '$(srcdir)/'`pixman-utils.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-utils.Tpo $(DEPDIR)/libpixman_1_la-pixman-utils.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-utils.c' object='libpixman_1_la-pixman-utils.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-utils.lo `test -f 'pixman-utils.c' || echo '$(srcdir)/'`pixman-utils.c
-
-libpixman_1_la-pixman-edge.lo: pixman-edge.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-edge.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-edge.Tpo -c -o libpixman_1_la-pixman-edge.lo `test -f 'pixman-edge.c' || echo '$(srcdir)/'`pixman-edge.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-edge.Tpo $(DEPDIR)/libpixman_1_la-pixman-edge.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-edge.c' object='libpixman_1_la-pixman-edge.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-edge.lo `test -f 'pixman-edge.c' || echo '$(srcdir)/'`pixman-edge.c
-
-libpixman_1_la-pixman-edge-accessors.lo: pixman-edge-accessors.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-edge-accessors.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-edge-accessors.Tpo -c -o libpixman_1_la-pixman-edge-accessors.lo `test -f 'pixman-edge-accessors.c' || echo '$(srcdir)/'`pixman-edge-accessors.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-edge-accessors.Tpo $(DEPDIR)/libpixman_1_la-pixman-edge-accessors.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-edge-accessors.c' object='libpixman_1_la-pixman-edge-accessors.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-edge-accessors.lo `test -f 'pixman-edge-accessors.c' || echo '$(srcdir)/'`pixman-edge-accessors.c
-
-libpixman_1_la-pixman-trap.lo: pixman-trap.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-trap.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-trap.Tpo -c -o libpixman_1_la-pixman-trap.lo `test -f 'pixman-trap.c' || echo '$(srcdir)/'`pixman-trap.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-trap.Tpo $(DEPDIR)/libpixman_1_la-pixman-trap.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-trap.c' object='libpixman_1_la-pixman-trap.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-trap.lo `test -f 'pixman-trap.c' || echo '$(srcdir)/'`pixman-trap.c
-
-libpixman_1_la-pixman-timer.lo: pixman-timer.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-timer.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-timer.Tpo -c -o libpixman_1_la-pixman-timer.lo `test -f 'pixman-timer.c' || echo '$(srcdir)/'`pixman-timer.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-timer.Tpo $(DEPDIR)/libpixman_1_la-pixman-timer.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-timer.c' object='libpixman_1_la-pixman-timer.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-timer.lo `test -f 'pixman-timer.c' || echo '$(srcdir)/'`pixman-timer.c
-
-libpixman_1_la-pixman-matrix.lo: pixman-matrix.c
-@am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -MT libpixman_1_la-pixman-matrix.lo -MD -MP -MF $(DEPDIR)/libpixman_1_la-pixman-matrix.Tpo -c -o libpixman_1_la-pixman-matrix.lo `test -f 'pixman-matrix.c' || echo '$(srcdir)/'`pixman-matrix.c
-@am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_1_la-pixman-matrix.Tpo $(DEPDIR)/libpixman_1_la-pixman-matrix.Plo
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='pixman-matrix.c' object='libpixman_1_la-pixman-matrix.lo' libtool=yes @AMDEPBACKSLASH@
-@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
-@am__fastdepCC_FALSE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_1_la_CFLAGS) $(CFLAGS) -c -o libpixman_1_la-pixman-matrix.lo `test -f 'pixman-matrix.c' || echo '$(srcdir)/'`pixman-matrix.c
-
 libpixman_arm_neon_la-pixman-arm-neon.lo: pixman-arm-neon.c
 @am__fastdepCC_TRUE@	$(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(libpixman_arm_neon_la_CFLAGS) $(CFLAGS) -MT libpixman_arm_neon_la-pixman-arm-neon.lo -MD -MP -MF $(DEPDIR)/libpixman_arm_neon_la-pixman-arm-neon.Tpo -c -o libpixman_arm_neon_la-pixman-arm-neon.lo `test -f 'pixman-arm-neon.c' || echo '$(srcdir)/'`pixman-arm-neon.c
 @am__fastdepCC_TRUE@	mv -f $(DEPDIR)/libpixman_arm_neon_la-pixman-arm-neon.Tpo $(DEPDIR)/libpixman_arm_neon_la-pixman-arm-neon.Plo
@@ -754,7 +606,7 @@ ID: $(HEADERS) $(SOURCES) $(LISP) $(TAGS_FILES)
 	unique=`for i in $$list; do \
 	    if test -f "$$i"; then echo $$i; else echo $(srcdir)/$$i; fi; \
 	  done | \
-	  $(AWK) '{ files[$$0] = 1; nonempty = 1; } \
+	  $(AWK) '{ files[$$0] = 1; nonemtpy = 1; } \
 	      END { if (nonempty) { for (i in files) print i; }; }'`; \
 	mkid -fID $$unique
 tags: TAGS
@@ -945,6 +797,19 @@ pixman-combine64.c : pixman-combine.c.template pixman-combine64.h make-combine.p
 	$(PERL) $(srcdir)/make-combine.pl 16 < $(srcdir)/pixman-combine.c.template > $@ || ($(RM) $@; exit 1)
 pixman-combine64.h : pixman-combine.h.template make-combine.pl
 	$(PERL) $(srcdir)/make-combine.pl 16 < $(srcdir)/pixman-combine.h.template > $@ || ($(RM) $@; exit 1)
+
+@USE_MMX_TRUE@$(libpixman_mmx_la_SOURCES:.c=.s) : ASM_CFLAGS=$(MMX_CFLAGS)
+
+@USE_VMX_TRUE@$(libpixman_vmx_la_SOURCES:.c=.s) : ASM_CFLAGS=$(VMX_CFLAGS)
+
+@USE_SSE2_TRUE@$(libpixman_sse2_la_SOURCES:.c=.s) : ASM_CFLAGS=$(SSE2_CFLAGS)
+
+@USE_ARM_SIMD_TRUE@$(libpixman_arm_simd_la_SOURCES:.c=.s) : ASM_CFLAGS=$(ARM_SIMD_CFLAGS)
+
+@USE_ARM_NEON_TRUE@$(libpixman_arm_neon_la_SOURCES:.c=.s) : ASM_CFLAGS=
+
+.c.s : $(libpixmaninclude_HEADERS) $(BUILT_SOURCES)
+	$(CC) $(CFLAGS) $(ASM_CFLAGS) -DHAVE_CONFIG_H -I$(srcdir) -I$(builddir) -I$(top_builddir) -S -o $@ $<
 # Tell versions [3.59,3.63) of GNU make to not export all variables.
 # Otherwise a system limit (for SysV at least) may be exceeded.
 .NOEXPORT:
diff --git a/pixman/pixman/makefile b/pixman/pixman/makefile
index d7380bad7..8209dc394 100644
--- a/pixman/pixman/makefile
+++ b/pixman/pixman/makefile
@@ -1,5 +1,4 @@
 LIBRARY = libpixman-1
-DEFINES += PIXMAN_DISABLE_DEPRECATED
 
 CSRCS =			\
 	pixman-access.c				\
diff --git a/pixman/pixman/pixman-access.c b/pixman/pixman/pixman-access.c
index d9fd38c15..389cf2a76 100644
--- a/pixman/pixman/pixman-access.c
+++ b/pixman/pixman/pixman-access.c
@@ -180,11 +180,11 @@ fetch_scanline_b8g8r8a8 (pixman_image_t *image,
     const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
     const uint32_t *pixel = (uint32_t *)bits + x;
     const uint32_t *end = pixel + width;
-    
+
     while (pixel < end)
     {
 	uint32_t p = READ (image, pixel++);
-	
+
 	*buffer++ = (((p & 0xff000000) >> 24)	|
 	             ((p & 0x00ff0000) >> 8)	|
 	             ((p & 0x0000ff00) << 8)	|
@@ -731,23 +731,27 @@ fetch_scanline_b2g3r3 (pixman_image_t *image,
     const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
     const uint8_t *pixel = (const uint8_t *)bits + x;
     const uint8_t *end = pixel + width;
-    
+
     while (pixel < end)
     {
 	uint32_t p = READ (image, pixel++);
 	uint32_t r, g, b;
-	
-	b = (((p & 0xc0)     ) |
-	     ((p & 0xc0) >> 2) |
-	     ((p & 0xc0) >> 4) |
-	     ((p & 0xc0) >> 6));
-	
-	g = ((p & 0x38) | ((p & 0x38) >> 3) | ((p & 0x30) << 2)) << 8;
-	
-	r = (((p & 0x07)     ) |
-	     ((p & 0x07) << 3) |
-	     ((p & 0x06) << 6)) << 16;
-	
+
+	b  = p & 0xc0;
+	b |= b >> 2;
+	b |= b >> 4;
+	b &= 0xff;
+
+	g  = (p & 0x38) << 10;
+	g |= g >> 3;
+	g |= g >> 6;
+	g &= 0xff00;
+
+	r  = (p & 0x7) << 21;
+	r |= r >> 3;
+	r |= r >> 6;
+	r &= 0xff0000;
+
 	*buffer++ = 0xff000000 | r | g | b;
     }
 }
@@ -798,7 +802,7 @@ fetch_scanline_a2b2g2r2 (pixman_image_t *image,
 	uint32_t a, r, g, b;
 	
 	a = ((p & 0xc0) * 0x55) << 18;
-	b = ((p & 0x30) * 0x55) >> 6;
+	b = ((p & 0x30) * 0x55) >> 4;
 	g = ((p & 0x0c) * 0x55) << 6;
 	r = ((p & 0x03) * 0x55) << 16;
 	
@@ -840,20 +844,22 @@ fetch_scanline_x4a4 (pixman_image_t *image,
     const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
     const uint8_t *pixel = (const uint8_t *)bits + x;
     const uint8_t *end = pixel + width;
-    
+   
     while (pixel < end)
     {
 	uint8_t p = READ (image, pixel++) & 0xf;
-	
+
 	*buffer++ = (p | (p << 4)) << 24;
     }
 }
 
-#define FETCH_8(img,l,o)    (READ (img, (uint8_t *)(l) + ((o) >> 2)))
+#define FETCH_8(img,l,o)    (READ (img, (((uint8_t *)(l)) + ((o) >> 3))))
 #ifdef WORDS_BIGENDIAN
-#define FETCH_4(img,l,o)    ((o) & 2 ? FETCH_8 (img,l,o) & 0xf : FETCH_8 (img,l,o) >> 4)
+#define FETCH_4(img,l,o)						\
+    (((4 * (o)) & 4) ? (FETCH_8 (img,l, 4 * (o)) & 0xf) : (FETCH_8 (img,l,(4 * (o))) >> 4))
 #else
-#define FETCH_4(img,l,o)    ((o) & 2 ? FETCH_8 (img,l,o) >> 4 : FETCH_8 (img,l,o) & 0xf)
+#define FETCH_4(img,l,o)						\
+    (((4 * (o)) & 4) ? (FETCH_8 (img, l, 4 * (o)) >> 4) : (FETCH_8 (img, l, (4 * (o))) & 0xf))
 #endif
 
 static void
@@ -867,13 +873,13 @@ fetch_scanline_a4 (pixman_image_t *image,
 {
     const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
     int i;
-    
+
     for (i = 0; i < width; ++i)
     {
 	uint32_t p = FETCH_4 (image, bits, i + x);
-	
+
 	p |= p << 4;
-	
+
 	*buffer++ = p << 24;
     }
 }
@@ -923,7 +929,7 @@ fetch_scanline_b1g2r1 (pixman_image_t *image,
 	b = ((p & 0x8) * 0xff) >> 3;
 	g = ((p & 0x6) * 0x55) << 7;
 	r = ((p & 0x1) * 0xff) << 16;
-	
+
 	*buffer++ = 0xff000000 | r | g | b;
     }
 }
@@ -940,16 +946,16 @@ fetch_scanline_a1r1g1b1 (pixman_image_t *image,
     uint32_t a, r, g, b;
     const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
     int i;
-    
+
     for (i = 0; i < width; ++i)
     {
 	uint32_t p = FETCH_4 (image, bits, i + x);
-	
+
 	a = ((p & 0x8) * 0xff) << 21;
 	r = ((p & 0x4) * 0xff) << 14;
 	g = ((p & 0x2) * 0xff) << 7;
 	b = ((p & 0x1) * 0xff);
-	
+
 	*buffer++ = a | r | g | b;
     }
 }
@@ -965,17 +971,17 @@ fetch_scanline_a1b1g1r1 (pixman_image_t *image,
 {
     const uint32_t *bits = image->bits.bits + y * image->bits.rowstride;
     int i;
-    
+
     for (i = 0; i < width; ++i)
     {
 	uint32_t p = FETCH_4 (image, bits, i + x);
 	uint32_t a, r, g, b;
-	
+
 	a = ((p & 0x8) * 0xff) << 21;
-	r = ((p & 0x4) * 0xff) >> 3;
+	b = ((p & 0x4) * 0xff) >> 2;
 	g = ((p & 0x2) * 0xff) << 7;
-	b = ((p & 0x1) * 0xff) << 16;
-	
+	r = ((p & 0x1) * 0xff) << 16;
+
 	*buffer++ = a | r | g | b;
     }
 }
@@ -1546,23 +1552,25 @@ fetch_pixel_b2g3r3 (bits_image_t *image,
 		    int           line)
 {
     uint32_t *bits = image->bits + line * image->rowstride;
-    uint32_t pixel = READ (image, (uint8_t *) bits + offset);
+    uint32_t p = READ (image, (uint8_t *) bits + offset);
     uint32_t r, g, b;
-    
-    b = ((pixel & 0xc0)         |
-	 ((pixel & 0xc0) >> 2)  |
-	 ((pixel & 0xc0) >> 4)  |
-	 ((pixel & 0xc0) >> 6));
-    
-    g = ((pixel & 0x38)         |
-	 ((pixel & 0x38) >> 3)  |
-	 ((pixel & 0x30) << 2)) << 8;
-    
-    r = ((pixel & 0x07)         |
-	 ((pixel & 0x07) << 3)  |
-	 ((pixel & 0x06) << 6)) << 16;
-    
-    return (0xff000000 | r | g | b);
+
+    b  = p & 0xc0;
+    b |= b >> 2;
+    b |= b >> 4;
+    b &= 0xff;
+
+    g  = (p & 0x38) << 10;
+    g |= g >> 3;
+    g |= g >> 6;
+    g &= 0xff00;
+
+    r  = (p & 0x7) << 21;
+    r |= r >> 3;
+    r |= r >> 6;
+    r &= 0xff0000;
+
+    return 0xff000000 | r | g | b;
 }
 
 static uint32_t
@@ -1592,7 +1600,7 @@ fetch_pixel_a2b2g2r2 (bits_image_t *image,
     uint32_t a, r, g, b;
     
     a = ((pixel & 0xc0) * 0x55) << 18;
-    b = ((pixel & 0x30) * 0x55) >> 6;
+    b = ((pixel & 0x30) * 0x55) >> 4;
     g = ((pixel & 0x0c) * 0x55) << 6;
     r = ((pixel & 0x03) * 0x55) << 16;
     
@@ -1674,12 +1682,12 @@ fetch_pixel_a1r1g1b1 (bits_image_t *image,
     uint32_t *bits = image->bits + line * image->rowstride;
     uint32_t pixel = FETCH_4 (image, bits, offset);
     uint32_t a, r, g, b;
-    
+
     a = ((pixel & 0x8) * 0xff) << 21;
     r = ((pixel & 0x4) * 0xff) << 14;
     g = ((pixel & 0x2) * 0xff) << 7;
     b = ((pixel & 0x1) * 0xff);
-    
+
     return a | r | g | b;
 }
 
@@ -1691,12 +1699,12 @@ fetch_pixel_a1b1g1r1 (bits_image_t *image,
     uint32_t *bits = image->bits + line * image->rowstride;
     uint32_t pixel = FETCH_4 (image, bits, offset);
     uint32_t a, r, g, b;
-    
+
     a = ((pixel & 0x8) * 0xff) << 21;
-    r = ((pixel & 0x4) * 0xff) >> 3;
+    b = ((pixel & 0x4) * 0xff) >> 2;
     g = ((pixel & 0x2) * 0xff) << 7;
-    b = ((pixel & 0x1) * 0xff) << 16;
-    
+    r = ((pixel & 0x1) * 0xff) << 16;
+
     return a | r | g | b;
 }
 
@@ -1708,7 +1716,7 @@ fetch_pixel_c4 (bits_image_t *image,
     uint32_t *bits = image->bits + line * image->rowstride;
     uint32_t pixel = FETCH_4 (image, bits, offset);
     const pixman_indexed_t * indexed = image->indexed;
-    
+
     return indexed->rgba[pixel];
 }
 
@@ -2425,22 +2433,32 @@ store_scanline_x4a4 (bits_image_t *  image,
     uint32_t *bits = image->bits + image->rowstride * y;
     uint8_t   *pixel = ((uint8_t *) bits) + x;
     int i;
-    
+
     for (i = 0; i < width; ++i)
 	WRITE (image, pixel++, values[i] >> 28);
 }
 
 #define STORE_8(img,l,o,v)  (WRITE (img, (uint8_t *)(l) + ((o) >> 3), (v)))
 #ifdef WORDS_BIGENDIAN
-#define STORE_4(img,l,o,v)					    \
-    STORE_8 (img,l,o,((o) & 4 ?					    \
-                      (FETCH_8 (img,l,o) & 0xf0) | (v) :            \
-                      (FETCH_8 (img,l,o) & 0x0f) | ((v) << 4)))
+
+#define STORE_4(img,l,o,v)						\
+    do									\
+    {									\
+	int bo = 4 * (o);						\
+	STORE_8 (img, l, bo, (bo & 4 ?					\
+			      (FETCH_8 (img, l, bo) & 0xf0) | (v) :	\
+			      (FETCH_8 (img, l, bo) & 0x0f) | ((v) << 4))); \
+    } while (0)
 #else
-#define STORE_4(img,l,o,v)					\
-    STORE_8 (img,l,o,((o) & 4 ?					\
-                      (FETCH_8 (img,l,o) & 0x0f) | ((v) << 4) : \
-                      (FETCH_8 (img,l,o) & 0xf0) | (v)))
+
+#define STORE_4(img,l,o,v)						\
+    do									\
+    {									\
+	int bo = 4 * (o);						\
+	STORE_8 (img, l, bo, (bo & 4 ?					\
+			      (FETCH_8 (img, l, bo) & 0x0f) | ((v) << 4) : \
+			      (FETCH_8 (img, l, bo) & 0xf0) | (v)));	\
+    } while (0)
 #endif
 
 static void
@@ -2452,7 +2470,7 @@ store_scanline_a4 (bits_image_t *  image,
 {
     uint32_t *bits = image->bits + image->rowstride * y;
     int i;
-    
+
     for (i = 0; i < width; ++i)
 	STORE_4 (image, bits, i + x, values[i] >> 28);
 }
@@ -2488,11 +2506,11 @@ store_scanline_b1g2r1 (bits_image_t *  image,
 {
     uint32_t *bits = image->bits + image->rowstride * y;
     int i;
-    
+
     for (i = 0; i < width; ++i)
     {
 	uint32_t pixel;
-	
+
 	SPLIT (values[i]);
 	pixel = (((b >> 4) & 0x8) |
 	         ((g >> 5) & 0x6) |
@@ -2510,16 +2528,17 @@ store_scanline_a1r1g1b1 (bits_image_t *  image,
 {
     uint32_t *bits = image->bits + image->rowstride * y;
     int i;
-    
+
     for (i = 0; i < width; ++i)
     {
 	uint32_t pixel;
-	
+
 	SPLIT_A (values[i]);
 	pixel = (((a >> 4) & 0x8) |
 	         ((r >> 5) & 0x4) |
 	         ((g >> 6) & 0x2) |
 	         ((b >> 7)      ));
+
 	STORE_4 (image, bits, i + x, pixel);
     }
 }
@@ -2533,16 +2552,17 @@ store_scanline_a1b1g1r1 (bits_image_t *  image,
 {
     uint32_t *bits = image->bits + image->rowstride * y;
     int i;
-    
+
     for (i = 0; i < width; ++i)
     {
 	uint32_t pixel;
-	
+
 	SPLIT_A (values[i]);
 	pixel = (((a >> 4) & 0x8) |
 	         ((b >> 5) & 0x4) |
 	         ((g >> 6) & 0x2) |
 	         ((r >> 7)      ));
+
 	STORE_4 (image, bits, i + x, pixel);
     }
 }
diff --git a/pixman/pixman/pixman-arm-neon-asm.S b/pixman/pixman/pixman-arm-neon-asm.S
new file mode 100644
index 000000000..e8ccf77a9
--- /dev/null
+++ b/pixman/pixman/pixman-arm-neon-asm.S
@@ -0,0 +1,1051 @@
+/*
+ * Copyright © 2009 Nokia Corporation
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and its
+ * documentation for any purpose is hereby granted without fee, provided that
+ * the above copyright notice appear in all copies and that both that
+ * copyright notice and this permission notice appear in supporting
+ * documentation, and that the name of Nokia Corporation not be used in
+ * advertising or publicity pertaining to distribution of the software without
+ * specific, written prior permission.  Nokia Corporation makes no
+ * representations about the suitability of this software for any purpose.
+ * It is provided "as is" without express or implied warranty.
+ *
+ * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
+ * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
+ * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
+ * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
+ * SOFTWARE.
+ *
+ * Author:  Siarhei Siamashka (siarhei.siamashka@nokia.com)
+ */
+
+/*
+ * This file contains implementations of NEON optimized pixel processing
+ * functions. There is no full and detailed tutorial, but some functions
+ * (those which are exposing some new or interesting features) are
+ * extensively commented and can be used as examples.
+ *
+ * You may want to have a look at the comments for following functions:
+ *  - pixman_composite_over_8888_0565_asm_neon
+ *  - pixman_composite_over_n_8_0565_asm_neon
+ */
+
+/* Prevent the stack from becoming executable for no reason... */
+#if defined(__linux__) && defined(__ELF__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+    .text
+    .fpu neon
+    .altmacro
+
+#include "pixman-arm-neon-asm.h"
+
+/* Global configuration options and preferences */
+
+/*
+ * The code can optionally make use of unaligned memory accesses to improve
+ * performance of handling leading/trailing pixels for each scanline.
+ * Configuration variable RESPECT_STRICT_ALIGNMENT can be set to 0 for
+ * example in linux if unaligned memory accesses are not configured to
+ * generate.exceptions.
+ */
+.set RESPECT_STRICT_ALIGNMENT, 1
+
+/*
+ * Set default prefetch type. There is a choice between the following options:
+ *
+ * PREFETCH_TYPE_NONE (may be useful for the ARM cores where PLD is set to work
+ * as NOP to workaround some HW bugs or for whatever other reason)
+ *
+ * PREFETCH_TYPE_SIMPLE (may be useful for simple single-issue ARM cores where
+ * advanced prefetch intruduces heavy overhead)
+ *
+ * PREFETCH_TYPE_ADVANCED (useful for superscalar cores such as ARM Cortex-A8
+ * which can run ARM and NEON instructions simultaneously so that extra ARM
+ * instructions do not add (many) extra cycles, but improve prefetch efficiency)
+ *
+ * Note: some types of function can't support advanced prefetch and fallback
+ *       to simple one (those which handle 24bpp pixels)
+ */
+.set PREFETCH_TYPE_DEFAULT, PREFETCH_TYPE_ADVANCED
+
+/* Prefetch distance in pixels for simple prefetch */
+.set PREFETCH_DISTANCE_SIMPLE, 64
+
+/*
+ * Implementation of pixman_composite_over_8888_0565_asm_neon
+ *
+ * This function takes a8r8g8b8 source buffer, r5g6b5 destination buffer and
+ * performs OVER compositing operation. Function fast_composite_over_8888_0565
+ * from pixman-fast-path.c does the same in C and can be used as a reference.
+ *
+ * First we need to have some NEON assembly code which can do the actual
+ * operation on the pixels and provide it to the template macro.
+ *
+ * Template macro quite conveniently takes care of emitting all the necessary
+ * code for memory reading and writing (including quite tricky cases of
+ * handling unaligned leading/trailing pixels), so we only need to deal with
+ * the data in NEON registers.
+ *
+ * NEON registers allocation in general is recommented to be the following:
+ * d0,  d1,  d2,  d3  - contain loaded source pixel data
+ * d4,  d5,  d6,  d7  - contain loaded destination pixels (if they are needed)
+ * d24, d25, d26, d27 - contain loading mask pixel data (if mask is used)
+ * d28, d29, d30, d31 - place for storing the result (destination pixels)
+ *
+ * As can be seen above, four 64-bit NEON registers are used for keeping
+ * intermediate pixel data and up to 8 pixels can be processed in one step
+ * for 32bpp formats (16 pixels for 16bpp, 32 pixels for 8bpp).
+ *
+ * This particular function uses the following registers allocation:
+ * d0,  d1,  d2,  d3  - contain loaded source pixel data
+ * d4,  d5            - contain loaded destination pixels (they are needed)
+ * d28, d29           - place for storing the result (destination pixels)
+ */
+
+/*
+ * Step one. We need to have some code to do some arithmetics on pixel data.
+ * This is implemented as a pair of macros: '*_head' and '*_tail'. When used
+ * back-to-back, they take pixel data from {d0, d1, d2, d3} and {d4, d5},
+ * perform all the needed calculations and write the result to {d28, d29}.
+ * The rationale for having two macros and not just one will be explained
+ * later. In practice, any single monolitic function which does the work can
+ * be split into two parts in any arbitrary way without affecting correctness.
+ *
+ * There is one special trick here too. Common template macro can optionally
+ * make our life a bit easier by doing R, G, B, A color components
+ * deinterleaving for 32bpp pixel formats (and this feature is used in
+ * 'pixman_composite_over_8888_0565_asm_neon' function). So it means that
+ * instead of having 8 packed pixels in {d0, d1, d2, d3} registers, we
+ * actually use d0 register for blue channel (a vector of eight 8-bit
+ * values), d1 register for green, d2 for red and d3 for alpha. This
+ * simple conversion can be also done with a few NEON instructions:
+ *
+ * Packed to planar conversion:
+ *  vuzp.8 d0, d1
+ *  vuzp.8 d2, d3
+ *  vuzp.8 d1, d3
+ *  vuzp.8 d0, d2
+ *
+ * Planar to packed conversion:
+ *  vzip.8 d0, d2
+ *  vzip.8 d1, d3
+ *  vzip.8 d2, d3
+ *  vzip.8 d0, d1
+ *
+ * But pixel can be loaded directly in planar format using VLD4.8 NEON
+ * instruction. It is 1 cycle slower than VLD1.32, so this is not always
+ * desirable, that's why deinterleaving is optional.
+ *
+ * But anyway, here is the code:
+ */
+.macro pixman_composite_over_8888_0565_process_pixblock_head
+    /* convert 8 r5g6b5 pixel data from {d4, d5} to planar 8-bit format
+       and put data into d6 - red, d7 - green, d30 - blue */
+    vshrn.u16   d6, q2, #8
+    vshrn.u16   d7, q2, #3
+    vsli.u16    q2, q2, #5
+    vsri.u8     d6, d6, #5
+    vmvn.8      d3, d3      /* invert source alpha */
+    vsri.u8     d7, d7, #6
+    vshrn.u16   d30, q2, #2
+    /* now do alpha blending, storing results in 8-bit planar format
+       into d16 - red, d19 - green, d18 - blue */
+    vmull.u8    q10, d3, d6
+    vmull.u8    q11, d3, d7
+    vmull.u8    q12, d3, d30
+    vrshr.u16   q13, q10, #8
+    vrshr.u16   q3, q11, #8
+    vrshr.u16   q15, q12, #8
+    vraddhn.u16 d20, q10, q13
+    vraddhn.u16 d23, q11, q3
+    vraddhn.u16 d22, q12, q15
+.endm
+
+.macro pixman_composite_over_8888_0565_process_pixblock_tail
+    /* ... continue alpha blending */
+    vqadd.u8    d16, d2, d20
+    vqadd.u8    q9, q0, q11
+    /* convert the result to r5g6b5 and store it into {d28, d29} */
+    vshll.u8    q14, d16, #8
+    vshll.u8    q8, d19, #8
+    vshll.u8    q9, d18, #8
+    vsri.u16    q14, q8, #5
+    vsri.u16    q14, q9, #11
+.endm
+
+/*
+ * OK, now we got almost everything that we need. Using the above two
+ * macros, the work can be done right. But now we want to optimize
+ * it a bit. ARM Cortex-A8 is an in-order core, and benefits really
+ * a lot from good code scheduling and software pipelining.
+ *
+ * Let's construct some code, which will run in the core main loop.
+ * Some pseudo-code of the main loop will look like this:
+ *   head
+ *   while (...) {
+ *     tail
+ *     head
+ *   }
+ *   tail
+ *
+ * It may look a bit weird, but this setup allows to hide instruction
+ * latencies better and also utilize dual-issue capability more
+ * efficiently (make pairs of load-store and ALU instructions).
+ *
+ * So what we need now is a '*_tail_head' macro, which will be used
+ * in the core main loop. A trivial straightforward implementation
+ * of this macro would look like this:
+ *
+ *   pixman_composite_over_8888_0565_process_pixblock_tail
+ *   vst1.16     {d28, d29}, [DST_W, :128]!
+ *   vld1.16     {d4, d5}, [DST_R, :128]!
+ *   vld4.32     {d0, d1, d2, d3}, [SRC]!
+ *   pixman_composite_over_8888_0565_process_pixblock_head
+ *   cache_preload 8, 8
+ *
+ * Now it also got some VLD/VST instructions. We simply can't move from
+ * processing one block of pixels to the other one with just arithmetics.
+ * The previously processed data needs to be written to memory and new
+ * data needs to be fetched. Fortunately, this main loop does not deal
+ * with partial leading/trailing pixels and can load/store a full block
+ * of pixels in a bulk. Additionally, destination buffer is already
+ * 16 bytes aligned here (which is good for performance).
+ *
+ * New things here are DST_R, DST_W, SRC and MASK identifiers. These
+ * are the aliases for ARM registers which are used as pointers for
+ * accessing data. We maintain separate pointers for reading and writing
+ * destination buffer (DST_R and DST_W).
+ *
+ * Another new thing is 'cache_preload' macro. It is used for prefetching
+ * data into CPU L2 cache and improve performance when dealing with large
+ * images which are far larger than cache size. It uses one argument
+ * (actually two, but they need to be the same here) - number of pixels
+ * in a block. Looking into 'pixman-arm-neon-asm.h' can provide some
+ * details about this macro. Moreover, if good performance is needed
+ * the code from this macro needs to be copied into '*_tail_head' macro
+ * and mixed with the rest of code for optimal instructions scheduling.
+ * We are actually doing it below.
+ *
+ * Now after all the explanations, here is the optimized code.
+ * Different instruction streams (originaling from '*_head', '*_tail'
+ * and 'cache_preload' macro) use different indentation levels for
+ * better readability. Actually taking the code from one of these
+ * indentation levels and ignoring a few VLD/VST instructions would
+ * result in exactly the code from '*_head', '*_tail' or 'cache_preload'
+ * macro!
+ */
+
+#if 1
+
+.macro pixman_composite_over_8888_0565_process_pixblock_tail_head
+        vqadd.u8    d16, d2, d20
+    vld1.16     {d4, d5}, [DST_R, :128]!
+        vqadd.u8    q9, q0, q11
+    vshrn.u16   d6, q2, #8
+    vld4.8      {d0, d1, d2, d3}, [SRC]!
+    vshrn.u16   d7, q2, #3
+    vsli.u16    q2, q2, #5
+        vshll.u8    q14, d16, #8
+                                    PF add PF_X, PF_X, #8
+        vshll.u8    q8, d19, #8
+                                    PF tst PF_CTL, #0xF
+    vsri.u8     d6, d6, #5
+                                    PF addne PF_X, PF_X, #8
+    vmvn.8      d3, d3
+                                    PF subne PF_CTL, PF_CTL, #1
+    vsri.u8     d7, d7, #6
+    vshrn.u16   d30, q2, #2
+    vmull.u8    q10, d3, d6
+                                    PF pld, [PF_SRC, PF_X, lsl #src_bpp_shift]
+    vmull.u8    q11, d3, d7
+    vmull.u8    q12, d3, d30
+                                    PF pld, [PF_DST, PF_X, lsl #dst_bpp_shift]
+        vsri.u16    q14, q8, #5
+                                    PF cmp PF_X, ORIG_W
+        vshll.u8    q9, d18, #8
+    vrshr.u16   q13, q10, #8
+                                    PF subge PF_X, PF_X, ORIG_W
+    vrshr.u16   q3, q11, #8
+    vrshr.u16   q15, q12, #8
+                                    PF subges PF_CTL, PF_CTL, #0x10
+        vsri.u16    q14, q9, #11
+                                    PF ldrgeb DUMMY, [PF_SRC, SRC_STRIDE, lsl #src_bpp_shift]!
+    vraddhn.u16 d20, q10, q13
+    vraddhn.u16 d23, q11, q3
+                                    PF ldrgeb DUMMY, [PF_DST, DST_STRIDE, lsl #dst_bpp_shift]!
+    vraddhn.u16 d22, q12, q15
+        vst1.16     {d28, d29}, [DST_W, :128]!
+.endm
+
+#else
+
+/* If we did not care much about the performance, we would just use this... */
+.macro pixman_composite_over_8888_0565_process_pixblock_tail_head
+    pixman_composite_over_8888_0565_process_pixblock_tail
+    vst1.16     {d28, d29}, [DST_W, :128]!
+    vld1.16     {d4, d5}, [DST_R, :128]!
+    vld4.32     {d0, d1, d2, d3}, [SRC]!
+    pixman_composite_over_8888_0565_process_pixblock_head
+    cache_preload 8, 8
+.endm
+
+#endif
+
+/*
+ * And now the final part. We are using 'generate_composite_function' macro
+ * to put all the stuff together. We are specifying the name of the function
+ * which we want to get, number of bits per pixel for the source, mask and
+ * destination (0 if unused, like mask in this case). Next come some bit
+ * flags:
+ *   FLAG_DST_READWRITE      - tells that the destination buffer is both read
+ *                             and written, for write-only buffer we would use
+ *                             FLAG_DST_WRITEONLY flag instead
+ *   FLAG_DEINTERLEAVE_32BPP - tells that we prefer to work with planar data
+ *                             and separate color channels for 32bpp format.
+ * The next things are:
+ *  - the number of pixels processed per iteration (8 in this case, because
+ *    that's the maximum what can fit into four 64-bit NEON registers).
+ *  - prefetch distance, measured in pixel blocks. In this case it is 5 times
+ *    by 8 pixels. That would be 40 pixels, or up to 160 bytes. Optimal
+ *    prefetch distance can be selected by running some benchmarks.
+ *
+ * After that we specify some macros, these are 'default_init',
+ * 'default_cleanup' here which are empty (but it is possible to have custom
+ * init/cleanup macros to be able to save/restore some extra NEON registers
+ * like d8-d15 or do anything else) followed by
+ * 'pixman_composite_over_8888_0565_process_pixblock_head',
+ * 'pixman_composite_over_8888_0565_process_pixblock_tail' and
+ * 'pixman_composite_over_8888_0565_process_pixblock_tail_head'
+ * which we got implemented above.
+ *
+ * The last part is the NEON registers allocation scheme.
+ */
+generate_composite_function \
+    pixman_composite_over_8888_0565_asm_neon, 32, 0, 16, \
+    FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+    8, /* number of pixels, processed in a single block */ \
+    5, /* prefetch distance */ \
+    default_init, \
+    default_cleanup, \
+    pixman_composite_over_8888_0565_process_pixblock_head, \
+    pixman_composite_over_8888_0565_process_pixblock_tail, \
+    pixman_composite_over_8888_0565_process_pixblock_tail_head, \
+    28, /* dst_w_basereg */ \
+    4,  /* dst_r_basereg */ \
+    0,  /* src_basereg   */ \
+    24  /* mask_basereg  */
+
+/******************************************************************************/
+
+.macro pixman_composite_src_8888_0565_process_pixblock_head
+    vshll.u8    q8, d1, #8
+    vshll.u8    q14, d2, #8
+    vshll.u8    q9, d0, #8
+.endm
+
+.macro pixman_composite_src_8888_0565_process_pixblock_tail
+    vsri.u16    q14, q8, #5
+    vsri.u16    q14, q9, #11
+.endm
+
+.macro pixman_composite_src_8888_0565_process_pixblock_tail_head
+        vsri.u16    q14, q8, #5
+                                    PF add PF_X, PF_X, #8
+                                    PF tst PF_CTL, #0xF
+    vld4.8      {d0, d1, d2, d3}, [SRC]!
+                                    PF addne PF_X, PF_X, #8
+                                    PF subne PF_CTL, PF_CTL, #1
+        vsri.u16    q14, q9, #11
+                                    PF cmp PF_X, ORIG_W
+                                    PF pld, [PF_SRC, PF_X, lsl #src_bpp_shift]
+    vshll.u8    q8, d1, #8
+        vst1.16     {d28, d29}, [DST_W, :128]!
+                                    PF subge PF_X, PF_X, ORIG_W
+                                    PF subges PF_CTL, PF_CTL, #0x10
+    vshll.u8    q14, d2, #8
+                                    PF ldrgeb DUMMY, [PF_SRC, SRC_STRIDE, lsl #src_bpp_shift]!
+    vshll.u8    q9, d0, #8
+.endm
+
+generate_composite_function \
+    pixman_composite_src_8888_0565_asm_neon, 32, 0, 16, \
+    FLAG_DST_WRITEONLY | FLAG_DEINTERLEAVE_32BPP, \
+    8, /* number of pixels, processed in a single block */ \
+    10, /* prefetch distance */ \
+    default_init, \
+    default_cleanup, \
+    pixman_composite_src_8888_0565_process_pixblock_head, \
+    pixman_composite_src_8888_0565_process_pixblock_tail, \
+    pixman_composite_src_8888_0565_process_pixblock_tail_head
+
+/******************************************************************************/
+
+.macro pixman_composite_add_8000_8000_process_pixblock_head
+    vqadd.u8    q14, q0, q2
+    vqadd.u8    q15, q1, q3
+.endm
+
+.macro pixman_composite_add_8000_8000_process_pixblock_tail
+.endm
+
+.macro pixman_composite_add_8000_8000_process_pixblock_tail_head
+    vld1.8      {d0, d1, d2, d3}, [SRC]!
+                                    PF add PF_X, PF_X, #32
+                                    PF tst PF_CTL, #0xF
+    vld1.8      {d4, d5, d6, d7}, [DST_R, :128]!
+                                    PF addne PF_X, PF_X, #32
+                                    PF subne PF_CTL, PF_CTL, #1
+        vst1.8      {d28, d29, d30, d31}, [DST_W, :128]!
+                                    PF cmp PF_X, ORIG_W
+                                    PF pld, [PF_SRC, PF_X, lsl #src_bpp_shift]
+                                    PF pld, [PF_DST, PF_X, lsl #dst_bpp_shift]
+                                    PF subge PF_X, PF_X, ORIG_W
+                                    PF subges PF_CTL, PF_CTL, #0x10
+    vqadd.u8    q14, q0, q2
+                                    PF ldrgeb DUMMY, [PF_SRC, SRC_STRIDE, lsl #src_bpp_shift]!
+                                    PF ldrgeb DUMMY, [PF_DST, DST_STRIDE, lsl #dst_bpp_shift]!
+    vqadd.u8    q15, q1, q3
+.endm
+
+generate_composite_function \
+    pixman_composite_add_8000_8000_asm_neon, 8, 0, 8, \
+    FLAG_DST_READWRITE, \
+    32, /* number of pixels, processed in a single block */ \
+    10, /* prefetch distance */ \
+    default_init, \
+    default_cleanup, \
+    pixman_composite_add_8000_8000_process_pixblock_head, \
+    pixman_composite_add_8000_8000_process_pixblock_tail, \
+    pixman_composite_add_8000_8000_process_pixblock_tail_head
+
+/******************************************************************************/
+
+.macro pixman_composite_over_8888_8888_process_pixblock_head
+    vmvn.8      d24, d3  /* get inverted alpha */
+    /* do alpha blending */
+    vmull.u8    q8, d24, d4
+    vmull.u8    q9, d24, d5
+    vmull.u8    q10, d24, d6
+    vmull.u8    q11, d24, d7
+.endm
+
+.macro pixman_composite_over_8888_8888_process_pixblock_tail
+    vrshr.u16   q14, q8, #8
+    vrshr.u16   q15, q9, #8
+    vrshr.u16   q12, q10, #8
+    vrshr.u16   q13, q11, #8
+    vraddhn.u16 d28, q14, q8
+    vraddhn.u16 d29, q15, q9
+    vraddhn.u16 d30, q12, q10
+    vraddhn.u16 d31, q13, q11
+    vqadd.u8    q14, q0, q14
+    vqadd.u8    q15, q1, q15
+.endm
+
+.macro pixman_composite_over_8888_8888_process_pixblock_tail_head
+    vld4.8      {d4, d5, d6, d7}, [DST_R, :128]!
+        vrshr.u16   q14, q8, #8
+                                    PF add PF_X, PF_X, #8
+                                    PF tst PF_CTL, #0xF
+        vrshr.u16   q15, q9, #8
+        vrshr.u16   q12, q10, #8
+        vrshr.u16   q13, q11, #8
+                                    PF addne PF_X, PF_X, #8
+                                    PF subne PF_CTL, PF_CTL, #1
+        vraddhn.u16 d28, q14, q8
+        vraddhn.u16 d29, q15, q9
+                                    PF cmp PF_X, ORIG_W
+        vraddhn.u16 d30, q12, q10
+        vraddhn.u16 d31, q13, q11
+        vqadd.u8    q14, q0, q14
+        vqadd.u8    q15, q1, q15
+    vld4.8      {d0, d1, d2, d3}, [SRC]!
+                                    PF pld, [PF_SRC, PF_X, lsl #src_bpp_shift]
+    vmvn.8      d22, d3
+                                    PF pld, [PF_DST, PF_X, lsl #dst_bpp_shift]
+        vst4.8      {d28, d29, d30, d31}, [DST_W, :128]!
+                                    PF subge PF_X, PF_X, ORIG_W
+    vmull.u8    q8, d22, d4
+                                    PF subges PF_CTL, PF_CTL, #0x10
+    vmull.u8    q9, d22, d5
+                                    PF ldrgeb DUMMY, [PF_SRC, SRC_STRIDE, lsl #src_bpp_shift]!
+    vmull.u8    q10, d22, d6
+                                    PF ldrgeb DUMMY, [PF_DST, DST_STRIDE, lsl #dst_bpp_shift]!
+    vmull.u8    q11, d22, d7
+.endm
+
+generate_composite_function \
+    pixman_composite_over_8888_8888_asm_neon, 32, 0, 32, \
+    FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+    8, /* number of pixels, processed in a single block */ \
+    5, /* prefetch distance */ \
+    default_init, \
+    default_cleanup, \
+    pixman_composite_over_8888_8888_process_pixblock_head, \
+    pixman_composite_over_8888_8888_process_pixblock_tail, \
+    pixman_composite_over_8888_8888_process_pixblock_tail_head
+
+/******************************************************************************/
+
+.macro pixman_composite_over_n_8_0565_process_pixblock_head
+    /* in */
+    vmull.u8    q0, d24, d8
+    vmull.u8    q1, d24, d9
+    vmull.u8    q6, d24, d10
+    vmull.u8    q7, d24, d11
+    vrshr.u16   q10, q0, #8
+    vrshr.u16   q11, q1, #8
+    vrshr.u16   q12, q6, #8
+    vrshr.u16   q13, q7, #8
+    vraddhn.u16 d0, q0, q10
+    vraddhn.u16 d1, q1, q11
+    vraddhn.u16 d2, q6, q12
+    vraddhn.u16 d3, q7, q13
+
+    vshrn.u16   d6, q2, #8
+    vshrn.u16   d7, q2, #3
+    vsli.u16    q2, q2, #5
+    vsri.u8     d6, d6, #5
+    vmvn.8      d3, d3
+    vsri.u8     d7, d7, #6
+    vshrn.u16   d30, q2, #2
+    /* now do alpha blending */
+    vmull.u8    q10, d3, d6
+    vmull.u8    q11, d3, d7
+    vmull.u8    q12, d3, d30
+    vrshr.u16   q13, q10, #8
+    vrshr.u16   q3, q11, #8
+    vrshr.u16   q15, q12, #8
+    vraddhn.u16 d20, q10, q13
+    vraddhn.u16 d23, q11, q3
+    vraddhn.u16 d22, q12, q15
+.endm
+
+.macro pixman_composite_over_n_8_0565_process_pixblock_tail
+    vqadd.u8    d16, d2, d20
+    vqadd.u8    q9, q0, q11
+    /* convert to r5g6b5 */
+    vshll.u8    q14, d16, #8
+    vshll.u8    q8, d19, #8
+    vshll.u8    q9, d18, #8
+    vsri.u16    q14, q8, #5
+    vsri.u16    q14, q9, #11
+.endm
+
+/* TODO: expand macros and do better instructions scheduling */
+.macro pixman_composite_over_n_8_0565_process_pixblock_tail_head
+    pixman_composite_over_n_8_0565_process_pixblock_tail
+    vst1.16     {d28, d29}, [DST_W, :128]!
+    vld1.16     {d4, d5}, [DST_R, :128]!
+    vld1.8      {d24}, [MASK]!
+    cache_preload 8, 8
+    pixman_composite_over_n_8_0565_process_pixblock_head
+.endm
+
+/*
+ * This function needs a special initialization of solid mask.
+ * Solid source pixel data is fetched from stack at ARGS_STACK_OFFSET
+ * offset, split into color components and replicated in d8-d11
+ * registers. Additionally, this function needs all the NEON registers,
+ * so it has to save d8-d15 registers which are callee saved according
+ * to ABI. These registers are restored from 'cleanup' macro. All the
+ * other NEON registers are caller saved, so can be clobbered freely
+ * without introducing any problems.
+ */
+.macro pixman_composite_over_n_8_0565_init
+    add         DUMMY, sp, #ARGS_STACK_OFFSET
+    vpush       {d8-d15}
+    vld1.32     {d11[0]}, [DUMMY]
+    vdup.8      d8, d11[0]
+    vdup.8      d9, d11[1]
+    vdup.8      d10, d11[2]
+    vdup.8      d11, d11[3]
+.endm
+
+.macro pixman_composite_over_n_8_0565_cleanup
+    vpop        {d8-d15}
+.endm
+
+generate_composite_function \
+    pixman_composite_over_n_8_0565_asm_neon, 0, 8, 16, \
+    FLAG_DST_READWRITE, \
+    8, /* number of pixels, processed in a single block */ \
+    5, /* prefetch distance */ \
+    pixman_composite_over_n_8_0565_init, \
+    pixman_composite_over_n_8_0565_cleanup, \
+    pixman_composite_over_n_8_0565_process_pixblock_head, \
+    pixman_composite_over_n_8_0565_process_pixblock_tail, \
+    pixman_composite_over_n_8_0565_process_pixblock_tail_head
+
+/******************************************************************************/
+
+.macro pixman_composite_src_0565_0565_process_pixblock_head
+.endm
+
+.macro pixman_composite_src_0565_0565_process_pixblock_tail
+.endm
+
+.macro pixman_composite_src_0565_0565_process_pixblock_tail_head
+    vst1.16 {d0, d1, d2, d3}, [DST_W, :128]!
+    vld1.16 {d0, d1, d2, d3}, [SRC]!
+    cache_preload 16, 16
+.endm
+
+generate_composite_function \
+    pixman_composite_src_0565_0565_asm_neon, 16, 0, 16, \
+    FLAG_DST_WRITEONLY, \
+    16, /* number of pixels, processed in a single block */ \
+    10, /* prefetch distance */ \
+    default_init, \
+    default_cleanup, \
+    pixman_composite_src_0565_0565_process_pixblock_head, \
+    pixman_composite_src_0565_0565_process_pixblock_tail, \
+    pixman_composite_src_0565_0565_process_pixblock_tail_head, \
+    0, /* dst_w_basereg */ \
+    0, /* dst_r_basereg */ \
+    0, /* src_basereg   */ \
+    0  /* mask_basereg  */
+
+/******************************************************************************/
+
+.macro pixman_composite_src_n_8_process_pixblock_head
+.endm
+
+.macro pixman_composite_src_n_8_process_pixblock_tail
+.endm
+
+.macro pixman_composite_src_n_8_process_pixblock_tail_head
+    vst1.8  {d0, d1, d2, d3}, [DST_W, :128]!
+.endm
+
+.macro pixman_composite_src_n_8_init
+    add         DUMMY, sp, #ARGS_STACK_OFFSET
+    vld1.32     {d0[0]}, [DUMMY]
+    vsli.u64    d0, d0, #8
+    vsli.u64    d0, d0, #16
+    vsli.u64    d0, d0, #32
+    vmov        d1, d0
+    vmov        q1, q0
+.endm
+
+.macro pixman_composite_src_n_8_cleanup
+.endm
+
+generate_composite_function \
+    pixman_composite_src_n_8_asm_neon, 0, 0, 8, \
+    FLAG_DST_WRITEONLY, \
+    32, /* number of pixels, processed in a single block */ \
+    0,  /* prefetch distance */ \
+    pixman_composite_src_n_8_init, \
+    pixman_composite_src_n_8_cleanup, \
+    pixman_composite_src_n_8_process_pixblock_head, \
+    pixman_composite_src_n_8_process_pixblock_tail, \
+    pixman_composite_src_n_8_process_pixblock_tail_head, \
+    0, /* dst_w_basereg */ \
+    0, /* dst_r_basereg */ \
+    0, /* src_basereg   */ \
+    0  /* mask_basereg  */
+
+/******************************************************************************/
+
+.macro pixman_composite_src_n_0565_process_pixblock_head
+.endm
+
+.macro pixman_composite_src_n_0565_process_pixblock_tail
+.endm
+
+.macro pixman_composite_src_n_0565_process_pixblock_tail_head
+    vst1.16 {d0, d1, d2, d3}, [DST_W, :128]!
+.endm
+
+.macro pixman_composite_src_n_0565_init
+    add         DUMMY, sp, #ARGS_STACK_OFFSET
+    vld1.32     {d0[0]}, [DUMMY]
+    vsli.u64    d0, d0, #16
+    vsli.u64    d0, d0, #32
+    vmov        d1, d0
+    vmov        q1, q0
+.endm
+
+.macro pixman_composite_src_n_0565_cleanup
+.endm
+
+generate_composite_function \
+    pixman_composite_src_n_0565_asm_neon, 0, 0, 16, \
+    FLAG_DST_WRITEONLY, \
+    16, /* number of pixels, processed in a single block */ \
+    0,  /* prefetch distance */ \
+    pixman_composite_src_n_0565_init, \
+    pixman_composite_src_n_0565_cleanup, \
+    pixman_composite_src_n_0565_process_pixblock_head, \
+    pixman_composite_src_n_0565_process_pixblock_tail, \
+    pixman_composite_src_n_0565_process_pixblock_tail_head, \
+    0, /* dst_w_basereg */ \
+    0, /* dst_r_basereg */ \
+    0, /* src_basereg   */ \
+    0  /* mask_basereg  */
+
+/******************************************************************************/
+
+.macro pixman_composite_src_n_8888_process_pixblock_head
+.endm
+
+.macro pixman_composite_src_n_8888_process_pixblock_tail
+.endm
+
+.macro pixman_composite_src_n_8888_process_pixblock_tail_head
+    vst1.32 {d0, d1, d2, d3}, [DST_W, :128]!
+.endm
+
+.macro pixman_composite_src_n_8888_init
+    add         DUMMY, sp, #ARGS_STACK_OFFSET
+    vld1.32     {d0[0]}, [DUMMY]
+    vsli.u64    d0, d0, #32
+    vmov        d1, d0
+    vmov        q1, q0
+.endm
+
+.macro pixman_composite_src_n_8888_cleanup
+.endm
+
+generate_composite_function \
+    pixman_composite_src_n_8888_asm_neon, 0, 0, 32, \
+    FLAG_DST_WRITEONLY, \
+    8, /* number of pixels, processed in a single block */ \
+    0, /* prefetch distance */ \
+    pixman_composite_src_n_8888_init, \
+    pixman_composite_src_n_8888_cleanup, \
+    pixman_composite_src_n_8888_process_pixblock_head, \
+    pixman_composite_src_n_8888_process_pixblock_tail, \
+    pixman_composite_src_n_8888_process_pixblock_tail_head, \
+    0, /* dst_w_basereg */ \
+    0, /* dst_r_basereg */ \
+    0, /* src_basereg   */ \
+    0  /* mask_basereg  */
+
+/******************************************************************************/
+
+.macro pixman_composite_src_8888_8888_process_pixblock_head
+.endm
+
+.macro pixman_composite_src_8888_8888_process_pixblock_tail
+.endm
+
+.macro pixman_composite_src_8888_8888_process_pixblock_tail_head
+    vst1.32 {d0, d1, d2, d3}, [DST_W, :128]!
+    vld1.32 {d0, d1, d2, d3}, [SRC]!
+    cache_preload 8, 8
+.endm
+
+generate_composite_function \
+    pixman_composite_src_8888_8888_asm_neon, 32, 0, 32, \
+    FLAG_DST_WRITEONLY, \
+    8, /* number of pixels, processed in a single block */ \
+    10, /* prefetch distance */ \
+    default_init, \
+    default_cleanup, \
+    pixman_composite_src_8888_8888_process_pixblock_head, \
+    pixman_composite_src_8888_8888_process_pixblock_tail, \
+    pixman_composite_src_8888_8888_process_pixblock_tail_head, \
+    0, /* dst_w_basereg */ \
+    0, /* dst_r_basereg */ \
+    0, /* src_basereg   */ \
+    0  /* mask_basereg  */
+
+/******************************************************************************/
+
+.macro pixman_composite_over_n_8_8888_process_pixblock_head
+    /* expecting deinterleaved source data in {d8, d9, d10, d11} */
+    /* d8 - blue, d9 - green, d10 - red, d11 - alpha */
+    /* and destination data in {d4, d5, d6, d7} */
+    /* mask is in d24 (d25, d26, d27 are unused) */
+
+    /* in */
+    vmull.u8    q0, d24, d8
+    vmull.u8    q1, d24, d9
+    vmull.u8    q6, d24, d10
+    vmull.u8    q7, d24, d11
+    vrshr.u16   q10, q0, #8
+    vrshr.u16   q11, q1, #8
+    vrshr.u16   q12, q6, #8
+    vrshr.u16   q13, q7, #8
+    vraddhn.u16 d0, q0, q10
+    vraddhn.u16 d1, q1, q11
+    vraddhn.u16 d2, q6, q12
+    vraddhn.u16 d3, q7, q13
+    vmvn.8      d24, d3  /* get inverted alpha */
+    /* source:      d0 - blue, d1 - green, d2 - red, d3 - alpha */
+    /* destination: d4 - blue, d5 - green, d6 - red, d7 - alpha */
+    /* now do alpha blending */
+    vmull.u8    q8, d24, d4
+    vmull.u8    q9, d24, d5
+    vmull.u8    q10, d24, d6
+    vmull.u8    q11, d24, d7
+.endm
+
+.macro pixman_composite_over_n_8_8888_process_pixblock_tail
+    vrshr.u16   q14, q8, #8
+    vrshr.u16   q15, q9, #8
+    vrshr.u16   q12, q10, #8
+    vrshr.u16   q13, q11, #8
+    vraddhn.u16 d28, q14, q8
+    vraddhn.u16 d29, q15, q9
+    vraddhn.u16 d30, q12, q10
+    vraddhn.u16 d31, q13, q11
+    vqadd.u8    q14, q0, q14
+    vqadd.u8    q15, q1, q15
+.endm
+
+/* TODO: expand macros and do better instructions scheduling */
+.macro pixman_composite_over_n_8_8888_process_pixblock_tail_head
+    pixman_composite_over_n_8_8888_process_pixblock_tail
+    vst4.8      {d28, d29, d30, d31}, [DST_W, :128]!
+    vld4.8      {d4, d5, d6, d7}, [DST_R, :128]!
+    vld1.8      {d24}, [MASK]!
+    cache_preload 8, 8
+    pixman_composite_over_n_8_8888_process_pixblock_head
+.endm
+
+.macro pixman_composite_over_n_8_8888_init
+    add         DUMMY, sp, #ARGS_STACK_OFFSET
+    vpush       {d8-d15}
+    vld1.32     {d11[0]}, [DUMMY]
+    vdup.8      d8, d11[0]
+    vdup.8      d9, d11[1]
+    vdup.8      d10, d11[2]
+    vdup.8      d11, d11[3]
+.endm
+
+.macro pixman_composite_over_n_8_8888_cleanup
+    vpop        {d8-d15}
+.endm
+
+generate_composite_function \
+    pixman_composite_over_n_8_8888_asm_neon, 0, 8, 32, \
+    FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+    8, /* number of pixels, processed in a single block */ \
+    5, /* prefetch distance */ \
+    pixman_composite_over_n_8_8888_init, \
+    pixman_composite_over_n_8_8888_cleanup, \
+    pixman_composite_over_n_8_8888_process_pixblock_head, \
+    pixman_composite_over_n_8_8888_process_pixblock_tail, \
+    pixman_composite_over_n_8_8888_process_pixblock_tail_head
+
+/******************************************************************************/
+
+.macro pixman_composite_add_n_8_8_process_pixblock_head
+    /* expecting source data in {d8, d9, d10, d11} */
+    /* d8 - blue, d9 - green, d10 - red, d11 - alpha */
+    /* and destination data in {d4, d5, d6, d7} */
+    /* mask is in d24, d25, d26, d27 */
+    vmull.u8    q0, d24, d11
+    vmull.u8    q1, d25, d11
+    vmull.u8    q6, d26, d11
+    vmull.u8    q7, d27, d11
+    vrshr.u16   q10, q0, #8
+    vrshr.u16   q11, q1, #8
+    vrshr.u16   q12, q6, #8
+    vrshr.u16   q13, q7, #8
+    vraddhn.u16 d0, q0, q10
+    vraddhn.u16 d1, q1, q11
+    vraddhn.u16 d2, q6, q12
+    vraddhn.u16 d3, q7, q13
+    vqadd.u8    q14, q0, q2
+    vqadd.u8    q15, q1, q3
+.endm
+
+.macro pixman_composite_add_n_8_8_process_pixblock_tail
+.endm
+
+/* TODO: expand macros and do better instructions scheduling */
+.macro pixman_composite_add_n_8_8_process_pixblock_tail_head
+    pixman_composite_add_n_8_8_process_pixblock_tail
+    vst1.8      {d28, d29, d30, d31}, [DST_W, :128]!
+    vld1.8      {d4, d5, d6, d7}, [DST_R, :128]!
+    vld1.8      {d24, d25, d26, d27}, [MASK]!
+    cache_preload 32, 32
+    pixman_composite_add_n_8_8_process_pixblock_head
+.endm
+
+.macro pixman_composite_add_n_8_8_init
+    add         DUMMY, sp, #ARGS_STACK_OFFSET
+    vpush       {d8-d15}
+    vld1.32     {d11[0]}, [DUMMY]
+    vdup.8      d11, d11[3]
+.endm
+
+.macro pixman_composite_add_n_8_8_cleanup
+    vpop        {d8-d15}
+.endm
+
+generate_composite_function \
+    pixman_composite_add_n_8_8_asm_neon, 0, 8, 8, \
+    FLAG_DST_READWRITE, \
+    32, /* number of pixels, processed in a single block */ \
+    5, /* prefetch distance */ \
+    pixman_composite_add_n_8_8_init, \
+    pixman_composite_add_n_8_8_cleanup, \
+    pixman_composite_add_n_8_8_process_pixblock_head, \
+    pixman_composite_add_n_8_8_process_pixblock_tail, \
+    pixman_composite_add_n_8_8_process_pixblock_tail_head
+
+/******************************************************************************/
+
+.macro pixman_composite_add_8_8_8_process_pixblock_head
+    /* expecting source data in {d0, d1, d2, d3} */
+    /* destination data in {d4, d5, d6, d7} */
+    /* mask in {d24, d25, d26, d27} */
+    vmull.u8    q8, d24, d0
+    vmull.u8    q9, d25, d1
+    vmull.u8    q10, d26, d2
+    vmull.u8    q11, d27, d3
+    vrshr.u16   q0, q8, #8
+    vrshr.u16   q1, q9, #8
+    vrshr.u16   q12, q10, #8
+    vrshr.u16   q13, q11, #8
+    vraddhn.u16 d0, q0, q8
+    vraddhn.u16 d1, q1, q9
+    vraddhn.u16 d2, q12, q10
+    vraddhn.u16 d3, q13, q11
+    vqadd.u8    q14, q0, q2
+    vqadd.u8    q15, q1, q3
+.endm
+
+.macro pixman_composite_add_8_8_8_process_pixblock_tail
+.endm
+
+/* TODO: expand macros and do better instructions scheduling */
+.macro pixman_composite_add_8_8_8_process_pixblock_tail_head
+    pixman_composite_add_8_8_8_process_pixblock_tail
+    vst1.8      {d28, d29, d30, d31}, [DST_W, :128]!
+    vld1.8      {d4, d5, d6, d7}, [DST_R, :128]!
+    vld1.8      {d24, d25, d26, d27}, [MASK]!
+    vld1.8      {d0, d1, d2, d3}, [SRC]!
+    cache_preload 32, 32
+    pixman_composite_add_8_8_8_process_pixblock_head
+.endm
+
+.macro pixman_composite_add_8_8_8_init
+.endm
+
+.macro pixman_composite_add_8_8_8_cleanup
+.endm
+
+generate_composite_function \
+    pixman_composite_add_8_8_8_asm_neon, 8, 8, 8, \
+    FLAG_DST_READWRITE, \
+    32, /* number of pixels, processed in a single block */ \
+    5, /* prefetch distance */ \
+    pixman_composite_add_8_8_8_init, \
+    pixman_composite_add_8_8_8_cleanup, \
+    pixman_composite_add_8_8_8_process_pixblock_head, \
+    pixman_composite_add_8_8_8_process_pixblock_tail, \
+    pixman_composite_add_8_8_8_process_pixblock_tail_head
+
+/******************************************************************************/
+
+.macro pixman_composite_over_8888_n_8888_process_pixblock_head
+    /* expecting source data in {d0, d1, d2, d3} */
+    /* destination data in {d4, d5, d6, d7} */
+    /* solid mask is in d15 */
+
+    /* 'in' */
+    vmull.u8    q8, d15, d3
+    vmull.u8    q6, d15, d2
+    vmull.u8    q5, d15, d1
+    vmull.u8    q4, d15, d0
+    vrshr.u16   q13, q8, #8
+    vrshr.u16   q12, q6, #8
+    vrshr.u16   q11, q5, #8
+    vrshr.u16   q10, q4, #8
+    vraddhn.u16 d3, q8, q13
+    vraddhn.u16 d2, q6, q12
+    vraddhn.u16 d1, q5, q11
+    vraddhn.u16 d0, q4, q10
+    vmvn.8      d24, d3  /* get inverted alpha */
+    /* now do alpha blending */
+    vmull.u8    q8, d24, d4
+    vmull.u8    q9, d24, d5
+    vmull.u8    q10, d24, d6
+    vmull.u8    q11, d24, d7
+.endm
+
+.macro pixman_composite_over_8888_n_8888_process_pixblock_tail
+    vrshr.u16   q14, q8, #8
+    vrshr.u16   q15, q9, #8
+    vrshr.u16   q12, q10, #8
+    vrshr.u16   q13, q11, #8
+    vraddhn.u16 d28, q14, q8
+    vraddhn.u16 d29, q15, q9
+    vraddhn.u16 d30, q12, q10
+    vraddhn.u16 d31, q13, q11
+    vqadd.u8    q14, q0, q14
+    vqadd.u8    q15, q1, q15
+.endm
+
+/* TODO: expand macros and do better instructions scheduling */
+.macro pixman_composite_over_8888_n_8888_process_pixblock_tail_head
+    vld4.8     {d4, d5, d6, d7}, [DST_R, :128]!
+    pixman_composite_over_8888_n_8888_process_pixblock_tail
+    vld4.8     {d0, d1, d2, d3}, [SRC]!
+    cache_preload 8, 8
+    pixman_composite_over_8888_n_8888_process_pixblock_head
+    vst4.8     {d28, d29, d30, d31}, [DST_W, :128]!
+.endm
+
+.macro pixman_composite_over_8888_n_8888_init
+    add         DUMMY, sp, #48
+    vpush       {d8-d15}
+    vld1.32     {d15[0]}, [DUMMY]
+    vdup.8      d15, d15[3]
+.endm
+
+.macro pixman_composite_over_8888_n_8888_cleanup
+    vpop        {d8-d15}
+.endm
+
+generate_composite_function \
+    pixman_composite_over_8888_n_8888_asm_neon, 32, 0, 32, \
+    FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+    8, /* number of pixels, processed in a single block */ \
+    5, /* prefetch distance */ \
+    pixman_composite_over_8888_n_8888_init, \
+    pixman_composite_over_8888_n_8888_cleanup, \
+    pixman_composite_over_8888_n_8888_process_pixblock_head, \
+    pixman_composite_over_8888_n_8888_process_pixblock_tail, \
+    pixman_composite_over_8888_n_8888_process_pixblock_tail_head
+
+/******************************************************************************/
+
+.macro pixman_composite_src_0888_0888_process_pixblock_head
+.endm
+
+.macro pixman_composite_src_0888_0888_process_pixblock_tail
+.endm
+
+.macro pixman_composite_src_0888_0888_process_pixblock_tail_head
+    vst3.8 {d0, d1, d2}, [DST_W]!
+    vld3.8 {d0, d1, d2}, [SRC]!
+    cache_preload 8, 8
+.endm
+
+generate_composite_function \
+    pixman_composite_src_0888_0888_asm_neon, 24, 0, 24, \
+    FLAG_DST_WRITEONLY, \
+    8, /* number of pixels, processed in a single block */ \
+    10, /* prefetch distance */ \
+    default_init, \
+    default_cleanup, \
+    pixman_composite_src_0888_0888_process_pixblock_head, \
+    pixman_composite_src_0888_0888_process_pixblock_tail, \
+    pixman_composite_src_0888_0888_process_pixblock_tail_head, \
+    0, /* dst_w_basereg */ \
+    0, /* dst_r_basereg */ \
+    0, /* src_basereg   */ \
+    0  /* mask_basereg  */
diff --git a/pixman/pixman/pixman-arm-neon-asm.h b/pixman/pixman/pixman-arm-neon-asm.h
new file mode 100644
index 000000000..e7be5cddf
--- /dev/null
+++ b/pixman/pixman/pixman-arm-neon-asm.h
@@ -0,0 +1,787 @@
+/*
+ * Copyright © 2009 Nokia Corporation
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and its
+ * documentation for any purpose is hereby granted without fee, provided that
+ * the above copyright notice appear in all copies and that both that
+ * copyright notice and this permission notice appear in supporting
+ * documentation, and that the name of Nokia Corporation not be used in
+ * advertising or publicity pertaining to distribution of the software without
+ * specific, written prior permission.  Nokia Corporation makes no
+ * representations about the suitability of this software for any purpose.
+ * It is provided "as is" without express or implied warranty.
+ *
+ * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
+ * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
+ * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
+ * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
+ * SOFTWARE.
+ *
+ * Author:  Siarhei Siamashka (siarhei.siamashka@nokia.com)
+ */
+
+/*
+ * This file contains a macro ('generate_composite_function') which can
+ * construct 2D image processing functions, based on a common template.
+ * Any combinations of source, destination and mask images with 8bpp,
+ * 16bpp, 24bpp, 32bpp color formats are supported.
+ *
+ * This macro takes care of:
+ *  - handling of leading and trailing unaligned pixels
+ *  - doing most of the work related to L2 cache preload
+ *  - encourages the use of software pipelining for better instructions
+ *    scheduling
+ *
+ * The user of this macro has to provide some configuration parameters
+ * (bit depths for the images, prefetch distance, etc.) and a set of
+ * macros, which should implement basic code chunks responsible for
+ * pixels processing. See 'pixman-arm-neon-asm.S' file for the usage
+ * examples.
+ *
+ * TODO:
+ *  - try overlapped pixel method (from Ian Rickards) when processing
+ *    exactly two blocks of pixels
+ *  - maybe add an option to do reverse scanline processing
+ */
+
+/*
+ * Bit flags for 'generate_composite_function' macro which are used
+ * to tune generated functions behavior.
+ */
+.set FLAG_DST_WRITEONLY,       0
+.set FLAG_DST_READWRITE,       1
+.set FLAG_DEINTERLEAVE_32BPP,  2
+
+/*
+ * Offset in stack where mask and source pointer/stride can be accessed
+ * from 'init' macro. This is useful for doing special handling for solid mask.
+ */
+.set ARGS_STACK_OFFSET,        40
+
+/*
+ * Constants for selecting preferable prefetch type.
+ */
+.set PREFETCH_TYPE_NONE,       0 /* No prefetch at all */
+.set PREFETCH_TYPE_SIMPLE,     1 /* A simple, fixed-distance-ahead prefetch */
+.set PREFETCH_TYPE_ADVANCED,   2 /* Advanced fine-grained prefetch */
+
+/*
+ * Definitions of supplementary pixld/pixst macros (for partial load/store of
+ * pixel data).
+ */
+
+.macro pixldst1 op, elem_size, reg1, mem_operand, abits
+.if abits > 0
+    op&.&elem_size {d&reg1}, [&mem_operand&, :&abits&]!
+.else
+    op&.&elem_size {d&reg1}, [&mem_operand&]!
+.endif
+.endm
+
+.macro pixldst2 op, elem_size, reg1, reg2, mem_operand, abits
+.if abits > 0
+    op&.&elem_size {d&reg1, d&reg2}, [&mem_operand&, :&abits&]!
+.else
+    op&.&elem_size {d&reg1, d&reg2}, [&mem_operand&]!
+.endif
+.endm
+
+.macro pixldst4 op, elem_size, reg1, reg2, reg3, reg4, mem_operand, abits
+.if abits > 0
+    op&.&elem_size {d&reg1, d&reg2, d&reg3, d&reg4}, [&mem_operand&, :&abits&]!
+.else
+    op&.&elem_size {d&reg1, d&reg2, d&reg3, d&reg4}, [&mem_operand&]!
+.endif
+.endm
+
+.macro pixldst0 op, elem_size, reg1, idx, mem_operand, abits
+    op&.&elem_size {d&reg1[idx]}, [&mem_operand&]!
+.endm
+
+.macro pixldst3 op, elem_size, reg1, reg2, reg3, mem_operand
+    op&.&elem_size {d&reg1, d&reg2, d&reg3}, [&mem_operand&]!
+.endm
+
+.macro pixldst30 op, elem_size, reg1, reg2, reg3, idx, mem_operand
+    op&.&elem_size {d&reg1[idx], d&reg2[idx], d&reg3[idx]}, [&mem_operand&]!
+.endm
+
+.macro pixldst numbytes, op, elem_size, basereg, mem_operand, abits
+.if numbytes == 32
+    pixldst4 op, elem_size, %(basereg+4), %(basereg+5), \
+                              %(basereg+6), %(basereg+7), mem_operand, abits
+.elseif numbytes == 16
+    pixldst2 op, elem_size, %(basereg+2), %(basereg+3), mem_operand, abits
+.elseif numbytes == 8
+    pixldst1 op, elem_size, %(basereg+1), mem_operand, abits
+.elseif numbytes == 4
+    .if !RESPECT_STRICT_ALIGNMENT || (elem_size == 32)
+        pixldst0 op, 32, %(basereg+0), 1, mem_operand, abits
+    .elseif elem_size == 16
+        pixldst0 op, 16, %(basereg+0), 2, mem_operand, abits
+        pixldst0 op, 16, %(basereg+0), 3, mem_operand, abits
+    .else
+        pixldst0 op, 8, %(basereg+0), 4, mem_operand, abits
+        pixldst0 op, 8, %(basereg+0), 5, mem_operand, abits
+        pixldst0 op, 8, %(basereg+0), 6, mem_operand, abits
+        pixldst0 op, 8, %(basereg+0), 7, mem_operand, abits
+    .endif
+.elseif numbytes == 2
+    .if !RESPECT_STRICT_ALIGNMENT || (elem_size == 16)
+        pixldst0 op, 16, %(basereg+0), 1, mem_operand, abits
+    .else
+        pixldst0 op, 8, %(basereg+0), 2, mem_operand, abits
+        pixldst0 op, 8, %(basereg+0), 3, mem_operand, abits
+    .endif
+.elseif numbytes == 1
+    pixldst0 op, 8, %(basereg+0), 1, mem_operand, abits
+.else
+    .error "unsupported size: numbytes"
+.endif
+.endm
+
+.macro pixld numpix, bpp, basereg, mem_operand, abits=0
+.if bpp > 0
+.if (bpp == 32) && (numpix == 8) && (DEINTERLEAVE_32BPP_ENABLED != 0)
+    pixldst4 vld4, 8, %(basereg+4), %(basereg+5), \
+                      %(basereg+6), %(basereg+7), mem_operand, abits
+.elseif (bpp == 24) && (numpix == 8)
+    pixldst3 vld3, 8, %(basereg+3), %(basereg+4), %(basereg+5), mem_operand
+.elseif (bpp == 24) && (numpix == 4)
+    pixldst30 vld3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 4, mem_operand
+    pixldst30 vld3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 5, mem_operand
+    pixldst30 vld3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 6, mem_operand
+    pixldst30 vld3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 7, mem_operand
+.elseif (bpp == 24) && (numpix == 2)
+    pixldst30 vld3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 2, mem_operand
+    pixldst30 vld3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 3, mem_operand
+.elseif (bpp == 24) && (numpix == 1)
+    pixldst30 vld3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 1, mem_operand
+.else
+    pixldst %(numpix * bpp / 8), vld1, %(bpp), basereg, mem_operand, abits
+.endif
+.endif
+.endm
+
+.macro pixst numpix, bpp, basereg, mem_operand, abits=0
+.if bpp > 0
+.if (bpp == 32) && (numpix == 8) && (DEINTERLEAVE_32BPP_ENABLED != 0)
+    pixldst4 vst4, 8, %(basereg+4), %(basereg+5), \
+                      %(basereg+6), %(basereg+7), mem_operand, abits
+.elseif (bpp == 24) && (numpix == 8)
+    pixldst3 vst3, 8, %(basereg+3), %(basereg+4), %(basereg+5), mem_operand
+.elseif (bpp == 24) && (numpix == 4)
+    pixldst30 vst3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 4, mem_operand
+    pixldst30 vst3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 5, mem_operand
+    pixldst30 vst3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 6, mem_operand
+    pixldst30 vst3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 7, mem_operand
+.elseif (bpp == 24) && (numpix == 2)
+    pixldst30 vst3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 2, mem_operand
+    pixldst30 vst3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 3, mem_operand
+.elseif (bpp == 24) && (numpix == 1)
+    pixldst30 vst3, 8, %(basereg+0), %(basereg+1), %(basereg+2), 1, mem_operand
+.else
+    pixldst %(numpix * bpp / 8), vst1, %(bpp), basereg, mem_operand, abits
+.endif
+.endif
+.endm
+
+.macro pixld_a numpix, bpp, basereg, mem_operand
+.if (bpp * numpix) <= 128
+    pixld numpix, bpp, basereg, mem_operand, %(bpp * numpix)
+.else
+    pixld numpix, bpp, basereg, mem_operand, 128
+.endif
+.endm
+
+.macro pixst_a numpix, bpp, basereg, mem_operand
+.if (bpp * numpix) <= 128
+    pixst numpix, bpp, basereg, mem_operand, %(bpp * numpix)
+.else
+    pixst numpix, bpp, basereg, mem_operand, 128
+.endif
+.endm
+
+.macro vuzp8 reg1, reg2
+    vuzp.8 d&reg1, d&reg2
+.endm
+
+.macro vzip8 reg1, reg2
+    vzip.8 d&reg1, d&reg2
+.endm
+
+/* deinterleave B, G, R, A channels for eight 32bpp pixels in 4 registers */
+.macro pixdeinterleave bpp, basereg
+.if (bpp == 32) && (DEINTERLEAVE_32BPP_ENABLED != 0)
+    vuzp8 %(basereg+0), %(basereg+1)
+    vuzp8 %(basereg+2), %(basereg+3)
+    vuzp8 %(basereg+1), %(basereg+3)
+    vuzp8 %(basereg+0), %(basereg+2)
+.endif
+.endm
+
+/* interleave B, G, R, A channels for eight 32bpp pixels in 4 registers */
+.macro pixinterleave bpp, basereg
+.if (bpp == 32) && (DEINTERLEAVE_32BPP_ENABLED != 0)
+    vzip8 %(basereg+0), %(basereg+2)
+    vzip8 %(basereg+1), %(basereg+3)
+    vzip8 %(basereg+2), %(basereg+3)
+    vzip8 %(basereg+0), %(basereg+1)
+.endif
+.endm
+
+/*
+ * This is a macro for implementing cache preload. The main idea is that
+ * cache preload logic is mostly independent from the rest of pixels
+ * processing code. It starts at the top left pixel and moves forward
+ * across pixels and can jump across scanlines. Prefetch distance is
+ * handled in an 'incremental' way: it starts from 0 and advances to the
+ * optimal distance over time. After reaching optimal prefetch distance,
+ * it is kept constant. There are some checks which prevent prefetching
+ * unneeded pixel lines below the image (but it still can prefetch a bit
+ * more data on the right side of the image - not a big issue and may
+ * be actually helpful when rendering text glyphs). Additional trick is
+ * the use of LDR instruction for prefetch instead of PLD when moving to
+ * the next line, the point is that we have a high chance of getting TLB
+ * miss in this case, and PLD would be useless.
+ *
+ * This sounds like it may introduce a noticeable overhead (when working with
+ * fully cached data). But in reality, due to having a separate pipeline and
+ * instruction queue for NEON unit in ARM Cortex-A8, normal ARM code can
+ * execute simultaneously with NEON and be completely shadowed by it. Thus
+ * we get no performance overhead at all (*). This looks like a very nice
+ * feature of Cortex-A8, if used wisely. We don't have a hardware prefetcher,
+ * but still can implement some rather advanced prefetch logic in sofware
+ * for almost zero cost!
+ *
+ * (*) The overhead of the prefetcher is visible when running some trivial
+ * pixels processing like simple copy. Anyway, having prefetch is a must
+ * when working with the graphics data.
+ */
+.macro PF a, x:vararg
+.if (PREFETCH_TYPE_CURRENT == PREFETCH_TYPE_ADVANCED)
+    a x
+.endif
+.endm
+
+.macro cache_preload std_increment, boost_increment
+.if (src_bpp_shift >= 0) || (dst_r_bpp != 0) || (mask_bpp_shift >= 0)
+.if regs_shortage
+    PF ldr ORIG_W, [sp] /* If we are short on regs, ORIG_W is kept on stack */
+.endif
+.if std_increment != 0
+    PF add PF_X, PF_X, #std_increment
+.endif
+    PF tst PF_CTL, #0xF
+    PF addne PF_X, PF_X, #boost_increment
+    PF subne PF_CTL, PF_CTL, #1
+    PF cmp PF_X, ORIG_W
+.if src_bpp_shift >= 0
+    PF pld, [PF_SRC, PF_X, lsl #src_bpp_shift]
+.endif
+.if dst_r_bpp != 0
+    PF pld, [PF_DST, PF_X, lsl #dst_bpp_shift]
+.endif
+.if mask_bpp_shift >= 0
+    PF pld, [PF_MASK, PF_X, lsl #mask_bpp_shift]
+.endif
+    PF subge PF_X, PF_X, ORIG_W
+    PF subges PF_CTL, PF_CTL, #0x10
+.if src_bpp_shift >= 0
+    PF ldrgeb DUMMY, [PF_SRC, SRC_STRIDE, lsl #src_bpp_shift]!
+.endif
+.if dst_r_bpp != 0
+    PF ldrgeb DUMMY, [PF_DST, DST_STRIDE, lsl #dst_bpp_shift]!
+.endif
+.if mask_bpp_shift >= 0
+    PF ldrgeb DUMMY, [PF_MASK, MASK_STRIDE, lsl #mask_bpp_shift]!
+.endif
+.endif
+.endm
+
+.macro cache_preload_simple
+.if (PREFETCH_TYPE_CURRENT == PREFETCH_TYPE_SIMPLE)
+.if src_bpp > 0
+    pld [SRC, #(PREFETCH_DISTANCE_SIMPLE * src_bpp / 8)]
+.endif
+.if dst_r_bpp > 0
+    pld [DST_R, #(PREFETCH_DISTANCE_SIMPLE * dst_r_bpp / 8)]
+.endif
+.if mask_bpp > 0
+    pld [MASK, #(PREFETCH_DISTANCE_SIMPLE * mask_bpp / 8)]
+.endif
+.endif
+.endm
+
+/*
+ * Macro which is used to process leading pixels until destination
+ * pointer is properly aligned (at 16 bytes boundary). When destination
+ * buffer uses 16bpp format, this is unnecessary, or even pointless.
+ */
+.macro ensure_destination_ptr_alignment process_pixblock_head, \
+                                        process_pixblock_tail, \
+                                        process_pixblock_tail_head
+.if dst_w_bpp != 24
+    tst         DST_R, #0xF
+    beq         2f
+
+.irp lowbit, 1, 2, 4, 8, 16
+local skip1
+.if (dst_w_bpp <= (lowbit * 8)) && ((lowbit * 8) < (pixblock_size * dst_w_bpp))
+.if lowbit < 16 /* we don't need more than 16-byte alignment */
+    tst         DST_R, #lowbit
+    beq         1f
+.endif
+    pixld       (lowbit * 8 / dst_w_bpp), src_bpp, src_basereg, SRC
+    pixld       (lowbit * 8 / dst_w_bpp), mask_bpp, mask_basereg, MASK
+.if dst_r_bpp > 0
+    pixld_a     (lowbit * 8 / dst_r_bpp), dst_r_bpp, dst_r_basereg, DST_R
+.else
+    add         DST_R, DST_R, #lowbit
+.endif
+    PF add      PF_X, PF_X, #(lowbit * 8 / dst_w_bpp)
+    sub         W, W, #(lowbit * 8 / dst_w_bpp)
+1:
+.endif
+.endr
+    pixdeinterleave src_bpp, src_basereg
+    pixdeinterleave mask_bpp, mask_basereg
+    pixdeinterleave dst_r_bpp, dst_r_basereg
+
+    process_pixblock_head
+    cache_preload 0, pixblock_size
+    cache_preload_simple
+    process_pixblock_tail
+
+    pixinterleave dst_w_bpp, dst_w_basereg
+.irp lowbit, 1, 2, 4, 8, 16
+.if (dst_w_bpp <= (lowbit * 8)) && ((lowbit * 8) < (pixblock_size * dst_w_bpp))
+.if lowbit < 16 /* we don't need more than 16-byte alignment */
+    tst         DST_W, #lowbit
+    beq         1f
+.endif
+    pixst_a     (lowbit * 8 / dst_w_bpp), dst_w_bpp, dst_w_basereg, DST_W
+1:
+.endif
+.endr
+.endif
+2:
+.endm
+
+/*
+ * Special code for processing up to (pixblock_size - 1) remaining
+ * trailing pixels. As SIMD processing performs operation on
+ * pixblock_size pixels, anything smaller than this has to be loaded
+ * and stored in a special way. Loading and storing of pixel data is
+ * performed in such a way that we fill some 'slots' in the NEON
+ * registers (some slots naturally are unused), then perform compositing
+ * operation as usual. In the end, the data is taken from these 'slots'
+ * and saved to memory.
+ *
+ * cache_preload_flag - allows to suppress prefetch if
+ *                      set to 0
+ * dst_aligned_flag   - selects whether destination buffer
+ *                      is aligned
+ */
+.macro process_trailing_pixels cache_preload_flag, \
+                               dst_aligned_flag, \
+                               process_pixblock_head, \
+                               process_pixblock_tail, \
+                               process_pixblock_tail_head
+    tst         W, #(pixblock_size - 1)
+    beq         2f
+.irp chunk_size, 16, 8, 4, 2, 1
+.if pixblock_size > chunk_size
+    tst         W, #chunk_size
+    beq         1f
+    pixld       chunk_size, src_bpp, src_basereg, SRC
+    pixld       chunk_size, mask_bpp, mask_basereg, MASK
+.if dst_aligned_flag != 0
+    pixld_a     chunk_size, dst_r_bpp, dst_r_basereg, DST_R
+.else
+    pixld       chunk_size, dst_r_bpp, dst_r_basereg, DST_R
+.endif
+.if cache_preload_flag != 0
+    PF add      PF_X, PF_X, #chunk_size
+.endif
+1:
+.endif
+.endr
+    pixdeinterleave src_bpp, src_basereg
+    pixdeinterleave mask_bpp, mask_basereg
+    pixdeinterleave dst_r_bpp, dst_r_basereg
+
+    process_pixblock_head
+.if cache_preload_flag != 0
+    cache_preload 0, pixblock_size
+    cache_preload_simple
+.endif
+    process_pixblock_tail
+    pixinterleave dst_w_bpp, dst_w_basereg
+.irp chunk_size, 16, 8, 4, 2, 1
+.if pixblock_size > chunk_size
+    tst         W, #chunk_size
+    beq         1f
+.if dst_aligned_flag != 0
+    pixst_a     chunk_size, dst_w_bpp, dst_w_basereg, DST_W
+.else
+    pixst       chunk_size, dst_w_bpp, dst_w_basereg, DST_W
+.endif
+1:
+.endif
+.endr
+2:
+.endm
+
+/*
+ * Macro, which performs all the needed operations to switch to the next
+ * scanline and start the next loop iteration unless all the scanlines
+ * are already processed.
+ */
+.macro advance_to_next_scanline start_of_loop_label
+.if regs_shortage
+    ldrd        W, [sp] /* load W and H (width and height) from stack */
+.else
+    mov         W, ORIG_W
+.endif
+    add         DST_W, DST_W, DST_STRIDE, lsl #dst_bpp_shift
+.if src_bpp != 0
+    add         SRC, SRC, SRC_STRIDE, lsl #src_bpp_shift
+.endif
+.if mask_bpp != 0
+    add         MASK, MASK, MASK_STRIDE, lsl #mask_bpp_shift
+.endif
+.if (dst_w_bpp != 24)
+    sub         DST_W, DST_W, W, lsl #dst_bpp_shift
+.endif
+.if (src_bpp != 24) && (src_bpp != 0)
+    sub         SRC, SRC, W, lsl #src_bpp_shift
+.endif
+.if (mask_bpp != 24) && (mask_bpp != 0)
+    sub         MASK, MASK, W, lsl #mask_bpp_shift
+.endif
+    subs        H, H, #1
+    mov         DST_R, DST_W
+.if regs_shortage
+    str         H, [sp, #4] /* save updated height to stack */
+.endif
+    bge         start_of_loop_label
+.endm
+
+/*
+ * Registers are allocated in the following way by default:
+ * d0, d1, d2, d3     - reserved for loading source pixel data
+ * d4, d5, d6, d7     - reserved for loading destination pixel data
+ * d24, d25, d26, d27 - reserved for loading mask pixel data
+ * d28, d29, d30, d31 - final destination pixel data for writeback to memory
+ */
+.macro generate_composite_function fname, \
+                                   src_bpp_, \
+                                   mask_bpp_, \
+                                   dst_w_bpp_, \
+                                   flags, \
+                                   pixblock_size_, \
+                                   prefetch_distance, \
+                                   init, \
+                                   cleanup, \
+                                   process_pixblock_head, \
+                                   process_pixblock_tail, \
+                                   process_pixblock_tail_head, \
+                                   dst_w_basereg_ = 28, \
+                                   dst_r_basereg_ = 4, \
+                                   src_basereg_   = 0, \
+                                   mask_basereg_  = 24
+
+    .func fname
+    .global fname
+    /* For ELF format also set function visibility to hidden */
+#ifdef __ELF__
+    .hidden fname
+    .type fname, %function
+#endif
+fname:
+    push        {r4-r12, lr}        /* save all registers */
+
+/*
+ * Select prefetch type for this function. If prefetch distance is
+ * set to 0 or one of the color formats is 24bpp, SIMPLE prefetch
+ * has to be used instead of ADVANCED.
+ */
+    .set PREFETCH_TYPE_CURRENT, PREFETCH_TYPE_DEFAULT
+.if prefetch_distance == 0
+    .set PREFETCH_TYPE_CURRENT, PREFETCH_TYPE_NONE
+.elseif (PREFETCH_TYPE_CURRENT > PREFETCH_TYPE_SIMPLE) && \
+        ((src_bpp_ == 24) || (mask_bpp_ == 24) || (dst_w_bpp_ == 24))
+    .set PREFETCH_TYPE_CURRENT, PREFETCH_TYPE_SIMPLE
+.endif
+
+/*
+ * Make some macro arguments globally visible and accessible
+ * from other macros
+ */
+    .set src_bpp, src_bpp_
+    .set mask_bpp, mask_bpp_
+    .set dst_w_bpp, dst_w_bpp_
+    .set pixblock_size, pixblock_size_
+    .set dst_w_basereg, dst_w_basereg_
+    .set dst_r_basereg, dst_r_basereg_
+    .set src_basereg, src_basereg_
+    .set mask_basereg, mask_basereg_
+
+/*
+ * Assign symbolic names to registers
+ */
+    W           .req        r0      /* width (is updated during processing) */
+    H           .req        r1      /* height (is updated during processing) */
+    DST_W       .req        r2      /* destination buffer pointer for writes */
+    DST_STRIDE  .req        r3      /* destination image stride */
+    SRC         .req        r4      /* source buffer pointer */
+    SRC_STRIDE  .req        r5      /* source image stride */
+    DST_R       .req        r6      /* destination buffer pointer for reads */
+
+    MASK        .req        r7      /* mask pointer */
+    MASK_STRIDE .req        r8      /* mask stride */
+
+    PF_CTL      .req        r9      /* combined lines counter and prefetch */
+                                    /* distance increment counter */
+    PF_X        .req        r10     /* pixel index in a scanline for current */
+                                    /* pretetch position */
+    PF_SRC      .req        r11     /* pointer to source scanline start */
+                                    /* for prefetch purposes */
+    PF_DST      .req        r12     /* pointer to destination scanline start */
+                                    /* for prefetch purposes */
+    PF_MASK     .req        r14     /* pointer to mask scanline start */
+                                    /* for prefetch purposes */
+/*
+ * Check whether we have enough registers for all the local variables.
+ * If we don't have enough registers, original width and height are
+ * kept on top of stack (and 'regs_shortage' variable is set to indicate
+ * this for the rest of code). Even if there are enough registers, the
+ * allocation scheme may be a bit different depending on whether source
+ * or mask is not used.
+ */
+.if (PREFETCH_TYPE_CURRENT < PREFETCH_TYPE_ADVANCED)
+    ORIG_W      .req        r10     /* saved original width */
+    DUMMY       .req        r12     /* temporary register */
+    .set        regs_shortage, 0
+.elseif mask_bpp == 0
+    ORIG_W      .req        r7      /* saved original width */
+    DUMMY       .req        r8      /* temporary register */
+    .set        regs_shortage, 0
+.elseif src_bpp == 0
+    ORIG_W      .req        r4      /* saved original width */
+    DUMMY       .req        r5      /* temporary register */
+    .set        regs_shortage, 0
+.else
+    ORIG_W      .req        r1      /* saved original width */
+    DUMMY       .req        r1      /* temporary register */
+    .set        regs_shortage, 1
+.endif
+
+    .set mask_bpp_shift, -1
+.if src_bpp == 32
+    .set src_bpp_shift, 2
+.elseif src_bpp == 24
+    .set src_bpp_shift, 0
+.elseif src_bpp == 16
+    .set src_bpp_shift, 1
+.elseif src_bpp == 8
+    .set src_bpp_shift, 0
+.elseif src_bpp == 0
+    .set src_bpp_shift, -1
+.else
+    .error "requested src bpp (src_bpp) is not supported"
+.endif
+.if mask_bpp == 32
+    .set mask_bpp_shift, 2
+.elseif mask_bpp == 24
+    .set mask_bpp_shift, 0
+.elseif mask_bpp == 8
+    .set mask_bpp_shift, 0
+.elseif mask_bpp == 0
+    .set mask_bpp_shift, -1
+.else
+    .error "requested mask bpp (mask_bpp) is not supported"
+.endif
+.if dst_w_bpp == 32
+    .set dst_bpp_shift, 2
+.elseif dst_w_bpp == 24
+    .set dst_bpp_shift, 0
+.elseif dst_w_bpp == 16
+    .set dst_bpp_shift, 1
+.elseif dst_w_bpp == 8
+    .set dst_bpp_shift, 0
+.else
+    .error "requested dst bpp (dst_w_bpp) is not supported"
+.endif
+
+.if (((flags) & FLAG_DST_READWRITE) != 0)
+    .set dst_r_bpp, dst_w_bpp
+.else
+    .set dst_r_bpp, 0
+.endif
+.if (((flags) & FLAG_DEINTERLEAVE_32BPP) != 0)
+    .set DEINTERLEAVE_32BPP_ENABLED, 1
+.else
+    .set DEINTERLEAVE_32BPP_ENABLED, 0
+.endif
+
+.if prefetch_distance < 0 || prefetch_distance > 15
+    .error "invalid prefetch distance (prefetch_distance)"
+.endif
+
+.if src_bpp > 0
+    ldr         SRC, [sp, #40]
+.endif
+.if mask_bpp > 0
+    ldr         MASK, [sp, #48]
+.endif
+    PF mov      PF_X, #0
+.if src_bpp > 0
+    ldr         SRC_STRIDE, [sp, #44]
+.endif
+.if mask_bpp > 0
+    ldr         MASK_STRIDE, [sp, #52]
+.endif
+    mov         DST_R, DST_W
+
+.if src_bpp == 24
+    sub         SRC_STRIDE, SRC_STRIDE, W
+    sub         SRC_STRIDE, SRC_STRIDE, W, lsl #1
+.endif
+.if mask_bpp == 24
+    sub         MASK_STRIDE, MASK_STRIDE, W
+    sub         MASK_STRIDE, MASK_STRIDE, W, lsl #1
+.endif
+.if dst_w_bpp == 24
+    sub         DST_STRIDE, DST_STRIDE, W
+    sub         DST_STRIDE, DST_STRIDE, W, lsl #1
+.endif
+
+/*
+ * Setup advanced prefetcher initial state
+ */
+    PF mov      PF_SRC, SRC
+    PF mov      PF_DST, DST_R
+    PF mov      PF_MASK, MASK
+    /* PF_CTL = prefetch_distance | ((h - 1) << 4) */
+    PF mov      PF_CTL, H, lsl #4
+    PF add      PF_CTL, #(prefetch_distance - 0x10)
+
+    init
+.if regs_shortage
+    push        {r0, r1}
+.endif
+    subs        H, H, #1
+.if regs_shortage
+    str         H, [sp, #4] /* save updated height to stack */
+.else
+    mov         ORIG_W, W
+.endif
+    blt         9f
+    cmp         W, #(pixblock_size * 2)
+    blt         8f
+/*
+ * This is the start of the pipelined loop, which if optimized for
+ * long scanlines
+ */
+0:
+    ensure_destination_ptr_alignment process_pixblock_head, \
+                                     process_pixblock_tail, \
+                                     process_pixblock_tail_head
+
+    /* Implement "head (tail_head) ... (tail_head) tail" loop pattern */
+    pixld_a     pixblock_size, dst_r_bpp, \
+                (dst_r_basereg - pixblock_size * dst_r_bpp / 64), DST_R
+    pixld       pixblock_size, src_bpp, \
+                (src_basereg - pixblock_size * src_bpp / 64), SRC
+    pixld       pixblock_size, mask_bpp, \
+                (mask_basereg - pixblock_size * mask_bpp / 64), MASK
+    PF add      PF_X, PF_X, #pixblock_size
+    process_pixblock_head
+    cache_preload 0, pixblock_size
+    cache_preload_simple
+    subs        W, W, #(pixblock_size * 2)
+    blt         2f
+1:
+    process_pixblock_tail_head
+    cache_preload_simple
+    subs        W, W, #pixblock_size
+    bge         1b
+2:
+    process_pixblock_tail
+    pixst_a     pixblock_size, dst_w_bpp, \
+                (dst_w_basereg - pixblock_size * dst_w_bpp / 64), DST_W
+
+    /* Process the remaining trailing pixels in the scanline */
+    process_trailing_pixels 1, 1, \
+                            process_pixblock_head, \
+                            process_pixblock_tail, \
+                            process_pixblock_tail_head
+    advance_to_next_scanline 0b
+
+.if regs_shortage
+    pop         {r0, r1}
+.endif
+    cleanup
+    pop         {r4-r12, pc}  /* exit */
+/*
+ * This is the start of the loop, designed to process images with small width
+ * (less than pixblock_size * 2 pixels). In this case neither pipelining
+ * nor prefetch are used.
+ */
+8:
+    /* Process exactly pixblock_size pixels if needed */
+    tst         W, #pixblock_size
+    beq         1f
+    pixld       pixblock_size, dst_r_bpp, \
+                (dst_r_basereg - pixblock_size * dst_r_bpp / 64), DST_R
+    pixld       pixblock_size, src_bpp, \
+                (src_basereg - pixblock_size * src_bpp / 64), SRC
+    pixld       pixblock_size, mask_bpp, \
+                (mask_basereg - pixblock_size * mask_bpp / 64), MASK
+    process_pixblock_head
+    process_pixblock_tail
+    pixst       pixblock_size, dst_w_bpp, \
+                (dst_w_basereg - pixblock_size * dst_w_bpp / 64), DST_W
+1:
+    /* Process the remaining trailing pixels in the scanline */
+    process_trailing_pixels 0, 0, \
+                            process_pixblock_head, \
+                            process_pixblock_tail, \
+                            process_pixblock_tail_head
+    advance_to_next_scanline 8b
+9:
+.if regs_shortage
+    pop         {r0, r1}
+.endif
+    cleanup
+    pop         {r4-r12, pc}  /* exit */
+
+    .unreq      SRC
+    .unreq      MASK
+    .unreq      DST_R
+    .unreq      DST_W
+    .unreq      ORIG_W
+    .unreq      W
+    .unreq      H
+    .unreq      SRC_STRIDE
+    .unreq      DST_STRIDE
+    .unreq      MASK_STRIDE
+    .unreq      PF_CTL
+    .unreq      PF_X
+    .unreq      PF_SRC
+    .unreq      PF_DST
+    .unreq      PF_MASK
+    .unreq      DUMMY
+    .endfunc
+.endm
+
+.macro default_init
+.endm
+
+.macro default_cleanup
+.endm
diff --git a/pixman/pixman/pixman-arm-neon.c b/pixman/pixman/pixman-arm-neon.c
index 8a2d72ea3..2ed8b4bdf 100644
--- a/pixman/pixman/pixman-arm-neon.c
+++ b/pixman/pixman/pixman-arm-neon.c
@@ -30,1670 +30,223 @@
 #include <config.h>
 #endif
 
-#include <arm_neon.h>
 #include <string.h>
 #include "pixman-private.h"
 
-/* Deal with an intrinsic that is defined differently in GCC */
-#if !defined(__ARMCC_VERSION) && !defined(__pld)
-#define __pld(_x) __builtin_prefetch (_x)
-#endif
-
-static force_inline uint8x8x4_t
-unpack0565 (uint16x8_t rgb)
-{
-    uint16x8_t gb, b;
-    uint8x8x4_t res;
-
-    res.val[3] = vdup_n_u8 (0);
-    gb = vshrq_n_u16 (rgb, 5);
-    b = vshrq_n_u16 (rgb, 5 + 6);
-
-    res.val[0] = vmovn_u16 (rgb);  /* get low 5 bits */
-    res.val[1] = vmovn_u16 (gb);   /* get mid 6 bits */
-    res.val[2] = vmovn_u16 (b);    /* get top 5 bits */
-
-    res.val[0] = vshl_n_u8 (res.val[0], 3); /* shift to top */
-    res.val[1] = vshl_n_u8 (res.val[1], 2); /* shift to top */
-    res.val[2] = vshl_n_u8 (res.val[2], 3); /* shift to top */
-
-    res.val[0] = vsri_n_u8 (res.val[0], res.val[0], 5);
-    res.val[1] = vsri_n_u8 (res.val[1], res.val[1], 6);
-    res.val[2] = vsri_n_u8 (res.val[2], res.val[2], 5);
-
-    return res;
-}
-
-#ifdef USE_GCC_INLINE_ASM
-/* Some versions of gcc have problems with vshll_n_u8 intrinsic (Bug 23576) */
-#define vshll_n_u8(a, n) ({ uint16x8_t r; \
-    asm ("vshll.u8 %q0, %P1, %2\n" : "=w" (r) : "w" (a), "i" (n)); r; })
-#endif
-
-static force_inline uint16x8_t
-pack0565 (uint8x8x4_t s)
-{
-    uint16x8_t rgb, val_g, val_r;
-
-    rgb = vshll_n_u8 (s.val[2], 8);
-    val_g = vshll_n_u8 (s.val[1], 8);
-    val_r = vshll_n_u8 (s.val[0], 8);
-    rgb = vsriq_n_u16 (rgb, val_g, 5);
-    rgb = vsriq_n_u16 (rgb, val_r, 5 + 6);
-
-    return rgb;
-}
-
-static force_inline uint8x8_t
-neon2mul (uint8x8_t x,
-          uint8x8_t alpha)
-{
-    uint16x8_t tmp, tmp2;
-    uint8x8_t res;
-
-    tmp = vmull_u8 (x, alpha);
-    tmp2 = vrshrq_n_u16 (tmp, 8);
-    res = vraddhn_u16 (tmp, tmp2);
-
-    return res;
-}
-
-static force_inline uint8x8x4_t
-neon8mul (uint8x8x4_t x,
-          uint8x8_t   alpha)
-{
-    uint16x8x4_t tmp;
-    uint8x8x4_t res;
-    uint16x8_t qtmp1, qtmp2;
-
-    tmp.val[0] = vmull_u8 (x.val[0], alpha);
-    tmp.val[1] = vmull_u8 (x.val[1], alpha);
-    tmp.val[2] = vmull_u8 (x.val[2], alpha);
-    tmp.val[3] = vmull_u8 (x.val[3], alpha);
-
-    qtmp1 = vrshrq_n_u16 (tmp.val[0], 8);
-    qtmp2 = vrshrq_n_u16 (tmp.val[1], 8);
-    res.val[0] = vraddhn_u16 (tmp.val[0], qtmp1);
-    qtmp1 = vrshrq_n_u16 (tmp.val[2], 8);
-    res.val[1] = vraddhn_u16 (tmp.val[1], qtmp2);
-    qtmp2 = vrshrq_n_u16 (tmp.val[3], 8);
-    res.val[2] = vraddhn_u16 (tmp.val[2], qtmp1);
-    res.val[3] = vraddhn_u16 (tmp.val[3], qtmp2);
-
-    return res;
-}
-
-static force_inline uint8x8x4_t
-neon8qadd (uint8x8x4_t x,
-           uint8x8x4_t y)
-{
-    uint8x8x4_t res;
-
-    res.val[0] = vqadd_u8 (x.val[0], y.val[0]);
-    res.val[1] = vqadd_u8 (x.val[1], y.val[1]);
-    res.val[2] = vqadd_u8 (x.val[2], y.val[2]);
-    res.val[3] = vqadd_u8 (x.val[3], y.val[3]);
-
-    return res;
-}
-
-static void
-neon_composite_add_8000_8000 (pixman_implementation_t * impl,
-                              pixman_op_t               op,
-                              pixman_image_t *          src_image,
-                              pixman_image_t *          mask_image,
-                              pixman_image_t *          dst_image,
-                              int32_t                   src_x,
-                              int32_t                   src_y,
-                              int32_t                   mask_x,
-                              int32_t                   mask_y,
-                              int32_t                   dest_x,
-                              int32_t                   dest_y,
-                              int32_t                   width,
-                              int32_t                   height)
-{
-    uint8_t     *dst_line, *dst;
-    uint8_t     *src_line, *src;
-    int dst_stride, src_stride;
-    uint16_t w;
-
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint8_t, src_stride, src_line, 1);
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
-
-    if (width >= 8)
-    {
-	/* Use overlapping 8-pixel method */
-	while (height--)
-	{
-	    uint8_t *keep_dst = 0;
-	    uint8x8_t sval, dval, temp;
-
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    src = src_line;
-	    src_line += src_stride;
-	    w = width;
-
-#ifndef USE_GCC_INLINE_ASM
-	    sval = vld1_u8 ((void *)src);
-	    dval = vld1_u8 ((void *)dst);
-	    keep_dst = dst;
-
-	    temp = vqadd_u8 (dval, sval);
-
-	    src += (w & 7);
-	    dst += (w & 7);
-	    w -= (w & 7);
-
-	    while (w)
-	    {
-		sval = vld1_u8 ((void *)src);
-		dval = vld1_u8 ((void *)dst);
-
-		vst1_u8 ((void *)keep_dst, temp);
-		keep_dst = dst;
-
-		temp = vqadd_u8 (dval, sval);
-
-		src += 8;
-		dst += 8;
-		w -= 8;
-	    }
-
-	    vst1_u8 ((void *)keep_dst, temp);
-#else
-	    asm volatile (
-/* avoid using d8-d15 (q4-q7) aapcs callee-save registers */
-	        "vld1.8  {d0}, [%[src]]\n\t"
-	        "vld1.8  {d4}, [%[dst]]\n\t"
-	        "mov     %[keep_dst], %[dst]\n\t"
-
-	        "and ip, %[w], #7\n\t"
-	        "add %[src], %[src], ip\n\t"
-	        "add %[dst], %[dst], ip\n\t"
-	        "subs %[w], %[w], ip\n\t"
-	        "b 9f\n\t"
-/* LOOP */
-	        "2:\n\t"
-	        "vld1.8  {d0}, [%[src]]!\n\t"
-	        "vld1.8  {d4}, [%[dst]]!\n\t"
-	        "vst1.8  {d20}, [%[keep_dst]]\n\t"
-	        "sub     %[keep_dst], %[dst], #8\n\t"
-	        "subs %[w], %[w], #8\n\t"
-	        "9:\n\t"
-	        "vqadd.u8 d20, d0, d4\n\t"
-
-	        "bne 2b\n\t"
-
-	        "1:\n\t"
-	        "vst1.8  {d20}, [%[keep_dst]]\n\t"
-
-		: [w] "+r" (w), [src] "+r" (src), [dst] "+r" (dst), [keep_dst] "=r" (keep_dst)
-		:
-		: "ip", "cc", "memory", "d0", "d4",
-	        "d20"
-	        );
-#endif
-	}
-    }
-    else
-    {
-	const uint8_t nil = 0;
-	const uint8x8_t vnil = vld1_dup_u8 (&nil);
-
-	while (height--)
-	{
-	    uint8x8_t sval = vnil, dval = vnil;
-	    uint8_t *dst4 = 0, *dst2 = 0;
-
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    src = src_line;
-	    src_line += src_stride;
-	    w = width;
-
-	    if (w & 4)
-	    {
-		sval = vreinterpret_u8_u32 (
-		    vld1_lane_u32 ((void *)src, vreinterpret_u32_u8 (sval), 1));
-		dval = vreinterpret_u8_u32 (
-		    vld1_lane_u32 ((void *)dst, vreinterpret_u32_u8 (dval), 1));
-
-		dst4 = dst;
-		src += 4;
-		dst += 4;
-	    }
-
-	    if (w & 2)
-	    {
-		sval = vreinterpret_u8_u16 (
-		    vld1_lane_u16 ((void *)src, vreinterpret_u16_u8 (sval), 1));
-		dval = vreinterpret_u8_u16 (
-		    vld1_lane_u16 ((void *)dst, vreinterpret_u16_u8 (dval), 1));
-
-		dst2 = dst;
-		src += 2;
-		dst += 2;
-	    }
-
-	    if (w & 1)
-	    {
-		sval = vld1_lane_u8 (src, sval, 1);
-		dval = vld1_lane_u8 (dst, dval, 1);
-	    }
-
-	    dval = vqadd_u8 (dval, sval);
-
-	    if (w & 1)
-		vst1_lane_u8 (dst, dval, 1);
-
-	    if (w & 2)
-		vst1_lane_u16 ((void *)dst2, vreinterpret_u16_u8 (dval), 1);
-
-	    if (w & 4)
-		vst1_lane_u32 ((void *)dst4, vreinterpret_u32_u8 (dval), 1);
-	}
-    }
-}
-
-static void
-neon_composite_over_8888_8888 (pixman_implementation_t * impl,
-                               pixman_op_t               op,
-                               pixman_image_t *          src_image,
-                               pixman_image_t *          mask_image,
-                               pixman_image_t *          dst_image,
-                               int32_t                   src_x,
-                               int32_t                   src_y,
-                               int32_t                   mask_x,
-                               int32_t                   mask_y,
-                               int32_t                   dest_x,
-                               int32_t                   dest_y,
-                               int32_t                   width,
-                               int32_t                   height)
-{
-    uint32_t    *dst_line, *dst;
-    uint32_t    *src_line, *src;
-    int dst_stride, src_stride;
-    uint32_t w;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-    if (width >= 8)
-    {
-	/* Use overlapping 8-pixel method */
-	while (height--)
-	{
-	    uint32_t *keep_dst = 0;
-	    uint8x8x4_t sval, dval, temp;
-
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    src = src_line;
-	    src_line += src_stride;
-	    w = width;
-
-#ifndef USE_GCC_INLINE_ASM
-	    sval = vld4_u8 ((void *)src);
-	    dval = vld4_u8 ((void *)dst);
-	    keep_dst = dst;
-
-	    temp = neon8mul (dval, vmvn_u8 (sval.val[3]));
-	    temp = neon8qadd (sval, temp);
-
-	    src += (w & 7);
-	    dst += (w & 7);
-	    w -= (w & 7);
-
-	    while (w)
-	    {
-		sval = vld4_u8 ((void *)src);
-		dval = vld4_u8 ((void *)dst);
-
-		vst4_u8 ((void *)keep_dst, temp);
-		keep_dst = dst;
-
-		temp = neon8mul (dval, vmvn_u8 (sval.val[3]));
-		temp = neon8qadd (sval, temp);
-
-		src += 8;
-		dst += 8;
-		w -= 8;
-	    }
-
-	    vst4_u8 ((void *)keep_dst, temp);
-#else
-	    asm volatile (
-/* avoid using d8-d15 (q4-q7) aapcs callee-save registers */
-	        "vld4.8  {d0-d3}, [%[src]]\n\t"
-	        "vld4.8  {d4-d7}, [%[dst]]\n\t"
-	        "mov     %[keep_dst], %[dst]\n\t"
-
-	        "and ip, %[w], #7\n\t"
-	        "add %[src], %[src], ip, LSL#2\n\t"
-	        "add %[dst], %[dst], ip, LSL#2\n\t"
-	        "subs %[w], %[w], ip\n\t"
-	        "b 9f\n\t"
-/* LOOP */
-	        "2:\n\t"
-	        "vld4.8  {d0-d3}, [%[src]]!\n\t"
-	        "vld4.8  {d4-d7}, [%[dst]]!\n\t"
-	        "vst4.8  {d20-d23}, [%[keep_dst]]\n\t"
-	        "sub     %[keep_dst], %[dst], #8*4\n\t"
-	        "subs %[w], %[w], #8\n\t"
-	        "9:\n\t"
-	        "vmvn.8  d31, d3\n\t"
-	        "vmull.u8 q10, d31, d4\n\t"
-	        "vmull.u8 q11, d31, d5\n\t"
-	        "vmull.u8 q12, d31, d6\n\t"
-	        "vmull.u8 q13, d31, d7\n\t"
-	        "vrshr.u16 q8, q10, #8\n\t"
-	        "vrshr.u16 q9, q11, #8\n\t"
-	        "vraddhn.u16 d20, q10, q8\n\t"
-	        "vraddhn.u16 d21, q11, q9\n\t"
-	        "vrshr.u16 q8, q12, #8\n\t"
-	        "vrshr.u16 q9, q13, #8\n\t"
-	        "vraddhn.u16 d22, q12, q8\n\t"
-	        "vraddhn.u16 d23, q13, q9\n\t"
-/* result in d20-d23 */
-	        "vqadd.u8 d20, d0, d20\n\t"
-	        "vqadd.u8 d21, d1, d21\n\t"
-	        "vqadd.u8 d22, d2, d22\n\t"
-	        "vqadd.u8 d23, d3, d23\n\t"
-
-	        "bne 2b\n\t"
-
-	        "1:\n\t"
-	        "vst4.8  {d20-d23}, [%[keep_dst]]\n\t"
-
-		: [w] "+r" (w), [src] "+r" (src), [dst] "+r" (dst), [keep_dst] "=r" (keep_dst)
-		:
-		: "ip", "cc", "memory", "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7",
-	        "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23"
-	        );
-#endif
-	}
-    }
-    else
-    {
-	uint8x8_t alpha_selector = vreinterpret_u8_u64 (
-	    vcreate_u64 (0x0707070703030303ULL));
-
-	/* Handle width < 8 */
-	while (height--)
-	{
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    src = src_line;
-	    src_line += src_stride;
-	    w = width;
-
-	    while (w >= 2)
-	    {
-		uint8x8_t sval, dval;
-
-		/* two 32-bit pixels packed into D-reg; ad-hoc vectorization */
-		sval = vreinterpret_u8_u32 (vld1_u32 ((void *)src));
-		dval = vreinterpret_u8_u32 (vld1_u32 ((void *)dst));
-		dval = neon2mul (dval, vtbl1_u8 (vmvn_u8 (sval), alpha_selector));
-		vst1_u8 ((void *)dst, vqadd_u8 (sval, dval));
-
-		src += 2;
-		dst += 2;
-		w -= 2;
-	    }
-
-	    if (w)
-	    {
-		uint8x8_t sval, dval;
-
-		/* single 32-bit pixel in lane 0 */
-		sval = vreinterpret_u8_u32 (vld1_dup_u32 ((void *)src));  /* only interested in lane 0 */
-		dval = vreinterpret_u8_u32 (vld1_dup_u32 ((void *)dst));  /* only interested in lane 0 */
-		dval = neon2mul (dval, vtbl1_u8 (vmvn_u8 (sval), alpha_selector));
-		vst1_lane_u32 ((void *)dst, vreinterpret_u32_u8 (vqadd_u8 (sval, dval)), 0);
-	    }
-	}
-    }
-}
-
-static void
-neon_composite_over_8888_n_8888 (pixman_implementation_t * impl,
-                                 pixman_op_t               op,
-                                 pixman_image_t *          src_image,
-                                 pixman_image_t *          mask_image,
-                                 pixman_image_t *          dst_image,
-                                 int32_t                   src_x,
-                                 int32_t                   src_y,
-                                 int32_t                   mask_x,
-                                 int32_t                   mask_y,
-                                 int32_t                   dest_x,
-                                 int32_t                   dest_y,
-                                 int32_t                   width,
-                                 int32_t                   height)
-{
-    uint32_t    *dst_line, *dst;
-    uint32_t    *src_line, *src;
-    uint32_t mask;
-    int dst_stride, src_stride;
-    uint32_t w;
-    uint8x8_t mask_alpha;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);
-    mask_alpha = vdup_n_u8 ((mask) >> 24);
-
-    if (width >= 8)
-    {
-	/* Use overlapping 8-pixel method */
-	while (height--)
-	{
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    src = src_line;
-	    src_line += src_stride;
-	    w = width;
-
-	    uint32_t *keep_dst = 0;
-
-#ifndef USE_GCC_INLINE_ASM
-	    uint8x8x4_t sval, dval, temp;
-
-	    sval = vld4_u8 ((void *)src);
-	    dval = vld4_u8 ((void *)dst);
-	    keep_dst = dst;
-
-	    sval = neon8mul (sval, mask_alpha);
-	    temp = neon8mul (dval, vmvn_u8 (sval.val[3]));
-	    temp = neon8qadd (sval, temp);
-
-	    src += (w & 7);
-	    dst += (w & 7);
-	    w -= (w & 7);
-
-	    while (w)
-	    {
-		sval = vld4_u8 ((void *)src);
-		dval = vld4_u8 ((void *)dst);
-
-		vst4_u8 ((void *)keep_dst, temp);
-		keep_dst = dst;
-
-		sval = neon8mul (sval, mask_alpha);
-		temp = neon8mul (dval, vmvn_u8 (sval.val[3]));
-		temp = neon8qadd (sval, temp);
-
-		src += 8;
-		dst += 8;
-		w -= 8;
-	    }
-	    vst4_u8 ((void *)keep_dst, temp);
-#else
-	    asm volatile (
-/* avoid using d8-d15 (q4-q7) aapcs callee-save registers */
-	        "vdup.32      d30, %[mask]\n\t"
-	        "vdup.8       d30, d30[3]\n\t"
-
-	        "vld4.8       {d0-d3}, [%[src]]\n\t"
-	        "vld4.8       {d4-d7}, [%[dst]]\n\t"
-	        "mov  %[keep_dst], %[dst]\n\t"
-
-	        "and  ip, %[w], #7\n\t"
-	        "add  %[src], %[src], ip, LSL#2\n\t"
-	        "add  %[dst], %[dst], ip, LSL#2\n\t"
-	        "subs  %[w], %[w], ip\n\t"
-	        "b 9f\n\t"
-/* LOOP */
-	        "2:\n\t"
-	        "vld4.8       {d0-d3}, [%[src]]!\n\t"
-	        "vld4.8       {d4-d7}, [%[dst]]!\n\t"
-	        "vst4.8       {d20-d23}, [%[keep_dst]]\n\t"
-	        "sub  %[keep_dst], %[dst], #8*4\n\t"
-	        "subs  %[w], %[w], #8\n\t"
-
-	        "9:\n\t"
-	        "vmull.u8     q10, d30, d0\n\t"
-	        "vmull.u8     q11, d30, d1\n\t"
-	        "vmull.u8     q12, d30, d2\n\t"
-	        "vmull.u8     q13, d30, d3\n\t"
-	        "vrshr.u16    q8, q10, #8\n\t"
-	        "vrshr.u16    q9, q11, #8\n\t"
-	        "vraddhn.u16  d0, q10, q8\n\t"
-	        "vraddhn.u16  d1, q11, q9\n\t"
-	        "vrshr.u16    q9, q13, #8\n\t"
-	        "vrshr.u16    q8, q12, #8\n\t"
-	        "vraddhn.u16  d3, q13, q9\n\t"
-	        "vraddhn.u16  d2, q12, q8\n\t"
-
-	        "vmvn.8       d31, d3\n\t"
-	        "vmull.u8     q10, d31, d4\n\t"
-	        "vmull.u8     q11, d31, d5\n\t"
-	        "vmull.u8     q12, d31, d6\n\t"
-	        "vmull.u8     q13, d31, d7\n\t"
-	        "vrshr.u16    q8, q10, #8\n\t"
-	        "vrshr.u16    q9, q11, #8\n\t"
-	        "vraddhn.u16  d20, q10, q8\n\t"
-	        "vrshr.u16    q8, q12, #8\n\t"
-	        "vraddhn.u16  d21, q11, q9\n\t"
-	        "vrshr.u16    q9, q13, #8\n\t"
-	        "vraddhn.u16  d22, q12, q8\n\t"
-	        "vraddhn.u16  d23, q13, q9\n\t"
-
-/* result in d20-d23 */
-	        "vqadd.u8     d20, d0, d20\n\t"
-	        "vqadd.u8     d21, d1, d21\n\t"
-	        "vqadd.u8     d22, d2, d22\n\t"
-	        "vqadd.u8     d23, d3, d23\n\t"
-
-	        "bne  2b\n\t"
-
-	        "1:\n\t"
-	        "vst4.8       {d20-d23}, [%[keep_dst]]\n\t"
-
-		: [w] "+r" (w), [src] "+r" (src), [dst] "+r" (dst), [keep_dst] "=r" (keep_dst)
-		: [mask] "r" (mask)
-		: "ip", "cc", "memory", "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7",
-	        "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", "d24", "d25", "d26", "d27",
-	        "d30", "d31"
-	        );
-#endif
-	}
-    }
-    else
-    {
-	uint8x8_t alpha_selector = vreinterpret_u8_u64 (vcreate_u64 (0x0707070703030303ULL));
-
-	/* Handle width < 8 */
-	while (height--)
-	{
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    src = src_line;
-	    src_line += src_stride;
-	    w = width;
-
-	    while (w >= 2)
-	    {
-		uint8x8_t sval, dval;
-
-		sval = vreinterpret_u8_u32 (vld1_u32 ((void *)src));
-		dval = vreinterpret_u8_u32 (vld1_u32 ((void *)dst));
-
-		/* sval * const alpha_mul */
-		sval = neon2mul (sval, mask_alpha);
-
-		/* dval * 255-(src alpha) */
-		dval = neon2mul (dval, vtbl1_u8 (vmvn_u8 (sval), alpha_selector));
-
-		vst1_u8 ((void *)dst, vqadd_u8 (sval, dval));
-
-		src += 2;
-		dst += 2;
-		w -= 2;
-	    }
-
-	    if (w)
-	    {
-		uint8x8_t sval, dval;
-
-		sval = vreinterpret_u8_u32 (vld1_dup_u32 ((void *)src));
-		dval = vreinterpret_u8_u32 (vld1_dup_u32 ((void *)dst));
-
-		/* sval * const alpha_mul */
-		sval = neon2mul (sval, mask_alpha);
-
-		/* dval * 255-(src alpha) */
-		dval = neon2mul (dval, vtbl1_u8 (vmvn_u8 (sval), alpha_selector));
-
-		vst1_lane_u32 ((void *)dst, vreinterpret_u32_u8 (vqadd_u8 (sval, dval)), 0);
-	    }
-	}
-    }
+#define BIND_SRC_NULL_DST(name, src_type, src_cnt, dst_type, dst_cnt)   \
+void                                                                    \
+pixman_##name##_asm_neon (int32_t   w,                                  \
+                          int32_t   h,                                  \
+                          dst_type *dst,                                \
+                          int32_t   dst_stride,                         \
+                          src_type *src,                                \
+                          int32_t   src_stride);                        \
+                                                                        \
+static void                                                             \
+neon_##name (pixman_implementation_t *imp,                              \
+             pixman_op_t              op,                               \
+             pixman_image_t *         src_image,                        \
+             pixman_image_t *         mask_image,                       \
+             pixman_image_t *         dst_image,                        \
+             int32_t                  src_x,                            \
+             int32_t                  src_y,                            \
+             int32_t                  mask_x,                           \
+             int32_t                  mask_y,                           \
+             int32_t                  dest_x,                           \
+             int32_t                  dest_y,                           \
+             int32_t                  width,                            \
+             int32_t                  height)                           \
+{                                                                       \
+    dst_type *dst_line;                                                 \
+    src_type *src_line;                                                 \
+    int32_t dst_stride, src_stride;                                     \
+                                                                        \
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, src_type,           \
+                           src_stride, src_line, src_cnt);              \
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, dst_type,         \
+                           dst_stride, dst_line, dst_cnt);              \
+                                                                        \
+    pixman_##name##_asm_neon (width, height,                            \
+                              dst_line, dst_stride,                     \
+                              src_line, src_stride);                    \
 }
 
-static void
-neon_composite_over_n_8_0565 (pixman_implementation_t * impl,
-			      pixman_op_t               op,
-			      pixman_image_t *          src_image,
-			      pixman_image_t *          mask_image,
-			      pixman_image_t *          dst_image,
-			      int32_t                   src_x,
-			      int32_t                   src_y,
-			      int32_t                   mask_x,
-			      int32_t                   mask_y,
-			      int32_t                   dest_x,
-			      int32_t                   dest_y,
-			      int32_t                   width,
-			      int32_t                   height)
-{
-    uint32_t     src, srca;
-    uint16_t    *dst_line, *dst;
-    uint8_t     *mask_line, *mask;
-    int          dst_stride, mask_stride;
-    uint32_t     w;
-    uint8x8_t    sval2;
-    uint8x8x4_t  sval8;
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    srca = src >> 24;
-    if (src == 0)
-	return;
-
-    sval2=vreinterpret_u8_u32 (vdup_n_u32 (src));
-    sval8.val[0]=vdup_lane_u8 (sval2,0);
-    sval8.val[1]=vdup_lane_u8 (sval2,1);
-    sval8.val[2]=vdup_lane_u8 (sval2,2);
-    sval8.val[3]=vdup_lane_u8 (sval2,3);
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-
-    if (width>=8)
-    {
-	/* Use overlapping 8-pixel method, modified to avoid rewritten dest being reused */
-	while (height--)
-	{
-	    uint16_t *keep_dst=0;
-
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    mask = mask_line;
-	    mask_line += mask_stride;
-	    w = width;
-
-#ifndef USE_GCC_INLINE_ASM
-	    uint8x8_t alpha;
-	    uint16x8_t dval, temp;
-	    uint8x8x4_t sval8temp;
-
-	    alpha = vld1_u8 ((void *)mask);
-	    dval = vld1q_u16 ((void *)dst);
-	    keep_dst = dst;
-
-	    sval8temp = neon8mul (sval8, alpha);
-	    temp = pack0565 (neon8qadd (sval8temp, neon8mul (unpack0565 (dval), vmvn_u8 (sval8temp.val[3]))));
-
-	    mask += (w & 7);
-	    dst += (w & 7);
-	    w -= (w & 7);
-
-	    while (w)
-	    {
-		dval = vld1q_u16 ((void *)dst);
-		alpha = vld1_u8 ((void *)mask);
-
-		vst1q_u16 ((void *)keep_dst, temp);
-		keep_dst = dst;
-
-		sval8temp = neon8mul (sval8, alpha);
-		temp = pack0565 (neon8qadd (sval8temp, neon8mul (unpack0565 (dval), vmvn_u8 (sval8temp.val[3]))));
-
-		mask+=8;
-		dst+=8;
-		w-=8;
-	    }
-	    vst1q_u16 ((void *)keep_dst, temp);
-#else
-	    asm volatile (
-		"vdup.32      d0, %[src]\n\t"
-		"vdup.8       d1, d0[1]\n\t"
-		"vdup.8       d2, d0[2]\n\t"
-		"vdup.8       d3, d0[3]\n\t"
-		"vdup.8       d0, d0[0]\n\t"
-
-		"vld1.8       {q12}, [%[dst]]\n\t"
-		"vld1.8       {d31}, [%[mask]]\n\t"
-		"mov  %[keep_dst], %[dst]\n\t"
-
-		"and  ip, %[w], #7\n\t"
-		"add  %[mask], %[mask], ip\n\t"
-		"add  %[dst], %[dst], ip, LSL#1\n\t"
-		"subs  %[w], %[w], ip\n\t"
-		"b  9f\n\t"
-/* LOOP */
-		"2:\n\t"
-
-		"vld1.16      {q12}, [%[dst]]!\n\t"
-		"vld1.8       {d31}, [%[mask]]!\n\t"
-		"vst1.16      {q10}, [%[keep_dst]]\n\t"
-		"sub  %[keep_dst], %[dst], #8*2\n\t"
-		"subs  %[w], %[w], #8\n\t"
-		"9:\n\t"
-/* expand 0565 q12 to 8888 {d4-d7} */
-		"vmovn.u16    d4, q12\t\n"
-		"vshr.u16     q11, q12, #5\t\n"
-		"vshr.u16     q10, q12, #6+5\t\n"
-		"vmovn.u16    d5, q11\t\n"
-		"vmovn.u16    d6, q10\t\n"
-		"vshl.u8      d4, d4, #3\t\n"
-		"vshl.u8      d5, d5, #2\t\n"
-		"vshl.u8      d6, d6, #3\t\n"
-		"vsri.u8      d4, d4, #5\t\n"
-		"vsri.u8      d5, d5, #6\t\n"
-		"vsri.u8      d6, d6, #5\t\n"
-
-		"vmull.u8     q10, d31, d0\n\t"
-		"vmull.u8     q11, d31, d1\n\t"
-		"vmull.u8     q12, d31, d2\n\t"
-		"vmull.u8     q13, d31, d3\n\t"
-		"vrshr.u16    q8, q10, #8\n\t"
-		"vrshr.u16    q9, q11, #8\n\t"
-		"vraddhn.u16  d20, q10, q8\n\t"
-		"vraddhn.u16  d21, q11, q9\n\t"
-		"vrshr.u16    q9, q13, #8\n\t"
-		"vrshr.u16    q8, q12, #8\n\t"
-		"vraddhn.u16  d23, q13, q9\n\t"
-		"vraddhn.u16  d22, q12, q8\n\t"
-
-/* duplicate in 4/2/1 & 8pix vsns */
-		"vmvn.8       d30, d23\n\t"
-		"vmull.u8     q14, d30, d6\n\t"
-		"vmull.u8     q13, d30, d5\n\t"
-		"vmull.u8     q12, d30, d4\n\t"
-		"vrshr.u16    q8, q14, #8\n\t"
-		"vrshr.u16    q9, q13, #8\n\t"
-		"vraddhn.u16  d6, q14, q8\n\t"
-		"vrshr.u16    q8, q12, #8\n\t"
-		"vraddhn.u16  d5, q13, q9\n\t"
-		"vqadd.u8     d6, d6, d22\n\t"  /* moved up */
-		"vraddhn.u16  d4, q12, q8\n\t"
-/* intentionally don't calculate alpha */
-/* result in d4-d6 */
-
-/*              "vqadd.u8     d6, d6, d22\n\t"  ** moved up */
-		"vqadd.u8     d5, d5, d21\n\t"
-		"vqadd.u8     d4, d4, d20\n\t"
-
-/* pack 8888 {d20-d23} to 0565 q10 */
-		"vshll.u8     q10, d6, #8\n\t"
-		"vshll.u8     q3, d5, #8\n\t"
-		"vshll.u8     q2, d4, #8\n\t"
-		"vsri.u16     q10, q3, #5\t\n"
-		"vsri.u16     q10, q2, #11\t\n"
-
-		"bne 2b\n\t"
-
-		"1:\n\t"
-		"vst1.16      {q10}, [%[keep_dst]]\n\t"
-
-		: [w] "+r" (w), [dst] "+r" (dst), [mask] "+r" (mask), [keep_dst] "=r" (keep_dst)
-		: [src] "r" (src)
-		: "ip", "cc", "memory", "d0","d1","d2","d3","d4","d5","d6","d7",
-		  "d16","d17","d18","d19","d20","d21","d22","d23","d24","d25","d26","d27","d28","d29",
-		  "d30","d31"
-		);
-#endif
-	}
-    }
-    else
-    {
-	while (height--)
-	{
-	    void *dst4=0, *dst2=0;
-
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    mask = mask_line;
-	    mask_line += mask_stride;
-	    w = width;
-
-
-#if 1 /* #ifndef USE_GCC_INLINE_ASM */
-	    uint8x8_t alpha;
-	    uint16x8_t dval, temp;
-	    uint8x8x4_t sval8temp;
-
-	    if (w&4)
-	    {
-		alpha = vreinterpret_u8_u32 (vld1_lane_u32 ((void *)mask, vreinterpret_u32_u8 (alpha),1));
-		dval = vreinterpretq_u16_u64 (vld1q_lane_u64 ((void *)dst, vreinterpretq_u64_u16 (dval),1));
-		dst4=dst;
-		mask+=4;
-		dst+=4;
-	    }
-	    if (w&2)
-	    {
-		alpha = vreinterpret_u8_u16 (vld1_lane_u16 ((void *)mask, vreinterpret_u16_u8 (alpha),1));
-		dval = vreinterpretq_u16_u32 (vld1q_lane_u32 ((void *)dst, vreinterpretq_u32_u16 (dval),1));
-		dst2=dst;
-		mask+=2;
-		dst+=2;
-	    }
-	    if (w&1)
-	    {
-		alpha = vld1_lane_u8 ((void *)mask, alpha,1);
-		dval = vld1q_lane_u16 ((void *)dst, dval,1);
-	    }
-
-	    sval8temp = neon8mul (sval8, alpha);
-	    temp = pack0565 (neon8qadd (sval8temp, neon8mul (unpack0565 (dval), vmvn_u8 (sval8temp.val[3]))));
-
-	    if (w&1)
-		vst1q_lane_u16 ((void *)dst, temp,1);
-	    if (w&2)
-		vst1q_lane_u32 ((void *)dst2, vreinterpretq_u32_u16 (temp),1);
-	    if (w&4)
-		vst1q_lane_u64 ((void *)dst4, vreinterpretq_u64_u16 (temp),1);
-#else
-	    /* this code has some bug (does not pass blitters-test) */
-	    asm volatile (
-		"vdup.32      d0, %[src]\n\t"
-		"vdup.8       d1, d0[1]\n\t"
-		"vdup.8       d2, d0[2]\n\t"
-		"vdup.8       d3, d0[3]\n\t"
-		"vdup.8       d0, d0[0]\n\t"
-
-		"tst  %[w], #4\t\n"
-		"beq  skip_load4\t\n"
-
-		"vld1.64      {d25}, [%[dst]]\n\t"
-		"vld1.32      {d31[1]}, [%[mask]]\n\t"
-		"mov  %[dst4], %[dst]\t\n"
-		"add  %[mask], %[mask], #4\t\n"
-		"add  %[dst], %[dst], #4*2\t\n"
-
-		"skip_load4:\t\n"
-		"tst  %[w], #2\t\n"
-		"beq  skip_load2\t\n"
-		"vld1.32      {d24[1]}, [%[dst]]\n\t"
-		"vld1.16      {d31[1]}, [%[mask]]\n\t"
-		"mov  %[dst2], %[dst]\t\n"
-		"add  %[mask], %[mask], #2\t\n"
-		"add  %[dst], %[dst], #2*2\t\n"
-
-		"skip_load2:\t\n"
-		"tst  %[w], #1\t\n"
-		"beq  skip_load1\t\n"
-		"vld1.16      {d24[1]}, [%[dst]]\n\t"
-		"vld1.8       {d31[1]}, [%[mask]]\n\t"
-
-		"skip_load1:\t\n"
-/* expand 0565 q12 to 8888 {d4-d7} */
-		"vmovn.u16    d4, q12\t\n"
-		"vshr.u16     q11, q12, #5\t\n"
-		"vshr.u16     q10, q12, #6+5\t\n"
-		"vmovn.u16    d5, q11\t\n"
-		"vmovn.u16    d6, q10\t\n"
-		"vshl.u8      d4, d4, #3\t\n"
-		"vshl.u8      d5, d5, #2\t\n"
-		"vshl.u8      d6, d6, #3\t\n"
-		"vsri.u8      d4, d4, #5\t\n"
-		"vsri.u8      d5, d5, #6\t\n"
-		"vsri.u8      d6, d6, #5\t\n"
-
-		"vmull.u8     q10, d31, d0\n\t"
-		"vmull.u8     q11, d31, d1\n\t"
-		"vmull.u8     q12, d31, d2\n\t"
-		"vmull.u8     q13, d31, d3\n\t"
-		"vrshr.u16    q8, q10, #8\n\t"
-		"vrshr.u16    q9, q11, #8\n\t"
-		"vraddhn.u16  d20, q10, q8\n\t"
-		"vraddhn.u16  d21, q11, q9\n\t"
-		"vrshr.u16    q9, q13, #8\n\t"
-		"vrshr.u16    q8, q12, #8\n\t"
-		"vraddhn.u16  d23, q13, q9\n\t"
-		"vraddhn.u16  d22, q12, q8\n\t"
-
-/* duplicate in 4/2/1 & 8pix vsns */
-		"vmvn.8       d30, d23\n\t"
-		"vmull.u8     q14, d30, d6\n\t"
-		"vmull.u8     q13, d30, d5\n\t"
-		"vmull.u8     q12, d30, d4\n\t"
-		"vrshr.u16    q8, q14, #8\n\t"
-		"vrshr.u16    q9, q13, #8\n\t"
-		"vraddhn.u16  d6, q14, q8\n\t"
-		"vrshr.u16    q8, q12, #8\n\t"
-		"vraddhn.u16  d5, q13, q9\n\t"
-		"vqadd.u8     d6, d6, d22\n\t"  /* moved up */
-		"vraddhn.u16  d4, q12, q8\n\t"
-/* intentionally don't calculate alpha */
-/* result in d4-d6 */
-
-/*              "vqadd.u8     d6, d6, d22\n\t"  ** moved up */
-		"vqadd.u8     d5, d5, d21\n\t"
-		"vqadd.u8     d4, d4, d20\n\t"
-
-/* pack 8888 {d20-d23} to 0565 q10 */
-		"vshll.u8     q10, d6, #8\n\t"
-		"vshll.u8     q3, d5, #8\n\t"
-		"vshll.u8     q2, d4, #8\n\t"
-		"vsri.u16     q10, q3, #5\t\n"
-		"vsri.u16     q10, q2, #11\t\n"
-
-		"tst  %[w], #1\n\t"
-		"beq skip_store1\t\n"
-		"vst1.16      {d20[1]}, [%[dst]]\t\n"
-		"skip_store1:\t\n"
-		"tst  %[w], #2\n\t"
-		"beq  skip_store2\t\n"
-		"vst1.32      {d20[1]}, [%[dst2]]\t\n"
-		"skip_store2:\t\n"
-		"tst  %[w], #4\n\t"
-		"beq skip_store4\t\n"
-		"vst1.16      {d21}, [%[dst4]]\t\n"
-		"skip_store4:\t\n"
-
-		: [w] "+r" (w), [dst] "+r" (dst), [mask] "+r" (mask), [dst4] "+r" (dst4), [dst2] "+r" (dst2)
-		: [src] "r" (src)
-		: "ip", "cc", "memory", "d0","d1","d2","d3","d4","d5","d6","d7",
-		  "d16","d17","d18","d19","d20","d21","d22","d23","d24","d25","d26","d27","d28","d29",
-		  "d30","d31"
-		);
-#endif
-	}
-    }
+#define BIND_N_MASK_DST(name, mask_type, mask_cnt, dst_type, dst_cnt)   \
+void                                                                    \
+pixman_##name##_asm_neon (int32_t    w,                                 \
+                          int32_t    h,                                 \
+                          dst_type  *dst,                               \
+                          int32_t    dst_stride,                        \
+                          uint32_t   src,                               \
+                          int32_t    unused,                            \
+                          mask_type *mask,                              \
+                          int32_t    mask_stride);                      \
+                                                                        \
+static void                                                             \
+neon_##name (pixman_implementation_t *imp,                              \
+             pixman_op_t              op,                               \
+             pixman_image_t *         src_image,                        \
+             pixman_image_t *         mask_image,                       \
+             pixman_image_t *         dst_image,                        \
+             int32_t                  src_x,                            \
+             int32_t                  src_y,                            \
+             int32_t                  mask_x,                           \
+             int32_t                  mask_y,                           \
+             int32_t                  dest_x,                           \
+             int32_t                  dest_y,                           \
+             int32_t                  width,                            \
+             int32_t                  height)                           \
+{                                                                       \
+    dst_type  *dst_line;                                                \
+    mask_type *mask_line;                                               \
+    int32_t    dst_stride, mask_stride;                                 \
+    uint32_t   src;                                                     \
+                                                                        \
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);  \
+                                                                        \
+    if (src == 0)                                                       \
+	return;                                                         \
+                                                                        \
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, dst_type,         \
+                           dst_stride, dst_line, dst_cnt);              \
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, mask_type,       \
+                           mask_stride, mask_line, mask_cnt);           \
+                                                                        \
+    pixman_##name##_asm_neon (width, height,                            \
+                              dst_line, dst_stride,                     \
+                              src, 0,                                   \
+                              mask_line, mask_stride);                  \
 }
 
-static void
-neon_composite_over_n_8_8888 (pixman_implementation_t * impl,
-                              pixman_op_t               op,
-                              pixman_image_t *          src_image,
-                              pixman_image_t *          mask_image,
-                              pixman_image_t *          dst_image,
-                              int32_t                   src_x,
-                              int32_t                   src_y,
-                              int32_t                   mask_x,
-                              int32_t                   mask_y,
-                              int32_t                   dest_x,
-                              int32_t                   dest_y,
-                              int32_t                   width,
-                              int32_t                   height)
-{
-    uint32_t src, srca;
-    uint32_t    *dst_line, *dst;
-    uint8_t     *mask_line, *mask;
-    int dst_stride, mask_stride;
-    uint32_t w;
-    uint8x8_t sval2;
-    uint8x8x4_t sval8;
-    uint8x8_t mask_selector = vreinterpret_u8_u64 (vcreate_u64 (0x0101010100000000ULL));
-    uint8x8_t alpha_selector = vreinterpret_u8_u64 (vcreate_u64 (0x0707070703030303ULL));
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-    
-    /* bail out if fully transparent */
-    srca = src >> 24;
-    if (src == 0)
-	return;
-
-    sval2 = vreinterpret_u8_u32 (vdup_n_u32 (src));
-    sval8.val[0] = vdup_lane_u8 (sval2, 0);
-    sval8.val[1] = vdup_lane_u8 (sval2, 1);
-    sval8.val[2] = vdup_lane_u8 (sval2, 2);
-    sval8.val[3] = vdup_lane_u8 (sval2, 3);
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-
-    if (width >= 8)
-    {
-	/* Use overlapping 8-pixel method, modified to avoid
-	 * rewritten dest being reused
-	 */
-	while (height--)
-	{
-	    uint32_t *keep_dst = 0;
-
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    mask = mask_line;
-	    mask_line += mask_stride;
-	    w = width;
-
-#ifndef USE_GCC_INLINE_ASM
-	    uint8x8_t alpha;
-	    uint8x8x4_t dval, temp;
-
-	    alpha = vld1_u8 ((void *)mask);
-	    dval = vld4_u8 ((void *)dst);
-	    keep_dst = dst;
-
-	    temp = neon8mul (sval8, alpha);
-	    dval = neon8mul (dval, vmvn_u8 (temp.val[3]));
-	    temp = neon8qadd (temp, dval);
-
-	    mask += (w & 7);
-	    dst += (w & 7);
-	    w -= (w & 7);
-
-	    while (w)
-	    {
-		alpha = vld1_u8 ((void *)mask);
-		dval = vld4_u8 ((void *)dst);
-
-		vst4_u8 ((void *)keep_dst, temp);
-		keep_dst = dst;
-
-		temp = neon8mul (sval8, alpha);
-		dval = neon8mul (dval, vmvn_u8 (temp.val[3]));
-		temp = neon8qadd (temp, dval);
-
-		mask += 8;
-		dst += 8;
-		w -= 8;
-	    }
-	    vst4_u8 ((void *)keep_dst, temp);
-#else
-	    asm volatile (
-	        "vdup.32      d0, %[src]\n\t"
-	        "vdup.8       d1, d0[1]\n\t"
-	        "vdup.8       d2, d0[2]\n\t"
-	        "vdup.8       d3, d0[3]\n\t"
-	        "vdup.8       d0, d0[0]\n\t"
-
-	        "vld4.8       {d4-d7}, [%[dst]]\n\t"
-	        "vld1.8       {d31}, [%[mask]]\n\t"
-	        "mov  %[keep_dst], %[dst]\n\t"
-
-	        "and  ip, %[w], #7\n\t"
-	        "add  %[mask], %[mask], ip\n\t"
-	        "add  %[dst], %[dst], ip, LSL#2\n\t"
-	        "subs  %[w], %[w], ip\n\t"
-	        "b 9f\n\t"
-/* LOOP */
-	        "2:\n\t"
-	        "vld4.8       {d4-d7}, [%[dst]]!\n\t"
-	        "vld1.8       {d31}, [%[mask]]!\n\t"
-	        "vst4.8       {d20-d23}, [%[keep_dst]]\n\t"
-	        "sub  %[keep_dst], %[dst], #8*4\n\t"
-	        "subs  %[w], %[w], #8\n\t"
-	        "9:\n\t"
-
-	        "vmull.u8     q10, d31, d0\n\t"
-	        "vmull.u8     q11, d31, d1\n\t"
-	        "vmull.u8     q12, d31, d2\n\t"
-	        "vmull.u8     q13, d31, d3\n\t"
-	        "vrshr.u16    q8, q10, #8\n\t"
-	        "vrshr.u16    q9, q11, #8\n\t"
-	        "vraddhn.u16  d20, q10, q8\n\t"
-	        "vraddhn.u16  d21, q11, q9\n\t"
-	        "vrshr.u16    q9, q13, #8\n\t"
-	        "vrshr.u16    q8, q12, #8\n\t"
-	        "vraddhn.u16  d23, q13, q9\n\t"
-	        "vraddhn.u16  d22, q12, q8\n\t"
-
-	        "vmvn.8       d30, d23\n\t"
-	        "vmull.u8     q12, d30, d4\n\t"
-	        "vmull.u8     q13, d30, d5\n\t"
-	        "vmull.u8     q14, d30, d6\n\t"
-	        "vmull.u8     q15, d30, d7\n\t"
-
-	        "vrshr.u16    q8, q12, #8\n\t"
-	        "vrshr.u16    q9, q13, #8\n\t"
-	        "vraddhn.u16  d4, q12, q8\n\t"
-	        "vrshr.u16    q8, q14, #8\n\t"
-	        "vraddhn.u16  d5, q13, q9\n\t"
-	        "vrshr.u16    q9, q15, #8\n\t"
-	        "vraddhn.u16  d6, q14, q8\n\t"
-	        "vraddhn.u16  d7, q15, q9\n\t"
-/* result in d4-d7 */
-
-	        "vqadd.u8     d20, d4, d20\n\t"
-	        "vqadd.u8     d21, d5, d21\n\t"
-	        "vqadd.u8     d22, d6, d22\n\t"
-	        "vqadd.u8     d23, d7, d23\n\t"
-
-	        "bne 2b\n\t"
-
-	        "1:\n\t"
-	        "vst4.8       {d20-d23}, [%[keep_dst]]\n\t"
-
-		: [w] "+r" (w), [dst] "+r" (dst), [mask] "+r" (mask), [keep_dst] "=r" (keep_dst)
-		: [src] "r" (src)
-		: "ip", "cc", "memory", "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7",
-	        "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", "d24", "d25", "d26", "d27", "d28", "d29",
-	        "d30", "d31"
-	        );
-#endif
-	}
-    }
-    else
-    {
-	while (height--)
-	{
-	    uint8x8_t alpha;
-
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    mask = mask_line;
-	    mask_line += mask_stride;
-	    w = width;
-
-	    while (w >= 2)
-	    {
-		uint8x8_t dval, temp, res;
-
-		alpha = vtbl1_u8 (
-		    vreinterpret_u8_u16 (vld1_dup_u16 ((void *)mask)), mask_selector);
-		dval = vld1_u8 ((void *)dst);
-
-		temp = neon2mul (sval2, alpha);
-		res = vqadd_u8 (
-		    temp, neon2mul (dval, vtbl1_u8 (vmvn_u8 (temp), alpha_selector)));
-
-		vst1_u8 ((void *)dst, res);
-
-		mask += 2;
-		dst += 2;
-		w -= 2;
-	    }
-
-	    if (w)
-	    {
-		uint8x8_t dval, temp, res;
-
-		alpha = vtbl1_u8 (vld1_dup_u8 ((void *)mask), mask_selector);
-		dval = vreinterpret_u8_u32 (vld1_dup_u32 ((void *)dst));
-
-		temp = neon2mul (sval2, alpha);
-		res = vqadd_u8 (
-		    temp, neon2mul (dval, vtbl1_u8 (vmvn_u8 (temp), alpha_selector)));
-
-		vst1_lane_u32 ((void *)dst, vreinterpret_u32_u8 (res), 0);
-	    }
-	}
-    }
-}
-
-static void
-neon_composite_add_8888_8_8 (pixman_implementation_t * impl,
-                             pixman_op_t               op,
-                             pixman_image_t *          src_image,
-                             pixman_image_t *          mask_image,
-                             pixman_image_t *          dst_image,
-                             int32_t                   src_x,
-                             int32_t                   src_y,
-                             int32_t                   mask_x,
-                             int32_t                   mask_y,
-                             int32_t                   dest_x,
-                             int32_t                   dest_y,
-                             int32_t                   width,
-                             int32_t                   height)
-{
-    uint8_t     *dst_line, *dst;
-    uint8_t     *mask_line, *mask;
-    int dst_stride, mask_stride;
-    uint32_t w;
-    uint32_t src;
-    uint8x8_t sa;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-    sa = vdup_n_u8 ((src) >> 24);
-
-    if (width >= 8)
-    {
-	/* Use overlapping 8-pixel method, modified to avoid rewritten dest being reused */
-	while (height--)
-	{
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    mask = mask_line;
-	    mask_line += mask_stride;
-	    w = width;
-
-	    uint8x8_t mval, dval, res;
-	    uint8_t     *keep_dst;
-
-	    mval = vld1_u8 ((void *)mask);
-	    dval = vld1_u8 ((void *)dst);
-	    keep_dst = dst;
-
-	    res = vqadd_u8 (neon2mul (mval, sa), dval);
-
-	    mask += (w & 7);
-	    dst += (w & 7);
-	    w -= w & 7;
-
-	    while (w)
-	    {
-		mval = vld1_u8 ((void *)mask);
-		dval = vld1_u8 ((void *)dst);
-		vst1_u8 ((void *)keep_dst, res);
-		keep_dst = dst;
-
-		res = vqadd_u8 (neon2mul (mval, sa), dval);
-
-		mask += 8;
-		dst += 8;
-		w -= 8;
-	    }
-	    vst1_u8 ((void *)keep_dst, res);
-	}
-    }
-    else
-    {
-	/* Use 4/2/1 load/store method to handle 1-7 pixels */
-	while (height--)
-	{
-	    dst = dst_line;
-	    dst_line += dst_stride;
-	    mask = mask_line;
-	    mask_line += mask_stride;
-	    w = width;
-
-	    uint8x8_t mval = sa, dval = sa, res;
-	    uint8_t *dst4 = 0, *dst2 = 0;
-
-	    if (w & 4)
-	    {
-		mval = vreinterpret_u8_u32 (
-		    vld1_lane_u32 ((void *)mask, vreinterpret_u32_u8 (mval), 1));
-		dval = vreinterpret_u8_u32 (
-		    vld1_lane_u32 ((void *)dst, vreinterpret_u32_u8 (dval), 1));
-
-		dst4 = dst;
-		mask += 4;
-		dst += 4;
-	    }
-
-	    if (w & 2)
-	    {
-		mval = vreinterpret_u8_u16 (
-		    vld1_lane_u16 ((void *)mask, vreinterpret_u16_u8 (mval), 1));
-		dval = vreinterpret_u8_u16 (
-		    vld1_lane_u16 ((void *)dst, vreinterpret_u16_u8 (dval), 1));
-		dst2 = dst;
-		mask += 2;
-		dst += 2;
-	    }
-
-	    if (w & 1)
-	    {
-		mval = vld1_lane_u8 (mask, mval, 1);
-		dval = vld1_lane_u8 (dst, dval, 1);
-	    }
-
-	    res = vqadd_u8 (neon2mul (mval, sa), dval);
-
-	    if (w & 1)
-		vst1_lane_u8 (dst, res, 1);
-	    if (w & 2)
-		vst1_lane_u16 ((void *)dst2, vreinterpret_u16_u8 (res), 1);
-	    if (w & 4)
-		vst1_lane_u32 ((void *)dst4, vreinterpret_u32_u8 (res), 1);
-	}
-    }
+#define BIND_SRC_N_DST(name, src_type, src_cnt, dst_type, dst_cnt)      \
+void                                                                    \
+pixman_##name##_asm_neon (int32_t    w,                                 \
+                          int32_t    h,                                 \
+                          dst_type  *dst,                               \
+                          int32_t    dst_stride,                        \
+                          src_type  *src,                               \
+                          int32_t    src_stride,                        \
+                          uint32_t   mask);                             \
+                                                                        \
+static void                                                             \
+neon_##name (pixman_implementation_t *imp,                              \
+             pixman_op_t              op,                               \
+             pixman_image_t *         src_image,                        \
+             pixman_image_t *         mask_image,                       \
+             pixman_image_t *         dst_image,                        \
+             int32_t                  src_x,                            \
+             int32_t                  src_y,                            \
+             int32_t                  mask_x,                           \
+             int32_t                  mask_y,                           \
+             int32_t                  dest_x,                           \
+             int32_t                  dest_y,                           \
+             int32_t                  width,                            \
+             int32_t                  height)                           \
+{                                                                       \
+    dst_type  *dst_line;                                                \
+    src_type  *src_line;                                                \
+    int32_t    dst_stride, src_stride;                                  \
+    uint32_t   mask;                                                    \
+                                                                        \
+    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);\
+                                                                        \
+    if (mask == 0)                                                      \
+	return;                                                         \
+                                                                        \
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, dst_type,         \
+                           dst_stride, dst_line, dst_cnt);              \
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, src_type,           \
+                           src_stride, src_line, src_cnt);              \
+                                                                        \
+    pixman_##name##_asm_neon (width, height,                            \
+                              dst_line, dst_stride,                     \
+                              src_line, src_stride,                     \
+                              mask);                                    \
 }
 
-#ifdef USE_GCC_INLINE_ASM
-
-static void
-neon_composite_src_16_16 (pixman_implementation_t * impl,
-                          pixman_op_t               op,
-                          pixman_image_t *          src_image,
-                          pixman_image_t *          mask_image,
-                          pixman_image_t *          dst_image,
-                          int32_t                   src_x,
-                          int32_t                   src_y,
-                          int32_t                   mask_x,
-                          int32_t                   mask_y,
-                          int32_t                   dest_x,
-                          int32_t                   dest_y,
-                          int32_t                   width,
-                          int32_t                   height)
-{
-    uint16_t    *dst_line, *src_line;
-    uint32_t dst_stride, src_stride;
-
-    if (!height || !width)
-	return;
-
-    /* We simply copy 16-bit-aligned pixels from one place to another. */
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint16_t, src_stride, src_line, 1);
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-
-    /* Preload the first input scanline */
-    {
-	uint16_t *src_ptr = src_line;
-	uint32_t count = width;
-
-	asm volatile (
-	    "0: @ loop							\n"
-	    "	subs    %[count], %[count], #32				\n"
-	    "	pld     [%[src]]					\n"
-	    "	add     %[src], %[src], #64				\n"
-	    "	bgt 0b							\n"
-
-	    /* Clobbered input registers marked as input/outputs */
-	    : [src] "+r" (src_ptr), [count] "+r" (count)
-	    :     /* no unclobbered inputs */
-	    : "cc"
-	    );
-    }
-
-    while (height--)
-    {
-	uint16_t *dst_ptr = dst_line;
-	uint16_t *src_ptr = src_line;
-	uint32_t count = width;
-	uint32_t tmp = 0;
-
-	/* Uses multi-register access and preloading to maximise bandwidth.
-	 * Each pixel is one halfword, so a quadword contains 8px.
-	 * Preload frequency assumed a 64-byte cacheline.
-	 */
-	asm volatile (
-	    "	cmp       %[count], #64				\n"
-	    "	blt 1f    @ skip oversized fragments		\n"
-	    "0: @ start with eight quadwords at a time		\n"
-	    /* preload from next scanline */
-	    "	pld       [%[src], %[src_stride], LSL #1]	\n"
-	    "	sub       %[count], %[count], #64		\n"
-	    "	vld1.16   {d16, d17, d18, d19}, [%[src]]!		\n"
-	    "	vld1.16   {d20, d21, d22, d23}, [%[src]]!		\n"
-	    /* preload from next scanline */
-	    "	pld       [%[src], %[src_stride], LSL #1]	\n"
-	    "	vld1.16   {d24, d25, d26, d27}, [%[src]]!		\n"
-	    "	vld1.16   {d28, d29, d30, d31}, [%[src]]!		\n"
-	    "	cmp       %[count], #64				\n"
-	    "	vst1.16   {d16, d17, d18, d19}, [%[dst]]!		\n"
-	    "	vst1.16   {d20, d21, d22, d23}, [%[dst]]!		\n"
-	    "	vst1.16   {d24, d25, d26, d27}, [%[dst]]!		\n"
-	    "	vst1.16   {d28, d29, d30, d31}, [%[dst]]!		\n"
-	    "	bge 0b						\n"
-	    "	cmp       %[count], #0				\n"
-	    "	beq 7f    @ aligned fastpath			\n"
-	    "1: @ four quadwords				\n"
-	    "	tst       %[count], #32				\n"
-	    "	beq 2f    @ skip oversized fragment		\n"
-	    /* preload from next scanline */
-	    "	pld       [%[src], %[src_stride], LSL #1]	\n"
-	    "	vld1.16   {d16, d17, d18, d19}, [%[src]]!		\n"
-	    "	vld1.16   {d20, d21, d22, d23}, [%[src]]!		\n"
-	    "	vst1.16   {d16, d17, d18, d19}, [%[dst]]!		\n"
-	    "	vst1.16   {d20, d21, d22, d23}, [%[dst]]!		\n"
-	    "2: @ two quadwords					\n"
-	    "	tst       %[count], #16				\n"
-	    "	beq 3f    @ skip oversized fragment		\n"
-	    /* preload from next scanline */
-	    "	pld       [%[src], %[src_stride], LSL #1]	\n"
-	    "	vld1.16   {d16, d17, d18, d19}, [%[src]]!		\n"
-	    "	vst1.16   {d16, d17, d18, d19}, [%[dst]]!		\n"
-	    "3: @ one quadword					\n"
-	    "	tst       %[count], #8				\n"
-	    "	beq 4f    @ skip oversized fragment		\n"
-	    "	vld1.16   {d16, d17}, [%[src]]!			\n"
-	    "	vst1.16   {d16, d17}, [%[dst]]!			\n"
-	    "4: @ one doubleword				\n"
-	    "	tst       %[count], #4				\n"
-	    "	beq 5f    @ skip oversized fragment		\n"
-	    "	vld1.16   {d16}, [%[src]]!			\n"
-	    "	vst1.16   {d16}, [%[dst]]!			\n"
-	    "5: @ one word					\n"
-	    "	tst       %[count], #2				\n"
-	    "	beq 6f    @ skip oversized fragment		\n"
-	    "	ldr       %[tmp], [%[src]], #4			\n"
-	    "	str       %[tmp], [%[dst]], #4			\n"
-	    "6: @ one halfword					\n"
-	    "	tst       %[count], #1				\n"
-	    "	beq 7f    @ skip oversized fragment		\n"
-	    "	ldrh      %[tmp], [%[src]]			\n"
-	    "	strh      %[tmp], [%[dst]]			\n"
-	    "7: @ end						\n"
-
-	    /* Clobbered input registers marked as input/outputs */
-	    : [dst] "+r" (dst_ptr), [src] "+r" (src_ptr),
-	      [count] "+r" (count), [tmp] "+r" (tmp)
-
-	      /* Unclobbered input */
-	    : [src_stride] "r" (src_stride)
-
-	      /* Clobbered vector registers */
-	    : "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23",
-	      "d24", "d25", "d26", "d27", "d28", "d29", "d30", "d31", "cc", "memory"
-	    );
-
-	src_line += src_stride;
-	dst_line += dst_stride;
-    }
+#define BIND_SRC_MASK_DST(name, src_type, src_cnt, mask_type, mask_cnt, \
+                          dst_type, dst_cnt)                            \
+void                                                                    \
+pixman_##name##_asm_neon (int32_t    w,                                 \
+                          int32_t    h,                                 \
+                          dst_type  *dst,                               \
+                          int32_t    dst_stride,                        \
+                          src_type  *src,                               \
+                          int32_t    src_stride,                        \
+                          mask_type *mask,                              \
+                          int32_t    mask_stride);                      \
+                                                                        \
+static void                                                             \
+neon_##name (pixman_implementation_t *imp,                              \
+             pixman_op_t              op,                               \
+             pixman_image_t *         src_image,                        \
+             pixman_image_t *         mask_image,                       \
+             pixman_image_t *         dst_image,                        \
+             int32_t                  src_x,                            \
+             int32_t                  src_y,                            \
+             int32_t                  mask_x,                           \
+             int32_t                  mask_y,                           \
+             int32_t                  dest_x,                           \
+             int32_t                  dest_y,                           \
+             int32_t                  width,                            \
+             int32_t                  height)                           \
+{                                                                       \
+    dst_type  *dst_line;                                                \
+    src_type  *src_line;                                                \
+    mask_type *mask_line;                                               \
+    int32_t    dst_stride, src_stride, mask_stride;                     \
+                                                                        \
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, dst_type,         \
+                           dst_stride, dst_line, dst_cnt);              \
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, src_type,           \
+                           src_stride, src_line, src_cnt);              \
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, mask_type,       \
+                           mask_stride, mask_line, mask_cnt);           \
+                                                                        \
+    pixman_##name##_asm_neon (width, height,                            \
+                              dst_line, dst_stride,                     \
+                              src_line, src_stride,                     \
+                              mask_line, mask_stride);                  \
 }
 
-#endif /* USE_GCC_INLINE_ASM */
-
-static void
-neon_composite_src_24_16 (pixman_implementation_t * impl,
-                          pixman_op_t               op,
-                          pixman_image_t *          src_image,
-                          pixman_image_t *          mask_image,
-                          pixman_image_t *          dst_image,
-                          int32_t                   src_x,
-                          int32_t                   src_y,
-                          int32_t                   mask_x,
-                          int32_t                   mask_y,
-                          int32_t                   dest_x,
-                          int32_t                   dest_y,
-                          int32_t                   width,
-                          int32_t                   height)
-{
-    uint16_t    *dst_line;
-    uint32_t    *src_line;
-    uint32_t dst_stride, src_stride;
-
-    if (!width || !height)
-	return;
-
-    /* We simply copy pixels from one place to another,
-     * assuming that the source's alpha is opaque.
-     */
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-
-    /* Preload the first input scanline */
-    {
-	uint8_t *src_ptr = (uint8_t*) src_line;
-	uint32_t count = (width + 15) / 16;
 
-#ifdef USE_GCC_INLINE_ASM
-	asm volatile (
-	    "0: @ loop						\n"
-	    "	subs    %[count], %[count], #1			\n"
-	    "	pld     [%[src]]				\n"
-	    "	add     %[src], %[src], #64			\n"
-	    "	bgt 0b						\n"
+BIND_SRC_NULL_DST(composite_src_8888_8888, uint32_t, 1, uint32_t, 1)
+BIND_SRC_NULL_DST(composite_src_0565_0565, uint16_t, 1, uint16_t, 1)
+BIND_SRC_NULL_DST(composite_src_0888_0888, uint8_t, 3, uint8_t, 3)
+BIND_SRC_NULL_DST(composite_src_8888_0565, uint32_t, 1, uint16_t, 1)
+BIND_SRC_NULL_DST(composite_add_8000_8000, uint8_t, 1, uint8_t, 1)
 
-	    /* Clobbered input registers marked as input/outputs */
-	    : [src] "+r" (src_ptr), [count] "+r" (count)
-	    :     /* no unclobbered inputs */
-	    : "cc"
-	    );
-#else
-	do
-	{
-	    __pld (src_ptr);
-	    src_ptr += 64;
-	}
-	while (--count);
-#endif
-    }
-
-    while (height--)
-    {
-	uint16_t *dst_ptr = dst_line;
-	uint32_t *src_ptr = src_line;
-	uint32_t count = width;
-	const uint32_t rb_mask = 0x1F;
-	const uint32_t g_mask = 0x3F;
+BIND_SRC_NULL_DST(composite_over_8888_0565, uint32_t, 1, uint16_t, 1)
+BIND_SRC_NULL_DST(composite_over_8888_8888, uint32_t, 1, uint32_t, 1)
 
-	/* If you're going to complain about a goto, take a long hard look
-	 * at the massive blocks of assembler this skips over.  ;-)
-	 */
-	if (count < 8)
-	    goto small_stuff;
+BIND_N_MASK_DST(composite_over_n_8_0565, uint8_t, 1, uint16_t, 1)
+BIND_N_MASK_DST(composite_over_n_8_8888, uint8_t, 1, uint32_t, 1)
+BIND_N_MASK_DST(composite_add_n_8_8, uint8_t, 1, uint8_t, 1)
 
-#ifdef USE_GCC_INLINE_ASM
+BIND_SRC_N_DST(composite_over_8888_n_8888, uint32_t, 1, uint32_t, 1)
 
-	/* This is not as aggressive as the RGB565-source case.
-	 * Generally the source is in cached RAM when the formats are
-	 * different, so we use preload.
-	 * 
-	 * We don't need to blend, so we are not reading from the
-	 * uncached framebuffer.
-	 */
-	asm volatile (
-	    "	cmp       %[count], #16				\n"
-	    "	blt 1f    @ skip oversized fragments		\n"
-	    "0: @ start with sixteen pixels at a time		\n"
-	    "	sub       %[count], %[count], #16		\n"
-	    "	pld      [%[src], %[src_stride], lsl #2]        @ preload from next scanline			\n"
-	    "	vld4.8    {d0, d1, d2, d3}, [%[src]]!		@ d3 is alpha and ignored, d2-0 are rgb.	\n"
-	    "	vld4.8    {d4, d5, d6, d7}, [%[src]]!		@ d7 is alpha and ignored, d6-4 are rgb.	\n"
-	    "	vshll.u8  q8, d2, #8				@ expand first red for repacking		\n"
-	    "	vshll.u8  q10, d1, #8				@ expand first green for repacking		\n"
-	    "	vshll.u8  q11, d0, #8				@ expand first blue for repacking		\n"
-	    "	vshll.u8  q9, d6, #8				@ expand second red for repacking		\n"
-	    "	vsri.u16  q8, q10, #5				@ insert first green after red			\n"
-	    "	vshll.u8  q10, d5, #8				@ expand second green for repacking		\n"
-	    "	vsri.u16  q8, q11, #11				@ insert first blue after green			\n"
-	    "	vshll.u8  q11, d4, #8				@ expand second blue for repacking		\n"
-	    "	vsri.u16  q9, q10, #5				@ insert second green after red			\n"
-	    "	vsri.u16  q9, q11, #11				@ insert second blue after green		\n"
-	    "	cmp       %[count], #16				\n"
-	    "	vst1.16   {d16, d17, d18, d19}, [%[dst]]!          @ store 16 pixels				\n"
-	    "	bge 0b						\n"
-	    "1: @ end of main loop				\n"
-	    "	cmp       %[count], #8				@ can we still do an 8-pixel block?		\n"
-	    "	blt 2f						\n"
-	    "	sub       %[count], %[count], #8		\n"
-	    "	pld      [%[src], %[src_stride], lsl #2]        @ preload from next scanline			\n"
-	    "	vld4.8    {d0, d1, d2, d3}, [%[src]]!		@ d3 is alpha and ignored, d2-0 are rgb.	\n"
-	    "	vshll.u8  q8, d2, #8				@ expand first red for repacking		\n"
-	    "	vshll.u8  q10, d1, #8				@ expand first green for repacking		\n"
-	    "	vshll.u8  q11, d0, #8				@ expand first blue for repacking		\n"
-	    "	vsri.u16  q8, q10, #5				@ insert first green after red			\n"
-	    "	vsri.u16  q8, q11, #11				@ insert first blue after green			\n"
-	    "	vst1.16   {d16, d17}, [%[dst]]!          @ store 8 pixels				\n"
-	    "2: @ end						\n"
+BIND_SRC_MASK_DST(composite_add_8_8_8, uint8_t, 1, uint8_t, 1, uint8_t, 1)
 
-	    /* Clobbered input and working registers marked as input/outputs */
-	    : [dst] "+r" (dst_ptr), [src] "+r" (src_ptr), [count] "+r" (count)
+void
+pixman_composite_src_n_8_asm_neon (int32_t   w,
+                                   int32_t   h,
+                                   uint8_t  *dst,
+                                   int32_t   dst_stride,
+                                   uint8_t   src);
 
-	      /* Unclobbered input */
-	    : [src_stride] "r" (src_stride)
+void
+pixman_composite_src_n_0565_asm_neon (int32_t   w,
+                                      int32_t   h,
+                                      uint16_t *dst,
+                                      int32_t   dst_stride,
+                                      uint16_t  src);
 
-	      /* Clobbered vector registers */
-
-	      /* NB: these are the quad aliases of the
-	       * double registers used in the asm
-	       */
-	    : "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d16", "d17",
-	      "d18", "d19", "d20", "d21", "d22", "d23", "cc", "memory"
-	    );
-#else
-	/* A copy of the above code, in intrinsics-form. */
-	while (count >= 16)
-	{
-	    uint8x8x4_t pixel_set_a, pixel_set_b;
-	    uint16x8_t red_a, green_a, blue_a;
-	    uint16x8_t red_b, green_b, blue_b;
-	    uint16x8_t dest_pixels_a, dest_pixels_b;
-
-	    count -= 16;
-	    __pld (src_ptr + src_stride);
-	    pixel_set_a = vld4_u8 ((uint8_t*)(src_ptr));
-	    pixel_set_b = vld4_u8 ((uint8_t*)(src_ptr + 8));
-	    src_ptr += 16;
-
-	    red_a   = vshll_n_u8 (pixel_set_a.val[2], 8);
-	    green_a = vshll_n_u8 (pixel_set_a.val[1], 8);
-	    blue_a  = vshll_n_u8 (pixel_set_a.val[0], 8);
-	    
-	    red_b   = vshll_n_u8 (pixel_set_b.val[2], 8);
-	    green_b = vshll_n_u8 (pixel_set_b.val[1], 8);
-	    blue_b  = vshll_n_u8 (pixel_set_b.val[0], 8);
-	    
-	    dest_pixels_a = vsriq_n_u16 (red_a, green_a, 5);
-	    dest_pixels_b = vsriq_n_u16 (red_b, green_b, 5);
-	    
-	    dest_pixels_a = vsriq_n_u16 (dest_pixels_a, blue_a, 11);
-	    dest_pixels_b = vsriq_n_u16 (dest_pixels_b, blue_b, 11);
-
-	    /* There doesn't seem to be an intrinsic for the
-	     * double-quadword variant
-	     */
-	    vst1q_u16 (dst_ptr, dest_pixels_a);
-	    vst1q_u16 (dst_ptr + 8, dest_pixels_b);
-	    dst_ptr += 16;
-	}
-
-	/* 8-pixel loop */
-	if (count >= 8)
-	{
-	    uint8x8x4_t pixel_set_a;
-	    uint16x8_t red_a, green_a, blue_a;
-	    uint16x8_t dest_pixels_a;
-
-	    __pld (src_ptr + src_stride);
-	    count -= 8;
-	    pixel_set_a = vld4_u8 ((uint8_t*)(src_ptr));
-	    src_ptr += 8;
-
-	    red_a   = vshll_n_u8 (pixel_set_a.val[2], 8);
-	    green_a = vshll_n_u8 (pixel_set_a.val[1], 8);
-	    blue_a  = vshll_n_u8 (pixel_set_a.val[0], 8);
-
-	    dest_pixels_a = vsriq_n_u16 (red_a, green_a, 5);
-	    dest_pixels_a = vsriq_n_u16 (dest_pixels_a, blue_a, 11);
-
-	    vst1q_u16 (dst_ptr, dest_pixels_a);
-	    dst_ptr += 8;
-	}
-
-#endif  /* USE_GCC_INLINE_ASM */
-
-    small_stuff:
-	if (count)
-	    __pld (src_ptr + src_stride);
-
-	while (count >= 2)
-	{
-	    uint32_t src_pixel_a = *src_ptr++;
-	    uint32_t src_pixel_b = *src_ptr++;
-
-	    /* ARM is really good at shift-then-ALU ops. */
-	    /* This should be a total of six shift-ANDs and five shift-ORs. */
-	    uint32_t dst_pixels_a;
-	    uint32_t dst_pixels_b;
-
-	    dst_pixels_a  = ((src_pixel_a >>  3) & rb_mask);
-	    dst_pixels_a |= ((src_pixel_a >> 10) &  g_mask) << 5;
-	    dst_pixels_a |= ((src_pixel_a >> 19) & rb_mask) << 11;
-
-	    dst_pixels_b  = ((src_pixel_b >>  3) & rb_mask);
-	    dst_pixels_b |= ((src_pixel_b >> 10) &  g_mask) << 5;
-	    dst_pixels_b |= ((src_pixel_b >> 19) & rb_mask) << 11;
-
-	    /* little-endian mode only */
-	    *((uint32_t*) dst_ptr) = dst_pixels_a | (dst_pixels_b << 16);
-	    dst_ptr += 2;
-	    count -= 2;
-	}
-
-	if (count)
-	{
-	    uint32_t src_pixel = *src_ptr++;
-
-	    /* ARM is really good at shift-then-ALU ops.
-	     * This block should end up as three shift-ANDs
-	     * and two shift-ORs.
-	     */
-	    uint32_t tmp_blue  = (src_pixel >>  3) & rb_mask;
-	    uint32_t tmp_green = (src_pixel >> 10) & g_mask;
-	    uint32_t tmp_red   = (src_pixel >> 19) & rb_mask;
-	    uint16_t dst_pixel = (tmp_red << 11) | (tmp_green << 5) | tmp_blue;
-
-	    *dst_ptr++ = dst_pixel;
-	    count--;
-	}
-
-	src_line += src_stride;
-	dst_line += dst_stride;
-    }
-}
+void
+pixman_composite_src_n_8888_asm_neon (int32_t   w,
+                                      int32_t   h,
+                                      uint32_t *dst,
+                                      int32_t   dst_stride,
+                                      uint32_t  src);
 
 static pixman_bool_t
 pixman_fill_neon (uint32_t *bits,
@@ -1705,936 +258,70 @@ pixman_fill_neon (uint32_t *bits,
                   int       height,
                   uint32_t  _xor)
 {
-    uint32_t byte_stride, color;
-    char *dst;
-
     /* stride is always multiple of 32bit units in pixman */
-    byte_stride = stride * sizeof(uint32_t);
+    uint32_t byte_stride = stride * sizeof(uint32_t);
 
     switch (bpp)
     {
     case 8:
-	dst = ((char *) bits) + y * byte_stride + x;
-	_xor &= 0xff;
-	color = _xor << 24 | _xor << 16 | _xor << 8 | _xor;
-	break;
-
+	pixman_composite_src_n_8_asm_neon (
+		width,
+		height,
+		(uint8_t *)(((char *) bits) + y * byte_stride + x),
+		byte_stride,
+		_xor & 0xff);
+	return TRUE;
     case 16:
-	dst = ((char *) bits) + y * byte_stride + x * 2;
-	_xor &= 0xffff;
-	color = _xor << 16 | _xor;
-	width *= 2;         /* width to bytes */
-	break;
-
+	pixman_composite_src_n_0565_asm_neon (
+		width,
+		height,
+		(uint16_t *)(((char *) bits) + y * byte_stride + x * 2),
+		byte_stride / 2,
+		_xor & 0xffff);
+	return TRUE;
     case 32:
-	dst = ((char *) bits) + y * byte_stride + x * 4;
-	color = _xor;
-	width *= 4;         /* width to bytes */
-	break;
-
+	pixman_composite_src_n_8888_asm_neon (
+		width,
+		height,
+		(uint32_t *)(((char *) bits) + y * byte_stride + x * 4),
+		byte_stride / 4,
+		_xor);
+	return TRUE;
     default:
 	return FALSE;
     }
-
-#ifdef USE_GCC_INLINE_ASM
-    if (width < 16)
-    {
-	/* We have a special case for such small widths that don't allow
-	 * us to use wide 128-bit stores anyway. We don't waste time
-	 * trying to align writes, since there are only very few of them anyway
-	 */
-	asm volatile (
-	    "cmp		%[height], #0\n"/* Check if empty fill */
-	    "beq		3f\n"
-	    "vdup.32	d0, %[color]\n"/* Fill the color to neon req */
-
-	    /* Check if we have a such width that can easily be handled by single
-	     * operation for each scanline. This significantly reduces the number
-	     * of test/branch instructions for each scanline
-	     */
-	    "cmp		%[width], #8\n"
-	    "beq		4f\n"
-	    "cmp		%[width], #4\n"
-	    "beq		5f\n"
-	    "cmp		%[width], #2\n"
-	    "beq		6f\n"
-
-	    /* Loop starts here for each scanline */
-	    "1:\n"
-	    "mov		r4, %[dst]\n" /* Starting address of the current line */
-	    "tst		%[width], #8\n"
-	    "beq		2f\n"
-	    "vst1.8		{d0}, [r4]!\n"
-	    "2:\n"
-	    "tst		%[width], #4\n"
-	    "beq		2f\n"
-	    "str		%[color], [r4], #4\n"
-	    "2:\n"
-	    "tst		%[width], #2\n"
-	    "beq		2f\n"
-	    "strh		%[color], [r4], #2\n"
-	    "2:\n"
-	    "tst		%[width], #1\n"
-	    "beq		2f\n"
-	    "strb		%[color], [r4], #1\n"
-	    "2:\n"
-
-	    "subs		%[height], %[height], #1\n"
-	    "add		%[dst], %[dst], %[byte_stride]\n"
-	    "bne		1b\n"
-	    "b		3f\n"
-
-	    /* Special fillers for those widths that we can do with single operation */
-	    "4:\n"
-	    "subs		%[height], %[height], #1\n"
-	    "vst1.8		{d0}, [%[dst]]\n"
-	    "add		%[dst], %[dst], %[byte_stride]\n"
-	    "bne		4b\n"
-	    "b		3f\n"
-
-	    "5:\n"
-	    "subs		%[height], %[height], #1\n"
-	    "str		%[color], [%[dst]]\n"
-	    "add		%[dst], %[dst], %[byte_stride]\n"
-	    "bne		5b\n"
-	    "b		3f\n"
-
-	    "6:\n"
-	    "subs		%[height], %[height], #1\n"
-	    "strh		%[color], [%[dst]]\n"
-	    "add		%[dst], %[dst], %[byte_stride]\n"
-	    "bne		6b\n"
-
-	    "3:\n"
-	    : [height] "+r" (height), [dst] "+r" (dst)
-	    : [color] "r" (color), [width] "r" (width),
-	      [byte_stride] "r" (byte_stride)
-	    : "memory", "cc", "d0", "r4");
-    }
-    else
-    {
-	asm volatile (
-	    "cmp		%[height], #0\n"/* Check if empty fill */
-	    "beq		5f\n"
-	    "vdup.32	q0, %[color]\n"/* Fill the color to neon req */
-
-	    /* Loop starts here for each scanline */
-	    "1:\n"
-	    "mov		r4, %[dst]\n"/* Starting address of the current line */
-	    "mov		r5, %[width]\n"/* We're going to write this many bytes */
-	    "ands		r6, r4, #15\n"/* Are we at the 128-bit aligned address? */
-	    "beq		2f\n"/* Jump to the best case */
-
-	    /* We're not 128-bit aligned: However, we know that we can get to the
-	       next aligned location, since the fill is at least 16 bytes wide */
-	    "rsb                r6, r6, #16\n" /* We would need to go forward this much */
-	    "sub		r5, r5, r6\n"/* Update bytes left */
-	    "tst		r6, #1\n"
-	    "beq		6f\n"
-	    "vst1.8		{d0[0]}, [r4]!\n"/* Store byte, now we are word aligned */
-	    "6:\n"
-	    "tst		r6, #2\n"
-	    "beq		6f\n"
-	    "vst1.16	{d0[0]}, [r4, :16]!\n"/* Store half word, now we are 16-bit aligned */
-	    "6:\n"
-	    "tst		r6, #4\n"
-	    "beq		6f\n"
-	    "vst1.32	{d0[0]}, [r4, :32]!\n"/* Store word, now we're 32-bit aligned */
-	    "6:\n"
-	    "tst		r6, #8\n"
-	    "beq		2f\n"
-	    "vst1.64	{d0}, [r4, :64]!\n"/* Store qword now we're 64-bit aligned */
-
-	    /* The good case: We're 128-bit aligned for this scanline */
-	    "2:\n"
-	    "and		r6, r5, #15\n"/* Number of tailing bytes */
-	    "cmp		r5, r6\n"/* Do we have at least one qword to write? */
-	    "beq		6f\n"/* No, we just write the tail */
-	    "lsr		r5, r5, #4\n"/* This many full qwords to write */
-
-	    /* The main block: Do 128-bit aligned writes */
-	    "3:\n"
-	    "subs		r5, r5, #1\n"
-	    "vst1.64	{d0, d1}, [r4, :128]!\n"
-	    "bne		3b\n"
-
-	    /* Handle the tailing bytes: Do 64, 32, 16 and 8-bit aligned writes as needed.
-	       We know that we're currently at 128-bit aligned address, so we can just
-	       pick the biggest operations that the remaining write width allows */
-	    "6:\n"
-	    "cmp		r6, #0\n"
-	    "beq		4f\n"
-	    "tst		r6, #8\n"
-	    "beq		6f\n"
-	    "vst1.64	{d0}, [r4, :64]!\n"
-	    "6:\n"
-	    "tst		r6, #4\n"
-	    "beq		6f\n"
-	    "vst1.32	{d0[0]}, [r4, :32]!\n"
-	    "6:\n"
-	    "tst		r6, #2\n"
-	    "beq		6f\n"
-	    "vst1.16	{d0[0]}, [r4, :16]!\n"
-	    "6:\n"
-	    "tst		r6, #1\n"
-	    "beq		4f\n"
-	    "vst1.8		{d0[0]}, [r4]!\n"
-	    "4:\n"
-
-	    /* Handle the next scanline */
-	    "subs		%[height], %[height], #1\n"
-	    "add		%[dst], %[dst], %[byte_stride]\n"
-	    "bne		1b\n"
-	    "5:\n"
-	    : [height] "+r" (height), [dst] "+r" (dst)
-	    : [color] "r" (color), [width] "r" (width),
-	      [byte_stride] "r" (byte_stride)
-	    : "memory", "cc", "d0", "d1", "r4", "r5", "r6");
-    }
-    return TRUE;
-
-#else
-
-    /* TODO: intrinsic version for armcc */
-    return FALSE;
-
-#endif
-}
-
-/* TODO: is there a more generic way of doing this being introduced? */
-#define NEON_SCANLINE_BUFFER_PIXELS (1024)
-
-static inline void
-neon_quadword_copy (void *   dst,
-		    void *   src,
-		    uint32_t count,         /* of quadwords */
-		    uint32_t trailer_count  /* of bytes */)
-{
-    uint8_t *t_dst = dst, *t_src = src;
-
-    /* Uses aligned multi-register loads to maximise read bandwidth
-     * on uncached memory such as framebuffers
-     * The accesses do not have the aligned qualifiers, so that the copy
-     * may convert between aligned-uncached and unaligned-cached memory.
-     * It is assumed that the CPU can infer alignedness from the address.
-     */
-
-#ifdef USE_GCC_INLINE_ASM
-
-    asm volatile (
-        "	cmp       %[count], #8				\n"
-        "	blt 1f    @ skip oversized fragments		\n"
-        "0: @ start with eight quadwords at a time		\n"
-        "	sub       %[count], %[count], #8		\n"
-        "	vld1.8    {d16, d17, d18, d19}, [%[src]]!		\n"
-        "	vld1.8    {d20, d21, d22, d23}, [%[src]]!		\n"
-        "	vld1.8    {d24, d25, d26, d27}, [%[src]]!		\n"
-        "	vld1.8    {d28, d29, d30, d31}, [%[src]]!		\n"
-        "	cmp       %[count], #8				\n"
-        "	vst1.8    {d16, d17, d18, d19}, [%[dst]]!		\n"
-        "	vst1.8    {d20, d21, d22, d23}, [%[dst]]!		\n"
-        "	vst1.8    {d24, d25, d26, d27}, [%[dst]]!		\n"
-        "	vst1.8    {d28, d29, d30, d31}, [%[dst]]!		\n"
-        "	bge 0b						\n"
-        "1: @ four quadwords					\n"
-        "	tst       %[count], #4				\n"
-        "	beq 2f    @ skip oversized fragment		\n"
-        "	vld1.8    {d16, d17, d18, d19}, [%[src]]!		\n"
-        "	vld1.8    {d20, d21, d22, d23}, [%[src]]!		\n"
-        "	vst1.8    {d16, d17, d18, d19}, [%[dst]]!		\n"
-        "	vst1.8    {d20, d21, d22, d23}, [%[dst]]!		\n"
-        "2: @ two quadwords					\n"
-        "	tst       %[count], #2				\n"
-        "	beq 3f    @ skip oversized fragment		\n"
-        "	vld1.8    {d16, d17, d18, d19}, [%[src]]!		\n"
-        "	vst1.8    {d16, d17, d18, d19}, [%[dst]]!		\n"
-        "3: @ one quadword					\n"
-        "	tst       %[count], #1				\n"
-        "	beq 4f    @ skip oversized fragment		\n"
-        "	vld1.8    {d16, d17}, [%[src]]!			\n"
-        "	vst1.8    {d16, d17}, [%[dst]]!			\n"
-        "4: @ end						\n"
-
-        /* Clobbered input registers marked as input/outputs */
-	: [dst] "+r" (t_dst), [src] "+r" (t_src), [count] "+r" (count)
-
-	  /* No unclobbered inputs */
-	:
-
-        /* Clobbered vector registers */
-	: "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", "d24", "d25",
-	  "d26", "d27", "d28", "d29", "d30", "d31", "cc", "memory");
-
-#else
-
-    while (count >= 8)
-    {
-	uint8x16x4_t t1 = vld4q_u8 (t_src);
-	uint8x16x4_t t2 = vld4q_u8 (t_src + sizeof(uint8x16x4_t));
-	
-	t_src += sizeof(uint8x16x4_t) * 2;
-	vst4q_u8 (t_dst, t1);
-	vst4q_u8 (t_dst + sizeof(uint8x16x4_t), t2);
-	t_dst += sizeof(uint8x16x4_t) * 2;
-	count -= 8;
-    }
-
-    if (count & 4)
-    {
-	uint8x16x4_t t1 = vld4q_u8 (t_src);
-	
-	t_src += sizeof(uint8x16x4_t);
-	vst4q_u8 (t_dst, t1);
-	t_dst += sizeof(uint8x16x4_t);
-    }
-
-    if (count & 2)
-    {
-	uint8x8x4_t t1 = vld4_u8 (t_src);
-	
-	t_src += sizeof(uint8x8x4_t);
-	vst4_u8 (t_dst, t1);
-	t_dst += sizeof(uint8x8x4_t);
-    }
-
-    if (count & 1)
-    {
-	uint8x16_t t1 = vld1q_u8 (t_src);
-	
-	t_src += sizeof(uint8x16_t);
-	vst1q_u8 (t_dst, t1);
-	t_dst += sizeof(uint8x16_t);
-    }
-
-#endif  /* !USE_GCC_INLINE_ASM */
-
-    if (trailer_count)
-    {
-	if (trailer_count & 8)
-	{
-	    uint8x8_t t1 = vld1_u8 (t_src);
-	    
-	    t_src += sizeof(uint8x8_t);
-	    vst1_u8 (t_dst, t1);
-	    t_dst += sizeof(uint8x8_t);
-	}
-
-	if (trailer_count & 4)
-	{
-	    *((uint32_t*) t_dst) = *((uint32_t*) t_src);
-	    
-	    t_dst += 4;
-	    t_src += 4;
-	}
-
-	if (trailer_count & 2)
-	{
-	    *((uint16_t*) t_dst) = *((uint16_t*) t_src);
-	    
-	    t_dst += 2;
-	    t_src += 2;
-	}
-
-	if (trailer_count & 1)
-	{
-	    *t_dst++ = *t_src++;
-	}
-    }
-}
-
-static inline void
-solid_over_565_8_pix_neon (uint32_t  glyph_colour,
-                           uint16_t *dest,
-                           uint8_t * in_mask,
-                           uint32_t  dest_stride,    /* bytes, not elements */
-                           uint32_t  mask_stride,
-                           uint32_t  count           /* 8-pixel groups */)
-{
-    /* Inner loop of glyph blitter (solid colour, alpha mask) */
-
-#ifdef USE_GCC_INLINE_ASM
-
-    asm volatile (
-        "	vld4.8 {d20[], d21[], d22[], d23[]}, [%[glyph_colour]]  @ splat solid colour components	\n"
-        "0:	@ loop																				\n"
-        "	vld1.16   {d0, d1}, [%[dest]]         @ load first pixels from framebuffer			\n"
-        "	vld1.8    {d17}, [%[in_mask]]         @ load alpha mask of glyph						\n"
-        "	vmull.u8  q9, d17, d23               @ apply glyph colour alpha to mask				\n"
-        "	vshrn.u16 d17, q9, #8                @ reformat it to match original mask			\n"
-        "	vmvn      d18, d17                   @ we need the inverse mask for the background	\n"
-        "	vsli.u16  q3, q0, #5                 @ duplicate framebuffer blue bits				\n"
-        "	vshrn.u16 d2, q0, #8                 @ unpack red from framebuffer pixels			\n"
-        "	vshrn.u16 d4, q0, #3                 @ unpack green									\n"
-        "	vsri.u8   d2, d2, #5                 @ duplicate red bits (extend 5 to 8)			\n"
-        "	vshrn.u16 d6, q3, #2                 @ unpack extended blue (truncate 10 to 8)		\n"
-        "	vsri.u8   d4, d4, #6                 @ duplicate green bits (extend 6 to 8)			\n"
-        "	vmull.u8  q1, d2, d18                @ apply inverse mask to background red...		\n"
-        "	vmull.u8  q2, d4, d18                @ ...green...									\n"
-        "	vmull.u8  q3, d6, d18                @ ...blue										\n"
-        "	subs      %[count], %[count], #1     @ decrement/test loop counter					\n"
-        "	vmlal.u8  q1, d17, d22               @ add masked foreground red...					\n"
-        "	vmlal.u8  q2, d17, d21               @ ...green...									\n"
-        "	vmlal.u8  q3, d17, d20               @ ...blue										\n"
-        "	add %[in_mask], %[in_mask], %[mask_stride] @ advance mask pointer, while we wait		\n"
-        "	vsri.16   q1, q2, #5                 @ pack green behind red						\n"
-        "	vsri.16   q1, q3, #11                @ pack blue into pixels						\n"
-        "	vst1.16   {d2, d3}, [%[dest]]         @ store composited pixels						\n"
-        "	add %[dest], %[dest], %[dest_stride]  @ advance framebuffer pointer					\n"
-        "	bne 0b                               @ next please									\n"
-
-	/* Clobbered registers marked as input/outputs */
-	: [dest] "+r" (dest), [in_mask] "+r" (in_mask), [count] "+r" (count)
-	  
-	  /* Inputs */
-	: [dest_stride] "r" (dest_stride), [mask_stride] "r" (mask_stride), [glyph_colour] "r" (&glyph_colour)
-
-	  /* Clobbers, including the inputs we modify, and potentially lots of memory */
-	: "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d17", "d18", "d19",
-	  "d20", "d21", "d22", "d23", "d24", "d25", "cc", "memory"
-        );
-
-#else
-
-    uint8x8x4_t solid_colour = vld4_dup_u8 ((uint8_t*) &glyph_colour);
-
-    while (count--)
-    {
-	uint16x8_t pixels = vld1q_u16 (dest);
-	uint8x8_t mask = vshrn_n_u16 (vmull_u8 (solid_colour.val[3], vld1_u8 (in_mask)), 8);
-	uint8x8_t mask_image = vmvn_u8 (mask);
-
-	uint8x8_t t_red   = vshrn_n_u16 (pixels, 8);
-	uint8x8_t t_green = vshrn_n_u16 (pixels, 3);
-	uint8x8_t t_blue  = vshrn_n_u16 (vsli_n_u8 (pixels, pixels, 5), 2);
-
-	uint16x8_t s_red   = vmull_u8 (vsri_n_u8 (t_red, t_red, 5), mask_image);
-	uint16x8_t s_green = vmull_u8 (vsri_n_u8 (t_green, t_green, 6), mask_image);
-	uint16x8_t s_blue  = vmull_u8 (t_blue, mask_image);
-
-	s_red   = vmlal (s_red, mask, solid_colour.val[2]);
-	s_green = vmlal (s_green, mask, solid_colour.val[1]);
-	s_blue  = vmlal (s_blue, mask, solid_colour.val[0]);
-
-	pixels = vsri_n_u16 (s_red, s_green, 5);
-	pixels = vsri_n_u16 (pixels, s_blue, 11);
-	vst1q_u16 (dest, pixels);
-
-	dest += dest_stride;
-	mask += mask_stride;
-    }
-
-#endif
 }
 
-#if 0 /* this is broken currently */
-static void
-neon_composite_over_n_8_0565 (pixman_implementation_t * impl,
-                              pixman_op_t               op,
-                              pixman_image_t *          src_image,
-                              pixman_image_t *          mask_image,
-                              pixman_image_t *          dst_image,
-                              int32_t                   src_x,
-                              int32_t                   src_y,
-                              int32_t                   mask_x,
-                              int32_t                   mask_y,
-                              int32_t                   dest_x,
-                              int32_t                   dest_y,
-                              int32_t                   width,
-                              int32_t                   height)
-{
-    uint32_t  src, srca;
-    uint16_t *dst_line, *aligned_line;
-    uint8_t  *mask_line;
-    uint32_t  dst_stride, mask_stride;
-    uint32_t  kernel_count, copy_count, copy_tail;
-    uint8_t   kernel_offset, copy_offset;
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    /* bail out if fully transparent or degenerate */
-    srca = src >> 24;
-    if (src == 0)
-	return;
-
-    if (width == 0 || height == 0)
-	return;
-
-    if (width > NEON_SCANLINE_BUFFER_PIXELS)
-    {
-	/* split the blit, so we can use a fixed-size scanline buffer
-	 * TODO: there must be a more elegant way of doing this.
-	 */
-	int x;
-	for (x = 0; x < width; x += NEON_SCANLINE_BUFFER_PIXELS)
-	{
-	    neon_composite_over_n_8_0565 (
-		impl, op,
-		src_image, mask_image, dst_image,
-		src_x + x, src_y, mask_x + x, mask_y, dest_x + x, dest_y,
-		(x + NEON_SCANLINE_BUFFER_PIXELS > width) ? width - x : NEON_SCANLINE_BUFFER_PIXELS, height);
-	}
-
-	return;
-    }
-    
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-
-    /* keep within minimum number of aligned quadwords on width
-     * while also keeping the minimum number of columns to process
-     */
-    {
-	unsigned long aligned_left = (unsigned long)(dst_line) & ~0xF;
-	unsigned long aligned_right = (((unsigned long)(dst_line + width)) + 0xF) & ~0xF;
-	unsigned long ceiling_length = (((unsigned long) width) * sizeof(*dst_line) + 0xF) & ~0xF;
-
-	/* the fast copy should be quadword aligned */
-	copy_offset = dst_line - ((uint16_t*) aligned_left);
-	aligned_line = dst_line - copy_offset;
-	copy_count = (uint32_t) ((aligned_right - aligned_left) >> 4);
-	copy_tail = 0;
-
-	if (aligned_right - aligned_left > ceiling_length)
-	{
-	    /* unaligned routine is tightest */
-	    kernel_count = (uint32_t) (ceiling_length >> 4);
-	    kernel_offset = copy_offset;
-	}
-	else
-	{
-	    /* aligned routine is equally tight, so it is safer to align */
-	    kernel_count = copy_count;
-	    kernel_offset = 0;
-	}
-
-	/* We should avoid reading beyond scanline ends for safety */
-	if (aligned_line < (dst_line - dest_x) ||
-	    (aligned_line + (copy_count * 16 / sizeof(*dst_line))) > ((dst_line - dest_x) + dst_image->bits.width))
-	{
-	    /* switch to precise read */
-	    copy_offset = kernel_offset = 0;
-	    aligned_line = dst_line;
-	    kernel_count = (uint32_t) (ceiling_length >> 4);
-	    copy_count = (width * sizeof(*dst_line)) >> 4;
-	    copy_tail = (width * sizeof(*dst_line)) & 0xF;
-	}
-    }
-
-    {
-	uint16_t scan_line[NEON_SCANLINE_BUFFER_PIXELS + 8];         /* deliberately not initialised */
-	uint8_t glyph_line[NEON_SCANLINE_BUFFER_PIXELS + 8];
-	int y = height;
-
-	/* row-major order */
-	/* left edge, middle block, right edge */
-	for ( ; y--; mask_line += mask_stride, aligned_line += dst_stride, dst_line += dst_stride)
-	{
-	    /* We don't want to overrun the edges of the glyph,
-	     * so realign the edge data into known buffers
-	     */
-	    neon_quadword_copy (glyph_line + copy_offset, mask_line, width >> 4, width & 0xF);
-
-	    /* Uncached framebuffer access is really, really slow
-	     * if we do it piecemeal. It should be much faster if we
-	     * grab it all at once. One scanline should easily fit in
-	     * L1 cache, so this should not waste RAM bandwidth.
-	     */
-	    neon_quadword_copy (scan_line, aligned_line, copy_count, copy_tail);
-
-	    /* Apply the actual filter */
-	    solid_over_565_8_pix_neon (
-		src, scan_line + kernel_offset,
-		glyph_line + kernel_offset, 8 * sizeof(*dst_line),
-		8, kernel_count);
-
-	    /* Copy the modified scanline back */
-	    neon_quadword_copy (dst_line, scan_line + copy_offset,
-				width >> 3, (width & 7) * 2);
-	}
-    }
-}
-#endif
-
-#ifdef USE_GCC_INLINE_ASM
-
-static inline void
-plain_over_565_8_pix_neon (uint32_t  colour,
-			   uint16_t *dest,
-			   uint32_t  dest_stride,     /* bytes, not elements */
-			   uint32_t  count            /* 8-pixel groups */)
-{
-    /* Inner loop for plain translucent rects
-     * (solid colour without alpha mask)
-     */
-    asm volatile (
-        "	vld4.8   {d20[], d21[], d22[], d23[]}, [%[colour]]  @ solid colour load/splat \n"
-        "	vmull.u8  q12, d23, d22              @ premultiply alpha red   \n"
-        "	vmull.u8  q13, d23, d21              @ premultiply alpha green \n"
-        "	vmull.u8  q14, d23, d20              @ premultiply alpha blue  \n"
-        "	vmvn      d18, d23                   @ inverse alpha for background \n"
-        "0:	@ loop\n"
-        "	vld1.16   {d0, d1}, [%[dest]]         @ load first pixels from framebuffer	\n"
-        "	vshrn.u16 d2, q0, #8                 @ unpack red from framebuffer pixels	\n"
-        "	vshrn.u16 d4, q0, #3                 @ unpack green				\n"
-        "	vsli.u16  q3, q0, #5                 @ duplicate framebuffer blue bits		\n"
-        "	vsri.u8   d2, d2, #5                 @ duplicate red bits (extend 5 to 8)	\n"
-        "	vsri.u8   d4, d4, #6                 @ duplicate green bits (extend 6 to 8)	\n"
-        "	vshrn.u16 d6, q3, #2                 @ unpack extended blue (truncate 10 to 8)	\n"
-        "	vmov      q0, q12                    @ retrieve foreground red   \n"
-        "	vmlal.u8  q0, d2, d18                @ blend red - my kingdom for a four-operand MLA \n"
-        "	vmov      q1, q13                    @ retrieve foreground green \n"
-        "	vmlal.u8  q1, d4, d18                @ blend green               \n"
-        "	vmov      q2, q14                    @ retrieve foreground blue  \n"
-        "	vmlal.u8  q2, d6, d18                @ blend blue                \n"
-        "	subs      %[count], %[count], #1     @ decrement/test loop counter		\n"
-        "	vsri.16   q0, q1, #5                 @ pack green behind red			\n"
-        "	vsri.16   q0, q2, #11                @ pack blue into pixels			\n"
-        "	vst1.16   {d0, d1}, [%[dest]]         @ store composited pixels			\n"
-        "	add %[dest], %[dest], %[dest_stride]  @ advance framebuffer pointer		\n"
-        "	bne 0b                               @ next please				\n"
-
-        /* Clobbered registers marked as input/outputs */
-	: [dest] "+r" (dest), [count] "+r" (count)
-
-	  /* Inputs */
-	: [dest_stride] "r" (dest_stride), [colour] "r" (&colour)
-
-	  /* Clobbers, including the inputs we modify, and
-	   * potentially lots of memory
-	   */
-	: "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d18", "d19",
-	  "d20", "d21", "d22", "d23", "d24", "d25", "d26", "d27", "d28", "d29",
-	  "cc", "memory"
-        );
-}
-
-static void
-neon_composite_over_n_0565 (pixman_implementation_t * impl,
-                            pixman_op_t               op,
-                            pixman_image_t *          src_image,
-                            pixman_image_t *          mask_image,
-                            pixman_image_t *          dst_image,
-                            int32_t                   src_x,
-                            int32_t                   src_y,
-                            int32_t                   mask_x,
-                            int32_t                   mask_y,
-                            int32_t                   dest_x,
-                            int32_t                   dest_y,
-                            int32_t                   width,
-                            int32_t                   height)
-{
-    uint32_t src, srca;
-    uint16_t    *dst_line, *aligned_line;
-    uint32_t dst_stride;
-    uint32_t kernel_count, copy_count, copy_tail;
-    uint8_t kernel_offset, copy_offset;
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    /* bail out if fully transparent */
-    srca = src >> 24;
-    if (src == 0)
-	return;
-    
-    if (width == 0 || height == 0)
-	return;
-
-    if (width > NEON_SCANLINE_BUFFER_PIXELS)
-    {
-	/* split the blit, so we can use a fixed-size scanline buffer *
-	 * TODO: there must be a more elegant way of doing this.
-	 */
-	int x;
-	
-	for (x = 0; x < width; x += NEON_SCANLINE_BUFFER_PIXELS)
-	{
-	    neon_composite_over_n_0565 (
-		impl, op,
-		src_image, mask_image, dst_image,
-		src_x + x, src_y, mask_x + x, mask_y, dest_x + x, dest_y,
-		(x + NEON_SCANLINE_BUFFER_PIXELS > width) ? width - x : NEON_SCANLINE_BUFFER_PIXELS, height);
-	}
-	return;
-    }
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-
-    /* keep within minimum number of aligned quadwords on width
-     * while also keeping the minimum number of columns to process
-     */
-    {
-	unsigned long aligned_left = (unsigned long)(dst_line) & ~0xF;
-	unsigned long aligned_right = (((unsigned long)(dst_line + width)) + 0xF) & ~0xF;
-	unsigned long ceiling_length = (((unsigned long) width) * sizeof(*dst_line) + 0xF) & ~0xF;
-
-	/* the fast copy should be quadword aligned */
-	copy_offset = dst_line - ((uint16_t*) aligned_left);
-	aligned_line = dst_line - copy_offset;
-	copy_count = (uint32_t) ((aligned_right - aligned_left) >> 4);
-	copy_tail = 0;
-
-	if (aligned_right - aligned_left > ceiling_length)
-	{
-	    /* unaligned routine is tightest */
-	    kernel_count = (uint32_t) (ceiling_length >> 4);
-	    kernel_offset = copy_offset;
-	}
-	else
-	{
-	    /* aligned routine is equally tight, so it is safer to align */
-	    kernel_count = copy_count;
-	    kernel_offset = 0;
-	}
-
-	/* We should avoid reading beyond scanline ends for safety */
-	if (aligned_line < (dst_line - dest_x) ||
-	    (aligned_line + (copy_count * 16 / sizeof(*dst_line))) > ((dst_line - dest_x) + dst_image->bits.width))
-	{
-	    /* switch to precise read */
-	    copy_offset = kernel_offset = 0;
-	    aligned_line = dst_line;
-	    kernel_count = (uint32_t) (ceiling_length >> 4);
-	    copy_count = (width * sizeof(*dst_line)) >> 4;
-	    copy_tail = (width * sizeof(*dst_line)) & 0xF;
-	}
-    }
-
-    {
-	uint16_t scan_line[NEON_SCANLINE_BUFFER_PIXELS + 8];  /* deliberately not initialised */
-
-	/* row-major order */
-	/* left edge, middle block, right edge */
-	for ( ; height--; aligned_line += dst_stride, dst_line += dst_stride)
-	{
-	    /* Uncached framebuffer access is really, really slow if we do it piecemeal.
-	     * It should be much faster if we grab it all at once.
-	     * One scanline should easily fit in L1 cache, so this should
-	     * not waste RAM bandwidth.
-	     */
-	    neon_quadword_copy (scan_line, aligned_line, copy_count, copy_tail);
-
-	    /* Apply the actual filter */
-	    plain_over_565_8_pix_neon (
-		src, scan_line + kernel_offset, 8 * sizeof(*dst_line), kernel_count);
-
-	    /* Copy the modified scanline back */
-	    neon_quadword_copy (
-		dst_line, scan_line + copy_offset, width >> 3, (width & 7) * 2);
-	}
-    }
-}
-
-static inline void
-ARGB8_over_565_8_pix_neon (uint32_t *src,
-                           uint16_t *dest,
-                           uint32_t  src_stride,     /* bytes, not elements */
-                           uint32_t  count           /* 8-pixel groups */)
-{
-    asm volatile (
-        "0:	@ loop\n"
-        "	pld   [%[src], %[src_stride]]         @ preload from next scanline	\n"
-        "	vld1.16   {d0, d1}, [%[dest]]         @ load pixels from framebuffer	\n"
-        "	vld4.8   {d20, d21, d22, d23},[%[src]]! @ load source image pixels		\n"
-        "	vsli.u16  q3, q0, #5                 @ duplicate framebuffer blue bits		\n"
-        "	vshrn.u16 d2, q0, #8                 @ unpack red from framebuffer pixels	\n"
-        "	vshrn.u16 d4, q0, #3                 @ unpack green				\n"
-        "	vmvn      d18, d23                   @ we need the inverse alpha for the background	\n"
-        "	vsri.u8   d2, d2, #5                 @ duplicate red bits (extend 5 to 8)	\n"
-        "	vshrn.u16 d6, q3, #2                 @ unpack extended blue (truncate 10 to 8)	\n"
-        "	vsri.u8   d4, d4, #6                 @ duplicate green bits (extend 6 to 8)	\n"
-        "	vmull.u8  q1, d2, d18                @ apply inverse alpha to background red...	\n"
-        "	vmull.u8  q2, d4, d18                @ ...green...				\n"
-        "	vmull.u8  q3, d6, d18                @ ...blue					\n"
-        "	subs      %[count], %[count], #1     @ decrement/test loop counter		\n"
-        "	vmlal.u8  q1, d23, d22               @ add blended foreground red...		\n"
-        "	vmlal.u8  q2, d23, d21               @ ...green...				\n"
-        "	vmlal.u8  q3, d23, d20               @ ...blue					\n"
-        "	vsri.16   q1, q2, #5                 @ pack green behind red			\n"
-        "	vsri.16   q1, q3, #11                @ pack blue into pixels			\n"
-        "	vst1.16   {d2, d3}, [%[dest]]!        @ store composited pixels			\n"
-        "	bne 0b                               @ next please				\n"
-
-        /* Clobbered registers marked as input/outputs */
-	: [dest] "+r" (dest), [src] "+r" (src), [count] "+r" (count)
-
-	  /* Inputs */
-	: [src_stride] "r" (src_stride)
-
-	  /* Clobbers, including the inputs we modify, and potentially lots of memory */
-	: "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", "d17", "d18", "d20",
-	  "d21", "d22", "d23", "cc", "memory"
-        );
-}
-
-static void
-neon_composite_over_8888_0565 (pixman_implementation_t * impl,
-                               pixman_op_t               op,
-                               pixman_image_t *          src_image,
-                               pixman_image_t *          mask_image,
-                               pixman_image_t *          dst_image,
-                               int32_t                   src_x,
-                               int32_t                   src_y,
-                               int32_t                   mask_x,
-                               int32_t                   mask_y,
-                               int32_t                   dest_x,
-                               int32_t                   dest_y,
-                               int32_t                   width,
-                               int32_t                   height)
-{
-    uint32_t    *src_line;
-    uint16_t    *dst_line, *aligned_line;
-    uint32_t dst_stride, src_stride;
-    uint32_t kernel_count, copy_count, copy_tail;
-    uint8_t kernel_offset, copy_offset;
-
-    /* we assume mask is opaque 
-     * so the only alpha to deal with is embedded in src
-     */
-    if (width > NEON_SCANLINE_BUFFER_PIXELS)
-    {
-	/* split the blit, so we can use a fixed-size scanline buffer */
-	int x;
-	for (x = 0; x < width; x += NEON_SCANLINE_BUFFER_PIXELS)
-	{
-	    neon_composite_over_8888_0565 (
-		impl, op,
-		src_image, mask_image, dst_image,
-		src_x + x, src_y, mask_x + x, mask_y, dest_x + x, dest_y,
-		(x + NEON_SCANLINE_BUFFER_PIXELS > width) ? width - x : NEON_SCANLINE_BUFFER_PIXELS, height);
-	}
-	return;
-    }
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-    /* keep within minimum number of aligned quadwords on width
-     * while also keeping the minimum number of columns to process
-     */
-    {
-	unsigned long aligned_left = (unsigned long)(dst_line) & ~0xF;
-	unsigned long aligned_right = (((unsigned long)(dst_line + width)) + 0xF) & ~0xF;
-	unsigned long ceiling_length = (((unsigned long) width) * sizeof(*dst_line) + 0xF) & ~0xF;
-
-	/* the fast copy should be quadword aligned */
-	copy_offset = dst_line - ((uint16_t*) aligned_left);
-	aligned_line = dst_line - copy_offset;
-	copy_count = (uint32_t) ((aligned_right - aligned_left) >> 4);
-	copy_tail = 0;
-
-	if (aligned_right - aligned_left > ceiling_length)
-	{
-	    /* unaligned routine is tightest */
-	    kernel_count = (uint32_t) (ceiling_length >> 4);
-	    kernel_offset = copy_offset;
-	}
-	else
-	{
-	    /* aligned routine is equally tight, so it is safer to align */
-	    kernel_count = copy_count;
-	    kernel_offset = 0;
-	}
-
-	/* We should avoid reading beyond scanline ends for safety */
-	if (aligned_line < (dst_line - dest_x) ||
-	    (aligned_line + (copy_count * 16 / sizeof(*dst_line))) > ((dst_line - dest_x) + dst_image->bits.width))
-	{
-	    /* switch to precise read */
-	    copy_offset = kernel_offset = 0;
-	    aligned_line = dst_line;
-	    kernel_count = (uint32_t) (ceiling_length >> 4);
-	    copy_count = (width * sizeof(*dst_line)) >> 4;
-	    copy_tail = (width * sizeof(*dst_line)) & 0xF;
-	}
-    }
-
-    /* Preload the first input scanline */
-    {
-	uint8_t *src_ptr = (uint8_t*) src_line;
-	uint32_t count = (width + 15) / 16;
-
-#ifdef USE_GCC_INLINE_ASM
-	asm volatile (
-	    "0: @ loop						\n"
-	    "	subs    %[count], %[count], #1			\n"
-	    "	pld     [%[src]]				\n"
-	    "	add     %[src], %[src], #64			\n"
-	    "	bgt 0b						\n"
-
-	    /* Clobbered input registers marked as input/outputs */
-	    : [src] "+r" (src_ptr), [count] "+r" (count)
-	    :     /* no unclobbered inputs */
-	    : "cc"
-	    );
-#else
-	do
-	{
-	    __pld (src_ptr);
-	    src_ptr += 64;
-	}
-	while (--count);
-#endif
-    }
-
-    {
-	uint16_t scan_line[NEON_SCANLINE_BUFFER_PIXELS + 8]; /* deliberately not initialised */
-
-	/* row-major order */
-	/* left edge, middle block, right edge */
-	for ( ; height--; src_line += src_stride, aligned_line += dst_stride)
-	{
-	    /* Uncached framebuffer access is really, really slow if we do
-	     * it piecemeal. It should be much faster if we grab it all at
-	     * once. One scanline should easily fit in L1 cache, so this
-	     * should not waste RAM bandwidth.
-	     */
-	    neon_quadword_copy (scan_line, aligned_line, copy_count, copy_tail);
-
-	    /* Apply the actual filter */
-	    ARGB8_over_565_8_pix_neon (
-		src_line, scan_line + kernel_offset,
-		src_stride * sizeof(*src_line), kernel_count);
-
-	    /* Copy the modified scanline back */
-	    neon_quadword_copy (dst_line,
-				scan_line + copy_offset,
-				width >> 3, (width & 7) * 2);
-	}
-    }
-}
-
-#endif  /* USE_GCC_INLINE_ASM */
-
 static const pixman_fast_path_t arm_neon_fast_path_array[] =
 {
-    { PIXMAN_OP_ADD,  PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8,       neon_composite_add_8888_8_8,     0 },
-    { PIXMAN_OP_ADD,  PIXMAN_a8,       PIXMAN_null,     PIXMAN_a8,       neon_composite_add_8000_8000,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r5g6b5,   neon_composite_over_n_8_0565,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b5g6r5,   neon_composite_over_n_8_0565,    0 },
-    { PIXMAN_OP_SRC,  PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_src_24_16,        0 },
-    { PIXMAN_OP_SRC,  PIXMAN_x8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_src_24_16,        0 },
-    { PIXMAN_OP_SRC,  PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_src_24_16,        0 },
-    { PIXMAN_OP_SRC,  PIXMAN_x8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_src_24_16,        0 },
-#ifdef USE_GCC_INLINE_ASM
-    { PIXMAN_OP_SRC,  PIXMAN_r5g6b5,   PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_src_16_16,        0 },
-    { PIXMAN_OP_SRC,  PIXMAN_b5g6r5,   PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_src_16_16,        0 },
-#if 0 /* this code has some bugs */
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_over_n_0565,      0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_over_n_0565,      0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_over_8888_0565,   0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_over_8888_0565,   0 },
-#endif
-#endif
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, neon_composite_over_8888_8888,   0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, neon_composite_over_8888_8888,   0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, neon_composite_over_8888_8888,   0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, neon_composite_over_8888_8888,   0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, neon_composite_over_8888_n_8888, NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, neon_composite_over_8888_n_8888, NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, neon_composite_over_n_8_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, neon_composite_over_n_8_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, neon_composite_over_n_8_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, neon_composite_over_n_8_8888,    0 },
+    { PIXMAN_OP_SRC,  PIXMAN_r5g6b5,   PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_src_0565_0565    },
+    { PIXMAN_OP_SRC,  PIXMAN_b5g6r5,   PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_src_0565_0565    },
+    { PIXMAN_OP_SRC,  PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_src_8888_0565    },
+    { PIXMAN_OP_SRC,  PIXMAN_x8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_src_8888_0565    },
+    { PIXMAN_OP_SRC,  PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_src_8888_0565    },
+    { PIXMAN_OP_SRC,  PIXMAN_x8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_src_8888_0565    },
+    { PIXMAN_OP_SRC,  PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, neon_composite_src_8888_8888    },
+    { PIXMAN_OP_SRC,  PIXMAN_x8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, neon_composite_src_8888_8888    },
+    { PIXMAN_OP_SRC,  PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, neon_composite_src_8888_8888    },
+    { PIXMAN_OP_SRC,  PIXMAN_x8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, neon_composite_src_8888_8888    },
+    { PIXMAN_OP_SRC,  PIXMAN_r8g8b8,   PIXMAN_null,     PIXMAN_r8g8b8,   neon_composite_src_0888_0888    },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r5g6b5,   neon_composite_over_n_8_0565    },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b5g6r5,   neon_composite_over_n_8_0565    },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, neon_composite_over_n_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, neon_composite_over_n_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, neon_composite_over_n_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, neon_composite_over_n_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_solid,    PIXMAN_a8r8g8b8, neon_composite_over_8888_n_8888 },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_solid,    PIXMAN_x8r8g8b8, neon_composite_over_8888_n_8888 },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   neon_composite_over_8888_0565   },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   neon_composite_over_8888_0565   },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, neon_composite_over_8888_8888   },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, neon_composite_over_8888_8888   },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, neon_composite_over_8888_8888   },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, neon_composite_over_8888_8888   },
+    { PIXMAN_OP_ADD,  PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8,       neon_composite_add_n_8_8        },
+    { PIXMAN_OP_ADD,  PIXMAN_a8,       PIXMAN_a8,       PIXMAN_a8,       neon_composite_add_8_8_8        },
+    { PIXMAN_OP_ADD,  PIXMAN_a8,       PIXMAN_null,     PIXMAN_a8,       neon_composite_add_8000_8000    },
     { PIXMAN_OP_NONE },
 };
 
@@ -2674,79 +361,6 @@ arm_neon_composite (pixman_implementation_t *imp,
 }
 
 static pixman_bool_t
-pixman_blt_neon (void *src_bits,
-                 void *dst_bits,
-                 int   src_stride,
-                 int   dst_stride,
-                 int   src_bpp,
-                 int   dst_bpp,
-                 int   src_x,
-                 int   src_y,
-                 int   dst_x,
-                 int   dst_y,
-                 int   width,
-                 int   height)
-{
-    if (!width || !height)
-	return TRUE;
-
-    /* accelerate only straight copies involving complete bytes */
-    if (src_bpp != dst_bpp || (src_bpp & 7))
-	return FALSE;
-
-    {
-	uint32_t bytes_per_pixel = src_bpp >> 3;
-	uint32_t byte_width = width * bytes_per_pixel;
-	/* parameter is in words for some reason */
-	int32_t src_stride_bytes = src_stride * 4;
-	int32_t dst_stride_bytes = dst_stride * 4;
-	uint8_t *src_bytes = ((uint8_t*) src_bits) +
-	    src_y * src_stride_bytes + src_x * bytes_per_pixel;
-	uint8_t *dst_bytes = ((uint8_t*) dst_bits) +
-	    dst_y * dst_stride_bytes + dst_x * bytes_per_pixel;
-	uint32_t quadword_count = byte_width / 16;
-	uint32_t offset         = byte_width % 16;
-
-	while (height--)
-	{
-	    neon_quadword_copy (dst_bytes, src_bytes, quadword_count, offset);
-	    src_bytes += src_stride_bytes;
-	    dst_bytes += dst_stride_bytes;
-	}
-    }
-
-    return TRUE;
-}
-
-static pixman_bool_t
-arm_neon_blt (pixman_implementation_t *imp,
-              uint32_t *               src_bits,
-              uint32_t *               dst_bits,
-              int                      src_stride,
-              int                      dst_stride,
-              int                      src_bpp,
-              int                      dst_bpp,
-              int                      src_x,
-              int                      src_y,
-              int                      dst_x,
-              int                      dst_y,
-              int                      width,
-              int                      height)
-{
-    if (pixman_blt_neon (
-            src_bits, dst_bits, src_stride, dst_stride, src_bpp, dst_bpp,
-            src_x, src_y, dst_x, dst_y, width, height))
-    {
-	return TRUE;
-    }
-
-    return _pixman_implementation_blt (
-               imp->delegate,
-               src_bits, dst_bits, src_stride, dst_stride, src_bpp, dst_bpp,
-               src_x, src_y, dst_x, dst_y, width, height);
-}
-
-static pixman_bool_t
 arm_neon_fill (pixman_implementation_t *imp,
                uint32_t *               bits,
                int                      stride,
@@ -2771,11 +385,7 @@ _pixman_implementation_create_arm_neon (void)
     pixman_implementation_t *imp = _pixman_implementation_create (general);
 
     imp->composite = arm_neon_composite;
-#if 0 /* this code has some bugs */
-    imp->blt = arm_neon_blt;
-#endif
     imp->fill = arm_neon_fill;
 
     return imp;
 }
-
diff --git a/pixman/pixman/pixman-arm-simd.c b/pixman/pixman/pixman-arm-simd.c
index fb7bf3da8..265c820b4 100644
--- a/pixman/pixman/pixman-arm-simd.c
+++ b/pixman/pixman/pixman-arm-simd.c
@@ -221,7 +221,7 @@ arm_composite_over_8888_n_8888 (
     PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
     PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
 
-    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);
+    mask = _pixman_image_get_solid (mask_image, PIXMAN_a8r8g8b8);
     mask = (mask) >> 24;
 
     while (height--)
@@ -421,19 +421,19 @@ arm_composite_over_n_8_8888 (pixman_implementation_t * impl,
 
 static const pixman_fast_path_t arm_simd_fast_path_array[] =
 {
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, arm_composite_over_8888_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, arm_composite_over_8888_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, arm_composite_over_8888_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, arm_composite_over_8888_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, arm_composite_over_8888_n_8888,  NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, arm_composite_over_8888_n_8888,  NEED_SOLID_MASK },
-
-    { PIXMAN_OP_ADD, PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       arm_composite_add_8000_8000,     0 },
-
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, arm_composite_over_n_8_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, arm_composite_over_n_8_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, arm_composite_over_n_8_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, arm_composite_over_n_8_8888,     0 },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, arm_composite_over_8888_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, arm_composite_over_8888_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, arm_composite_over_8888_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, arm_composite_over_8888_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_solid,    PIXMAN_a8r8g8b8, arm_composite_over_8888_n_8888  },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_solid,    PIXMAN_x8r8g8b8, arm_composite_over_8888_n_8888  },
+
+    { PIXMAN_OP_ADD, PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       arm_composite_add_8000_8000     },
+
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, arm_composite_over_n_8_8888     },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, arm_composite_over_n_8_8888     },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, arm_composite_over_n_8_8888     },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, arm_composite_over_n_8_8888     },
 
     { PIXMAN_OP_NONE },
 };
diff --git a/pixman/pixman/pixman-bits-image.c b/pixman/pixman/pixman-bits-image.c
index 4e78ce107..5a5a69057 100644
--- a/pixman/pixman/pixman-bits-image.c
+++ b/pixman/pixman/pixman-bits-image.c
@@ -4,6 +4,7 @@
  *             2008 Aaron Plattner, NVIDIA Corporation
  * Copyright © 2000 SuSE, Inc.
  * Copyright © 2007, 2009 Red Hat, Inc.
+ * Copyright © 2008 André Tupinambá <andrelrt@gmail.com>
  *
  * Permission to use, copy, modify, distribute, and sell this software and its
  * documentation for any purpose is hereby granted without fee, provided that
@@ -28,6 +29,7 @@
 #ifdef HAVE_CONFIG_H
 #include <config.h>
 #endif
+#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include "pixman-private.h"
@@ -157,6 +159,9 @@ repeat (pixman_repeat_t repeat, int size, int *coord)
 
     case PIXMAN_REPEAT_NONE:
 	break;
+
+    default:
+        break;
     }
 }
 
@@ -181,6 +186,97 @@ bits_image_fetch_pixel_nearest (bits_image_t   *image,
     }
 }
 
+#if SIZEOF_LONG > 4
+
+static force_inline uint32_t
+bilinear_interpolation (uint32_t tl, uint32_t tr,
+			uint32_t bl, uint32_t br,
+			int distx, int disty)
+{
+    uint64_t distxy, distxiy, distixy, distixiy;
+    uint64_t tl64, tr64, bl64, br64;
+    uint64_t f, r;
+
+    distxy = distx * disty;
+    distxiy = distx * (256 - disty);
+    distixy = (256 - distx) * disty;
+    distixiy = (256 - distx) * (256 - disty);
+
+    /* Alpha and Blue */
+    tl64 = tl & 0xff0000ff;
+    tr64 = tr & 0xff0000ff;
+    bl64 = bl & 0xff0000ff;
+    br64 = br & 0xff0000ff;
+
+    f = tl64 * distixiy + tr64 * distxiy + bl64 * distixy + br64 * distxy;
+    r = f & 0x0000ff0000ff0000ull;
+
+    /* Red and Green */
+    tl64 = tl;
+    tl64 = ((tl64 << 16) & 0x000000ff00000000ull) | (tl64 & 0x0000ff00ull);
+
+    tr64 = tr;
+    tr64 = ((tr64 << 16) & 0x000000ff00000000ull) | (tr64 & 0x0000ff00ull);
+
+    bl64 = bl;
+    bl64 = ((bl64 << 16) & 0x000000ff00000000ull) | (bl64 & 0x0000ff00ull);
+
+    br64 = br;
+    br64 = ((br64 << 16) & 0x000000ff00000000ull) | (br64 & 0x0000ff00ull);
+
+    f = tl64 * distixiy + tr64 * distxiy + bl64 * distixy + br64 * distxy;
+    r |= ((f >> 16) & 0x000000ff00000000ull) | (f & 0xff000000ull);
+
+    return (uint32_t)(r >> 16);
+}
+
+#else
+
+static force_inline uint32_t
+bilinear_interpolation (uint32_t tl, uint32_t tr,
+			uint32_t bl, uint32_t br,
+			int distx, int disty)
+{
+    int distxy, distxiy, distixy, distixiy;
+    uint32_t f, r;
+
+    distxy = distx * disty;
+    distxiy = (distx << 8) - distxy;	/* distx * (256 - disty) */
+    distixy = (disty << 8) - distxy;	/* disty * (256 - distx) */
+    distixiy =
+	256 * 256 - (disty << 8) -
+	(distx << 8) + distxy;		/* (256 - distx) * (256 - disty) */
+
+    /* Blue */
+    r = (tl & 0x000000ff) * distixiy + (tr & 0x000000ff) * distxiy
+      + (bl & 0x000000ff) * distixy  + (br & 0x000000ff) * distxy;
+
+    /* Green */
+    f = (tl & 0x0000ff00) * distixiy + (tr & 0x0000ff00) * distxiy
+      + (bl & 0x0000ff00) * distixy  + (br & 0x0000ff00) * distxy;
+    r |= f & 0xff000000;
+
+    tl >>= 16;
+    tr >>= 16;
+    bl >>= 16;
+    br >>= 16;
+    r >>= 16;
+
+    /* Red */
+    f = (tl & 0x000000ff) * distixiy + (tr & 0x000000ff) * distxiy
+      + (bl & 0x000000ff) * distixy  + (br & 0x000000ff) * distxy;
+    r |= f & 0x00ff0000;
+
+    /* Alpha */
+    f = (tl & 0x0000ff00) * distixiy + (tr & 0x0000ff00) * distxiy
+      + (bl & 0x0000ff00) * distixy  + (br & 0x0000ff00) * distxy;
+    r |= f & 0xff000000;
+
+    return r;
+}
+
+#endif
+
 static force_inline uint32_t
 bits_image_fetch_pixel_bilinear (bits_image_t   *image,
 				 pixman_fixed_t  x,
@@ -190,9 +286,8 @@ bits_image_fetch_pixel_bilinear (bits_image_t   *image,
     int width = image->width;
     int height = image->height;
     int x1, y1, x2, y2;
-    uint32_t tl, tr, bl, br, r;
-    int32_t distx, disty, idistx, idisty;
-    uint32_t ft, fb;
+    uint32_t tl, tr, bl, br;
+    int32_t distx, disty;
 
     x1 = x - pixman_fixed_1 / 2;
     y1 = y - pixman_fixed_1 / 2;
@@ -211,7 +306,7 @@ bits_image_fetch_pixel_bilinear (bits_image_t   *image,
 	repeat (repeat_mode, height, &y1);
 	repeat (repeat_mode, width, &x2);
 	repeat (repeat_mode, height, &y2);
-	
+
 	tl = get_pixel (image, x1, y1, FALSE);
 	bl = get_pixel (image, x1, y2, FALSE);
 	tr = get_pixel (image, x2, y1, FALSE);
@@ -225,24 +320,218 @@ bits_image_fetch_pixel_bilinear (bits_image_t   *image,
 	br = get_pixel (image, x2, y2, TRUE);
     }
 
-    idistx = 256 - distx;
-    idisty = 256 - disty;
+    return bilinear_interpolation (tl, tr, bl, br, distx, disty);
+}
+
+static void
+bits_image_fetch_bilinear_no_repeat_8888 (pixman_image_t * ima,
+					  int              offset,
+					  int              line,
+					  int              width,
+					  uint32_t *       buffer,
+					  const uint32_t * mask,
+					  uint32_t         mask_bits)
+{
+    bits_image_t *bits = &ima->bits;
+    pixman_fixed_t x_top, x_bottom, x;
+    pixman_fixed_t ux_top, ux_bottom, ux;
+    pixman_vector_t v;
+    uint32_t top_mask, bottom_mask;
+    uint32_t *top_row;
+    uint32_t *bottom_row;
+    uint32_t *end;
+    uint32_t zero[2] = { 0, 0 };
+    int y, y1, y2;
+    int disty;
+    int mask_inc;
+    int w;
+
+    /* reference point is the center of the pixel */
+    v.vector[0] = pixman_int_to_fixed (offset) + pixman_fixed_1 / 2;
+    v.vector[1] = pixman_int_to_fixed (line) + pixman_fixed_1 / 2;
+    v.vector[2] = pixman_fixed_1;
 
-#define GET8(v, i)   ((uint16_t) (uint8_t) ((v) >> i))
-    ft = GET8 (tl, 0) * idistx + GET8 (tr, 0) * distx;
-    fb = GET8 (bl, 0) * idistx + GET8 (br, 0) * distx;
-    r = (((ft * idisty + fb * disty) >> 16) & 0xff);
-    ft = GET8 (tl, 8) * idistx + GET8 (tr, 8) * distx;
-    fb = GET8 (bl, 8) * idistx + GET8 (br, 8) * distx;
-    r |= (((ft * idisty + fb * disty) >> 8) & 0xff00);
-    ft = GET8 (tl, 16) * idistx + GET8 (tr, 16) * distx;
-    fb = GET8 (bl, 16) * idistx + GET8 (br, 16) * distx;
-    r |= (((ft * idisty + fb * disty)) & 0xff0000);
-    ft = GET8 (tl, 24) * idistx + GET8 (tr, 24) * distx;
-    fb = GET8 (bl, 24) * idistx + GET8 (br, 24) * distx;
-    r |= (((ft * idisty + fb * disty) << 8) & 0xff000000);
+    if (!pixman_transform_point_3d (bits->common.transform, &v))
+	return;
 
-    return r;
+    ux = ux_top = ux_bottom = bits->common.transform->matrix[0][0];
+    x = x_top = x_bottom = v.vector[0] - pixman_fixed_1/2;
+
+    y = v.vector[1] - pixman_fixed_1/2;
+    disty = (y >> 8) & 0xff;
+
+    /* Load the pointers to the first and second lines from the source
+     * image that bilinear code must read.
+     *
+     * The main trick in this code is about the check if any line are
+     * outside of the image;
+     *
+     * When I realize that a line (any one) is outside, I change
+     * the pointer to a dummy area with zeros. Once I change this, I
+     * must be sure the pointer will not change, so I set the
+     * variables to each pointer increments inside the loop.
+     */
+    y1 = pixman_fixed_to_int (y);
+    y2 = y1 + 1;
+
+    if (y1 < 0 || y1 >= bits->height)
+    {
+	top_row = zero;
+	x_top = 0;
+	ux_top = 0;
+    }
+    else
+    {
+	top_row = bits->bits + y1 * bits->rowstride;
+	x_top = x;
+	ux_top = ux;
+    }
+
+    if (y2 < 0 || y2 >= bits->height)
+    {
+	bottom_row = zero;
+	x_bottom = 0;
+	ux_bottom = 0;
+    }
+    else
+    {
+	bottom_row = bits->bits + y2 * bits->rowstride;
+	x_bottom = x;
+	ux_bottom = ux;
+    }
+
+    /* Instead of checking whether the operation uses the mast in
+     * each loop iteration, verify this only once and prepare the
+     * variables to make the code smaller inside the loop.
+     */
+    if (!mask)
+    {
+        mask_inc = 0;
+        mask_bits = 1;
+        mask = &mask_bits;
+    }
+    else
+    {
+        /* If have a mask, prepare the variables to check it */
+        mask_inc = 1;
+    }
+
+    /* If both are zero, then the whole thing is zero */
+    if (top_row == zero && bottom_row == zero)
+    {
+	memset (buffer, 0, width * sizeof (uint32_t));
+	return;
+    }
+    else if (bits->format == PIXMAN_x8r8g8b8)
+    {
+	if (top_row == zero)
+	{
+	    top_mask = 0;
+	    bottom_mask = 0xff000000;
+	}
+	else if (bottom_row == zero)
+	{
+	    top_mask = 0xff000000;
+	    bottom_mask = 0;
+	}
+	else
+	{
+	    top_mask = 0xff000000;
+	    bottom_mask = 0xff000000;
+	}
+    }
+    else
+    {
+	top_mask = 0;
+	bottom_mask = 0;
+    }
+
+    end = buffer + width;
+
+    /* Zero fill to the left of the image */
+    while (buffer < end && x < pixman_fixed_minus_1)
+    {
+	*buffer++ = 0;
+	x += ux;
+	x_top += ux_top;
+	x_bottom += ux_bottom;
+	mask += mask_inc;
+    }
+
+    /* Left edge
+     */
+    while (buffer < end && x < 0)
+    {
+	uint32_t tr, br;
+	int32_t distx;
+
+	tr = top_row[pixman_fixed_to_int (x_top) + 1] | top_mask;
+	br = bottom_row[pixman_fixed_to_int (x_bottom) + 1] | bottom_mask;
+
+	distx = (x >> 8) & 0xff;
+
+	*buffer++ = bilinear_interpolation (0, tr, 0, br, distx, disty);
+
+	x += ux;
+	x_top += ux_top;
+	x_bottom += ux_bottom;
+	mask += mask_inc;
+    }
+
+    /* Main part */
+    w = pixman_int_to_fixed (bits->width - 1);
+
+    while (buffer < end  &&  x < w)
+    {
+	if (*mask)
+	{
+	    uint32_t tl, tr, bl, br;
+	    int32_t distx;
+
+	    tl = top_row [pixman_fixed_to_int (x_top)] | top_mask;
+	    tr = top_row [pixman_fixed_to_int (x_top) + 1] | top_mask;
+	    bl = bottom_row [pixman_fixed_to_int (x_bottom)] | bottom_mask;
+	    br = bottom_row [pixman_fixed_to_int (x_bottom) + 1] | bottom_mask;
+
+	    distx = (x >> 8) & 0xff;
+
+	    *buffer = bilinear_interpolation (tl, tr, bl, br, distx, disty);
+	}
+
+	buffer++;
+	x += ux;
+	x_top += ux_top;
+	x_bottom += ux_bottom;
+	mask += mask_inc;
+    }
+
+    /* Right Edge */
+    w = pixman_int_to_fixed (bits->width);
+    while (buffer < end  &&  x < w)
+    {
+	if (*mask)
+	{
+	    uint32_t tl, bl;
+	    int32_t distx;
+
+	    tl = top_row [pixman_fixed_to_int (x_top)] | top_mask;
+	    bl = bottom_row [pixman_fixed_to_int (x_bottom)] | bottom_mask;
+
+	    distx = (x >> 8) & 0xff;
+
+	    *buffer = bilinear_interpolation (tl, 0, bl, 0, distx, disty);
+	}
+
+	buffer++;
+	x += ux;
+	x_top += ux_top;
+	x_bottom += ux_bottom;
+	mask += mask_inc;
+    }
+
+    /* Zero fill to the left of the image */
+    while (buffer < end)
+	*buffer++ = 0;
 }
 
 static force_inline uint32_t
@@ -339,6 +628,9 @@ bits_image_fetch_pixel_filtered (bits_image_t *image,
     case PIXMAN_FILTER_CONVOLUTION:
 	return bits_image_fetch_pixel_convolution (image, x, y);
 	break;
+
+    default:
+        break;
     }
 
     return 0;
@@ -664,6 +956,24 @@ bits_image_property_changed (pixman_image_t *image)
 	image->common.get_scanline_64 = bits_image_fetch_untransformed_64;
 	image->common.get_scanline_32 = bits_image_fetch_untransformed_32;
     }
+    else if (bits->common.transform					&&
+	     bits->common.transform->matrix[2][0] == 0			&&
+	     bits->common.transform->matrix[2][1] == 0			&&
+	     bits->common.transform->matrix[2][2] == pixman_fixed_1	&&
+	     bits->common.transform->matrix[0][0] > 0			&&
+	     bits->common.transform->matrix[1][0] == 0			&&
+	     (bits->common.filter == PIXMAN_FILTER_BILINEAR ||
+	      bits->common.filter == PIXMAN_FILTER_GOOD	    ||
+	      bits->common.filter == PIXMAN_FILTER_BEST)		&&
+	     bits->common.repeat == PIXMAN_REPEAT_NONE			&&
+	     (bits->format == PIXMAN_a8r8g8b8	||
+	      bits->format == PIXMAN_x8r8g8b8))
+    {
+	image->common.get_scanline_64 =
+	    _pixman_image_get_scanline_generic_64;
+	image->common.get_scanline_32 =
+	    bits_image_fetch_bilinear_no_repeat_8888;
+    }
     else
     {
 	image->common.get_scanline_64 =
diff --git a/pixman/pixman/pixman-edge.c b/pixman/pixman/pixman-edge.c
index 81a2e960a..ab9bea0f8 100644
--- a/pixman/pixman/pixman-edge.c
+++ b/pixman/pixman/pixman-edge.c
@@ -358,6 +358,9 @@ PIXMAN_RASTERIZE_EDGES (pixman_image_t *image,
     case 8:
 	rasterize_edges_8 (image, l, r, t, b);
 	break;
+
+    default:
+        break;
     }
 }
 
diff --git a/pixman/pixman/pixman-fast-path.c b/pixman/pixman/pixman-fast-path.c
index 5ab8d8c99..75a0c1e07 100644
--- a/pixman/pixman/pixman-fast-path.c
+++ b/pixman/pixman/pixman-fast-path.c
@@ -723,6 +723,7 @@ fast_composite_over_8888_8888 (pixman_implementation_t *imp,
     }
 }
 
+#if 0
 static void
 fast_composite_over_8888_0888 (pixman_implementation_t *imp,
 			       pixman_op_t              op,
@@ -773,6 +774,7 @@ fast_composite_over_8888_0888 (pixman_implementation_t *imp,
 	}
     }
 }
+#endif
 
 static void
 fast_composite_over_8888_0565 (pixman_implementation_t *imp,
@@ -971,19 +973,19 @@ fast_composite_add_8888_8888 (pixman_implementation_t *imp,
 }
 
 static void
-fast_composite_add_8888_8_8 (pixman_implementation_t *imp,
-                             pixman_op_t              op,
-                             pixman_image_t *         src_image,
-                             pixman_image_t *         mask_image,
-                             pixman_image_t *         dst_image,
-                             int32_t                  src_x,
-                             int32_t                  src_y,
-                             int32_t                  mask_x,
-                             int32_t                  mask_y,
-                             int32_t                  dest_x,
-                             int32_t                  dest_y,
-                             int32_t                  width,
-                             int32_t                  height)
+fast_composite_add_n_8_8 (pixman_implementation_t *imp,
+			  pixman_op_t              op,
+			  pixman_image_t *         src_image,
+			  pixman_image_t *         mask_image,
+			  pixman_image_t *         dst_image,
+			  int32_t                  src_x,
+			  int32_t                  src_y,
+			  int32_t                  mask_x,
+			  int32_t                  mask_y,
+			  int32_t                  dest_x,
+			  int32_t                  dest_y,
+			  int32_t                  width,
+			  int32_t                  height)
 {
     uint8_t     *dst_line, *dst;
     uint8_t     *mask_line, *mask;
@@ -1023,6 +1025,254 @@ fast_composite_add_8888_8_8 (pixman_implementation_t *imp,
     }
 }
 
+#ifdef WORDS_BIGENDIAN
+#define CREATE_BITMASK(n) (0x80000000 >> (n))
+#define UPDATE_BITMASK(n) ((n) >> 1)
+#else
+#define CREATE_BITMASK(n) (1 << (n))
+#define UPDATE_BITMASK(n) ((n) << 1)
+#endif
+
+#define TEST_BIT(p, n) \
+	(*((p) + ((n) >> 5)) & CREATE_BITMASK ((n) & 31))
+#define SET_BIT(p, n) \
+	do { *((p) + ((n) >> 5)) |= CREATE_BITMASK ((n) & 31); } while (0);
+
+static void
+fast_composite_add_1000_1000 (pixman_implementation_t *imp,
+                              pixman_op_t              op,
+                              pixman_image_t *         src_image,
+                              pixman_image_t *         mask_image,
+                              pixman_image_t *         dst_image,
+                              int32_t                  src_x,
+                              int32_t                  src_y,
+                              int32_t                  mask_x,
+                              int32_t                  mask_y,
+                              int32_t                  dest_x,
+                              int32_t                  dest_y,
+                              int32_t                  width,
+                              int32_t                  height)
+{
+    uint32_t     *dst_line, *dst;
+    uint32_t     *src_line, *src;
+    int           dst_stride, src_stride;
+    int32_t       w;
+
+    PIXMAN_IMAGE_GET_LINE (src_image, 0, src_y, uint32_t,
+                           src_stride, src_line, 1);
+    PIXMAN_IMAGE_GET_LINE (dst_image, 0, dest_y, uint32_t,
+                           dst_stride, dst_line, 1);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	while (w--)
+	{
+	    /*
+	     * TODO: improve performance by processing uint32_t data instead
+	     *       of individual bits
+	     */
+	    if (TEST_BIT (src, src_x + w))
+		SET_BIT (dst, dest_x + w);
+	}
+    }
+}
+
+static void
+fast_composite_over_n_1_8888 (pixman_implementation_t *imp,
+                              pixman_op_t              op,
+                              pixman_image_t *         src_image,
+                              pixman_image_t *         mask_image,
+                              pixman_image_t *         dst_image,
+                              int32_t                  src_x,
+                              int32_t                  src_y,
+                              int32_t                  mask_x,
+                              int32_t                  mask_y,
+                              int32_t                  dest_x,
+                              int32_t                  dest_y,
+                              int32_t                  width,
+                              int32_t                  height)
+{
+    uint32_t     src, srca;
+    uint32_t    *dst, *dst_line;
+    uint32_t    *mask, *mask_line;
+    int          mask_stride, dst_stride;
+    uint32_t     bitcache, bitmask;
+    int32_t      w;
+
+    if (width <= 0)
+	return;
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+    srca = src >> 24;
+    if (src == 0)
+	return;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t,
+                           dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, 0, mask_y, uint32_t,
+                           mask_stride, mask_line, 1);
+    mask_line += mask_x >> 5;
+
+    if (srca == 0xff)
+    {
+	while (height--)
+	{
+	    dst = dst_line;
+	    dst_line += dst_stride;
+	    mask = mask_line;
+	    mask_line += mask_stride;
+	    w = width;
+
+	    bitcache = *mask++;
+	    bitmask = CREATE_BITMASK (mask_x & 31);
+
+	    while (w--)
+	    {
+		if (bitmask == 0)
+		{
+		    bitcache = *mask++;
+		    bitmask = CREATE_BITMASK (0);
+		}
+		if (bitcache & bitmask)
+		    *dst = src;
+		bitmask = UPDATE_BITMASK (bitmask);
+		dst++;
+	    }
+	}
+    }
+    else
+    {
+	while (height--)
+	{
+	    dst = dst_line;
+	    dst_line += dst_stride;
+	    mask = mask_line;
+	    mask_line += mask_stride;
+	    w = width;
+
+	    bitcache = *mask++;
+	    bitmask = CREATE_BITMASK (mask_x & 31);
+
+	    while (w--)
+	    {
+		if (bitmask == 0)
+		{
+		    bitcache = *mask++;
+		    bitmask = CREATE_BITMASK (0);
+		}
+		if (bitcache & bitmask)
+		    *dst = over (src, *dst);
+		bitmask = UPDATE_BITMASK (bitmask);
+		dst++;
+	    }
+	}
+    }
+}
+
+static void
+fast_composite_over_n_1_0565 (pixman_implementation_t *imp,
+                              pixman_op_t              op,
+                              pixman_image_t *         src_image,
+                              pixman_image_t *         mask_image,
+                              pixman_image_t *         dst_image,
+                              int32_t                  src_x,
+                              int32_t                  src_y,
+                              int32_t                  mask_x,
+                              int32_t                  mask_y,
+                              int32_t                  dest_x,
+                              int32_t                  dest_y,
+                              int32_t                  width,
+                              int32_t                  height)
+{
+    uint32_t     src, srca;
+    uint16_t    *dst, *dst_line;
+    uint32_t    *mask, *mask_line;
+    int          mask_stride, dst_stride;
+    uint32_t     bitcache, bitmask;
+    int32_t      w;
+    uint32_t     d;
+    uint16_t     src565;
+
+    if (width <= 0)
+	return;
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+    srca = src >> 24;
+    if (src == 0)
+	return;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t,
+                           dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, 0, mask_y, uint32_t,
+                           mask_stride, mask_line, 1);
+    mask_line += mask_x >> 5;
+
+    if (srca == 0xff)
+    {
+	src565 = CONVERT_8888_TO_0565 (src);
+	while (height--)
+	{
+	    dst = dst_line;
+	    dst_line += dst_stride;
+	    mask = mask_line;
+	    mask_line += mask_stride;
+	    w = width;
+
+	    bitcache = *mask++;
+	    bitmask = CREATE_BITMASK (mask_x & 31);
+
+	    while (w--)
+	    {
+		if (bitmask == 0)
+		{
+		    bitcache = *mask++;
+		    bitmask = CREATE_BITMASK (0);
+		}
+		if (bitcache & bitmask)
+		    *dst = src565;
+		bitmask = UPDATE_BITMASK (bitmask);
+		dst++;
+	    }
+	}
+    }
+    else
+    {
+	while (height--)
+	{
+	    dst = dst_line;
+	    dst_line += dst_stride;
+	    mask = mask_line;
+	    mask_line += mask_stride;
+	    w = width;
+
+	    bitcache = *mask++;
+	    bitmask = CREATE_BITMASK (mask_x & 31);
+
+	    while (w--)
+	    {
+		if (bitmask == 0)
+		{
+		    bitcache = *mask++;
+		    bitmask = CREATE_BITMASK (0);
+		}
+		if (bitcache & bitmask)
+		{
+		    d = over (src, CONVERT_0565_TO_0888 (*dst));
+		    *dst = CONVERT_8888_TO_0565 (d);
+		}
+		bitmask = UPDATE_BITMASK (bitmask);
+		dst++;
+	    }
+	}
+    }
+}
+
 /*
  * Simple bitblt
  */
@@ -1097,51 +1347,58 @@ fast_composite_src_8888_x888 (pixman_implementation_t *imp,
 
 static const pixman_fast_path_t c_fast_paths[] =
 {
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r5g6b5,   fast_composite_over_n_8_0565, 0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b5g6r5,   fast_composite_over_n_8_0565, 0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r8g8b8,   fast_composite_over_n_8_0888, 0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b8g8r8,   fast_composite_over_n_8_0888, 0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, fast_composite_over_n_8_8888, 0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, fast_composite_over_n_8_8888, 0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, fast_composite_over_n_8_8888, 0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, fast_composite_over_n_8_8888, 0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_a8r8g8b8, fast_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_x8r8g8b8, fast_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_r5g6b5,   fast_composite_over_n_8888_0565_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_a8b8g8r8, fast_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_x8b8g8r8, fast_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_b5g6r5,   fast_composite_over_n_8888_0565_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, fast_composite_over_x888_8_8888,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, fast_composite_over_x888_8_8888,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, fast_composite_over_x888_8_8888,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, fast_composite_over_x888_8_8888,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, fast_composite_over_8888_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, fast_composite_over_8888_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   fast_composite_over_8888_0565,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, fast_composite_over_8888_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, fast_composite_over_8888_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   fast_composite_over_8888_0565,    0 },
-    { PIXMAN_OP_ADD, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_a8r8g8b8, fast_composite_add_8888_8888,   0 },
-    { PIXMAN_OP_ADD, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_a8b8g8r8, fast_composite_add_8888_8888,   0 },
-    { PIXMAN_OP_ADD, PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       fast_composite_add_8000_8000,   0 },
-    { PIXMAN_OP_ADD, PIXMAN_solid,     PIXMAN_a8r8g8b8, PIXMAN_a8r8g8b8, fast_composite_add_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_ADD, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       fast_composite_add_8888_8_8,    0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_a8r8g8b8, fast_composite_solid_fill, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_x8r8g8b8, fast_composite_solid_fill, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_a8b8g8r8, fast_composite_solid_fill, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_x8b8g8r8, fast_composite_solid_fill, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_a8,       fast_composite_solid_fill, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_r5g6b5,   fast_composite_solid_fill, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, fast_composite_src_8888_x888, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_x8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, fast_composite_src_8888_x888, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, fast_composite_src_8888_x888, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_x8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, fast_composite_src_8888_x888, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_r5g6b5,   fast_composite_src_x888_0565, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_x8r8g8b8,  PIXMAN_null,     PIXMAN_r5g6b5,   fast_composite_src_x888_0565, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_b5g6r5,   fast_composite_src_x888_0565, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_x8b8g8r8,  PIXMAN_null,     PIXMAN_b5g6r5,   fast_composite_src_x888_0565, 0 },
-    { PIXMAN_OP_IN,  PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       fast_composite_in_8_8,   0 },
-    { PIXMAN_OP_IN,  PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       fast_composite_in_n_8_8, 0 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r5g6b5,   fast_composite_over_n_8_0565 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b5g6r5,   fast_composite_over_n_8_0565 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r8g8b8,   fast_composite_over_n_8_0888 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b8g8r8,   fast_composite_over_n_8_0888 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, fast_composite_over_n_8_8888 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, fast_composite_over_n_8_8888 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, fast_composite_over_n_8_8888 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, fast_composite_over_n_8_8888 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a1,       PIXMAN_a8r8g8b8, fast_composite_over_n_1_8888, },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a1,       PIXMAN_x8r8g8b8, fast_composite_over_n_1_8888, },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a1,       PIXMAN_a8b8g8r8, fast_composite_over_n_1_8888, },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a1,       PIXMAN_x8b8g8r8, fast_composite_over_n_1_8888, },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a1,       PIXMAN_r5g6b5,   fast_composite_over_n_1_0565 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a1,       PIXMAN_b5g6r5,   fast_composite_over_n_1_0565 },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_a8r8g8b8, fast_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_x8r8g8b8, fast_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_r5g6b5,   fast_composite_over_n_8888_0565_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_a8b8g8r8, fast_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_x8b8g8r8, fast_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_b5g6r5,   fast_composite_over_n_8888_0565_ca },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, fast_composite_over_x888_8_8888,      },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, fast_composite_over_x888_8_8888,      },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, fast_composite_over_x888_8_8888,      },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, fast_composite_over_x888_8_8888,      },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, fast_composite_over_8888_8888,   },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, fast_composite_over_8888_8888,   },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   fast_composite_over_8888_0565,   },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, fast_composite_over_8888_8888,   },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, fast_composite_over_8888_8888,   },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   fast_composite_over_8888_0565,   },
+    { PIXMAN_OP_ADD, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_a8r8g8b8, fast_composite_add_8888_8888,  },
+    { PIXMAN_OP_ADD, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_a8b8g8r8, fast_composite_add_8888_8888,  },
+    { PIXMAN_OP_ADD, PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       fast_composite_add_8000_8000,  },
+    { PIXMAN_OP_ADD, PIXMAN_a1,        PIXMAN_null,     PIXMAN_a1,       fast_composite_add_1000_1000,   },
+    { PIXMAN_OP_ADD, PIXMAN_solid,   PIXMAN_a8r8g8b8_ca, PIXMAN_a8r8g8b8, fast_composite_add_n_8888_8888_ca },
+    { PIXMAN_OP_ADD, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       fast_composite_add_n_8_8,   },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_a8r8g8b8, fast_composite_solid_fill },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_x8r8g8b8, fast_composite_solid_fill },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_a8b8g8r8, fast_composite_solid_fill },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_x8b8g8r8, fast_composite_solid_fill },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_a8,       fast_composite_solid_fill },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_null,     PIXMAN_r5g6b5,   fast_composite_solid_fill },
+    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, fast_composite_src_8888_x888 },
+    { PIXMAN_OP_SRC, PIXMAN_x8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, fast_composite_src_8888_x888 },
+    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, fast_composite_src_8888_x888 },
+    { PIXMAN_OP_SRC, PIXMAN_x8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, fast_composite_src_8888_x888 },
+    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_r5g6b5,   fast_composite_src_x888_0565 },
+    { PIXMAN_OP_SRC, PIXMAN_x8r8g8b8,  PIXMAN_null,     PIXMAN_r5g6b5,   fast_composite_src_x888_0565 },
+    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_b5g6r5,   fast_composite_src_x888_0565 },
+    { PIXMAN_OP_SRC, PIXMAN_x8b8g8r8,  PIXMAN_null,     PIXMAN_b5g6r5,   fast_composite_src_x888_0565 },
+    { PIXMAN_OP_IN,  PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       fast_composite_in_8_8,  },
+    { PIXMAN_OP_IN,  PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       fast_composite_in_n_8_8 },
     { PIXMAN_OP_NONE },
 };
 
diff --git a/pixman/pixman/pixman-general.c b/pixman/pixman/pixman-general.c
index 3ead3dac7..c96a3f97e 100644
--- a/pixman/pixman/pixman-general.c
+++ b/pixman/pixman/pixman-general.c
@@ -133,15 +133,27 @@ general_composite_rect  (pixman_implementation_t *imp,
     /* Skip the store step and composite directly into the
      * destination if the output format of the compose func matches
      * the destination format.
+     *
+     * If the destination format is a8r8g8b8 then we can always do
+     * this. If it is x8r8g8b8, then we can only do it if the
+     * operator doesn't make use of destination alpha.
      */
-    if (!wide &&
-        !dest->common.alpha_map &&
-        !dest->bits.write_func &&
-        (op == PIXMAN_OP_ADD || op == PIXMAN_OP_OVER) &&
-        (dest->bits.format == PIXMAN_a8r8g8b8 ||
-         dest->bits.format == PIXMAN_x8r8g8b8))
+    if ((dest->bits.format == PIXMAN_a8r8g8b8)	||
+	(dest->bits.format == PIXMAN_x8r8g8b8	&&
+	 (op == PIXMAN_OP_OVER		||
+	  op == PIXMAN_OP_ADD		||
+	  op == PIXMAN_OP_SRC		||
+	  op == PIXMAN_OP_CLEAR		||
+	  op == PIXMAN_OP_IN_REVERSE	||
+	  op == PIXMAN_OP_OUT_REVERSE	||
+	  op == PIXMAN_OP_DST)))
     {
-	store = NULL;
+	if (!wide &&
+	    !dest->common.alpha_map &&
+	    !dest->bits.write_func)
+	{
+	    store = NULL;
+	}
     }
 
     if (!store)
diff --git a/pixman/pixman/pixman-image.c b/pixman/pixman/pixman-image.c
index bba5fd68c..e55ba2c15 100644
--- a/pixman/pixman/pixman-image.c
+++ b/pixman/pixman/pixman-image.c
@@ -48,8 +48,6 @@ _pixman_init_gradient (gradient_t *                  gradient,
     gradient->n_stops = n_stops;
 
     gradient->stop_range = 0xffff;
-    gradient->color_table = NULL;
-    gradient->color_table_size = 0;
     gradient->common.class = SOURCE_IMAGE_CLASS_UNKNOWN;
 
     return TRUE;
@@ -593,6 +591,10 @@ _pixman_image_is_opaque (pixman_image_t *image)
 	if (ALPHA_8 (image->solid.color) != 0xff)
 	    return FALSE;
 	break;
+
+    default:
+        return FALSE;
+        break;
     }
 
     /* Convolution filters can introduce translucency if the sum of the
diff --git a/pixman/pixman/pixman-implementation.c b/pixman/pixman/pixman-implementation.c
index bcda9fe85..6488332b0 100644
--- a/pixman/pixman/pixman-implementation.c
+++ b/pixman/pixman/pixman-implementation.c
@@ -156,7 +156,7 @@ _pixman_implementation_create (pixman_implementation_t *delegate)
     imp->blt = delegate_blt;
     imp->fill = delegate_fill;
 
-    for (i = 0; i < PIXMAN_OP_LAST; ++i)
+    for (i = 0; i < PIXMAN_N_OPERATORS; ++i)
     {
 	imp->combine_32[i] = delegate_combine_32;
 	imp->combine_64[i] = delegate_combine_64;
diff --git a/pixman/pixman/pixman-mmx.c b/pixman/pixman/pixman-mmx.c
index 319c5a79d..afece1178 100644
--- a/pixman/pixman/pixman-mmx.c
+++ b/pixman/pixman/pixman-mmx.c
@@ -1,3418 +1,3414 @@
-/*
- * Copyright © 2004, 2005 Red Hat, Inc.
- * Copyright © 2004 Nicholas Miell
- * Copyright © 2005 Trolltech AS
- *
- * Permission to use, copy, modify, distribute, and sell this software and its
- * documentation for any purpose is hereby granted without fee, provided that
- * the above copyright notice appear in all copies and that both that
- * copyright notice and this permission notice appear in supporting
- * documentation, and that the name of Red Hat not be used in advertising or
- * publicity pertaining to distribution of the software without specific,
- * written prior permission.  Red Hat makes no representations about the
- * suitability of this software for any purpose.  It is provided "as is"
- * without express or implied warranty.
- *
- * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
- * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
- * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
- * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
- * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
- * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
- * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
- * SOFTWARE.
- *
- * Author:  Søren Sandmann (sandmann@redhat.com)
- * Minor Improvements: Nicholas Miell (nmiell@gmail.com)
- * MMX code paths for fbcompose.c by Lars Knoll (lars@trolltech.com)
- *
- * Based on work by Owen Taylor
- */
-
-#ifdef HAVE_CONFIG_H
-#include <config.h>
-#endif
-
-#ifdef USE_MMX
-
-#include <mmintrin.h>
-#include "pixman-private.h"
-#include "pixman-combine32.h"
-
-#define no_vERBOSE
-
-#ifdef VERBOSE
-#define CHECKPOINT() error_f ("at %s %d\n", __FUNCTION__, __LINE__)
-#else
-#define CHECKPOINT()
-#endif
-
-/* Notes about writing mmx code
- *
- * give memory operands as the second operand. If you give it as the
- * first, gcc will first load it into a register, then use that
- * register
- *
- *   ie. use
- *
- *         _mm_mullo_pi16 (x, mmx_constant);
- *
- *   not
- *
- *         _mm_mullo_pi16 (mmx_constant, x);
- *
- * Also try to minimize dependencies. i.e. when you need a value, try
- * to calculate it from a value that was calculated as early as
- * possible.
- */
-
-/* --------------- MMX primitives ------------------------------------- */
-
-#ifdef __GNUC__
-typedef uint64_t mmxdatafield;
-#else
-typedef __m64 mmxdatafield;
-/* If __m64 is defined as a struct or union, define M64_MEMBER to be the
-   name of the member used to access the data */
-# ifdef _MSC_VER
-#  define M64_MEMBER m64_u64
-# elif defined(__SUNPRO_C)
-#  define M64_MEMBER l_
-# endif
-#endif
-
-typedef struct
-{
-    mmxdatafield mmx_4x00ff;
-    mmxdatafield mmx_4x0080;
-    mmxdatafield mmx_565_rgb;
-    mmxdatafield mmx_565_unpack_multiplier;
-    mmxdatafield mmx_565_r;
-    mmxdatafield mmx_565_g;
-    mmxdatafield mmx_565_b;
-    mmxdatafield mmx_mask_0;
-    mmxdatafield mmx_mask_1;
-    mmxdatafield mmx_mask_2;
-    mmxdatafield mmx_mask_3;
-    mmxdatafield mmx_full_alpha;
-    mmxdatafield mmx_ffff0000ffff0000;
-    mmxdatafield mmx_0000ffff00000000;
-    mmxdatafield mmx_000000000000ffff;
-} mmx_data_t;
-
-#if defined(_MSC_VER)
-# define MMXDATA_INIT(field, val) { val ## UI64 }
-#elif defined(M64_MEMBER)       /* __m64 is a struct, not an integral type */
-# define MMXDATA_INIT(field, val) field =   { val ## ULL }
-#else                           /* __m64 is an integral type */
-# define MMXDATA_INIT(field, val) field =   val ## ULL
-#endif
-
-static const mmx_data_t c =
-{
-    MMXDATA_INIT (.mmx_4x00ff,                   0x00ff00ff00ff00ff),
-    MMXDATA_INIT (.mmx_4x0080,                   0x0080008000800080),
-    MMXDATA_INIT (.mmx_565_rgb,                  0x000001f0003f001f),
-    MMXDATA_INIT (.mmx_565_unpack_multiplier,    0x0000008404100840),
-    MMXDATA_INIT (.mmx_565_r,                    0x000000f800000000),
-    MMXDATA_INIT (.mmx_565_g,                    0x0000000000fc0000),
-    MMXDATA_INIT (.mmx_565_b,                    0x00000000000000f8),
-    MMXDATA_INIT (.mmx_mask_0,                   0xffffffffffff0000),
-    MMXDATA_INIT (.mmx_mask_1,                   0xffffffff0000ffff),
-    MMXDATA_INIT (.mmx_mask_2,                   0xffff0000ffffffff),
-    MMXDATA_INIT (.mmx_mask_3,                   0x0000ffffffffffff),
-    MMXDATA_INIT (.mmx_full_alpha,               0x00ff000000000000),
-    MMXDATA_INIT (.mmx_ffff0000ffff0000,         0xffff0000ffff0000),
-    MMXDATA_INIT (.mmx_0000ffff00000000,         0x0000ffff00000000),
-    MMXDATA_INIT (.mmx_000000000000ffff,         0x000000000000ffff),
-};
-
-#ifdef __GNUC__
-#    ifdef __ICC
-#        define MC(x)  M64 (c.mmx_ ## x)
-#    else
-#        define MC(x) ((__m64)c.mmx_ ## x)
-#    endif
-#else
-#    define MC(x) c.mmx_ ## x
-#endif
-
-static force_inline __m64
-M64 (uint64_t x)
-{
-#ifdef __ICC
-    return _mm_cvtsi64_m64 (x);
-#elif defined M64_MEMBER        /* __m64 is a struct, not an integral type */
-    __m64 res;
-
-    res.M64_MEMBER = x;
-    return res;
-#else                           /* __m64 is an integral type */
-    return (__m64)x;
-#endif
-}
-
-static force_inline uint64_t
-UINT64 (__m64 x)
-{
-#ifdef __ICC
-    return _mm_cvtm64_si64 (x);
-#elif defined M64_MEMBER        /* __m64 is a struct, not an integral type */
-    uint64_t res = x.M64_MEMBER;
-    return res;
-#else                           /* __m64 is an integral type */
-    return (uint64_t)x;
-#endif
-}
-
-static force_inline __m64
-shift (__m64 v,
-       int   s)
-{
-    if (s > 0)
-	return _mm_slli_si64 (v, s);
-    else if (s < 0)
-	return _mm_srli_si64 (v, -s);
-    else
-	return v;
-}
-
-static force_inline __m64
-negate (__m64 mask)
-{
-    return _mm_xor_si64 (mask, MC (4x00ff));
-}
-
-static force_inline __m64
-pix_multiply (__m64 a, __m64 b)
-{
-    __m64 res;
-
-    res = _mm_mullo_pi16 (a, b);
-    res = _mm_adds_pu16 (res, MC (4x0080));
-    res = _mm_adds_pu16 (res, _mm_srli_pi16 (res, 8));
-    res = _mm_srli_pi16 (res, 8);
-
-    return res;
-}
-
-static force_inline __m64
-pix_add (__m64 a, __m64 b)
-{
-    return _mm_adds_pu8 (a, b);
-}
-
-static force_inline __m64
-expand_alpha (__m64 pixel)
-{
-    __m64 t1, t2;
-
-    t1 = shift (pixel, -48);
-    t2 = shift (t1, 16);
-    t1 = _mm_or_si64 (t1, t2);
-    t2 = shift (t1, 32);
-    t1 = _mm_or_si64 (t1, t2);
-
-    return t1;
-}
-
-static force_inline __m64
-expand_alpha_rev (__m64 pixel)
-{
-    __m64 t1, t2;
-
-    /* move alpha to low 16 bits and zero the rest */
-    t1 = shift (pixel,  48);
-    t1 = shift (t1, -48);
-
-    t2 = shift (t1, 16);
-    t1 = _mm_or_si64 (t1, t2);
-    t2 = shift (t1, 32);
-    t1 = _mm_or_si64 (t1, t2);
-
-    return t1;
-}
-
-static force_inline __m64
-invert_colors (__m64 pixel)
-{
-    __m64 x, y, z;
-
-    x = y = z = pixel;
-
-    x = _mm_and_si64 (x, MC (ffff0000ffff0000));
-    y = _mm_and_si64 (y, MC (000000000000ffff));
-    z = _mm_and_si64 (z, MC (0000ffff00000000));
-
-    y = shift (y, 32);
-    z = shift (z, -32);
-
-    x = _mm_or_si64 (x, y);
-    x = _mm_or_si64 (x, z);
-
-    return x;
-}
-
-static force_inline __m64
-over (__m64 src,
-      __m64 srca,
-      __m64 dest)
-{
-    return _mm_adds_pu8 (src, pix_multiply (dest, negate (srca)));
-}
-
-static force_inline __m64
-over_rev_non_pre (__m64 src, __m64 dest)
-{
-    __m64 srca = expand_alpha (src);
-    __m64 srcfaaa = _mm_or_si64 (srca, MC (full_alpha));
-
-    return over (pix_multiply (invert_colors (src), srcfaaa), srca, dest);
-}
-
-static force_inline __m64
-in (__m64 src, __m64 mask)
-{
-    return pix_multiply (src, mask);
-}
-
-static force_inline __m64
-in_over_full_src_alpha (__m64 src, __m64 mask, __m64 dest)
-{
-    src = _mm_or_si64 (src, MC (full_alpha));
-
-    return over (in (src, mask), mask, dest);
-}
-
-#ifndef _MSC_VER
-static force_inline __m64
-in_over (__m64 src, __m64 srca, __m64 mask, __m64 dest)
-{
-    return over (in (src, mask), pix_multiply (srca, mask), dest);
-}
-
-#else
-
-#define in_over(src, srca, mask, dest)					\
-    over (in (src, mask), pix_multiply (srca, mask), dest)
-
-#endif
-
-static force_inline __m64
-load8888 (uint32_t v)
-{
-    return _mm_unpacklo_pi8 (_mm_cvtsi32_si64 (v), _mm_setzero_si64 ());
-}
-
-static force_inline __m64
-pack8888 (__m64 lo, __m64 hi)
-{
-    return _mm_packs_pu16 (lo, hi);
-}
-
-static force_inline uint32_t
-store8888 (__m64 v)
-{
-    return _mm_cvtsi64_si32 (pack8888 (v, _mm_setzero_si64 ()));
-}
-
-/* Expand 16 bits positioned at @pos (0-3) of a mmx register into
- *
- *    00RR00GG00BB
- *
- * --- Expanding 565 in the low word ---
- *
- * m = (m << (32 - 3)) | (m << (16 - 5)) | m;
- * m = m & (01f0003f001f);
- * m = m * (008404100840);
- * m = m >> 8;
- *
- * Note the trick here - the top word is shifted by another nibble to
- * avoid it bumping into the middle word
- */
-static force_inline __m64
-expand565 (__m64 pixel, int pos)
-{
-    __m64 p = pixel;
-    __m64 t1, t2;
-
-    /* move pixel to low 16 bit and zero the rest */
-    p = shift (shift (p, (3 - pos) * 16), -48);
-
-    t1 = shift (p, 36 - 11);
-    t2 = shift (p, 16 - 5);
-
-    p = _mm_or_si64 (t1, p);
-    p = _mm_or_si64 (t2, p);
-    p = _mm_and_si64 (p, MC (565_rgb));
-
-    pixel = _mm_mullo_pi16 (p, MC (565_unpack_multiplier));
-    return _mm_srli_pi16 (pixel, 8);
-}
-
-static force_inline __m64
-expand8888 (__m64 in, int pos)
-{
-    if (pos == 0)
-	return _mm_unpacklo_pi8 (in, _mm_setzero_si64 ());
-    else
-	return _mm_unpackhi_pi8 (in, _mm_setzero_si64 ());
-}
-
-static force_inline __m64
-expandx888 (__m64 in, int pos)
-{
-    return _mm_or_si64 (expand8888 (in, pos), MC (full_alpha));
-}
-
-static force_inline __m64
-pack_565 (__m64 pixel, __m64 target, int pos)
-{
-    __m64 p = pixel;
-    __m64 t = target;
-    __m64 r, g, b;
-
-    r = _mm_and_si64 (p, MC (565_r));
-    g = _mm_and_si64 (p, MC (565_g));
-    b = _mm_and_si64 (p, MC (565_b));
-
-    r = shift (r, -(32 - 8) + pos * 16);
-    g = shift (g, -(16 - 3) + pos * 16);
-    b = shift (b, -(0  + 3) + pos * 16);
-
-    if (pos == 0)
-	t = _mm_and_si64 (t, MC (mask_0));
-    else if (pos == 1)
-	t = _mm_and_si64 (t, MC (mask_1));
-    else if (pos == 2)
-	t = _mm_and_si64 (t, MC (mask_2));
-    else if (pos == 3)
-	t = _mm_and_si64 (t, MC (mask_3));
-
-    p = _mm_or_si64 (r, t);
-    p = _mm_or_si64 (g, p);
-
-    return _mm_or_si64 (b, p);
-}
-
-#ifndef _MSC_VER
-
-static force_inline __m64
-pix_add_mul (__m64 x, __m64 a, __m64 y, __m64 b)
-{
-    x = pix_multiply (x, a);
-    y = pix_multiply (y, b);
-
-    return pix_add (x, y);
-}
-
-#else
-
-#define pix_add_mul(x, a, y, b)	 \
-    ( x = pix_multiply (x, a),	 \
-      y = pix_multiply (y, a),	 \
-      pix_add (x, y) )
-
-#endif
-
-/* --------------- MMX code patch for fbcompose.c --------------------- */
-
-static force_inline uint32_t
-combine (const uint32_t *src, const uint32_t *mask)
-{
-    uint32_t ssrc = *src;
-
-    if (mask)
-    {
-	__m64 m = load8888 (*mask);
-	__m64 s = load8888 (ssrc);
-
-	m = expand_alpha (m);
-	s = pix_multiply (s, m);
-
-	ssrc = store8888 (s);
-    }
-
-    return ssrc;
-}
-
-static void
-mmx_combine_over_u (pixman_implementation_t *imp,
-                    pixman_op_t              op,
-                    uint32_t *               dest,
-                    const uint32_t *         src,
-                    const uint32_t *         mask,
-                    int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	uint32_t ssrc = combine (src, mask);
-	uint32_t a = ssrc >> 24;
-
-	if (a == 0xff)
-	{
-	    *dest = ssrc;
-	}
-	else if (ssrc)
-	{
-	    __m64 s, sa;
-	    s = load8888 (ssrc);
-	    sa = expand_alpha (s);
-	    *dest = store8888 (over (s, sa, load8888 (*dest)));
-	}
-
-	++dest;
-	++src;
-	if (mask)
-	    ++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_over_reverse_u (pixman_implementation_t *imp,
-                            pixman_op_t              op,
-                            uint32_t *               dest,
-                            const uint32_t *         src,
-                            const uint32_t *         mask,
-                            int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 d, da;
-	uint32_t s = combine (src, mask);
-	
-	d = load8888 (*dest);
-	da = expand_alpha (d);
-	*dest = store8888 (over (d, da, load8888 (s)));
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_in_u (pixman_implementation_t *imp,
-                  pixman_op_t              op,
-                  uint32_t *               dest,
-                  const uint32_t *         src,
-                  const uint32_t *         mask,
-                  int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 x, a;
-	
-	x = load8888 (combine (src, mask));
-	a = load8888 (*dest);
-	a = expand_alpha (a);
-	x = pix_multiply (x, a);
-	
-	*dest = store8888 (x);
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_in_reverse_u (pixman_implementation_t *imp,
-                          pixman_op_t              op,
-                          uint32_t *               dest,
-                          const uint32_t *         src,
-                          const uint32_t *         mask,
-                          int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 x, a;
-	
-	x = load8888 (*dest);
-	a = load8888 (combine (src, mask));
-	a = expand_alpha (a);
-	x = pix_multiply (x, a);
-	*dest = store8888 (x);
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_out_u (pixman_implementation_t *imp,
-                   pixman_op_t              op,
-                   uint32_t *               dest,
-                   const uint32_t *         src,
-                   const uint32_t *         mask,
-                   int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 x, a;
-	
-	x = load8888 (combine (src, mask));
-	a = load8888 (*dest);
-	a = expand_alpha (a);
-	a = negate (a);
-	x = pix_multiply (x, a);
-	*dest = store8888 (x);
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_out_reverse_u (pixman_implementation_t *imp,
-                           pixman_op_t              op,
-                           uint32_t *               dest,
-                           const uint32_t *         src,
-                           const uint32_t *         mask,
-                           int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 x, a;
-	
-	x = load8888 (*dest);
-	a = load8888 (combine (src, mask));
-	a = expand_alpha (a);
-	a = negate (a);
-	x = pix_multiply (x, a);
-
-	*dest = store8888 (x);
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_atop_u (pixman_implementation_t *imp,
-                    pixman_op_t              op,
-                    uint32_t *               dest,
-                    const uint32_t *         src,
-                    const uint32_t *         mask,
-                    int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 s, da, d, sia;
-	
-	s = load8888 (combine (src, mask));
-	d = load8888 (*dest);
-	sia = expand_alpha (s);
-	sia = negate (sia);
-	da = expand_alpha (d);
-	s = pix_add_mul (s, da, d, sia);
-	*dest = store8888 (s);
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_atop_reverse_u (pixman_implementation_t *imp,
-                            pixman_op_t              op,
-                            uint32_t *               dest,
-                            const uint32_t *         src,
-                            const uint32_t *         mask,
-                            int                      width)
-{
-    const uint32_t *end;
-
-    end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 s, dia, d, sa;
-	
-	s = load8888 (combine (src, mask));
-	d = load8888 (*dest);
-	sa = expand_alpha (s);
-	dia = expand_alpha (d);
-	dia = negate (dia);
-	s = pix_add_mul (s, dia, d, sa);
-	*dest = store8888 (s);
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_xor_u (pixman_implementation_t *imp,
-                   pixman_op_t              op,
-                   uint32_t *               dest,
-                   const uint32_t *         src,
-                   const uint32_t *         mask,
-                   int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 s, dia, d, sia;
-	
-	s = load8888 (combine (src, mask));
-	d = load8888 (*dest);
-	sia = expand_alpha (s);
-	dia = expand_alpha (d);
-	sia = negate (sia);
-	dia = negate (dia);
-	s = pix_add_mul (s, dia, d, sia);
-	*dest = store8888 (s);
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_add_u (pixman_implementation_t *imp,
-                   pixman_op_t              op,
-                   uint32_t *               dest,
-                   const uint32_t *         src,
-                   const uint32_t *         mask,
-                   int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	__m64 s, d;
-	
-	s = load8888 (combine (src, mask));
-	d = load8888 (*dest);
-	s = pix_add (s, d);
-	*dest = store8888 (s);
-
-	++dest;
-	++src;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_saturate_u (pixman_implementation_t *imp,
-                        pixman_op_t              op,
-                        uint32_t *               dest,
-                        const uint32_t *         src,
-                        const uint32_t *         mask,
-                        int                      width)
-{
-    const uint32_t *end = dest + width;
-
-    while (dest < end)
-    {
-	uint32_t s = combine (src, mask);
-	uint32_t d = *dest;
-	__m64 ms = load8888 (s);
-	__m64 md = load8888 (d);
-	uint32_t sa = s >> 24;
-	uint32_t da = ~d >> 24;
-
-	if (sa > da)
-	{
-	    __m64 msa = load8888 (DIV_UN8 (da, sa) << 24);
-	    msa = expand_alpha (msa);
-	    ms = pix_multiply (ms, msa);
-	}
-
-	md = pix_add (md, ms);
-	*dest = store8888 (md);
-
-	++src;
-	++dest;
-	if (mask)
-	    mask++;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_src_ca (pixman_implementation_t *imp,
-                    pixman_op_t              op,
-                    uint32_t *               dest,
-                    const uint32_t *         src,
-                    const uint32_t *         mask,
-                    int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	
-	s = pix_multiply (s, a);
-	*dest = store8888 (s);
-
-	++src;
-	++mask;
-	++dest;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_over_ca (pixman_implementation_t *imp,
-                     pixman_op_t              op,
-                     uint32_t *               dest,
-                     const uint32_t *         src,
-                     const uint32_t *         mask,
-                     int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 sa = expand_alpha (s);
-
-	*dest = store8888 (in_over (s, sa, a, d));
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_over_reverse_ca (pixman_implementation_t *imp,
-                             pixman_op_t              op,
-                             uint32_t *               dest,
-                             const uint32_t *         src,
-                             const uint32_t *         mask,
-                             int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 da = expand_alpha (d);
-
-	*dest = store8888 (over (d, da, in (s, a)));
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_in_ca (pixman_implementation_t *imp,
-                   pixman_op_t              op,
-                   uint32_t *               dest,
-                   const uint32_t *         src,
-                   const uint32_t *         mask,
-                   int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 da = expand_alpha (d);
-	
-	s = pix_multiply (s, a);
-	s = pix_multiply (s, da);
-	*dest = store8888 (s);
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_in_reverse_ca (pixman_implementation_t *imp,
-                           pixman_op_t              op,
-                           uint32_t *               dest,
-                           const uint32_t *         src,
-                           const uint32_t *         mask,
-                           int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 sa = expand_alpha (s);
-	
-	a = pix_multiply (a, sa);
-	d = pix_multiply (d, a);
-	*dest = store8888 (d);
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_out_ca (pixman_implementation_t *imp,
-                    pixman_op_t              op,
-                    uint32_t *               dest,
-                    const uint32_t *         src,
-                    const uint32_t *         mask,
-                    int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 da = expand_alpha (d);
-	
-	da = negate (da);
-	s = pix_multiply (s, a);
-	s = pix_multiply (s, da);
-	*dest = store8888 (s);
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_out_reverse_ca (pixman_implementation_t *imp,
-                            pixman_op_t              op,
-                            uint32_t *               dest,
-                            const uint32_t *         src,
-                            const uint32_t *         mask,
-                            int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 sa = expand_alpha (s);
-
-	a = pix_multiply (a, sa);
-	a = negate (a);
-	d = pix_multiply (d, a);
-	*dest = store8888 (d);
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_atop_ca (pixman_implementation_t *imp,
-                     pixman_op_t              op,
-                     uint32_t *               dest,
-                     const uint32_t *         src,
-                     const uint32_t *         mask,
-                     int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 da = expand_alpha (d);
-	__m64 sa = expand_alpha (s);
-
-	s = pix_multiply (s, a);
-	a = pix_multiply (a, sa);
-	a = negate (a);
-	d = pix_add_mul (d, a, s, da);
-	*dest = store8888 (d);
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_atop_reverse_ca (pixman_implementation_t *imp,
-                             pixman_op_t              op,
-                             uint32_t *               dest,
-                             const uint32_t *         src,
-                             const uint32_t *         mask,
-                             int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 da = expand_alpha (d);
-	__m64 sa = expand_alpha (s);
-
-	s = pix_multiply (s, a);
-	a = pix_multiply (a, sa);
-	da = negate (da);
-	d = pix_add_mul (d, a, s, da);
-	*dest = store8888 (d);
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_xor_ca (pixman_implementation_t *imp,
-                    pixman_op_t              op,
-                    uint32_t *               dest,
-                    const uint32_t *         src,
-                    const uint32_t *         mask,
-                    int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-	__m64 da = expand_alpha (d);
-	__m64 sa = expand_alpha (s);
-
-	s = pix_multiply (s, a);
-	a = pix_multiply (a, sa);
-	da = negate (da);
-	a = negate (a);
-	d = pix_add_mul (d, a, s, da);
-	*dest = store8888 (d);
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_combine_add_ca (pixman_implementation_t *imp,
-                    pixman_op_t              op,
-                    uint32_t *               dest,
-                    const uint32_t *         src,
-                    const uint32_t *         mask,
-                    int                      width)
-{
-    const uint32_t *end = src + width;
-
-    while (src < end)
-    {
-	__m64 a = load8888 (*mask);
-	__m64 s = load8888 (*src);
-	__m64 d = load8888 (*dest);
-
-	s = pix_multiply (s, a);
-	d = pix_add (s, d);
-	*dest = store8888 (d);
-
-	++src;
-	++dest;
-	++mask;
-    }
-    _mm_empty ();
-}
-
-/* ------------- MMX code paths called from fbpict.c -------------------- */
-
-static void
-mmx_composite_over_n_8888 (pixman_implementation_t *imp,
-                           pixman_op_t              op,
-                           pixman_image_t *         src_image,
-                           pixman_image_t *         mask_image,
-                           pixman_image_t *         dst_image,
-                           int32_t                  src_x,
-                           int32_t                  src_y,
-                           int32_t                  mask_x,
-                           int32_t                  mask_y,
-                           int32_t                  dest_x,
-                           int32_t                  dest_y,
-                           int32_t                  width,
-                           int32_t                  height)
-{
-    uint32_t src;
-    uint32_t    *dst_line, *dst;
-    uint16_t w;
-    int dst_stride;
-    __m64 vsrc, vsrca;
-
-    CHECKPOINT ();
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    if (src == 0)
-	return;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	w = width;
-
-	CHECKPOINT ();
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    *dst = store8888 (over (vsrc, vsrca, load8888 (*dst)));
-
-	    w--;
-	    dst++;
-	}
-
-	while (w >= 2)
-	{
-	    __m64 vdest;
-	    __m64 dest0, dest1;
-
-	    vdest = *(__m64 *)dst;
-
-	    dest0 = over (vsrc, vsrca, expand8888 (vdest, 0));
-	    dest1 = over (vsrc, vsrca, expand8888 (vdest, 1));
-
-	    *(__m64 *)dst = pack8888 (dest0, dest1);
-
-	    dst += 2;
-	    w -= 2;
-	}
-
-	CHECKPOINT ();
-
-	while (w)
-	{
-	    *dst = store8888 (over (vsrc, vsrca, load8888 (*dst)));
-
-	    w--;
-	    dst++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_n_0565 (pixman_implementation_t *imp,
-                           pixman_op_t              op,
-                           pixman_image_t *         src_image,
-                           pixman_image_t *         mask_image,
-                           pixman_image_t *         dst_image,
-                           int32_t                  src_x,
-                           int32_t                  src_y,
-                           int32_t                  mask_x,
-                           int32_t                  mask_y,
-                           int32_t                  dest_x,
-                           int32_t                  dest_y,
-                           int32_t                  width,
-                           int32_t                  height)
-{
-    uint32_t src;
-    uint16_t    *dst_line, *dst;
-    uint16_t w;
-    int dst_stride;
-    __m64 vsrc, vsrca;
-
-    CHECKPOINT ();
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    if (src == 0)
-	return;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	w = width;
-
-	CHECKPOINT ();
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    uint64_t d = *dst;
-	    __m64 vdest = expand565 (M64 (d), 0);
-	    
-	    vdest = pack_565 (over (vsrc, vsrca, vdest), vdest, 0);
-	    *dst = UINT64 (vdest);
-
-	    w--;
-	    dst++;
-	}
-
-	while (w >= 4)
-	{
-	    __m64 vdest;
-
-	    vdest = *(__m64 *)dst;
-
-	    vdest = pack_565 (over (vsrc, vsrca, expand565 (vdest, 0)), vdest, 0);
-	    vdest = pack_565 (over (vsrc, vsrca, expand565 (vdest, 1)), vdest, 1);
-	    vdest = pack_565 (over (vsrc, vsrca, expand565 (vdest, 2)), vdest, 2);
-	    vdest = pack_565 (over (vsrc, vsrca, expand565 (vdest, 3)), vdest, 3);
-
-	    *(__m64 *)dst = vdest;
-
-	    dst += 4;
-	    w -= 4;
-	}
-
-	CHECKPOINT ();
-
-	while (w)
-	{
-	    uint64_t d = *dst;
-	    __m64 vdest = expand565 (M64 (d), 0);
-	    
-	    vdest = pack_565 (over (vsrc, vsrca, vdest), vdest, 0);
-	    *dst = UINT64 (vdest);
-
-	    w--;
-	    dst++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
-                                   pixman_op_t              op,
-                                   pixman_image_t *         src_image,
-                                   pixman_image_t *         mask_image,
-                                   pixman_image_t *         dst_image,
-                                   int32_t                  src_x,
-                                   int32_t                  src_y,
-                                   int32_t                  mask_x,
-                                   int32_t                  mask_y,
-                                   int32_t                  dest_x,
-                                   int32_t                  dest_y,
-                                   int32_t                  width,
-                                   int32_t                  height)
-{
-    uint32_t src, srca;
-    uint32_t    *dst_line;
-    uint32_t    *mask_line;
-    int dst_stride, mask_stride;
-    __m64 vsrc, vsrca;
-
-    CHECKPOINT ();
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    srca = src >> 24;
-    if (src == 0)
-	return;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint32_t, mask_stride, mask_line, 1);
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    while (height--)
-    {
-	int twidth = width;
-	uint32_t *p = (uint32_t *)mask_line;
-	uint32_t *q = (uint32_t *)dst_line;
-
-	while (twidth && (unsigned long)q & 7)
-	{
-	    uint32_t m = *(uint32_t *)p;
-
-	    if (m)
-	    {
-		__m64 vdest = load8888 (*q);
-		vdest = in_over (vsrc, vsrca, load8888 (m), vdest);
-		*q = store8888 (vdest);
-	    }
-
-	    twidth--;
-	    p++;
-	    q++;
-	}
-
-	while (twidth >= 2)
-	{
-	    uint32_t m0, m1;
-	    m0 = *p;
-	    m1 = *(p + 1);
-
-	    if (m0 | m1)
-	    {
-		__m64 dest0, dest1;
-		__m64 vdest = *(__m64 *)q;
-
-		dest0 = in_over (vsrc, vsrca, load8888 (m0),
-		                 expand8888 (vdest, 0));
-		dest1 = in_over (vsrc, vsrca, load8888 (m1),
-		                 expand8888 (vdest, 1));
-
-		*(__m64 *)q = pack8888 (dest0, dest1);
-	    }
-
-	    p += 2;
-	    q += 2;
-	    twidth -= 2;
-	}
-
-	while (twidth)
-	{
-	    uint32_t m = *(uint32_t *)p;
-
-	    if (m)
-	    {
-		__m64 vdest = load8888 (*q);
-		vdest = in_over (vsrc, vsrca, load8888 (m), vdest);
-		*q = store8888 (vdest);
-	    }
-
-	    twidth--;
-	    p++;
-	    q++;
-	}
-
-	dst_line += dst_stride;
-	mask_line += mask_stride;
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_8888_n_8888 (pixman_implementation_t *imp,
-                                pixman_op_t              op,
-                                pixman_image_t *         src_image,
-                                pixman_image_t *         mask_image,
-                                pixman_image_t *         dst_image,
-                                int32_t                  src_x,
-                                int32_t                  src_y,
-                                int32_t                  mask_x,
-                                int32_t                  mask_y,
-                                int32_t                  dest_x,
-                                int32_t                  dest_y,
-                                int32_t                  width,
-                                int32_t                  height)
-{
-    uint32_t    *dst_line, *dst;
-    uint32_t    *src_line, *src;
-    uint32_t mask;
-    __m64 vmask;
-    int dst_stride, src_stride;
-    uint16_t w;
-    __m64 srca;
-
-    CHECKPOINT ();
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);
-    mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
-    vmask = load8888 (mask);
-    srca = MC (4x00ff);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    __m64 s = load8888 (*src);
-	    __m64 d = load8888 (*dst);
-
-	    *dst = store8888 (in_over (s, expand_alpha (s), vmask, d));
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-
-	while (w >= 2)
-	{
-	    __m64 vs = *(__m64 *)src;
-	    __m64 vd = *(__m64 *)dst;
-	    __m64 vsrc0 = expand8888 (vs, 0);
-	    __m64 vsrc1 = expand8888 (vs, 1);
-
-	    *(__m64 *)dst = pack8888 (
-	        in_over (vsrc0, expand_alpha (vsrc0), vmask, expand8888 (vd, 0)),
-	        in_over (vsrc1, expand_alpha (vsrc1), vmask, expand8888 (vd, 1)));
-
-	    w -= 2;
-	    dst += 2;
-	    src += 2;
-	}
-
-	while (w)
-	{
-	    __m64 s = load8888 (*src);
-	    __m64 d = load8888 (*dst);
-
-	    *dst = store8888 (in_over (s, expand_alpha (s), vmask, d));
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_x888_n_8888 (pixman_implementation_t *imp,
-                                pixman_op_t              op,
-                                pixman_image_t *         src_image,
-                                pixman_image_t *         mask_image,
-                                pixman_image_t *         dst_image,
-                                int32_t                  src_x,
-                                int32_t                  src_y,
-                                int32_t                  mask_x,
-                                int32_t                  mask_y,
-                                int32_t                  dest_x,
-                                int32_t                  dest_y,
-                                int32_t                  width,
-                                int32_t                  height)
-{
-    uint32_t *dst_line, *dst;
-    uint32_t *src_line, *src;
-    uint32_t mask;
-    __m64 vmask;
-    int dst_stride, src_stride;
-    uint16_t w;
-    __m64 srca;
-
-    CHECKPOINT ();
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);
-
-    mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
-    vmask = load8888 (mask);
-    srca = MC (4x00ff);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    __m64 s = load8888 (*src | 0xff000000);
-	    __m64 d = load8888 (*dst);
-
-	    *dst = store8888 (in_over (s, srca, vmask, d));
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-
-	while (w >= 16)
-	{
-	    __m64 vd0 = *(__m64 *)(dst + 0);
-	    __m64 vd1 = *(__m64 *)(dst + 2);
-	    __m64 vd2 = *(__m64 *)(dst + 4);
-	    __m64 vd3 = *(__m64 *)(dst + 6);
-	    __m64 vd4 = *(__m64 *)(dst + 8);
-	    __m64 vd5 = *(__m64 *)(dst + 10);
-	    __m64 vd6 = *(__m64 *)(dst + 12);
-	    __m64 vd7 = *(__m64 *)(dst + 14);
-
-	    __m64 vs0 = *(__m64 *)(src + 0);
-	    __m64 vs1 = *(__m64 *)(src + 2);
-	    __m64 vs2 = *(__m64 *)(src + 4);
-	    __m64 vs3 = *(__m64 *)(src + 6);
-	    __m64 vs4 = *(__m64 *)(src + 8);
-	    __m64 vs5 = *(__m64 *)(src + 10);
-	    __m64 vs6 = *(__m64 *)(src + 12);
-	    __m64 vs7 = *(__m64 *)(src + 14);
-
-	    vd0 = pack8888 (
-	        in_over (expandx888 (vs0, 0), srca, vmask, expand8888 (vd0, 0)),
-	        in_over (expandx888 (vs0, 1), srca, vmask, expand8888 (vd0, 1)));
-
-	    vd1 = pack8888 (
-	        in_over (expandx888 (vs1, 0), srca, vmask, expand8888 (vd1, 0)),
-	        in_over (expandx888 (vs1, 1), srca, vmask, expand8888 (vd1, 1)));
-
-	    vd2 = pack8888 (
-	        in_over (expandx888 (vs2, 0), srca, vmask, expand8888 (vd2, 0)),
-	        in_over (expandx888 (vs2, 1), srca, vmask, expand8888 (vd2, 1)));
-
-	    vd3 = pack8888 (
-	        in_over (expandx888 (vs3, 0), srca, vmask, expand8888 (vd3, 0)),
-	        in_over (expandx888 (vs3, 1), srca, vmask, expand8888 (vd3, 1)));
-
-	    vd4 = pack8888 (
-	        in_over (expandx888 (vs4, 0), srca, vmask, expand8888 (vd4, 0)),
-	        in_over (expandx888 (vs4, 1), srca, vmask, expand8888 (vd4, 1)));
-
-	    vd5 = pack8888 (
-	        in_over (expandx888 (vs5, 0), srca, vmask, expand8888 (vd5, 0)),
-	        in_over (expandx888 (vs5, 1), srca, vmask, expand8888 (vd5, 1)));
-
-	    vd6 = pack8888 (
-	        in_over (expandx888 (vs6, 0), srca, vmask, expand8888 (vd6, 0)),
-	        in_over (expandx888 (vs6, 1), srca, vmask, expand8888 (vd6, 1)));
-
-	    vd7 = pack8888 (
-	        in_over (expandx888 (vs7, 0), srca, vmask, expand8888 (vd7, 0)),
-	        in_over (expandx888 (vs7, 1), srca, vmask, expand8888 (vd7, 1)));
-
-	    *(__m64 *)(dst + 0) = vd0;
-	    *(__m64 *)(dst + 2) = vd1;
-	    *(__m64 *)(dst + 4) = vd2;
-	    *(__m64 *)(dst + 6) = vd3;
-	    *(__m64 *)(dst + 8) = vd4;
-	    *(__m64 *)(dst + 10) = vd5;
-	    *(__m64 *)(dst + 12) = vd6;
-	    *(__m64 *)(dst + 14) = vd7;
-
-	    w -= 16;
-	    dst += 16;
-	    src += 16;
-	}
-
-	while (w)
-	{
-	    __m64 s = load8888 (*src | 0xff000000);
-	    __m64 d = load8888 (*dst);
-
-	    *dst = store8888 (in_over (s, srca, vmask, d));
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_8888_8888 (pixman_implementation_t *imp,
-                              pixman_op_t              op,
-                              pixman_image_t *         src_image,
-                              pixman_image_t *         mask_image,
-                              pixman_image_t *         dst_image,
-                              int32_t                  src_x,
-                              int32_t                  src_y,
-                              int32_t                  mask_x,
-                              int32_t                  mask_y,
-                              int32_t                  dest_x,
-                              int32_t                  dest_y,
-                              int32_t                  width,
-                              int32_t                  height)
-{
-    uint32_t *dst_line, *dst;
-    uint32_t *src_line, *src;
-    uint32_t s;
-    int dst_stride, src_stride;
-    uint8_t a;
-    uint16_t w;
-
-    CHECKPOINT ();
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	while (w--)
-	{
-	    s = *src++;
-	    a = s >> 24;
-	    
-	    if (a == 0xff)
-	    {
-		*dst = s;
-	    }
-	    else if (s)
-	    {
-		__m64 ms, sa;
-		ms = load8888 (s);
-		sa = expand_alpha (ms);
-		*dst = store8888 (over (ms, sa, load8888 (*dst)));
-	    }
-	    
-	    dst++;
-	}
-    }
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_8888_0565 (pixman_implementation_t *imp,
-                              pixman_op_t              op,
-                              pixman_image_t *         src_image,
-                              pixman_image_t *         mask_image,
-                              pixman_image_t *         dst_image,
-                              int32_t                  src_x,
-                              int32_t                  src_y,
-                              int32_t                  mask_x,
-                              int32_t                  mask_y,
-                              int32_t                  dest_x,
-                              int32_t                  dest_y,
-                              int32_t                  width,
-                              int32_t                  height)
-{
-    uint16_t    *dst_line, *dst;
-    uint32_t    *src_line, *src;
-    int dst_stride, src_stride;
-    uint16_t w;
-
-    CHECKPOINT ();
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-#if 0
-    /* FIXME */
-    assert (src_image->drawable == mask_image->drawable);
-#endif
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	CHECKPOINT ();
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    __m64 vsrc = load8888 (*src);
-	    uint64_t d = *dst;
-	    __m64 vdest = expand565 (M64 (d), 0);
-
-	    vdest = pack_565 (
-		over (vsrc, expand_alpha (vsrc), vdest), vdest, 0);
-
-	    *dst = UINT64 (vdest) & 0xffff;
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-
-	CHECKPOINT ();
-
-	while (w >= 4)
-	{
-	    __m64 vsrc0, vsrc1, vsrc2, vsrc3;
-	    __m64 vdest;
-
-	    vsrc0 = load8888 (*(src + 0));
-	    vsrc1 = load8888 (*(src + 1));
-	    vsrc2 = load8888 (*(src + 2));
-	    vsrc3 = load8888 (*(src + 3));
-
-	    vdest = *(__m64 *)dst;
-
-	    vdest = pack_565 (over (vsrc0, expand_alpha (vsrc0), expand565 (vdest, 0)), vdest, 0);
-	    vdest = pack_565 (over (vsrc1, expand_alpha (vsrc1), expand565 (vdest, 1)), vdest, 1);
-	    vdest = pack_565 (over (vsrc2, expand_alpha (vsrc2), expand565 (vdest, 2)), vdest, 2);
-	    vdest = pack_565 (over (vsrc3, expand_alpha (vsrc3), expand565 (vdest, 3)), vdest, 3);
-
-	    *(__m64 *)dst = vdest;
-
-	    w -= 4;
-	    dst += 4;
-	    src += 4;
-	}
-
-	CHECKPOINT ();
-
-	while (w)
-	{
-	    __m64 vsrc = load8888 (*src);
-	    uint64_t d = *dst;
-	    __m64 vdest = expand565 (M64 (d), 0);
-
-	    vdest = pack_565 (over (vsrc, expand_alpha (vsrc), vdest), vdest, 0);
-
-	    *dst = UINT64 (vdest) & 0xffff;
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_n_8_8888 (pixman_implementation_t *imp,
-                             pixman_op_t              op,
-                             pixman_image_t *         src_image,
-                             pixman_image_t *         mask_image,
-                             pixman_image_t *         dst_image,
-                             int32_t                  src_x,
-                             int32_t                  src_y,
-                             int32_t                  mask_x,
-                             int32_t                  mask_y,
-                             int32_t                  dest_x,
-                             int32_t                  dest_y,
-                             int32_t                  width,
-                             int32_t                  height)
-{
-    uint32_t src, srca;
-    uint32_t *dst_line, *dst;
-    uint8_t *mask_line, *mask;
-    int dst_stride, mask_stride;
-    uint16_t w;
-    __m64 vsrc, vsrca;
-    uint64_t srcsrc;
-
-    CHECKPOINT ();
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    srca = src >> 24;
-    if (src == 0)
-	return;
-
-    srcsrc = (uint64_t)src << 32 | src;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	mask = mask_line;
-	mask_line += mask_stride;
-	w = width;
-
-	CHECKPOINT ();
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    uint64_t m = *mask;
-
-	    if (m)
-	    {
-		__m64 vdest = in_over (vsrc, vsrca,
-				       expand_alpha_rev (M64 (m)),
-				       load8888 (*dst));
-		
-		*dst = store8888 (vdest);
-	    }
-
-	    w--;
-	    mask++;
-	    dst++;
-	}
-
-	CHECKPOINT ();
-
-	while (w >= 2)
-	{
-	    uint64_t m0, m1;
-	    
-	    m0 = *mask;
-	    m1 = *(mask + 1);
-
-	    if (srca == 0xff && (m0 & m1) == 0xff)
-	    {
-		*(uint64_t *)dst = srcsrc;
-	    }
-	    else if (m0 | m1)
-	    {
-		__m64 vdest;
-		__m64 dest0, dest1;
-
-		vdest = *(__m64 *)dst;
-
-		dest0 = in_over (vsrc, vsrca, expand_alpha_rev (M64 (m0)),
-				 expand8888 (vdest, 0));
-		dest1 = in_over (vsrc, vsrca, expand_alpha_rev (M64 (m1)),
-				 expand8888 (vdest, 1));
-
-		*(__m64 *)dst = pack8888 (dest0, dest1);
-	    }
-
-	    mask += 2;
-	    dst += 2;
-	    w -= 2;
-	}
-
-	CHECKPOINT ();
-
-	while (w)
-	{
-	    uint64_t m = *mask;
-
-	    if (m)
-	    {
-		__m64 vdest = load8888 (*dst);
-
-		vdest = in_over (
-		    vsrc, vsrca, expand_alpha_rev (M64 (m)), vdest);
-		*dst = store8888 (vdest);
-	    }
-
-	    w--;
-	    mask++;
-	    dst++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-pixman_bool_t
-pixman_fill_mmx (uint32_t *bits,
-                 int       stride,
-                 int       bpp,
-                 int       x,
-                 int       y,
-                 int       width,
-                 int       height,
-                 uint32_t xor)
-{
-    uint64_t fill;
-    __m64 vfill;
-    uint32_t byte_width;
-    uint8_t     *byte_line;
-
-#ifdef __GNUC__
-    __m64 v1, v2, v3, v4, v5, v6, v7;
-#endif
-
-    if (bpp != 16 && bpp != 32 && bpp != 8)
-	return FALSE;
-
-    if (bpp == 16 && (xor >> 16 != (xor & 0xffff)))
-	return FALSE;
-
-    if (bpp == 8 &&
-        ((xor >> 16 != (xor & 0xffff)) ||
-         (xor >> 24 != (xor & 0x00ff) >> 16)))
-    {
-	return FALSE;
-    }
-
-    if (bpp == 8)
-    {
-	stride = stride * (int) sizeof (uint32_t) / 1;
-	byte_line = (uint8_t *)(((uint8_t *)bits) + stride * y + x);
-	byte_width = width;
-	stride *= 1;
-    }
-    else if (bpp == 16)
-    {
-	stride = stride * (int) sizeof (uint32_t) / 2;
-	byte_line = (uint8_t *)(((uint16_t *)bits) + stride * y + x);
-	byte_width = 2 * width;
-	stride *= 2;
-    }
-    else
-    {
-	stride = stride * (int) sizeof (uint32_t) / 4;
-	byte_line = (uint8_t *)(((uint32_t *)bits) + stride * y + x);
-	byte_width = 4 * width;
-	stride *= 4;
-    }
-
-    fill = ((uint64_t)xor << 32) | xor;
-    vfill = M64 (fill);
-
-#ifdef __GNUC__
-    __asm__ (
-        "movq		%7,	%0\n"
-        "movq		%7,	%1\n"
-        "movq		%7,	%2\n"
-        "movq		%7,	%3\n"
-        "movq		%7,	%4\n"
-        "movq		%7,	%5\n"
-        "movq		%7,	%6\n"
-	: "=y" (v1), "=y" (v2), "=y" (v3),
-        "=y" (v4), "=y" (v5), "=y" (v6), "=y" (v7)
-	: "y" (vfill));
-#endif
-
-    while (height--)
-    {
-	int w;
-	uint8_t *d = byte_line;
-	
-	byte_line += stride;
-	w = byte_width;
-
-	while (w >= 1 && ((unsigned long)d & 1))
-	{
-	    *(uint8_t *)d = (xor & 0xff);
-	    w--;
-	    d++;
-	}
-
-	while (w >= 2 && ((unsigned long)d & 3))
-	{
-	    *(uint16_t *)d = (xor & 0xffff);
-	    w -= 2;
-	    d += 2;
-	}
-
-	while (w >= 4 && ((unsigned long)d & 7))
-	{
-	    *(uint32_t *)d = xor;
-
-	    w -= 4;
-	    d += 4;
-	}
-
-	while (w >= 64)
-	{
-#ifdef __GNUC__
-	    __asm__ (
-	        "movq	%1,	  (%0)\n"
-	        "movq	%2,	 8(%0)\n"
-	        "movq	%3,	16(%0)\n"
-	        "movq	%4,	24(%0)\n"
-	        "movq	%5,	32(%0)\n"
-	        "movq	%6,	40(%0)\n"
-	        "movq	%7,	48(%0)\n"
-	        "movq	%8,	56(%0)\n"
-		:
-		: "r" (d),
-	        "y" (vfill), "y" (v1), "y" (v2), "y" (v3),
-	        "y" (v4), "y" (v5), "y" (v6), "y" (v7)
-		: "memory");
-#else
-	    *(__m64*) (d +  0) = vfill;
-	    *(__m64*) (d +  8) = vfill;
-	    *(__m64*) (d + 16) = vfill;
-	    *(__m64*) (d + 24) = vfill;
-	    *(__m64*) (d + 32) = vfill;
-	    *(__m64*) (d + 40) = vfill;
-	    *(__m64*) (d + 48) = vfill;
-	    *(__m64*) (d + 56) = vfill;
-#endif
-	    w -= 64;
-	    d += 64;
-	}
-
-	while (w >= 4)
-	{
-	    *(uint32_t *)d = xor;
-
-	    w -= 4;
-	    d += 4;
-	}
-	while (w >= 2)
-	{
-	    *(uint16_t *)d = (xor & 0xffff);
-	    w -= 2;
-	    d += 2;
-	}
-	while (w >= 1)
-	{
-	    *(uint8_t *)d = (xor & 0xff);
-	    w--;
-	    d++;
-	}
-
-    }
-
-    _mm_empty ();
-    return TRUE;
-}
-
-static void
-mmx_composite_src_n_8_8888 (pixman_implementation_t *imp,
-                            pixman_op_t              op,
-                            pixman_image_t *         src_image,
-                            pixman_image_t *         mask_image,
-                            pixman_image_t *         dst_image,
-                            int32_t                  src_x,
-                            int32_t                  src_y,
-                            int32_t                  mask_x,
-                            int32_t                  mask_y,
-                            int32_t                  dest_x,
-                            int32_t                  dest_y,
-                            int32_t                  width,
-                            int32_t                  height)
-{
-    uint32_t src, srca;
-    uint32_t    *dst_line, *dst;
-    uint8_t     *mask_line, *mask;
-    int dst_stride, mask_stride;
-    uint16_t w;
-    __m64 vsrc, vsrca;
-    uint64_t srcsrc;
-
-    CHECKPOINT ();
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    srca = src >> 24;
-    if (src == 0)
-    {
-	pixman_fill_mmx (dst_image->bits.bits, dst_image->bits.rowstride,
-			 PIXMAN_FORMAT_BPP (dst_image->bits.format),
-	                 dest_x, dest_y, width, height, 0);
-	return;
-    }
-
-    srcsrc = (uint64_t)src << 32 | src;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	mask = mask_line;
-	mask_line += mask_stride;
-	w = width;
-
-	CHECKPOINT ();
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    uint64_t m = *mask;
-
-	    if (m)
-	    {
-		__m64 vdest = in (vsrc, expand_alpha_rev (M64 (m)));
-		
-		*dst = store8888 (vdest);
-	    }
-	    else
-	    {
-		*dst = 0;
-	    }
-
-	    w--;
-	    mask++;
-	    dst++;
-	}
-
-	CHECKPOINT ();
-
-	while (w >= 2)
-	{
-	    uint64_t m0, m1;
-	    m0 = *mask;
-	    m1 = *(mask + 1);
-
-	    if (srca == 0xff && (m0 & m1) == 0xff)
-	    {
-		*(uint64_t *)dst = srcsrc;
-	    }
-	    else if (m0 | m1)
-	    {
-		__m64 vdest;
-		__m64 dest0, dest1;
-
-		vdest = *(__m64 *)dst;
-
-		dest0 = in (vsrc, expand_alpha_rev (M64 (m0)));
-		dest1 = in (vsrc, expand_alpha_rev (M64 (m1)));
-
-		*(__m64 *)dst = pack8888 (dest0, dest1);
-	    }
-	    else
-	    {
-		*(uint64_t *)dst = 0;
-	    }
-
-	    mask += 2;
-	    dst += 2;
-	    w -= 2;
-	}
-
-	CHECKPOINT ();
-
-	while (w)
-	{
-	    uint64_t m = *mask;
-
-	    if (m)
-	    {
-		__m64 vdest = load8888 (*dst);
-		
-		vdest = in (vsrc, expand_alpha_rev (M64 (m)));
-		*dst = store8888 (vdest);
-	    }
-	    else
-	    {
-		*dst = 0;
-	    }
-
-	    w--;
-	    mask++;
-	    dst++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_n_8_0565 (pixman_implementation_t *imp,
-                             pixman_op_t              op,
-                             pixman_image_t *         src_image,
-                             pixman_image_t *         mask_image,
-                             pixman_image_t *         dst_image,
-                             int32_t                  src_x,
-                             int32_t                  src_y,
-                             int32_t                  mask_x,
-                             int32_t                  mask_y,
-                             int32_t                  dest_x,
-                             int32_t                  dest_y,
-                             int32_t                  width,
-                             int32_t                  height)
-{
-    uint32_t src, srca;
-    uint16_t *dst_line, *dst;
-    uint8_t *mask_line, *mask;
-    int dst_stride, mask_stride;
-    uint16_t w;
-    __m64 vsrc, vsrca, tmp;
-    uint64_t srcsrcsrcsrc, src16;
-
-    CHECKPOINT ();
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    srca = src >> 24;
-    if (src == 0)
-	return;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    tmp = pack_565 (vsrc, _mm_setzero_si64 (), 0);
-    src16 = UINT64 (tmp);
-
-    srcsrcsrcsrc =
-	(uint64_t)src16 << 48 | (uint64_t)src16 << 32 |
-	(uint64_t)src16 << 16 | (uint64_t)src16;
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	mask = mask_line;
-	mask_line += mask_stride;
-	w = width;
-
-	CHECKPOINT ();
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    uint64_t m = *mask;
-
-	    if (m)
-	    {
-		uint64_t d = *dst;
-		__m64 vd = M64 (d);
-		__m64 vdest = in_over (
-		    vsrc, vsrca, expand_alpha_rev (M64 (m)), expand565 (vd, 0));
-		
-		vd = pack_565 (vdest, _mm_setzero_si64 (), 0);
-		*dst = UINT64 (vd);
-	    }
-
-	    w--;
-	    mask++;
-	    dst++;
-	}
-
-	CHECKPOINT ();
-
-	while (w >= 4)
-	{
-	    uint64_t m0, m1, m2, m3;
-	    m0 = *mask;
-	    m1 = *(mask + 1);
-	    m2 = *(mask + 2);
-	    m3 = *(mask + 3);
-
-	    if (srca == 0xff && (m0 & m1 & m2 & m3) == 0xff)
-	    {
-		*(uint64_t *)dst = srcsrcsrcsrc;
-	    }
-	    else if (m0 | m1 | m2 | m3)
-	    {
-		__m64 vdest;
-		__m64 vm0, vm1, vm2, vm3;
-
-		vdest = *(__m64 *)dst;
-
-		vm0 = M64 (m0);
-		vdest = pack_565 (in_over (vsrc, vsrca, expand_alpha_rev (vm0),
-					   expand565 (vdest, 0)), vdest, 0);
-		vm1 = M64 (m1);
-		vdest = pack_565 (in_over (vsrc, vsrca, expand_alpha_rev (vm1),
-					   expand565 (vdest, 1)), vdest, 1);
-		vm2 = M64 (m2);
-		vdest = pack_565 (in_over (vsrc, vsrca, expand_alpha_rev (vm2),
-					   expand565 (vdest, 2)), vdest, 2);
-		vm3 = M64 (m3);
-		vdest = pack_565 (in_over (vsrc, vsrca, expand_alpha_rev (vm3),
-					   expand565 (vdest, 3)), vdest, 3);
-
-		*(__m64 *)dst = vdest;
-	    }
-
-	    w -= 4;
-	    mask += 4;
-	    dst += 4;
-	}
-
-	CHECKPOINT ();
-
-	while (w)
-	{
-	    uint64_t m = *mask;
-
-	    if (m)
-	    {
-		uint64_t d = *dst;
-		__m64 vd = M64 (d);
-		__m64 vdest = in_over (vsrc, vsrca, expand_alpha_rev (M64 (m)),
-				       expand565 (vd, 0));
-		vd = pack_565 (vdest, _mm_setzero_si64 (), 0);
-		*dst = UINT64 (vd);
-	    }
-
-	    w--;
-	    mask++;
-	    dst++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_pixbuf_0565 (pixman_implementation_t *imp,
-                                pixman_op_t              op,
-                                pixman_image_t *         src_image,
-                                pixman_image_t *         mask_image,
-                                pixman_image_t *         dst_image,
-                                int32_t                  src_x,
-                                int32_t                  src_y,
-                                int32_t                  mask_x,
-                                int32_t                  mask_y,
-                                int32_t                  dest_x,
-                                int32_t                  dest_y,
-                                int32_t                  width,
-                                int32_t                  height)
-{
-    uint16_t    *dst_line, *dst;
-    uint32_t    *src_line, *src;
-    int dst_stride, src_stride;
-    uint16_t w;
-
-    CHECKPOINT ();
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-#if 0
-    /* FIXME */
-    assert (src_image->drawable == mask_image->drawable);
-#endif
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	CHECKPOINT ();
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    __m64 vsrc = load8888 (*src);
-	    uint64_t d = *dst;
-	    __m64 vdest = expand565 (M64 (d), 0);
-
-	    vdest = pack_565 (over_rev_non_pre (vsrc, vdest), vdest, 0);
-
-	    *dst = UINT64 (vdest);
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-
-	CHECKPOINT ();
-
-	while (w >= 4)
-	{
-	    uint32_t s0, s1, s2, s3;
-	    unsigned char a0, a1, a2, a3;
-
-	    s0 = *src;
-	    s1 = *(src + 1);
-	    s2 = *(src + 2);
-	    s3 = *(src + 3);
-
-	    a0 = (s0 >> 24);
-	    a1 = (s1 >> 24);
-	    a2 = (s2 >> 24);
-	    a3 = (s3 >> 24);
-
-	    if ((a0 & a1 & a2 & a3) == 0xFF)
-	    {
-		__m64 vdest;
-		vdest = pack_565 (invert_colors (load8888 (s0)), _mm_setzero_si64 (), 0);
-		vdest = pack_565 (invert_colors (load8888 (s1)), vdest, 1);
-		vdest = pack_565 (invert_colors (load8888 (s2)), vdest, 2);
-		vdest = pack_565 (invert_colors (load8888 (s3)), vdest, 3);
-
-		*(__m64 *)dst = vdest;
-	    }
-	    else if (s0 | s1 | s2 | s3)
-	    {
-		__m64 vdest = *(__m64 *)dst;
-
-		vdest = pack_565 (over_rev_non_pre (load8888 (s0), expand565 (vdest, 0)), vdest, 0);
-		vdest = pack_565 (over_rev_non_pre (load8888 (s1), expand565 (vdest, 1)), vdest, 1);
-		vdest = pack_565 (over_rev_non_pre (load8888 (s2), expand565 (vdest, 2)), vdest, 2);
-		vdest = pack_565 (over_rev_non_pre (load8888 (s3), expand565 (vdest, 3)), vdest, 3);
-
-		*(__m64 *)dst = vdest;
-	    }
-
-	    w -= 4;
-	    dst += 4;
-	    src += 4;
-	}
-
-	CHECKPOINT ();
-
-	while (w)
-	{
-	    __m64 vsrc = load8888 (*src);
-	    uint64_t d = *dst;
-	    __m64 vdest = expand565 (M64 (d), 0);
-
-	    vdest = pack_565 (over_rev_non_pre (vsrc, vdest), vdest, 0);
-
-	    *dst = UINT64 (vdest);
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_pixbuf_8888 (pixman_implementation_t *imp,
-                                pixman_op_t              op,
-                                pixman_image_t *         src_image,
-                                pixman_image_t *         mask_image,
-                                pixman_image_t *         dst_image,
-                                int32_t                  src_x,
-                                int32_t                  src_y,
-                                int32_t                  mask_x,
-                                int32_t                  mask_y,
-                                int32_t                  dest_x,
-                                int32_t                  dest_y,
-                                int32_t                  width,
-                                int32_t                  height)
-{
-    uint32_t    *dst_line, *dst;
-    uint32_t    *src_line, *src;
-    int dst_stride, src_stride;
-    uint16_t w;
-
-    CHECKPOINT ();
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-#if 0
-    /* FIXME */
-    assert (src_image->drawable == mask_image->drawable);
-#endif
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    __m64 s = load8888 (*src);
-	    __m64 d = load8888 (*dst);
-
-	    *dst = store8888 (over_rev_non_pre (s, d));
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-
-	while (w >= 2)
-	{
-	    uint64_t s0, s1;
-	    unsigned char a0, a1;
-	    __m64 d0, d1;
-
-	    s0 = *src;
-	    s1 = *(src + 1);
-
-	    a0 = (s0 >> 24);
-	    a1 = (s1 >> 24);
-
-	    if ((a0 & a1) == 0xFF)
-	    {
-		d0 = invert_colors (load8888 (s0));
-		d1 = invert_colors (load8888 (s1));
-
-		*(__m64 *)dst = pack8888 (d0, d1);
-	    }
-	    else if (s0 | s1)
-	    {
-		__m64 vdest = *(__m64 *)dst;
-
-		d0 = over_rev_non_pre (load8888 (s0), expand8888 (vdest, 0));
-		d1 = over_rev_non_pre (load8888 (s1), expand8888 (vdest, 1));
-
-		*(__m64 *)dst = pack8888 (d0, d1);
-	    }
-
-	    w -= 2;
-	    dst += 2;
-	    src += 2;
-	}
-
-	while (w)
-	{
-	    __m64 s = load8888 (*src);
-	    __m64 d = load8888 (*dst);
-
-	    *dst = store8888 (over_rev_non_pre (s, d));
-
-	    w--;
-	    dst++;
-	    src++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_over_n_8888_0565_ca (pixman_implementation_t *imp,
-                                   pixman_op_t              op,
-                                   pixman_image_t *         src_image,
-                                   pixman_image_t *         mask_image,
-                                   pixman_image_t *         dst_image,
-                                   int32_t                  src_x,
-                                   int32_t                  src_y,
-                                   int32_t                  mask_x,
-                                   int32_t                  mask_y,
-                                   int32_t                  dest_x,
-                                   int32_t                  dest_y,
-                                   int32_t                  width,
-                                   int32_t                  height)
-{
-    uint32_t src, srca;
-    uint16_t    *dst_line;
-    uint32_t    *mask_line;
-    int dst_stride, mask_stride;
-    __m64 vsrc, vsrca;
-
-    CHECKPOINT ();
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    srca = src >> 24;
-    if (src == 0)
-	return;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint32_t, mask_stride, mask_line, 1);
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    while (height--)
-    {
-	int twidth = width;
-	uint32_t *p = (uint32_t *)mask_line;
-	uint16_t *q = (uint16_t *)dst_line;
-
-	while (twidth && ((unsigned long)q & 7))
-	{
-	    uint32_t m = *(uint32_t *)p;
-
-	    if (m)
-	    {
-		uint64_t d = *q;
-		__m64 vdest = expand565 (M64 (d), 0);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
-		*q = UINT64 (vdest);
-	    }
-
-	    twidth--;
-	    p++;
-	    q++;
-	}
-
-	while (twidth >= 4)
-	{
-	    uint32_t m0, m1, m2, m3;
-
-	    m0 = *p;
-	    m1 = *(p + 1);
-	    m2 = *(p + 2);
-	    m3 = *(p + 3);
-
-	    if ((m0 | m1 | m2 | m3))
-	    {
-		__m64 vdest = *(__m64 *)q;
-
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m0), expand565 (vdest, 0)), vdest, 0);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m1), expand565 (vdest, 1)), vdest, 1);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m2), expand565 (vdest, 2)), vdest, 2);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m3), expand565 (vdest, 3)), vdest, 3);
-
-		*(__m64 *)q = vdest;
-	    }
-	    twidth -= 4;
-	    p += 4;
-	    q += 4;
-	}
-
-	while (twidth)
-	{
-	    uint32_t m;
-
-	    m = *(uint32_t *)p;
-	    if (m)
-	    {
-		uint64_t d = *q;
-		__m64 vdest = expand565 (M64 (d), 0);
-		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
-		*q = UINT64 (vdest);
-	    }
-
-	    twidth--;
-	    p++;
-	    q++;
-	}
-
-	mask_line += mask_stride;
-	dst_line += dst_stride;
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_in_n_8_8 (pixman_implementation_t *imp,
-                        pixman_op_t              op,
-                        pixman_image_t *         src_image,
-                        pixman_image_t *         mask_image,
-                        pixman_image_t *         dst_image,
-                        int32_t                  src_x,
-                        int32_t                  src_y,
-                        int32_t                  mask_x,
-                        int32_t                  mask_y,
-                        int32_t                  dest_x,
-                        int32_t                  dest_y,
-                        int32_t                  width,
-                        int32_t                  height)
-{
-    uint8_t *dst_line, *dst;
-    uint8_t *mask_line, *mask;
-    int dst_stride, mask_stride;
-    uint16_t w;
-    uint32_t src;
-    uint8_t sa;
-    __m64 vsrc, vsrca;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    sa = src >> 24;
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	mask = mask_line;
-	mask_line += mask_stride;
-	w = width;
-
-	if ((((unsigned long)dst_image & 3) == 0) &&
-	    (((unsigned long)src_image & 3) == 0))
-	{
-	    while (w >= 4)
-	    {
-		uint32_t m;
-		__m64 vmask;
-		__m64 vdest;
-
-		m = 0;
-
-		vmask = load8888 (*(uint32_t *)mask);
-		vdest = load8888 (*(uint32_t *)dst);
-
-		*(uint32_t *)dst = store8888 (in (in (vsrca, vmask), vdest));
-
-		dst += 4;
-		mask += 4;
-		w -= 4;
-	    }
-	}
-
-	while (w--)
-	{
-	    uint16_t tmp;
-	    uint8_t a;
-	    uint32_t m, d;
-
-	    a = *mask++;
-	    d = *dst;
-
-	    m = MUL_UN8 (sa, a, tmp);
-	    d = MUL_UN8 (m, d, tmp);
-
-	    *dst++ = d;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_in_8_8 (pixman_implementation_t *imp,
-                      pixman_op_t              op,
-                      pixman_image_t *         src_image,
-                      pixman_image_t *         mask_image,
-                      pixman_image_t *         dst_image,
-                      int32_t                  src_x,
-                      int32_t                  src_y,
-                      int32_t                  mask_x,
-                      int32_t                  mask_y,
-                      int32_t                  dest_x,
-                      int32_t                  dest_y,
-                      int32_t                  width,
-                      int32_t                  height)
-{
-    uint8_t     *dst_line, *dst;
-    uint8_t     *src_line, *src;
-    int src_stride, dst_stride;
-    uint16_t w;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint8_t, src_stride, src_line, 1);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	if ((((unsigned long)dst_image & 3) == 0) &&
-	    (((unsigned long)src_image & 3) == 0))
-	{
-	    while (w >= 4)
-	    {
-		uint32_t *s = (uint32_t *)src;
-		uint32_t *d = (uint32_t *)dst;
-
-		*d = store8888 (in (load8888 (*s), load8888 (*d)));
-
-		w -= 4;
-		dst += 4;
-		src += 4;
-	    }
-	}
-
-	while (w--)
-	{
-	    uint8_t s, d;
-	    uint16_t tmp;
-
-	    s = *src;
-	    d = *dst;
-
-	    *dst = MUL_UN8 (s, d, tmp);
-
-	    src++;
-	    dst++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_add_n_8_8 (pixman_implementation_t *imp,
-			 pixman_op_t              op,
-			 pixman_image_t *         src_image,
-			 pixman_image_t *         mask_image,
-			 pixman_image_t *         dst_image,
-			 int32_t                  src_x,
-			 int32_t                  src_y,
-			 int32_t                  mask_x,
-			 int32_t                  mask_y,
-			 int32_t                  dest_x,
-			 int32_t                  dest_y,
-			 int32_t                  width,
-			 int32_t                  height)
-{
-    uint8_t     *dst_line, *dst;
-    uint8_t     *mask_line, *mask;
-    int dst_stride, mask_stride;
-    uint16_t w;
-    uint32_t src;
-    uint8_t sa;
-    __m64 vsrc, vsrca;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-
-    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
-
-    sa = src >> 24;
-
-    if (src == 0)
-	return;
-
-    vsrc = load8888 (src);
-    vsrca = expand_alpha (vsrc);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	mask = mask_line;
-	mask_line += mask_stride;
-	w = width;
-
-	if ((((unsigned long)mask_image & 3) == 0) &&
-	    (((unsigned long)dst_image  & 3) == 0))
-	{
-	    while (w >= 4)
-	    {
-		__m64 vmask = load8888 (*(uint32_t *)mask);
-		__m64 vdest = load8888 (*(uint32_t *)dst);
-
-		*(uint32_t *)dst = store8888 (_mm_adds_pu8 (in (vsrca, vmask), vdest));
-
-		w -= 4;
-		dst += 4;
-		mask += 4;
-	    }
-	}
-
-	while (w--)
-	{
-	    uint16_t tmp;
-	    uint16_t a;
-	    uint32_t m, d;
-	    uint32_t r;
-
-	    a = *mask++;
-	    d = *dst;
-
-	    m = MUL_UN8 (sa, a, tmp);
-	    r = ADD_UN8 (m, d, tmp);
-
-	    *dst++ = r;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_add_8000_8000 (pixman_implementation_t *imp,
-                             pixman_op_t              op,
-                             pixman_image_t *         src_image,
-                             pixman_image_t *         mask_image,
-                             pixman_image_t *         dst_image,
-                             int32_t                  src_x,
-                             int32_t                  src_y,
-                             int32_t                  mask_x,
-                             int32_t                  mask_y,
-                             int32_t                  dest_x,
-                             int32_t                  dest_y,
-                             int32_t                  width,
-                             int32_t                  height)
-{
-    uint8_t *dst_line, *dst;
-    uint8_t *src_line, *src;
-    int dst_stride, src_stride;
-    uint16_t w;
-    uint8_t s, d;
-    uint16_t t;
-
-    CHECKPOINT ();
-
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint8_t, src_stride, src_line, 1);
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    s = *src;
-	    d = *dst;
-	    t = d + s;
-	    s = t | (0 - (t >> 8));
-	    *dst = s;
-
-	    dst++;
-	    src++;
-	    w--;
-	}
-
-	while (w >= 8)
-	{
-	    *(__m64*)dst = _mm_adds_pu8 (*(__m64*)src, *(__m64*)dst);
-	    dst += 8;
-	    src += 8;
-	    w -= 8;
-	}
-
-	while (w)
-	{
-	    s = *src;
-	    d = *dst;
-	    t = d + s;
-	    s = t | (0 - (t >> 8));
-	    *dst = s;
-
-	    dst++;
-	    src++;
-	    w--;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static void
-mmx_composite_add_8888_8888 (pixman_implementation_t *imp,
-                             pixman_op_t              op,
-                             pixman_image_t *         src_image,
-                             pixman_image_t *         mask_image,
-                             pixman_image_t *         dst_image,
-                             int32_t                  src_x,
-                             int32_t                  src_y,
-                             int32_t                  mask_x,
-                             int32_t                  mask_y,
-                             int32_t                  dest_x,
-                             int32_t                  dest_y,
-                             int32_t                  width,
-                             int32_t                  height)
-{
-    __m64 dst64;
-    uint32_t    *dst_line, *dst;
-    uint32_t    *src_line, *src;
-    int dst_stride, src_stride;
-    uint16_t w;
-
-    CHECKPOINT ();
-
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-
-    while (height--)
-    {
-	dst = dst_line;
-	dst_line += dst_stride;
-	src = src_line;
-	src_line += src_stride;
-	w = width;
-
-	while (w && (unsigned long)dst & 7)
-	{
-	    *dst = _mm_cvtsi64_si32 (_mm_adds_pu8 (_mm_cvtsi32_si64 (*src),
-	                                           _mm_cvtsi32_si64 (*dst)));
-	    dst++;
-	    src++;
-	    w--;
-	}
-
-	while (w >= 2)
-	{
-	    dst64 = _mm_adds_pu8 (*(__m64*)src, *(__m64*)dst);
-	    *(uint64_t*)dst = UINT64 (dst64);
-	    dst += 2;
-	    src += 2;
-	    w -= 2;
-	}
-
-	if (w)
-	{
-	    *dst = _mm_cvtsi64_si32 (_mm_adds_pu8 (_mm_cvtsi32_si64 (*src),
-	                                           _mm_cvtsi32_si64 (*dst)));
-
-	}
-    }
-
-    _mm_empty ();
-}
-
-static pixman_bool_t
-pixman_blt_mmx (uint32_t *src_bits,
-                uint32_t *dst_bits,
-                int       src_stride,
-                int       dst_stride,
-                int       src_bpp,
-                int       dst_bpp,
-                int       src_x,
-                int       src_y,
-                int       dst_x,
-                int       dst_y,
-                int       width,
-                int       height)
-{
-    uint8_t *   src_bytes;
-    uint8_t *   dst_bytes;
-    int byte_width;
-
-    if (src_bpp != dst_bpp)
-	return FALSE;
-
-    if (src_bpp == 16)
-    {
-	src_stride = src_stride * (int) sizeof (uint32_t) / 2;
-	dst_stride = dst_stride * (int) sizeof (uint32_t) / 2;
-	src_bytes = (uint8_t *)(((uint16_t *)src_bits) + src_stride * (src_y) + (src_x));
-	dst_bytes = (uint8_t *)(((uint16_t *)dst_bits) + dst_stride * (dst_y) + (dst_x));
-	byte_width = 2 * width;
-	src_stride *= 2;
-	dst_stride *= 2;
-    }
-    else if (src_bpp == 32)
-    {
-	src_stride = src_stride * (int) sizeof (uint32_t) / 4;
-	dst_stride = dst_stride * (int) sizeof (uint32_t) / 4;
-	src_bytes = (uint8_t *)(((uint32_t *)src_bits) + src_stride * (src_y) + (src_x));
-	dst_bytes = (uint8_t *)(((uint32_t *)dst_bits) + dst_stride * (dst_y) + (dst_x));
-	byte_width = 4 * width;
-	src_stride *= 4;
-	dst_stride *= 4;
-    }
-    else
-    {
-	return FALSE;
-    }
-
-    while (height--)
-    {
-	int w;
-	uint8_t *s = src_bytes;
-	uint8_t *d = dst_bytes;
-	src_bytes += src_stride;
-	dst_bytes += dst_stride;
-	w = byte_width;
-
-	while (w >= 2 && ((unsigned long)d & 3))
-	{
-	    *(uint16_t *)d = *(uint16_t *)s;
-	    w -= 2;
-	    s += 2;
-	    d += 2;
-	}
-
-	while (w >= 4 && ((unsigned long)d & 7))
-	{
-	    *(uint32_t *)d = *(uint32_t *)s;
-
-	    w -= 4;
-	    s += 4;
-	    d += 4;
-	}
-
-	while (w >= 64)
-	{
-#if defined (__GNUC__) || (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x590))
-	    __asm__ (
-	        "movq	  (%1),	  %%mm0\n"
-	        "movq	 8(%1),	  %%mm1\n"
-	        "movq	16(%1),	  %%mm2\n"
-	        "movq	24(%1),	  %%mm3\n"
-	        "movq	32(%1),	  %%mm4\n"
-	        "movq	40(%1),	  %%mm5\n"
-	        "movq	48(%1),	  %%mm6\n"
-	        "movq	56(%1),	  %%mm7\n"
-
-	        "movq	%%mm0,	  (%0)\n"
-	        "movq	%%mm1,	 8(%0)\n"
-	        "movq	%%mm2,	16(%0)\n"
-	        "movq	%%mm3,	24(%0)\n"
-	        "movq	%%mm4,	32(%0)\n"
-	        "movq	%%mm5,	40(%0)\n"
-	        "movq	%%mm6,	48(%0)\n"
-	        "movq	%%mm7,	56(%0)\n"
-		:
-		: "r" (d), "r" (s)
-		: "memory",
-	        "%mm0", "%mm1", "%mm2", "%mm3",
-	        "%mm4", "%mm5", "%mm6", "%mm7");
-#else
-	    __m64 v0 = *(__m64 *)(s + 0);
-	    __m64 v1 = *(__m64 *)(s + 8);
-	    __m64 v2 = *(__m64 *)(s + 16);
-	    __m64 v3 = *(__m64 *)(s + 24);
-	    __m64 v4 = *(__m64 *)(s + 32);
-	    __m64 v5 = *(__m64 *)(s + 40);
-	    __m64 v6 = *(__m64 *)(s + 48);
-	    __m64 v7 = *(__m64 *)(s + 56);
-	    *(__m64 *)(d + 0)  = v0;
-	    *(__m64 *)(d + 8)  = v1;
-	    *(__m64 *)(d + 16) = v2;
-	    *(__m64 *)(d + 24) = v3;
-	    *(__m64 *)(d + 32) = v4;
-	    *(__m64 *)(d + 40) = v5;
-	    *(__m64 *)(d + 48) = v6;
-	    *(__m64 *)(d + 56) = v7;
-#endif
-
-	    w -= 64;
-	    s += 64;
-	    d += 64;
-	}
-	while (w >= 4)
-	{
-	    *(uint32_t *)d = *(uint32_t *)s;
-
-	    w -= 4;
-	    s += 4;
-	    d += 4;
-	}
-	if (w >= 2)
-	{
-	    *(uint16_t *)d = *(uint16_t *)s;
-	    w -= 2;
-	    s += 2;
-	    d += 2;
-	}
-    }
-
-    _mm_empty ();
-
-    return TRUE;
-}
-
-static void
-mmx_composite_copy_area (pixman_implementation_t *imp,
-                         pixman_op_t              op,
-                         pixman_image_t *         src_image,
-                         pixman_image_t *         mask_image,
-                         pixman_image_t *         dst_image,
-                         int32_t                  src_x,
-                         int32_t                  src_y,
-                         int32_t                  mask_x,
-                         int32_t                  mask_y,
-                         int32_t                  dest_x,
-                         int32_t                  dest_y,
-                         int32_t                  width,
-                         int32_t                  height)
-{
-    pixman_blt_mmx (src_image->bits.bits,
-                    dst_image->bits.bits,
-                    src_image->bits.rowstride,
-                    dst_image->bits.rowstride,
-                    PIXMAN_FORMAT_BPP (src_image->bits.format),
-                    PIXMAN_FORMAT_BPP (dst_image->bits.format),
-                    src_x, src_y, dest_x, dest_y, width, height);
-}
-
-static void
-mmx_composite_over_x888_8_8888 (pixman_implementation_t *imp,
-                                pixman_op_t              op,
-                                pixman_image_t *         src_image,
-                                pixman_image_t *         mask_image,
-                                pixman_image_t *         dst_image,
-                                int32_t                  src_x,
-                                int32_t                  src_y,
-                                int32_t                  mask_x,
-                                int32_t                  mask_y,
-                                int32_t                  dest_x,
-                                int32_t                  dest_y,
-                                int32_t                  width,
-                                int32_t                  height)
-{
-    uint32_t    *src, *src_line;
-    uint32_t    *dst, *dst_line;
-    uint8_t     *mask, *mask_line;
-    int src_stride, mask_stride, dst_stride;
-    uint16_t w;
-
-    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
-    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
-    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
-
-    while (height--)
-    {
-	src = src_line;
-	src_line += src_stride;
-	dst = dst_line;
-	dst_line += dst_stride;
-	mask = mask_line;
-	mask_line += mask_stride;
-
-	w = width;
-
-	while (w--)
-	{
-	    uint64_t m = *mask;
-
-	    if (m)
-	    {
-		__m64 s = load8888 (*src | 0xff000000);
-
-		if (m == 0xff)
-		{
-		    *dst = store8888 (s);
-		}
-		else
-		{
-		    __m64 sa = expand_alpha (s);
-		    __m64 vm = expand_alpha_rev (M64 (m));
-		    __m64 vdest = in_over (s, sa, vm, load8888 (*dst));
-
-		    *dst = store8888 (vdest);
-		}
-	    }
-
-	    mask++;
-	    dst++;
-	    src++;
-	}
-    }
-
-    _mm_empty ();
-}
-
-static const pixman_fast_path_t mmx_fast_paths[] =
-{
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r5g6b5,   mmx_composite_over_n_8_0565,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b5g6r5,   mmx_composite_over_n_8_0565,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_over_n_8_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, mmx_composite_over_n_8_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, mmx_composite_over_n_8_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, mmx_composite_over_n_8_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_a8r8g8b8, mmx_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_x8r8g8b8, mmx_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_r5g6b5,   mmx_composite_over_n_8888_0565_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_a8b8g8r8, mmx_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_x8b8g8r8, mmx_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_b5g6r5,   mmx_composite_over_n_8888_0565_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8r8g8b8, PIXMAN_a8r8g8b8, mmx_composite_over_pixbuf_8888, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8b8g8r8, PIXMAN_a8r8g8b8, mmx_composite_over_pixbuf_8888, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8r8g8b8, PIXMAN_x8r8g8b8, mmx_composite_over_pixbuf_8888, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8b8g8r8, PIXMAN_x8r8g8b8, mmx_composite_over_pixbuf_8888, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8r8g8b8, PIXMAN_r5g6b5,   mmx_composite_over_pixbuf_0565, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8b8g8r8, PIXMAN_r5g6b5,   mmx_composite_over_pixbuf_0565, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8r8g8b8, PIXMAN_a8b8g8r8, mmx_composite_over_pixbuf_8888, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8b8g8r8, PIXMAN_a8b8g8r8, mmx_composite_over_pixbuf_8888, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8r8g8b8, PIXMAN_x8b8g8r8, mmx_composite_over_pixbuf_8888, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8b8g8r8, PIXMAN_x8b8g8r8, mmx_composite_over_pixbuf_8888, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8r8g8b8, PIXMAN_b5g6r5,   mmx_composite_over_pixbuf_0565, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8b8g8r8, PIXMAN_b5g6r5,   mmx_composite_over_pixbuf_0565, NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_over_x888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, mmx_composite_over_x888_n_8888,           NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, mmx_composite_over_x888_n_8888,           NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, mmx_composite_over_x888_n_8888,           NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_over_8888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, mmx_composite_over_8888_n_8888,           NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, mmx_composite_over_8888_n_8888,           NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, mmx_composite_over_8888_n_8888,           NEED_SOLID_MASK },
-#if 0
-    /* FIXME: This code is commented out since it's apparently not actually faster than the generic code. */
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, mmx_composite_over_x888_8_8888,   0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_over_x888_8_8888,   0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8r8g8, PIXMAN_a8,       PIXMAN_x8b8g8r8, mmx_composite_over_x888_8_8888,   0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8r8g8, PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_over_x888_8_8888,   0 },
-#endif
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_a8r8g8b8, mmx_composite_over_n_8888,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_over_n_8888,        0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_r5g6b5,   mmx_composite_over_n_0565,        0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_copy_area,          0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, mmx_composite_copy_area,          0 },
-
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, mmx_composite_over_8888_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_over_8888_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   mmx_composite_over_8888_0565,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, mmx_composite_over_8888_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, mmx_composite_over_8888_8888,     0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   mmx_composite_over_8888_0565,     0 },
-
-    { PIXMAN_OP_ADD, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_a8r8g8b8, mmx_composite_add_8888_8888,   0 },
-    { PIXMAN_OP_ADD, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_a8b8g8r8, mmx_composite_add_8888_8888,   0 },
-    { PIXMAN_OP_ADD, PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       mmx_composite_add_8000_8000,   0 },
-    { PIXMAN_OP_ADD, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       mmx_composite_add_n_8_8,    0 },
-
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_src_n_8_8888, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_x8r8g8b8, mmx_composite_src_n_8_8888, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8b8g8r8, mmx_composite_src_n_8_8888, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_x8b8g8r8, mmx_composite_src_n_8_8888, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_a8r8g8b8, mmx_composite_copy_area, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_a8b8g8r8, mmx_composite_copy_area, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_copy_area, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, mmx_composite_copy_area, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_x8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_copy_area, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_x8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, mmx_composite_copy_area, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_r5g6b5,    PIXMAN_null,     PIXMAN_r5g6b5,   mmx_composite_copy_area, 0 },
-    { PIXMAN_OP_SRC, PIXMAN_b5g6r5,    PIXMAN_null,     PIXMAN_b5g6r5,   mmx_composite_copy_area, 0 },
-
-    { PIXMAN_OP_IN,  PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       mmx_composite_in_8_8,   0 },
-    { PIXMAN_OP_IN,  PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       mmx_composite_in_n_8_8, 0 },
-
-    { PIXMAN_OP_NONE },
-};
-
-static void
-mmx_composite (pixman_implementation_t *imp,
-               pixman_op_t              op,
-               pixman_image_t *         src,
-               pixman_image_t *         mask,
-               pixman_image_t *         dest,
-               int32_t                  src_x,
-               int32_t                  src_y,
-               int32_t                  mask_x,
-               int32_t                  mask_y,
-               int32_t                  dest_x,
-               int32_t                  dest_y,
-               int32_t                  width,
-               int32_t                  height)
-{
-    if (_pixman_run_fast_path (mmx_fast_paths, imp,
-                               op, src, mask, dest,
-                               src_x, src_y,
-                               mask_x, mask_y,
-                               dest_x, dest_y,
-                               width, height))
-    {
-	return;
-    }
-
-    _pixman_implementation_composite (imp->delegate,
-                                      op, src, mask, dest, src_x, src_y,
-                                      mask_x, mask_y, dest_x, dest_y,
-                                      width, height);
-}
-
-static pixman_bool_t
-mmx_blt (pixman_implementation_t *imp,
-         uint32_t *               src_bits,
-         uint32_t *               dst_bits,
-         int                      src_stride,
-         int                      dst_stride,
-         int                      src_bpp,
-         int                      dst_bpp,
-         int                      src_x,
-         int                      src_y,
-         int                      dst_x,
-         int                      dst_y,
-         int                      width,
-         int                      height)
-{
-    if (!pixman_blt_mmx (
-            src_bits, dst_bits, src_stride, dst_stride, src_bpp, dst_bpp,
-            src_x, src_y, dst_x, dst_y, width, height))
-
-    {
-	return _pixman_implementation_blt (
-	           imp->delegate,
-	           src_bits, dst_bits, src_stride, dst_stride, src_bpp, dst_bpp,
-	           src_x, src_y, dst_x, dst_y, width, height);
-    }
-
-    return TRUE;
-}
-
-static pixman_bool_t
-mmx_fill (pixman_implementation_t *imp,
-          uint32_t *               bits,
-          int                      stride,
-          int                      bpp,
-          int                      x,
-          int                      y,
-          int                      width,
-          int                      height,
-          uint32_t xor)
-{
-    if (!pixman_fill_mmx (bits, stride, bpp, x, y, width, height, xor))
-    {
-	return _pixman_implementation_fill (
-	           imp->delegate, bits, stride, bpp, x, y, width, height, xor);
-    }
-
-    return TRUE;
-}
-
-pixman_implementation_t *
-_pixman_implementation_create_mmx (void)
-{
-    pixman_implementation_t *general = _pixman_implementation_create_fast_path ();
-    pixman_implementation_t *imp = _pixman_implementation_create (general);
-
-    imp->combine_32[PIXMAN_OP_OVER] = mmx_combine_over_u;
-    imp->combine_32[PIXMAN_OP_OVER_REVERSE] = mmx_combine_over_reverse_u;
-    imp->combine_32[PIXMAN_OP_IN] = mmx_combine_in_u;
-    imp->combine_32[PIXMAN_OP_IN_REVERSE] = mmx_combine_in_reverse_u;
-    imp->combine_32[PIXMAN_OP_OUT] = mmx_combine_out_u;
-    imp->combine_32[PIXMAN_OP_OUT_REVERSE] = mmx_combine_out_reverse_u;
-    imp->combine_32[PIXMAN_OP_ATOP] = mmx_combine_atop_u;
-    imp->combine_32[PIXMAN_OP_ATOP_REVERSE] = mmx_combine_atop_reverse_u;
-    imp->combine_32[PIXMAN_OP_XOR] = mmx_combine_xor_u;
-    imp->combine_32[PIXMAN_OP_ADD] = mmx_combine_add_u;
-    imp->combine_32[PIXMAN_OP_SATURATE] = mmx_combine_saturate_u;
-
-    imp->combine_32_ca[PIXMAN_OP_SRC] = mmx_combine_src_ca;
-    imp->combine_32_ca[PIXMAN_OP_OVER] = mmx_combine_over_ca;
-    imp->combine_32_ca[PIXMAN_OP_OVER_REVERSE] = mmx_combine_over_reverse_ca;
-    imp->combine_32_ca[PIXMAN_OP_IN] = mmx_combine_in_ca;
-    imp->combine_32_ca[PIXMAN_OP_IN_REVERSE] = mmx_combine_in_reverse_ca;
-    imp->combine_32_ca[PIXMAN_OP_OUT] = mmx_combine_out_ca;
-    imp->combine_32_ca[PIXMAN_OP_OUT_REVERSE] = mmx_combine_out_reverse_ca;
-    imp->combine_32_ca[PIXMAN_OP_ATOP] = mmx_combine_atop_ca;
-    imp->combine_32_ca[PIXMAN_OP_ATOP_REVERSE] = mmx_combine_atop_reverse_ca;
-    imp->combine_32_ca[PIXMAN_OP_XOR] = mmx_combine_xor_ca;
-    imp->combine_32_ca[PIXMAN_OP_ADD] = mmx_combine_add_ca;
-
-    imp->composite = mmx_composite;
-    imp->blt = mmx_blt;
-    imp->fill = mmx_fill;
-
-    return imp;
-}
-
-#endif /* USE_MMX */
+/*
+ * Copyright © 2004, 2005 Red Hat, Inc.
+ * Copyright © 2004 Nicholas Miell
+ * Copyright © 2005 Trolltech AS
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and its
+ * documentation for any purpose is hereby granted without fee, provided that
+ * the above copyright notice appear in all copies and that both that
+ * copyright notice and this permission notice appear in supporting
+ * documentation, and that the name of Red Hat not be used in advertising or
+ * publicity pertaining to distribution of the software without specific,
+ * written prior permission.  Red Hat makes no representations about the
+ * suitability of this software for any purpose.  It is provided "as is"
+ * without express or implied warranty.
+ *
+ * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
+ * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
+ * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
+ * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
+ * SOFTWARE.
+ *
+ * Author:  Søren Sandmann (sandmann@redhat.com)
+ * Minor Improvements: Nicholas Miell (nmiell@gmail.com)
+ * MMX code paths for fbcompose.c by Lars Knoll (lars@trolltech.com)
+ *
+ * Based on work by Owen Taylor
+ */
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#ifdef USE_MMX
+
+#include <mmintrin.h>
+#include "pixman-private.h"
+#include "pixman-combine32.h"
+
+#define no_vERBOSE
+
+#ifdef VERBOSE
+#define CHECKPOINT() error_f ("at %s %d\n", __FUNCTION__, __LINE__)
+#else
+#define CHECKPOINT()
+#endif
+
+/* Notes about writing mmx code
+ *
+ * give memory operands as the second operand. If you give it as the
+ * first, gcc will first load it into a register, then use that
+ * register
+ *
+ *   ie. use
+ *
+ *         _mm_mullo_pi16 (x, mmx_constant);
+ *
+ *   not
+ *
+ *         _mm_mullo_pi16 (mmx_constant, x);
+ *
+ * Also try to minimize dependencies. i.e. when you need a value, try
+ * to calculate it from a value that was calculated as early as
+ * possible.
+ */
+
+/* --------------- MMX primitives ------------------------------------- */
+
+#ifdef __GNUC__
+typedef uint64_t mmxdatafield;
+#else
+typedef __m64 mmxdatafield;
+/* If __m64 is defined as a struct or union, define M64_MEMBER to be the
+   name of the member used to access the data */
+# ifdef _MSC_VER
+#  define M64_MEMBER m64_u64
+# elif defined(__SUNPRO_C)
+#  define M64_MEMBER l_
+# endif
+#endif
+
+typedef struct
+{
+    mmxdatafield mmx_4x00ff;
+    mmxdatafield mmx_4x0080;
+    mmxdatafield mmx_565_rgb;
+    mmxdatafield mmx_565_unpack_multiplier;
+    mmxdatafield mmx_565_r;
+    mmxdatafield mmx_565_g;
+    mmxdatafield mmx_565_b;
+    mmxdatafield mmx_mask_0;
+    mmxdatafield mmx_mask_1;
+    mmxdatafield mmx_mask_2;
+    mmxdatafield mmx_mask_3;
+    mmxdatafield mmx_full_alpha;
+    mmxdatafield mmx_ffff0000ffff0000;
+    mmxdatafield mmx_0000ffff00000000;
+    mmxdatafield mmx_000000000000ffff;
+} mmx_data_t;
+
+#if defined(_MSC_VER)
+# define MMXDATA_INIT(field, val) { val ## UI64 }
+#elif defined(M64_MEMBER)       /* __m64 is a struct, not an integral type */
+# define MMXDATA_INIT(field, val) field =   { val ## ULL }
+#else                           /* __m64 is an integral type */
+# define MMXDATA_INIT(field, val) field =   val ## ULL
+#endif
+
+static const mmx_data_t c =
+{
+    MMXDATA_INIT (.mmx_4x00ff,                   0x00ff00ff00ff00ff),
+    MMXDATA_INIT (.mmx_4x0080,                   0x0080008000800080),
+    MMXDATA_INIT (.mmx_565_rgb,                  0x000001f0003f001f),
+    MMXDATA_INIT (.mmx_565_unpack_multiplier,    0x0000008404100840),
+    MMXDATA_INIT (.mmx_565_r,                    0x000000f800000000),
+    MMXDATA_INIT (.mmx_565_g,                    0x0000000000fc0000),
+    MMXDATA_INIT (.mmx_565_b,                    0x00000000000000f8),
+    MMXDATA_INIT (.mmx_mask_0,                   0xffffffffffff0000),
+    MMXDATA_INIT (.mmx_mask_1,                   0xffffffff0000ffff),
+    MMXDATA_INIT (.mmx_mask_2,                   0xffff0000ffffffff),
+    MMXDATA_INIT (.mmx_mask_3,                   0x0000ffffffffffff),
+    MMXDATA_INIT (.mmx_full_alpha,               0x00ff000000000000),
+    MMXDATA_INIT (.mmx_ffff0000ffff0000,         0xffff0000ffff0000),
+    MMXDATA_INIT (.mmx_0000ffff00000000,         0x0000ffff00000000),
+    MMXDATA_INIT (.mmx_000000000000ffff,         0x000000000000ffff),
+};
+
+#ifdef __GNUC__
+#    ifdef __ICC
+#        define MC(x)  M64 (c.mmx_ ## x)
+#    else
+#        define MC(x) ((__m64)c.mmx_ ## x)
+#    endif
+#else
+#    define MC(x) c.mmx_ ## x
+#endif
+
+static force_inline __m64
+M64 (uint64_t x)
+{
+#ifdef __ICC
+    return _mm_cvtsi64_m64 (x);
+#elif defined M64_MEMBER        /* __m64 is a struct, not an integral type */
+    __m64 res;
+
+    res.M64_MEMBER = x;
+    return res;
+#else                           /* __m64 is an integral type */
+    return (__m64)x;
+#endif
+}
+
+static force_inline uint64_t
+UINT64 (__m64 x)
+{
+#ifdef __ICC
+    return _mm_cvtm64_si64 (x);
+#elif defined M64_MEMBER        /* __m64 is a struct, not an integral type */
+    uint64_t res = x.M64_MEMBER;
+    return res;
+#else                           /* __m64 is an integral type */
+    return (uint64_t)x;
+#endif
+}
+
+static force_inline __m64
+shift (__m64 v,
+       int   s)
+{
+    if (s > 0)
+	return _mm_slli_si64 (v, s);
+    else if (s < 0)
+	return _mm_srli_si64 (v, -s);
+    else
+	return v;
+}
+
+static force_inline __m64
+negate (__m64 mask)
+{
+    return _mm_xor_si64 (mask, MC (4x00ff));
+}
+
+static force_inline __m64
+pix_multiply (__m64 a, __m64 b)
+{
+    __m64 res;
+
+    res = _mm_mullo_pi16 (a, b);
+    res = _mm_adds_pu16 (res, MC (4x0080));
+    res = _mm_adds_pu16 (res, _mm_srli_pi16 (res, 8));
+    res = _mm_srli_pi16 (res, 8);
+
+    return res;
+}
+
+static force_inline __m64
+pix_add (__m64 a, __m64 b)
+{
+    return _mm_adds_pu8 (a, b);
+}
+
+static force_inline __m64
+expand_alpha (__m64 pixel)
+{
+    __m64 t1, t2;
+
+    t1 = shift (pixel, -48);
+    t2 = shift (t1, 16);
+    t1 = _mm_or_si64 (t1, t2);
+    t2 = shift (t1, 32);
+    t1 = _mm_or_si64 (t1, t2);
+
+    return t1;
+}
+
+static force_inline __m64
+expand_alpha_rev (__m64 pixel)
+{
+    __m64 t1, t2;
+
+    /* move alpha to low 16 bits and zero the rest */
+    t1 = shift (pixel,  48);
+    t1 = shift (t1, -48);
+
+    t2 = shift (t1, 16);
+    t1 = _mm_or_si64 (t1, t2);
+    t2 = shift (t1, 32);
+    t1 = _mm_or_si64 (t1, t2);
+
+    return t1;
+}
+
+static force_inline __m64
+invert_colors (__m64 pixel)
+{
+    __m64 x, y, z;
+
+    x = y = z = pixel;
+
+    x = _mm_and_si64 (x, MC (ffff0000ffff0000));
+    y = _mm_and_si64 (y, MC (000000000000ffff));
+    z = _mm_and_si64 (z, MC (0000ffff00000000));
+
+    y = shift (y, 32);
+    z = shift (z, -32);
+
+    x = _mm_or_si64 (x, y);
+    x = _mm_or_si64 (x, z);
+
+    return x;
+}
+
+static force_inline __m64
+over (__m64 src,
+      __m64 srca,
+      __m64 dest)
+{
+    return _mm_adds_pu8 (src, pix_multiply (dest, negate (srca)));
+}
+
+static force_inline __m64
+over_rev_non_pre (__m64 src, __m64 dest)
+{
+    __m64 srca = expand_alpha (src);
+    __m64 srcfaaa = _mm_or_si64 (srca, MC (full_alpha));
+
+    return over (pix_multiply (invert_colors (src), srcfaaa), srca, dest);
+}
+
+static force_inline __m64
+in (__m64 src, __m64 mask)
+{
+    return pix_multiply (src, mask);
+}
+
+static force_inline __m64
+in_over_full_src_alpha (__m64 src, __m64 mask, __m64 dest)
+{
+    src = _mm_or_si64 (src, MC (full_alpha));
+
+    return over (in (src, mask), mask, dest);
+}
+
+#ifndef _MSC_VER
+static force_inline __m64
+in_over (__m64 src, __m64 srca, __m64 mask, __m64 dest)
+{
+    return over (in (src, mask), pix_multiply (srca, mask), dest);
+}
+
+#else
+
+#define in_over(src, srca, mask, dest)					\
+    over (in (src, mask), pix_multiply (srca, mask), dest)
+
+#endif
+
+static force_inline __m64
+load8888 (uint32_t v)
+{
+    return _mm_unpacklo_pi8 (_mm_cvtsi32_si64 (v), _mm_setzero_si64 ());
+}
+
+static force_inline __m64
+pack8888 (__m64 lo, __m64 hi)
+{
+    return _mm_packs_pu16 (lo, hi);
+}
+
+static force_inline uint32_t
+store8888 (__m64 v)
+{
+    return _mm_cvtsi64_si32 (pack8888 (v, _mm_setzero_si64 ()));
+}
+
+/* Expand 16 bits positioned at @pos (0-3) of a mmx register into
+ *
+ *    00RR00GG00BB
+ *
+ * --- Expanding 565 in the low word ---
+ *
+ * m = (m << (32 - 3)) | (m << (16 - 5)) | m;
+ * m = m & (01f0003f001f);
+ * m = m * (008404100840);
+ * m = m >> 8;
+ *
+ * Note the trick here - the top word is shifted by another nibble to
+ * avoid it bumping into the middle word
+ */
+static force_inline __m64
+expand565 (__m64 pixel, int pos)
+{
+    __m64 p = pixel;
+    __m64 t1, t2;
+
+    /* move pixel to low 16 bit and zero the rest */
+    p = shift (shift (p, (3 - pos) * 16), -48);
+
+    t1 = shift (p, 36 - 11);
+    t2 = shift (p, 16 - 5);
+
+    p = _mm_or_si64 (t1, p);
+    p = _mm_or_si64 (t2, p);
+    p = _mm_and_si64 (p, MC (565_rgb));
+
+    pixel = _mm_mullo_pi16 (p, MC (565_unpack_multiplier));
+    return _mm_srli_pi16 (pixel, 8);
+}
+
+static force_inline __m64
+expand8888 (__m64 in, int pos)
+{
+    if (pos == 0)
+	return _mm_unpacklo_pi8 (in, _mm_setzero_si64 ());
+    else
+	return _mm_unpackhi_pi8 (in, _mm_setzero_si64 ());
+}
+
+static force_inline __m64
+expandx888 (__m64 in, int pos)
+{
+    return _mm_or_si64 (expand8888 (in, pos), MC (full_alpha));
+}
+
+static force_inline __m64
+pack_565 (__m64 pixel, __m64 target, int pos)
+{
+    __m64 p = pixel;
+    __m64 t = target;
+    __m64 r, g, b;
+
+    r = _mm_and_si64 (p, MC (565_r));
+    g = _mm_and_si64 (p, MC (565_g));
+    b = _mm_and_si64 (p, MC (565_b));
+
+    r = shift (r, -(32 - 8) + pos * 16);
+    g = shift (g, -(16 - 3) + pos * 16);
+    b = shift (b, -(0  + 3) + pos * 16);
+
+    if (pos == 0)
+	t = _mm_and_si64 (t, MC (mask_0));
+    else if (pos == 1)
+	t = _mm_and_si64 (t, MC (mask_1));
+    else if (pos == 2)
+	t = _mm_and_si64 (t, MC (mask_2));
+    else if (pos == 3)
+	t = _mm_and_si64 (t, MC (mask_3));
+
+    p = _mm_or_si64 (r, t);
+    p = _mm_or_si64 (g, p);
+
+    return _mm_or_si64 (b, p);
+}
+
+#ifndef _MSC_VER
+
+static force_inline __m64
+pix_add_mul (__m64 x, __m64 a, __m64 y, __m64 b)
+{
+    x = pix_multiply (x, a);
+    y = pix_multiply (y, b);
+
+    return pix_add (x, y);
+}
+
+#else
+
+#define pix_add_mul(x, a, y, b)	 \
+    ( x = pix_multiply (x, a),	 \
+      y = pix_multiply (y, a),	 \
+      pix_add (x, y) )
+
+#endif
+
+/* --------------- MMX code patch for fbcompose.c --------------------- */
+
+static force_inline uint32_t
+combine (const uint32_t *src, const uint32_t *mask)
+{
+    uint32_t ssrc = *src;
+
+    if (mask)
+    {
+	__m64 m = load8888 (*mask);
+	__m64 s = load8888 (ssrc);
+
+	m = expand_alpha (m);
+	s = pix_multiply (s, m);
+
+	ssrc = store8888 (s);
+    }
+
+    return ssrc;
+}
+
+static void
+mmx_combine_over_u (pixman_implementation_t *imp,
+                    pixman_op_t              op,
+                    uint32_t *               dest,
+                    const uint32_t *         src,
+                    const uint32_t *         mask,
+                    int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	uint32_t ssrc = combine (src, mask);
+	uint32_t a = ssrc >> 24;
+
+	if (a == 0xff)
+	{
+	    *dest = ssrc;
+	}
+	else if (ssrc)
+	{
+	    __m64 s, sa;
+	    s = load8888 (ssrc);
+	    sa = expand_alpha (s);
+	    *dest = store8888 (over (s, sa, load8888 (*dest)));
+	}
+
+	++dest;
+	++src;
+	if (mask)
+	    ++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_over_reverse_u (pixman_implementation_t *imp,
+                            pixman_op_t              op,
+                            uint32_t *               dest,
+                            const uint32_t *         src,
+                            const uint32_t *         mask,
+                            int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 d, da;
+	uint32_t s = combine (src, mask);
+	
+	d = load8888 (*dest);
+	da = expand_alpha (d);
+	*dest = store8888 (over (d, da, load8888 (s)));
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_in_u (pixman_implementation_t *imp,
+                  pixman_op_t              op,
+                  uint32_t *               dest,
+                  const uint32_t *         src,
+                  const uint32_t *         mask,
+                  int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 x, a;
+	
+	x = load8888 (combine (src, mask));
+	a = load8888 (*dest);
+	a = expand_alpha (a);
+	x = pix_multiply (x, a);
+	
+	*dest = store8888 (x);
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_in_reverse_u (pixman_implementation_t *imp,
+                          pixman_op_t              op,
+                          uint32_t *               dest,
+                          const uint32_t *         src,
+                          const uint32_t *         mask,
+                          int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 x, a;
+	
+	x = load8888 (*dest);
+	a = load8888 (combine (src, mask));
+	a = expand_alpha (a);
+	x = pix_multiply (x, a);
+	*dest = store8888 (x);
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_out_u (pixman_implementation_t *imp,
+                   pixman_op_t              op,
+                   uint32_t *               dest,
+                   const uint32_t *         src,
+                   const uint32_t *         mask,
+                   int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 x, a;
+	
+	x = load8888 (combine (src, mask));
+	a = load8888 (*dest);
+	a = expand_alpha (a);
+	a = negate (a);
+	x = pix_multiply (x, a);
+	*dest = store8888 (x);
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_out_reverse_u (pixman_implementation_t *imp,
+                           pixman_op_t              op,
+                           uint32_t *               dest,
+                           const uint32_t *         src,
+                           const uint32_t *         mask,
+                           int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 x, a;
+	
+	x = load8888 (*dest);
+	a = load8888 (combine (src, mask));
+	a = expand_alpha (a);
+	a = negate (a);
+	x = pix_multiply (x, a);
+
+	*dest = store8888 (x);
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_atop_u (pixman_implementation_t *imp,
+                    pixman_op_t              op,
+                    uint32_t *               dest,
+                    const uint32_t *         src,
+                    const uint32_t *         mask,
+                    int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 s, da, d, sia;
+	
+	s = load8888 (combine (src, mask));
+	d = load8888 (*dest);
+	sia = expand_alpha (s);
+	sia = negate (sia);
+	da = expand_alpha (d);
+	s = pix_add_mul (s, da, d, sia);
+	*dest = store8888 (s);
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_atop_reverse_u (pixman_implementation_t *imp,
+                            pixman_op_t              op,
+                            uint32_t *               dest,
+                            const uint32_t *         src,
+                            const uint32_t *         mask,
+                            int                      width)
+{
+    const uint32_t *end;
+
+    end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 s, dia, d, sa;
+	
+	s = load8888 (combine (src, mask));
+	d = load8888 (*dest);
+	sa = expand_alpha (s);
+	dia = expand_alpha (d);
+	dia = negate (dia);
+	s = pix_add_mul (s, dia, d, sa);
+	*dest = store8888 (s);
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_xor_u (pixman_implementation_t *imp,
+                   pixman_op_t              op,
+                   uint32_t *               dest,
+                   const uint32_t *         src,
+                   const uint32_t *         mask,
+                   int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 s, dia, d, sia;
+	
+	s = load8888 (combine (src, mask));
+	d = load8888 (*dest);
+	sia = expand_alpha (s);
+	dia = expand_alpha (d);
+	sia = negate (sia);
+	dia = negate (dia);
+	s = pix_add_mul (s, dia, d, sia);
+	*dest = store8888 (s);
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_add_u (pixman_implementation_t *imp,
+                   pixman_op_t              op,
+                   uint32_t *               dest,
+                   const uint32_t *         src,
+                   const uint32_t *         mask,
+                   int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	__m64 s, d;
+	
+	s = load8888 (combine (src, mask));
+	d = load8888 (*dest);
+	s = pix_add (s, d);
+	*dest = store8888 (s);
+
+	++dest;
+	++src;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_saturate_u (pixman_implementation_t *imp,
+                        pixman_op_t              op,
+                        uint32_t *               dest,
+                        const uint32_t *         src,
+                        const uint32_t *         mask,
+                        int                      width)
+{
+    const uint32_t *end = dest + width;
+
+    while (dest < end)
+    {
+	uint32_t s = combine (src, mask);
+	uint32_t d = *dest;
+	__m64 ms = load8888 (s);
+	__m64 md = load8888 (d);
+	uint32_t sa = s >> 24;
+	uint32_t da = ~d >> 24;
+
+	if (sa > da)
+	{
+	    __m64 msa = load8888 (DIV_UN8 (da, sa) << 24);
+	    msa = expand_alpha (msa);
+	    ms = pix_multiply (ms, msa);
+	}
+
+	md = pix_add (md, ms);
+	*dest = store8888 (md);
+
+	++src;
+	++dest;
+	if (mask)
+	    mask++;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_src_ca (pixman_implementation_t *imp,
+                    pixman_op_t              op,
+                    uint32_t *               dest,
+                    const uint32_t *         src,
+                    const uint32_t *         mask,
+                    int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	
+	s = pix_multiply (s, a);
+	*dest = store8888 (s);
+
+	++src;
+	++mask;
+	++dest;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_over_ca (pixman_implementation_t *imp,
+                     pixman_op_t              op,
+                     uint32_t *               dest,
+                     const uint32_t *         src,
+                     const uint32_t *         mask,
+                     int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 sa = expand_alpha (s);
+
+	*dest = store8888 (in_over (s, sa, a, d));
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_over_reverse_ca (pixman_implementation_t *imp,
+                             pixman_op_t              op,
+                             uint32_t *               dest,
+                             const uint32_t *         src,
+                             const uint32_t *         mask,
+                             int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 da = expand_alpha (d);
+
+	*dest = store8888 (over (d, da, in (s, a)));
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_in_ca (pixman_implementation_t *imp,
+                   pixman_op_t              op,
+                   uint32_t *               dest,
+                   const uint32_t *         src,
+                   const uint32_t *         mask,
+                   int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 da = expand_alpha (d);
+	
+	s = pix_multiply (s, a);
+	s = pix_multiply (s, da);
+	*dest = store8888 (s);
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_in_reverse_ca (pixman_implementation_t *imp,
+                           pixman_op_t              op,
+                           uint32_t *               dest,
+                           const uint32_t *         src,
+                           const uint32_t *         mask,
+                           int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 sa = expand_alpha (s);
+	
+	a = pix_multiply (a, sa);
+	d = pix_multiply (d, a);
+	*dest = store8888 (d);
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_out_ca (pixman_implementation_t *imp,
+                    pixman_op_t              op,
+                    uint32_t *               dest,
+                    const uint32_t *         src,
+                    const uint32_t *         mask,
+                    int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 da = expand_alpha (d);
+	
+	da = negate (da);
+	s = pix_multiply (s, a);
+	s = pix_multiply (s, da);
+	*dest = store8888 (s);
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_out_reverse_ca (pixman_implementation_t *imp,
+                            pixman_op_t              op,
+                            uint32_t *               dest,
+                            const uint32_t *         src,
+                            const uint32_t *         mask,
+                            int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 sa = expand_alpha (s);
+
+	a = pix_multiply (a, sa);
+	a = negate (a);
+	d = pix_multiply (d, a);
+	*dest = store8888 (d);
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_atop_ca (pixman_implementation_t *imp,
+                     pixman_op_t              op,
+                     uint32_t *               dest,
+                     const uint32_t *         src,
+                     const uint32_t *         mask,
+                     int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 da = expand_alpha (d);
+	__m64 sa = expand_alpha (s);
+
+	s = pix_multiply (s, a);
+	a = pix_multiply (a, sa);
+	a = negate (a);
+	d = pix_add_mul (d, a, s, da);
+	*dest = store8888 (d);
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_atop_reverse_ca (pixman_implementation_t *imp,
+                             pixman_op_t              op,
+                             uint32_t *               dest,
+                             const uint32_t *         src,
+                             const uint32_t *         mask,
+                             int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 da = expand_alpha (d);
+	__m64 sa = expand_alpha (s);
+
+	s = pix_multiply (s, a);
+	a = pix_multiply (a, sa);
+	da = negate (da);
+	d = pix_add_mul (d, a, s, da);
+	*dest = store8888 (d);
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_xor_ca (pixman_implementation_t *imp,
+                    pixman_op_t              op,
+                    uint32_t *               dest,
+                    const uint32_t *         src,
+                    const uint32_t *         mask,
+                    int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+	__m64 da = expand_alpha (d);
+	__m64 sa = expand_alpha (s);
+
+	s = pix_multiply (s, a);
+	a = pix_multiply (a, sa);
+	da = negate (da);
+	a = negate (a);
+	d = pix_add_mul (d, a, s, da);
+	*dest = store8888 (d);
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_combine_add_ca (pixman_implementation_t *imp,
+                    pixman_op_t              op,
+                    uint32_t *               dest,
+                    const uint32_t *         src,
+                    const uint32_t *         mask,
+                    int                      width)
+{
+    const uint32_t *end = src + width;
+
+    while (src < end)
+    {
+	__m64 a = load8888 (*mask);
+	__m64 s = load8888 (*src);
+	__m64 d = load8888 (*dest);
+
+	s = pix_multiply (s, a);
+	d = pix_add (s, d);
+	*dest = store8888 (d);
+
+	++src;
+	++dest;
+	++mask;
+    }
+    _mm_empty ();
+}
+
+/* ------------- MMX code paths called from fbpict.c -------------------- */
+
+static void
+mmx_composite_over_n_8888 (pixman_implementation_t *imp,
+                           pixman_op_t              op,
+                           pixman_image_t *         src_image,
+                           pixman_image_t *         mask_image,
+                           pixman_image_t *         dst_image,
+                           int32_t                  src_x,
+                           int32_t                  src_y,
+                           int32_t                  mask_x,
+                           int32_t                  mask_y,
+                           int32_t                  dest_x,
+                           int32_t                  dest_y,
+                           int32_t                  width,
+                           int32_t                  height)
+{
+    uint32_t src;
+    uint32_t    *dst_line, *dst;
+    uint16_t w;
+    int dst_stride;
+    __m64 vsrc, vsrca;
+
+    CHECKPOINT ();
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    if (src == 0)
+	return;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	w = width;
+
+	CHECKPOINT ();
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    *dst = store8888 (over (vsrc, vsrca, load8888 (*dst)));
+
+	    w--;
+	    dst++;
+	}
+
+	while (w >= 2)
+	{
+	    __m64 vdest;
+	    __m64 dest0, dest1;
+
+	    vdest = *(__m64 *)dst;
+
+	    dest0 = over (vsrc, vsrca, expand8888 (vdest, 0));
+	    dest1 = over (vsrc, vsrca, expand8888 (vdest, 1));
+
+	    *(__m64 *)dst = pack8888 (dest0, dest1);
+
+	    dst += 2;
+	    w -= 2;
+	}
+
+	CHECKPOINT ();
+
+	while (w)
+	{
+	    *dst = store8888 (over (vsrc, vsrca, load8888 (*dst)));
+
+	    w--;
+	    dst++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_n_0565 (pixman_implementation_t *imp,
+                           pixman_op_t              op,
+                           pixman_image_t *         src_image,
+                           pixman_image_t *         mask_image,
+                           pixman_image_t *         dst_image,
+                           int32_t                  src_x,
+                           int32_t                  src_y,
+                           int32_t                  mask_x,
+                           int32_t                  mask_y,
+                           int32_t                  dest_x,
+                           int32_t                  dest_y,
+                           int32_t                  width,
+                           int32_t                  height)
+{
+    uint32_t src;
+    uint16_t    *dst_line, *dst;
+    uint16_t w;
+    int dst_stride;
+    __m64 vsrc, vsrca;
+
+    CHECKPOINT ();
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    if (src == 0)
+	return;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	w = width;
+
+	CHECKPOINT ();
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    uint64_t d = *dst;
+	    __m64 vdest = expand565 (M64 (d), 0);
+	    
+	    vdest = pack_565 (over (vsrc, vsrca, vdest), vdest, 0);
+	    *dst = UINT64 (vdest);
+
+	    w--;
+	    dst++;
+	}
+
+	while (w >= 4)
+	{
+	    __m64 vdest;
+
+	    vdest = *(__m64 *)dst;
+
+	    vdest = pack_565 (over (vsrc, vsrca, expand565 (vdest, 0)), vdest, 0);
+	    vdest = pack_565 (over (vsrc, vsrca, expand565 (vdest, 1)), vdest, 1);
+	    vdest = pack_565 (over (vsrc, vsrca, expand565 (vdest, 2)), vdest, 2);
+	    vdest = pack_565 (over (vsrc, vsrca, expand565 (vdest, 3)), vdest, 3);
+
+	    *(__m64 *)dst = vdest;
+
+	    dst += 4;
+	    w -= 4;
+	}
+
+	CHECKPOINT ();
+
+	while (w)
+	{
+	    uint64_t d = *dst;
+	    __m64 vdest = expand565 (M64 (d), 0);
+	    
+	    vdest = pack_565 (over (vsrc, vsrca, vdest), vdest, 0);
+	    *dst = UINT64 (vdest);
+
+	    w--;
+	    dst++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_n_8888_8888_ca (pixman_implementation_t *imp,
+                                   pixman_op_t              op,
+                                   pixman_image_t *         src_image,
+                                   pixman_image_t *         mask_image,
+                                   pixman_image_t *         dst_image,
+                                   int32_t                  src_x,
+                                   int32_t                  src_y,
+                                   int32_t                  mask_x,
+                                   int32_t                  mask_y,
+                                   int32_t                  dest_x,
+                                   int32_t                  dest_y,
+                                   int32_t                  width,
+                                   int32_t                  height)
+{
+    uint32_t src, srca;
+    uint32_t    *dst_line;
+    uint32_t    *mask_line;
+    int dst_stride, mask_stride;
+    __m64 vsrc, vsrca;
+
+    CHECKPOINT ();
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    srca = src >> 24;
+    if (src == 0)
+	return;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint32_t, mask_stride, mask_line, 1);
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    while (height--)
+    {
+	int twidth = width;
+	uint32_t *p = (uint32_t *)mask_line;
+	uint32_t *q = (uint32_t *)dst_line;
+
+	while (twidth && (unsigned long)q & 7)
+	{
+	    uint32_t m = *(uint32_t *)p;
+
+	    if (m)
+	    {
+		__m64 vdest = load8888 (*q);
+		vdest = in_over (vsrc, vsrca, load8888 (m), vdest);
+		*q = store8888 (vdest);
+	    }
+
+	    twidth--;
+	    p++;
+	    q++;
+	}
+
+	while (twidth >= 2)
+	{
+	    uint32_t m0, m1;
+	    m0 = *p;
+	    m1 = *(p + 1);
+
+	    if (m0 | m1)
+	    {
+		__m64 dest0, dest1;
+		__m64 vdest = *(__m64 *)q;
+
+		dest0 = in_over (vsrc, vsrca, load8888 (m0),
+		                 expand8888 (vdest, 0));
+		dest1 = in_over (vsrc, vsrca, load8888 (m1),
+		                 expand8888 (vdest, 1));
+
+		*(__m64 *)q = pack8888 (dest0, dest1);
+	    }
+
+	    p += 2;
+	    q += 2;
+	    twidth -= 2;
+	}
+
+	while (twidth)
+	{
+	    uint32_t m = *(uint32_t *)p;
+
+	    if (m)
+	    {
+		__m64 vdest = load8888 (*q);
+		vdest = in_over (vsrc, vsrca, load8888 (m), vdest);
+		*q = store8888 (vdest);
+	    }
+
+	    twidth--;
+	    p++;
+	    q++;
+	}
+
+	dst_line += dst_stride;
+	mask_line += mask_stride;
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_8888_n_8888 (pixman_implementation_t *imp,
+                                pixman_op_t              op,
+                                pixman_image_t *         src_image,
+                                pixman_image_t *         mask_image,
+                                pixman_image_t *         dst_image,
+                                int32_t                  src_x,
+                                int32_t                  src_y,
+                                int32_t                  mask_x,
+                                int32_t                  mask_y,
+                                int32_t                  dest_x,
+                                int32_t                  dest_y,
+                                int32_t                  width,
+                                int32_t                  height)
+{
+    uint32_t    *dst_line, *dst;
+    uint32_t    *src_line, *src;
+    uint32_t mask;
+    __m64 vmask;
+    int dst_stride, src_stride;
+    uint16_t w;
+    __m64 srca;
+
+    CHECKPOINT ();
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+
+    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);
+    mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
+    vmask = load8888 (mask);
+    srca = MC (4x00ff);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    __m64 s = load8888 (*src);
+	    __m64 d = load8888 (*dst);
+
+	    *dst = store8888 (in_over (s, expand_alpha (s), vmask, d));
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+
+	while (w >= 2)
+	{
+	    __m64 vs = *(__m64 *)src;
+	    __m64 vd = *(__m64 *)dst;
+	    __m64 vsrc0 = expand8888 (vs, 0);
+	    __m64 vsrc1 = expand8888 (vs, 1);
+
+	    *(__m64 *)dst = pack8888 (
+	        in_over (vsrc0, expand_alpha (vsrc0), vmask, expand8888 (vd, 0)),
+	        in_over (vsrc1, expand_alpha (vsrc1), vmask, expand8888 (vd, 1)));
+
+	    w -= 2;
+	    dst += 2;
+	    src += 2;
+	}
+
+	while (w)
+	{
+	    __m64 s = load8888 (*src);
+	    __m64 d = load8888 (*dst);
+
+	    *dst = store8888 (in_over (s, expand_alpha (s), vmask, d));
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_x888_n_8888 (pixman_implementation_t *imp,
+                                pixman_op_t              op,
+                                pixman_image_t *         src_image,
+                                pixman_image_t *         mask_image,
+                                pixman_image_t *         dst_image,
+                                int32_t                  src_x,
+                                int32_t                  src_y,
+                                int32_t                  mask_x,
+                                int32_t                  mask_y,
+                                int32_t                  dest_x,
+                                int32_t                  dest_y,
+                                int32_t                  width,
+                                int32_t                  height)
+{
+    uint32_t *dst_line, *dst;
+    uint32_t *src_line, *src;
+    uint32_t mask;
+    __m64 vmask;
+    int dst_stride, src_stride;
+    uint16_t w;
+    __m64 srca;
+
+    CHECKPOINT ();
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);
+
+    mask = mask | mask >> 8 | mask >> 16 | mask >> 24;
+    vmask = load8888 (mask);
+    srca = MC (4x00ff);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    __m64 s = load8888 (*src | 0xff000000);
+	    __m64 d = load8888 (*dst);
+
+	    *dst = store8888 (in_over (s, srca, vmask, d));
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+
+	while (w >= 16)
+	{
+	    __m64 vd0 = *(__m64 *)(dst + 0);
+	    __m64 vd1 = *(__m64 *)(dst + 2);
+	    __m64 vd2 = *(__m64 *)(dst + 4);
+	    __m64 vd3 = *(__m64 *)(dst + 6);
+	    __m64 vd4 = *(__m64 *)(dst + 8);
+	    __m64 vd5 = *(__m64 *)(dst + 10);
+	    __m64 vd6 = *(__m64 *)(dst + 12);
+	    __m64 vd7 = *(__m64 *)(dst + 14);
+
+	    __m64 vs0 = *(__m64 *)(src + 0);
+	    __m64 vs1 = *(__m64 *)(src + 2);
+	    __m64 vs2 = *(__m64 *)(src + 4);
+	    __m64 vs3 = *(__m64 *)(src + 6);
+	    __m64 vs4 = *(__m64 *)(src + 8);
+	    __m64 vs5 = *(__m64 *)(src + 10);
+	    __m64 vs6 = *(__m64 *)(src + 12);
+	    __m64 vs7 = *(__m64 *)(src + 14);
+
+	    vd0 = pack8888 (
+	        in_over (expandx888 (vs0, 0), srca, vmask, expand8888 (vd0, 0)),
+	        in_over (expandx888 (vs0, 1), srca, vmask, expand8888 (vd0, 1)));
+
+	    vd1 = pack8888 (
+	        in_over (expandx888 (vs1, 0), srca, vmask, expand8888 (vd1, 0)),
+	        in_over (expandx888 (vs1, 1), srca, vmask, expand8888 (vd1, 1)));
+
+	    vd2 = pack8888 (
+	        in_over (expandx888 (vs2, 0), srca, vmask, expand8888 (vd2, 0)),
+	        in_over (expandx888 (vs2, 1), srca, vmask, expand8888 (vd2, 1)));
+
+	    vd3 = pack8888 (
+	        in_over (expandx888 (vs3, 0), srca, vmask, expand8888 (vd3, 0)),
+	        in_over (expandx888 (vs3, 1), srca, vmask, expand8888 (vd3, 1)));
+
+	    vd4 = pack8888 (
+	        in_over (expandx888 (vs4, 0), srca, vmask, expand8888 (vd4, 0)),
+	        in_over (expandx888 (vs4, 1), srca, vmask, expand8888 (vd4, 1)));
+
+	    vd5 = pack8888 (
+	        in_over (expandx888 (vs5, 0), srca, vmask, expand8888 (vd5, 0)),
+	        in_over (expandx888 (vs5, 1), srca, vmask, expand8888 (vd5, 1)));
+
+	    vd6 = pack8888 (
+	        in_over (expandx888 (vs6, 0), srca, vmask, expand8888 (vd6, 0)),
+	        in_over (expandx888 (vs6, 1), srca, vmask, expand8888 (vd6, 1)));
+
+	    vd7 = pack8888 (
+	        in_over (expandx888 (vs7, 0), srca, vmask, expand8888 (vd7, 0)),
+	        in_over (expandx888 (vs7, 1), srca, vmask, expand8888 (vd7, 1)));
+
+	    *(__m64 *)(dst + 0) = vd0;
+	    *(__m64 *)(dst + 2) = vd1;
+	    *(__m64 *)(dst + 4) = vd2;
+	    *(__m64 *)(dst + 6) = vd3;
+	    *(__m64 *)(dst + 8) = vd4;
+	    *(__m64 *)(dst + 10) = vd5;
+	    *(__m64 *)(dst + 12) = vd6;
+	    *(__m64 *)(dst + 14) = vd7;
+
+	    w -= 16;
+	    dst += 16;
+	    src += 16;
+	}
+
+	while (w)
+	{
+	    __m64 s = load8888 (*src | 0xff000000);
+	    __m64 d = load8888 (*dst);
+
+	    *dst = store8888 (in_over (s, srca, vmask, d));
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_8888_8888 (pixman_implementation_t *imp,
+                              pixman_op_t              op,
+                              pixman_image_t *         src_image,
+                              pixman_image_t *         mask_image,
+                              pixman_image_t *         dst_image,
+                              int32_t                  src_x,
+                              int32_t                  src_y,
+                              int32_t                  mask_x,
+                              int32_t                  mask_y,
+                              int32_t                  dest_x,
+                              int32_t                  dest_y,
+                              int32_t                  width,
+                              int32_t                  height)
+{
+    uint32_t *dst_line, *dst;
+    uint32_t *src_line, *src;
+    uint32_t s;
+    int dst_stride, src_stride;
+    uint8_t a;
+    uint16_t w;
+
+    CHECKPOINT ();
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	while (w--)
+	{
+	    s = *src++;
+	    a = s >> 24;
+	    
+	    if (a == 0xff)
+	    {
+		*dst = s;
+	    }
+	    else if (s)
+	    {
+		__m64 ms, sa;
+		ms = load8888 (s);
+		sa = expand_alpha (ms);
+		*dst = store8888 (over (ms, sa, load8888 (*dst)));
+	    }
+	    
+	    dst++;
+	}
+    }
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_8888_0565 (pixman_implementation_t *imp,
+                              pixman_op_t              op,
+                              pixman_image_t *         src_image,
+                              pixman_image_t *         mask_image,
+                              pixman_image_t *         dst_image,
+                              int32_t                  src_x,
+                              int32_t                  src_y,
+                              int32_t                  mask_x,
+                              int32_t                  mask_y,
+                              int32_t                  dest_x,
+                              int32_t                  dest_y,
+                              int32_t                  width,
+                              int32_t                  height)
+{
+    uint16_t    *dst_line, *dst;
+    uint32_t    *src_line, *src;
+    int dst_stride, src_stride;
+    uint16_t w;
+
+    CHECKPOINT ();
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+
+#if 0
+    /* FIXME */
+    assert (src_image->drawable == mask_image->drawable);
+#endif
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	CHECKPOINT ();
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    __m64 vsrc = load8888 (*src);
+	    uint64_t d = *dst;
+	    __m64 vdest = expand565 (M64 (d), 0);
+
+	    vdest = pack_565 (
+		over (vsrc, expand_alpha (vsrc), vdest), vdest, 0);
+
+	    *dst = UINT64 (vdest) & 0xffff;
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+
+	CHECKPOINT ();
+
+	while (w >= 4)
+	{
+	    __m64 vsrc0, vsrc1, vsrc2, vsrc3;
+	    __m64 vdest;
+
+	    vsrc0 = load8888 (*(src + 0));
+	    vsrc1 = load8888 (*(src + 1));
+	    vsrc2 = load8888 (*(src + 2));
+	    vsrc3 = load8888 (*(src + 3));
+
+	    vdest = *(__m64 *)dst;
+
+	    vdest = pack_565 (over (vsrc0, expand_alpha (vsrc0), expand565 (vdest, 0)), vdest, 0);
+	    vdest = pack_565 (over (vsrc1, expand_alpha (vsrc1), expand565 (vdest, 1)), vdest, 1);
+	    vdest = pack_565 (over (vsrc2, expand_alpha (vsrc2), expand565 (vdest, 2)), vdest, 2);
+	    vdest = pack_565 (over (vsrc3, expand_alpha (vsrc3), expand565 (vdest, 3)), vdest, 3);
+
+	    *(__m64 *)dst = vdest;
+
+	    w -= 4;
+	    dst += 4;
+	    src += 4;
+	}
+
+	CHECKPOINT ();
+
+	while (w)
+	{
+	    __m64 vsrc = load8888 (*src);
+	    uint64_t d = *dst;
+	    __m64 vdest = expand565 (M64 (d), 0);
+
+	    vdest = pack_565 (over (vsrc, expand_alpha (vsrc), vdest), vdest, 0);
+
+	    *dst = UINT64 (vdest) & 0xffff;
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_n_8_8888 (pixman_implementation_t *imp,
+                             pixman_op_t              op,
+                             pixman_image_t *         src_image,
+                             pixman_image_t *         mask_image,
+                             pixman_image_t *         dst_image,
+                             int32_t                  src_x,
+                             int32_t                  src_y,
+                             int32_t                  mask_x,
+                             int32_t                  mask_y,
+                             int32_t                  dest_x,
+                             int32_t                  dest_y,
+                             int32_t                  width,
+                             int32_t                  height)
+{
+    uint32_t src, srca;
+    uint32_t *dst_line, *dst;
+    uint8_t *mask_line, *mask;
+    int dst_stride, mask_stride;
+    uint16_t w;
+    __m64 vsrc, vsrca;
+    uint64_t srcsrc;
+
+    CHECKPOINT ();
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    srca = src >> 24;
+    if (src == 0)
+	return;
+
+    srcsrc = (uint64_t)src << 32 | src;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	mask = mask_line;
+	mask_line += mask_stride;
+	w = width;
+
+	CHECKPOINT ();
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    uint64_t m = *mask;
+
+	    if (m)
+	    {
+		__m64 vdest = in_over (vsrc, vsrca,
+				       expand_alpha_rev (M64 (m)),
+				       load8888 (*dst));
+		
+		*dst = store8888 (vdest);
+	    }
+
+	    w--;
+	    mask++;
+	    dst++;
+	}
+
+	CHECKPOINT ();
+
+	while (w >= 2)
+	{
+	    uint64_t m0, m1;
+	    
+	    m0 = *mask;
+	    m1 = *(mask + 1);
+
+	    if (srca == 0xff && (m0 & m1) == 0xff)
+	    {
+		*(uint64_t *)dst = srcsrc;
+	    }
+	    else if (m0 | m1)
+	    {
+		__m64 vdest;
+		__m64 dest0, dest1;
+
+		vdest = *(__m64 *)dst;
+
+		dest0 = in_over (vsrc, vsrca, expand_alpha_rev (M64 (m0)),
+				 expand8888 (vdest, 0));
+		dest1 = in_over (vsrc, vsrca, expand_alpha_rev (M64 (m1)),
+				 expand8888 (vdest, 1));
+
+		*(__m64 *)dst = pack8888 (dest0, dest1);
+	    }
+
+	    mask += 2;
+	    dst += 2;
+	    w -= 2;
+	}
+
+	CHECKPOINT ();
+
+	while (w)
+	{
+	    uint64_t m = *mask;
+
+	    if (m)
+	    {
+		__m64 vdest = load8888 (*dst);
+
+		vdest = in_over (
+		    vsrc, vsrca, expand_alpha_rev (M64 (m)), vdest);
+		*dst = store8888 (vdest);
+	    }
+
+	    w--;
+	    mask++;
+	    dst++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+pixman_bool_t
+pixman_fill_mmx (uint32_t *bits,
+                 int       stride,
+                 int       bpp,
+                 int       x,
+                 int       y,
+                 int       width,
+                 int       height,
+                 uint32_t xor)
+{
+    uint64_t fill;
+    __m64 vfill;
+    uint32_t byte_width;
+    uint8_t     *byte_line;
+
+#ifdef __GNUC__
+    __m64 v1, v2, v3, v4, v5, v6, v7;
+#endif
+
+    if (bpp != 16 && bpp != 32 && bpp != 8)
+	return FALSE;
+
+    if (bpp == 16 && (xor >> 16 != (xor & 0xffff)))
+	return FALSE;
+
+    if (bpp == 8 &&
+        ((xor >> 16 != (xor & 0xffff)) ||
+         (xor >> 24 != (xor & 0x00ff) >> 16)))
+    {
+	return FALSE;
+    }
+
+    if (bpp == 8)
+    {
+	stride = stride * (int) sizeof (uint32_t) / 1;
+	byte_line = (uint8_t *)(((uint8_t *)bits) + stride * y + x);
+	byte_width = width;
+	stride *= 1;
+    }
+    else if (bpp == 16)
+    {
+	stride = stride * (int) sizeof (uint32_t) / 2;
+	byte_line = (uint8_t *)(((uint16_t *)bits) + stride * y + x);
+	byte_width = 2 * width;
+	stride *= 2;
+    }
+    else
+    {
+	stride = stride * (int) sizeof (uint32_t) / 4;
+	byte_line = (uint8_t *)(((uint32_t *)bits) + stride * y + x);
+	byte_width = 4 * width;
+	stride *= 4;
+    }
+
+    fill = ((uint64_t)xor << 32) | xor;
+    vfill = M64 (fill);
+
+#ifdef __GNUC__
+    __asm__ (
+        "movq		%7,	%0\n"
+        "movq		%7,	%1\n"
+        "movq		%7,	%2\n"
+        "movq		%7,	%3\n"
+        "movq		%7,	%4\n"
+        "movq		%7,	%5\n"
+        "movq		%7,	%6\n"
+	: "=y" (v1), "=y" (v2), "=y" (v3),
+        "=y" (v4), "=y" (v5), "=y" (v6), "=y" (v7)
+	: "y" (vfill));
+#endif
+
+    while (height--)
+    {
+	int w;
+	uint8_t *d = byte_line;
+	
+	byte_line += stride;
+	w = byte_width;
+
+	while (w >= 1 && ((unsigned long)d & 1))
+	{
+	    *(uint8_t *)d = (xor & 0xff);
+	    w--;
+	    d++;
+	}
+
+	while (w >= 2 && ((unsigned long)d & 3))
+	{
+	    *(uint16_t *)d = (xor & 0xffff);
+	    w -= 2;
+	    d += 2;
+	}
+
+	while (w >= 4 && ((unsigned long)d & 7))
+	{
+	    *(uint32_t *)d = xor;
+
+	    w -= 4;
+	    d += 4;
+	}
+
+	while (w >= 64)
+	{
+#ifdef __GNUC__
+	    __asm__ (
+	        "movq	%1,	  (%0)\n"
+	        "movq	%2,	 8(%0)\n"
+	        "movq	%3,	16(%0)\n"
+	        "movq	%4,	24(%0)\n"
+	        "movq	%5,	32(%0)\n"
+	        "movq	%6,	40(%0)\n"
+	        "movq	%7,	48(%0)\n"
+	        "movq	%8,	56(%0)\n"
+		:
+		: "r" (d),
+	        "y" (vfill), "y" (v1), "y" (v2), "y" (v3),
+	        "y" (v4), "y" (v5), "y" (v6), "y" (v7)
+		: "memory");
+#else
+	    *(__m64*) (d +  0) = vfill;
+	    *(__m64*) (d +  8) = vfill;
+	    *(__m64*) (d + 16) = vfill;
+	    *(__m64*) (d + 24) = vfill;
+	    *(__m64*) (d + 32) = vfill;
+	    *(__m64*) (d + 40) = vfill;
+	    *(__m64*) (d + 48) = vfill;
+	    *(__m64*) (d + 56) = vfill;
+#endif
+	    w -= 64;
+	    d += 64;
+	}
+
+	while (w >= 4)
+	{
+	    *(uint32_t *)d = xor;
+
+	    w -= 4;
+	    d += 4;
+	}
+	while (w >= 2)
+	{
+	    *(uint16_t *)d = (xor & 0xffff);
+	    w -= 2;
+	    d += 2;
+	}
+	while (w >= 1)
+	{
+	    *(uint8_t *)d = (xor & 0xff);
+	    w--;
+	    d++;
+	}
+
+    }
+
+    _mm_empty ();
+    return TRUE;
+}
+
+static void
+mmx_composite_src_n_8_8888 (pixman_implementation_t *imp,
+                            pixman_op_t              op,
+                            pixman_image_t *         src_image,
+                            pixman_image_t *         mask_image,
+                            pixman_image_t *         dst_image,
+                            int32_t                  src_x,
+                            int32_t                  src_y,
+                            int32_t                  mask_x,
+                            int32_t                  mask_y,
+                            int32_t                  dest_x,
+                            int32_t                  dest_y,
+                            int32_t                  width,
+                            int32_t                  height)
+{
+    uint32_t src, srca;
+    uint32_t    *dst_line, *dst;
+    uint8_t     *mask_line, *mask;
+    int dst_stride, mask_stride;
+    uint16_t w;
+    __m64 vsrc, vsrca;
+    uint64_t srcsrc;
+
+    CHECKPOINT ();
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    srca = src >> 24;
+    if (src == 0)
+    {
+	pixman_fill_mmx (dst_image->bits.bits, dst_image->bits.rowstride,
+			 PIXMAN_FORMAT_BPP (dst_image->bits.format),
+	                 dest_x, dest_y, width, height, 0);
+	return;
+    }
+
+    srcsrc = (uint64_t)src << 32 | src;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	mask = mask_line;
+	mask_line += mask_stride;
+	w = width;
+
+	CHECKPOINT ();
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    uint64_t m = *mask;
+
+	    if (m)
+	    {
+		__m64 vdest = in (vsrc, expand_alpha_rev (M64 (m)));
+		
+		*dst = store8888 (vdest);
+	    }
+	    else
+	    {
+		*dst = 0;
+	    }
+
+	    w--;
+	    mask++;
+	    dst++;
+	}
+
+	CHECKPOINT ();
+
+	while (w >= 2)
+	{
+	    uint64_t m0, m1;
+	    m0 = *mask;
+	    m1 = *(mask + 1);
+
+	    if (srca == 0xff && (m0 & m1) == 0xff)
+	    {
+		*(uint64_t *)dst = srcsrc;
+	    }
+	    else if (m0 | m1)
+	    {
+		__m64 vdest;
+		__m64 dest0, dest1;
+
+		vdest = *(__m64 *)dst;
+
+		dest0 = in (vsrc, expand_alpha_rev (M64 (m0)));
+		dest1 = in (vsrc, expand_alpha_rev (M64 (m1)));
+
+		*(__m64 *)dst = pack8888 (dest0, dest1);
+	    }
+	    else
+	    {
+		*(uint64_t *)dst = 0;
+	    }
+
+	    mask += 2;
+	    dst += 2;
+	    w -= 2;
+	}
+
+	CHECKPOINT ();
+
+	while (w)
+	{
+	    uint64_t m = *mask;
+
+	    if (m)
+	    {
+		__m64 vdest = load8888 (*dst);
+		
+		vdest = in (vsrc, expand_alpha_rev (M64 (m)));
+		*dst = store8888 (vdest);
+	    }
+	    else
+	    {
+		*dst = 0;
+	    }
+
+	    w--;
+	    mask++;
+	    dst++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_n_8_0565 (pixman_implementation_t *imp,
+                             pixman_op_t              op,
+                             pixman_image_t *         src_image,
+                             pixman_image_t *         mask_image,
+                             pixman_image_t *         dst_image,
+                             int32_t                  src_x,
+                             int32_t                  src_y,
+                             int32_t                  mask_x,
+                             int32_t                  mask_y,
+                             int32_t                  dest_x,
+                             int32_t                  dest_y,
+                             int32_t                  width,
+                             int32_t                  height)
+{
+    uint32_t src, srca;
+    uint16_t *dst_line, *dst;
+    uint8_t *mask_line, *mask;
+    int dst_stride, mask_stride;
+    uint16_t w;
+    __m64 vsrc, vsrca, tmp;
+    uint64_t srcsrcsrcsrc, src16;
+
+    CHECKPOINT ();
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    srca = src >> 24;
+    if (src == 0)
+	return;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    tmp = pack_565 (vsrc, _mm_setzero_si64 (), 0);
+    src16 = UINT64 (tmp);
+
+    srcsrcsrcsrc =
+	(uint64_t)src16 << 48 | (uint64_t)src16 << 32 |
+	(uint64_t)src16 << 16 | (uint64_t)src16;
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	mask = mask_line;
+	mask_line += mask_stride;
+	w = width;
+
+	CHECKPOINT ();
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    uint64_t m = *mask;
+
+	    if (m)
+	    {
+		uint64_t d = *dst;
+		__m64 vd = M64 (d);
+		__m64 vdest = in_over (
+		    vsrc, vsrca, expand_alpha_rev (M64 (m)), expand565 (vd, 0));
+		
+		vd = pack_565 (vdest, _mm_setzero_si64 (), 0);
+		*dst = UINT64 (vd);
+	    }
+
+	    w--;
+	    mask++;
+	    dst++;
+	}
+
+	CHECKPOINT ();
+
+	while (w >= 4)
+	{
+	    uint64_t m0, m1, m2, m3;
+	    m0 = *mask;
+	    m1 = *(mask + 1);
+	    m2 = *(mask + 2);
+	    m3 = *(mask + 3);
+
+	    if (srca == 0xff && (m0 & m1 & m2 & m3) == 0xff)
+	    {
+		*(uint64_t *)dst = srcsrcsrcsrc;
+	    }
+	    else if (m0 | m1 | m2 | m3)
+	    {
+		__m64 vdest;
+		__m64 vm0, vm1, vm2, vm3;
+
+		vdest = *(__m64 *)dst;
+
+		vm0 = M64 (m0);
+		vdest = pack_565 (in_over (vsrc, vsrca, expand_alpha_rev (vm0),
+					   expand565 (vdest, 0)), vdest, 0);
+		vm1 = M64 (m1);
+		vdest = pack_565 (in_over (vsrc, vsrca, expand_alpha_rev (vm1),
+					   expand565 (vdest, 1)), vdest, 1);
+		vm2 = M64 (m2);
+		vdest = pack_565 (in_over (vsrc, vsrca, expand_alpha_rev (vm2),
+					   expand565 (vdest, 2)), vdest, 2);
+		vm3 = M64 (m3);
+		vdest = pack_565 (in_over (vsrc, vsrca, expand_alpha_rev (vm3),
+					   expand565 (vdest, 3)), vdest, 3);
+
+		*(__m64 *)dst = vdest;
+	    }
+
+	    w -= 4;
+	    mask += 4;
+	    dst += 4;
+	}
+
+	CHECKPOINT ();
+
+	while (w)
+	{
+	    uint64_t m = *mask;
+
+	    if (m)
+	    {
+		uint64_t d = *dst;
+		__m64 vd = M64 (d);
+		__m64 vdest = in_over (vsrc, vsrca, expand_alpha_rev (M64 (m)),
+				       expand565 (vd, 0));
+		vd = pack_565 (vdest, _mm_setzero_si64 (), 0);
+		*dst = UINT64 (vd);
+	    }
+
+	    w--;
+	    mask++;
+	    dst++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_pixbuf_0565 (pixman_implementation_t *imp,
+                                pixman_op_t              op,
+                                pixman_image_t *         src_image,
+                                pixman_image_t *         mask_image,
+                                pixman_image_t *         dst_image,
+                                int32_t                  src_x,
+                                int32_t                  src_y,
+                                int32_t                  mask_x,
+                                int32_t                  mask_y,
+                                int32_t                  dest_x,
+                                int32_t                  dest_y,
+                                int32_t                  width,
+                                int32_t                  height)
+{
+    uint16_t    *dst_line, *dst;
+    uint32_t    *src_line, *src;
+    int dst_stride, src_stride;
+    uint16_t w;
+
+    CHECKPOINT ();
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+
+#if 0
+    /* FIXME */
+    assert (src_image->drawable == mask_image->drawable);
+#endif
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	CHECKPOINT ();
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    __m64 vsrc = load8888 (*src);
+	    uint64_t d = *dst;
+	    __m64 vdest = expand565 (M64 (d), 0);
+
+	    vdest = pack_565 (over_rev_non_pre (vsrc, vdest), vdest, 0);
+
+	    *dst = UINT64 (vdest);
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+
+	CHECKPOINT ();
+
+	while (w >= 4)
+	{
+	    uint32_t s0, s1, s2, s3;
+	    unsigned char a0, a1, a2, a3;
+
+	    s0 = *src;
+	    s1 = *(src + 1);
+	    s2 = *(src + 2);
+	    s3 = *(src + 3);
+
+	    a0 = (s0 >> 24);
+	    a1 = (s1 >> 24);
+	    a2 = (s2 >> 24);
+	    a3 = (s3 >> 24);
+
+	    if ((a0 & a1 & a2 & a3) == 0xFF)
+	    {
+		__m64 vdest;
+		vdest = pack_565 (invert_colors (load8888 (s0)), _mm_setzero_si64 (), 0);
+		vdest = pack_565 (invert_colors (load8888 (s1)), vdest, 1);
+		vdest = pack_565 (invert_colors (load8888 (s2)), vdest, 2);
+		vdest = pack_565 (invert_colors (load8888 (s3)), vdest, 3);
+
+		*(__m64 *)dst = vdest;
+	    }
+	    else if (s0 | s1 | s2 | s3)
+	    {
+		__m64 vdest = *(__m64 *)dst;
+
+		vdest = pack_565 (over_rev_non_pre (load8888 (s0), expand565 (vdest, 0)), vdest, 0);
+		vdest = pack_565 (over_rev_non_pre (load8888 (s1), expand565 (vdest, 1)), vdest, 1);
+		vdest = pack_565 (over_rev_non_pre (load8888 (s2), expand565 (vdest, 2)), vdest, 2);
+		vdest = pack_565 (over_rev_non_pre (load8888 (s3), expand565 (vdest, 3)), vdest, 3);
+
+		*(__m64 *)dst = vdest;
+	    }
+
+	    w -= 4;
+	    dst += 4;
+	    src += 4;
+	}
+
+	CHECKPOINT ();
+
+	while (w)
+	{
+	    __m64 vsrc = load8888 (*src);
+	    uint64_t d = *dst;
+	    __m64 vdest = expand565 (M64 (d), 0);
+
+	    vdest = pack_565 (over_rev_non_pre (vsrc, vdest), vdest, 0);
+
+	    *dst = UINT64 (vdest);
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_pixbuf_8888 (pixman_implementation_t *imp,
+                                pixman_op_t              op,
+                                pixman_image_t *         src_image,
+                                pixman_image_t *         mask_image,
+                                pixman_image_t *         dst_image,
+                                int32_t                  src_x,
+                                int32_t                  src_y,
+                                int32_t                  mask_x,
+                                int32_t                  mask_y,
+                                int32_t                  dest_x,
+                                int32_t                  dest_y,
+                                int32_t                  width,
+                                int32_t                  height)
+{
+    uint32_t    *dst_line, *dst;
+    uint32_t    *src_line, *src;
+    int dst_stride, src_stride;
+    uint16_t w;
+
+    CHECKPOINT ();
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+
+#if 0
+    /* FIXME */
+    assert (src_image->drawable == mask_image->drawable);
+#endif
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    __m64 s = load8888 (*src);
+	    __m64 d = load8888 (*dst);
+
+	    *dst = store8888 (over_rev_non_pre (s, d));
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+
+	while (w >= 2)
+	{
+	    uint64_t s0, s1;
+	    unsigned char a0, a1;
+	    __m64 d0, d1;
+
+	    s0 = *src;
+	    s1 = *(src + 1);
+
+	    a0 = (s0 >> 24);
+	    a1 = (s1 >> 24);
+
+	    if ((a0 & a1) == 0xFF)
+	    {
+		d0 = invert_colors (load8888 (s0));
+		d1 = invert_colors (load8888 (s1));
+
+		*(__m64 *)dst = pack8888 (d0, d1);
+	    }
+	    else if (s0 | s1)
+	    {
+		__m64 vdest = *(__m64 *)dst;
+
+		d0 = over_rev_non_pre (load8888 (s0), expand8888 (vdest, 0));
+		d1 = over_rev_non_pre (load8888 (s1), expand8888 (vdest, 1));
+
+		*(__m64 *)dst = pack8888 (d0, d1);
+	    }
+
+	    w -= 2;
+	    dst += 2;
+	    src += 2;
+	}
+
+	while (w)
+	{
+	    __m64 s = load8888 (*src);
+	    __m64 d = load8888 (*dst);
+
+	    *dst = store8888 (over_rev_non_pre (s, d));
+
+	    w--;
+	    dst++;
+	    src++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_over_n_8888_0565_ca (pixman_implementation_t *imp,
+                                   pixman_op_t              op,
+                                   pixman_image_t *         src_image,
+                                   pixman_image_t *         mask_image,
+                                   pixman_image_t *         dst_image,
+                                   int32_t                  src_x,
+                                   int32_t                  src_y,
+                                   int32_t                  mask_x,
+                                   int32_t                  mask_y,
+                                   int32_t                  dest_x,
+                                   int32_t                  dest_y,
+                                   int32_t                  width,
+                                   int32_t                  height)
+{
+    uint32_t src, srca;
+    uint16_t    *dst_line;
+    uint32_t    *mask_line;
+    int dst_stride, mask_stride;
+    __m64 vsrc, vsrca;
+
+    CHECKPOINT ();
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    srca = src >> 24;
+    if (src == 0)
+	return;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint16_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint32_t, mask_stride, mask_line, 1);
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    while (height--)
+    {
+	int twidth = width;
+	uint32_t *p = (uint32_t *)mask_line;
+	uint16_t *q = (uint16_t *)dst_line;
+
+	while (twidth && ((unsigned long)q & 7))
+	{
+	    uint32_t m = *(uint32_t *)p;
+
+	    if (m)
+	    {
+		uint64_t d = *q;
+		__m64 vdest = expand565 (M64 (d), 0);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
+		*q = UINT64 (vdest);
+	    }
+
+	    twidth--;
+	    p++;
+	    q++;
+	}
+
+	while (twidth >= 4)
+	{
+	    uint32_t m0, m1, m2, m3;
+
+	    m0 = *p;
+	    m1 = *(p + 1);
+	    m2 = *(p + 2);
+	    m3 = *(p + 3);
+
+	    if ((m0 | m1 | m2 | m3))
+	    {
+		__m64 vdest = *(__m64 *)q;
+
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m0), expand565 (vdest, 0)), vdest, 0);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m1), expand565 (vdest, 1)), vdest, 1);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m2), expand565 (vdest, 2)), vdest, 2);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m3), expand565 (vdest, 3)), vdest, 3);
+
+		*(__m64 *)q = vdest;
+	    }
+	    twidth -= 4;
+	    p += 4;
+	    q += 4;
+	}
+
+	while (twidth)
+	{
+	    uint32_t m;
+
+	    m = *(uint32_t *)p;
+	    if (m)
+	    {
+		uint64_t d = *q;
+		__m64 vdest = expand565 (M64 (d), 0);
+		vdest = pack_565 (in_over (vsrc, vsrca, load8888 (m), vdest), vdest, 0);
+		*q = UINT64 (vdest);
+	    }
+
+	    twidth--;
+	    p++;
+	    q++;
+	}
+
+	mask_line += mask_stride;
+	dst_line += dst_stride;
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_in_n_8_8 (pixman_implementation_t *imp,
+                        pixman_op_t              op,
+                        pixman_image_t *         src_image,
+                        pixman_image_t *         mask_image,
+                        pixman_image_t *         dst_image,
+                        int32_t                  src_x,
+                        int32_t                  src_y,
+                        int32_t                  mask_x,
+                        int32_t                  mask_y,
+                        int32_t                  dest_x,
+                        int32_t                  dest_y,
+                        int32_t                  width,
+                        int32_t                  height)
+{
+    uint8_t *dst_line, *dst;
+    uint8_t *mask_line, *mask;
+    int dst_stride, mask_stride;
+    uint16_t w;
+    uint32_t src;
+    uint8_t sa;
+    __m64 vsrc, vsrca;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    sa = src >> 24;
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	mask = mask_line;
+	mask_line += mask_stride;
+	w = width;
+
+	if ((((unsigned long)dst_image & 3) == 0) &&
+	    (((unsigned long)src_image & 3) == 0))
+	{
+	    while (w >= 4)
+	    {
+		uint32_t m;
+		__m64 vmask;
+		__m64 vdest;
+
+		m = 0;
+
+		vmask = load8888 (*(uint32_t *)mask);
+		vdest = load8888 (*(uint32_t *)dst);
+
+		*(uint32_t *)dst = store8888 (in (in (vsrca, vmask), vdest));
+
+		dst += 4;
+		mask += 4;
+		w -= 4;
+	    }
+	}
+
+	while (w--)
+	{
+	    uint16_t tmp;
+	    uint8_t a;
+	    uint32_t m, d;
+
+	    a = *mask++;
+	    d = *dst;
+
+	    m = MUL_UN8 (sa, a, tmp);
+	    d = MUL_UN8 (m, d, tmp);
+
+	    *dst++ = d;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_in_8_8 (pixman_implementation_t *imp,
+                      pixman_op_t              op,
+                      pixman_image_t *         src_image,
+                      pixman_image_t *         mask_image,
+                      pixman_image_t *         dst_image,
+                      int32_t                  src_x,
+                      int32_t                  src_y,
+                      int32_t                  mask_x,
+                      int32_t                  mask_y,
+                      int32_t                  dest_x,
+                      int32_t                  dest_y,
+                      int32_t                  width,
+                      int32_t                  height)
+{
+    uint8_t     *dst_line, *dst;
+    uint8_t     *src_line, *src;
+    int src_stride, dst_stride;
+    uint16_t w;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint8_t, src_stride, src_line, 1);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	if ((((unsigned long)dst_image & 3) == 0) &&
+	    (((unsigned long)src_image & 3) == 0))
+	{
+	    while (w >= 4)
+	    {
+		uint32_t *s = (uint32_t *)src;
+		uint32_t *d = (uint32_t *)dst;
+
+		*d = store8888 (in (load8888 (*s), load8888 (*d)));
+
+		w -= 4;
+		dst += 4;
+		src += 4;
+	    }
+	}
+
+	while (w--)
+	{
+	    uint8_t s, d;
+	    uint16_t tmp;
+
+	    s = *src;
+	    d = *dst;
+
+	    *dst = MUL_UN8 (s, d, tmp);
+
+	    src++;
+	    dst++;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_add_n_8_8 (pixman_implementation_t *imp,
+			 pixman_op_t              op,
+			 pixman_image_t *         src_image,
+			 pixman_image_t *         mask_image,
+			 pixman_image_t *         dst_image,
+			 int32_t                  src_x,
+			 int32_t                  src_y,
+			 int32_t                  mask_x,
+			 int32_t                  mask_y,
+			 int32_t                  dest_x,
+			 int32_t                  dest_y,
+			 int32_t                  width,
+			 int32_t                  height)
+{
+    uint8_t     *dst_line, *dst;
+    uint8_t     *mask_line, *mask;
+    int dst_stride, mask_stride;
+    uint16_t w;
+    uint32_t src;
+    uint8_t sa;
+    __m64 vsrc, vsrca;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
+
+    src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+    sa = src >> 24;
+
+    if (src == 0)
+	return;
+
+    vsrc = load8888 (src);
+    vsrca = expand_alpha (vsrc);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	mask = mask_line;
+	mask_line += mask_stride;
+	w = width;
+
+	if ((((unsigned long)mask_image & 3) == 0) &&
+	    (((unsigned long)dst_image  & 3) == 0))
+	{
+	    while (w >= 4)
+	    {
+		__m64 vmask = load8888 (*(uint32_t *)mask);
+		__m64 vdest = load8888 (*(uint32_t *)dst);
+
+		*(uint32_t *)dst = store8888 (_mm_adds_pu8 (in (vsrca, vmask), vdest));
+
+		w -= 4;
+		dst += 4;
+		mask += 4;
+	    }
+	}
+
+	while (w--)
+	{
+	    uint16_t tmp;
+	    uint16_t a;
+	    uint32_t m, d;
+	    uint32_t r;
+
+	    a = *mask++;
+	    d = *dst;
+
+	    m = MUL_UN8 (sa, a, tmp);
+	    r = ADD_UN8 (m, d, tmp);
+
+	    *dst++ = r;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_add_8000_8000 (pixman_implementation_t *imp,
+                             pixman_op_t              op,
+                             pixman_image_t *         src_image,
+                             pixman_image_t *         mask_image,
+                             pixman_image_t *         dst_image,
+                             int32_t                  src_x,
+                             int32_t                  src_y,
+                             int32_t                  mask_x,
+                             int32_t                  mask_y,
+                             int32_t                  dest_x,
+                             int32_t                  dest_y,
+                             int32_t                  width,
+                             int32_t                  height)
+{
+    uint8_t *dst_line, *dst;
+    uint8_t *src_line, *src;
+    int dst_stride, src_stride;
+    uint16_t w;
+    uint8_t s, d;
+    uint16_t t;
+
+    CHECKPOINT ();
+
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint8_t, src_stride, src_line, 1);
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    s = *src;
+	    d = *dst;
+	    t = d + s;
+	    s = t | (0 - (t >> 8));
+	    *dst = s;
+
+	    dst++;
+	    src++;
+	    w--;
+	}
+
+	while (w >= 8)
+	{
+	    *(__m64*)dst = _mm_adds_pu8 (*(__m64*)src, *(__m64*)dst);
+	    dst += 8;
+	    src += 8;
+	    w -= 8;
+	}
+
+	while (w)
+	{
+	    s = *src;
+	    d = *dst;
+	    t = d + s;
+	    s = t | (0 - (t >> 8));
+	    *dst = s;
+
+	    dst++;
+	    src++;
+	    w--;
+	}
+    }
+
+    _mm_empty ();
+}
+
+static void
+mmx_composite_add_8888_8888 (pixman_implementation_t *imp,
+                             pixman_op_t              op,
+                             pixman_image_t *         src_image,
+                             pixman_image_t *         mask_image,
+                             pixman_image_t *         dst_image,
+                             int32_t                  src_x,
+                             int32_t                  src_y,
+                             int32_t                  mask_x,
+                             int32_t                  mask_y,
+                             int32_t                  dest_x,
+                             int32_t                  dest_y,
+                             int32_t                  width,
+                             int32_t                  height)
+{
+    __m64 dst64;
+    uint32_t    *dst_line, *dst;
+    uint32_t    *src_line, *src;
+    int dst_stride, src_stride;
+    uint16_t w;
+
+    CHECKPOINT ();
+
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+
+    while (height--)
+    {
+	dst = dst_line;
+	dst_line += dst_stride;
+	src = src_line;
+	src_line += src_stride;
+	w = width;
+
+	while (w && (unsigned long)dst & 7)
+	{
+	    *dst = _mm_cvtsi64_si32 (_mm_adds_pu8 (_mm_cvtsi32_si64 (*src),
+	                                           _mm_cvtsi32_si64 (*dst)));
+	    dst++;
+	    src++;
+	    w--;
+	}
+
+	while (w >= 2)
+	{
+	    dst64 = _mm_adds_pu8 (*(__m64*)src, *(__m64*)dst);
+	    *(uint64_t*)dst = UINT64 (dst64);
+	    dst += 2;
+	    src += 2;
+	    w -= 2;
+	}
+
+	if (w)
+	{
+	    *dst = _mm_cvtsi64_si32 (_mm_adds_pu8 (_mm_cvtsi32_si64 (*src),
+	                                           _mm_cvtsi32_si64 (*dst)));
+
+	}
+    }
+
+    _mm_empty ();
+}
+
+static pixman_bool_t
+pixman_blt_mmx (uint32_t *src_bits,
+                uint32_t *dst_bits,
+                int       src_stride,
+                int       dst_stride,
+                int       src_bpp,
+                int       dst_bpp,
+                int       src_x,
+                int       src_y,
+                int       dst_x,
+                int       dst_y,
+                int       width,
+                int       height)
+{
+    uint8_t *   src_bytes;
+    uint8_t *   dst_bytes;
+    int byte_width;
+
+    if (src_bpp != dst_bpp)
+	return FALSE;
+
+    if (src_bpp == 16)
+    {
+	src_stride = src_stride * (int) sizeof (uint32_t) / 2;
+	dst_stride = dst_stride * (int) sizeof (uint32_t) / 2;
+	src_bytes = (uint8_t *)(((uint16_t *)src_bits) + src_stride * (src_y) + (src_x));
+	dst_bytes = (uint8_t *)(((uint16_t *)dst_bits) + dst_stride * (dst_y) + (dst_x));
+	byte_width = 2 * width;
+	src_stride *= 2;
+	dst_stride *= 2;
+    }
+    else if (src_bpp == 32)
+    {
+	src_stride = src_stride * (int) sizeof (uint32_t) / 4;
+	dst_stride = dst_stride * (int) sizeof (uint32_t) / 4;
+	src_bytes = (uint8_t *)(((uint32_t *)src_bits) + src_stride * (src_y) + (src_x));
+	dst_bytes = (uint8_t *)(((uint32_t *)dst_bits) + dst_stride * (dst_y) + (dst_x));
+	byte_width = 4 * width;
+	src_stride *= 4;
+	dst_stride *= 4;
+    }
+    else
+    {
+	return FALSE;
+    }
+
+    while (height--)
+    {
+	int w;
+	uint8_t *s = src_bytes;
+	uint8_t *d = dst_bytes;
+	src_bytes += src_stride;
+	dst_bytes += dst_stride;
+	w = byte_width;
+
+	while (w >= 2 && ((unsigned long)d & 3))
+	{
+	    *(uint16_t *)d = *(uint16_t *)s;
+	    w -= 2;
+	    s += 2;
+	    d += 2;
+	}
+
+	while (w >= 4 && ((unsigned long)d & 7))
+	{
+	    *(uint32_t *)d = *(uint32_t *)s;
+
+	    w -= 4;
+	    s += 4;
+	    d += 4;
+	}
+
+	while (w >= 64)
+	{
+#if defined (__GNUC__) || (defined(__SUNPRO_C) && (__SUNPRO_C >= 0x590))
+	    __asm__ (
+	        "movq	  (%1),	  %%mm0\n"
+	        "movq	 8(%1),	  %%mm1\n"
+	        "movq	16(%1),	  %%mm2\n"
+	        "movq	24(%1),	  %%mm3\n"
+	        "movq	32(%1),	  %%mm4\n"
+	        "movq	40(%1),	  %%mm5\n"
+	        "movq	48(%1),	  %%mm6\n"
+	        "movq	56(%1),	  %%mm7\n"
+
+	        "movq	%%mm0,	  (%0)\n"
+	        "movq	%%mm1,	 8(%0)\n"
+	        "movq	%%mm2,	16(%0)\n"
+	        "movq	%%mm3,	24(%0)\n"
+	        "movq	%%mm4,	32(%0)\n"
+	        "movq	%%mm5,	40(%0)\n"
+	        "movq	%%mm6,	48(%0)\n"
+	        "movq	%%mm7,	56(%0)\n"
+		:
+		: "r" (d), "r" (s)
+		: "memory",
+	        "%mm0", "%mm1", "%mm2", "%mm3",
+	        "%mm4", "%mm5", "%mm6", "%mm7");
+#else
+	    __m64 v0 = *(__m64 *)(s + 0);
+	    __m64 v1 = *(__m64 *)(s + 8);
+	    __m64 v2 = *(__m64 *)(s + 16);
+	    __m64 v3 = *(__m64 *)(s + 24);
+	    __m64 v4 = *(__m64 *)(s + 32);
+	    __m64 v5 = *(__m64 *)(s + 40);
+	    __m64 v6 = *(__m64 *)(s + 48);
+	    __m64 v7 = *(__m64 *)(s + 56);
+	    *(__m64 *)(d + 0)  = v0;
+	    *(__m64 *)(d + 8)  = v1;
+	    *(__m64 *)(d + 16) = v2;
+	    *(__m64 *)(d + 24) = v3;
+	    *(__m64 *)(d + 32) = v4;
+	    *(__m64 *)(d + 40) = v5;
+	    *(__m64 *)(d + 48) = v6;
+	    *(__m64 *)(d + 56) = v7;
+#endif
+
+	    w -= 64;
+	    s += 64;
+	    d += 64;
+	}
+	while (w >= 4)
+	{
+	    *(uint32_t *)d = *(uint32_t *)s;
+
+	    w -= 4;
+	    s += 4;
+	    d += 4;
+	}
+	if (w >= 2)
+	{
+	    *(uint16_t *)d = *(uint16_t *)s;
+	    w -= 2;
+	    s += 2;
+	    d += 2;
+	}
+    }
+
+    _mm_empty ();
+
+    return TRUE;
+}
+
+static void
+mmx_composite_copy_area (pixman_implementation_t *imp,
+                         pixman_op_t              op,
+                         pixman_image_t *         src_image,
+                         pixman_image_t *         mask_image,
+                         pixman_image_t *         dst_image,
+                         int32_t                  src_x,
+                         int32_t                  src_y,
+                         int32_t                  mask_x,
+                         int32_t                  mask_y,
+                         int32_t                  dest_x,
+                         int32_t                  dest_y,
+                         int32_t                  width,
+                         int32_t                  height)
+{
+    pixman_blt_mmx (src_image->bits.bits,
+                    dst_image->bits.bits,
+                    src_image->bits.rowstride,
+                    dst_image->bits.rowstride,
+                    PIXMAN_FORMAT_BPP (src_image->bits.format),
+                    PIXMAN_FORMAT_BPP (dst_image->bits.format),
+                    src_x, src_y, dest_x, dest_y, width, height);
+}
+
+#if 0
+static void
+mmx_composite_over_x888_8_8888 (pixman_implementation_t *imp,
+                                pixman_op_t              op,
+                                pixman_image_t *         src_image,
+                                pixman_image_t *         mask_image,
+                                pixman_image_t *         dst_image,
+                                int32_t                  src_x,
+                                int32_t                  src_y,
+                                int32_t                  mask_x,
+                                int32_t                  mask_y,
+                                int32_t                  dest_x,
+                                int32_t                  dest_y,
+                                int32_t                  width,
+                                int32_t                  height)
+{
+    uint32_t  *src, *src_line;
+    uint32_t  *dst, *dst_line;
+    uint8_t  *mask, *mask_line;
+    int src_stride, mask_stride, dst_stride;
+    uint16_t w;
+
+    PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
+    PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+
+    while (height--)
+    {
+	src = src_line;
+	src_line += src_stride;
+	dst = dst_line;
+	dst_line += dst_stride;
+	mask = mask_line;
+	mask_line += mask_stride;
+
+	w = width;
+
+	while (w--)
+	{
+	    uint64_t m = *mask;
+
+	    if (m)
+	    {
+		__m64 s = load8888 (*src | 0xff000000);
+
+		if (m == 0xff)
+		{
+		    *dst = store8888 (s);
+		}
+		else
+		{
+		    __m64 sa = expand_alpha (s);
+		    __m64 vm = expand_alpha_rev (M64 (m));
+		    __m64 vdest = in_over (s, sa, vm, load8888 (*dst));
+
+		    *dst = store8888 (vdest);
+		}
+	    }
+
+	    mask++;
+	    dst++;
+	    src++;
+	}
+    }
+
+    _mm_empty ();
+}
+#endif
+
+static const pixman_fast_path_t mmx_fast_paths[] =
+{
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r5g6b5,   mmx_composite_over_n_8_0565       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b5g6r5,   mmx_composite_over_n_8_0565       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_over_n_8_8888       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, mmx_composite_over_n_8_8888       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, mmx_composite_over_n_8_8888       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, mmx_composite_over_n_8_8888       },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_a8r8g8b8, mmx_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_x8r8g8b8, mmx_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_r5g6b5,   mmx_composite_over_n_8888_0565_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_a8b8g8r8, mmx_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_x8b8g8r8, mmx_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_b5g6r5,   mmx_composite_over_n_8888_0565_ca },
+    { PIXMAN_OP_OVER, PIXMAN_pixbuf,   PIXMAN_pixbuf,   PIXMAN_a8r8g8b8, mmx_composite_over_pixbuf_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_pixbuf,   PIXMAN_pixbuf,   PIXMAN_x8r8g8b8, mmx_composite_over_pixbuf_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_pixbuf,   PIXMAN_pixbuf,   PIXMAN_r5g6b5,   mmx_composite_over_pixbuf_0565    },
+    { PIXMAN_OP_OVER, PIXMAN_rpixbuf,  PIXMAN_rpixbuf,  PIXMAN_a8b8g8r8, mmx_composite_over_pixbuf_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_rpixbuf,  PIXMAN_rpixbuf,  PIXMAN_x8b8g8r8, mmx_composite_over_pixbuf_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_rpixbuf,  PIXMAN_rpixbuf,  PIXMAN_b5g6r5,   mmx_composite_over_pixbuf_0565    },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_solid,    PIXMAN_a8r8g8b8, mmx_composite_over_x888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_solid,    PIXMAN_x8r8g8b8, mmx_composite_over_x888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_solid,    PIXMAN_a8b8g8r8, mmx_composite_over_x888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_solid,    PIXMAN_x8b8g8r8, mmx_composite_over_x888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_solid,    PIXMAN_a8r8g8b8, mmx_composite_over_8888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_solid,    PIXMAN_x8r8g8b8, mmx_composite_over_8888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_solid,    PIXMAN_a8b8g8r8, mmx_composite_over_8888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_solid,    PIXMAN_x8b8g8r8, mmx_composite_over_8888_n_8888    },
+#if 0
+    /* FIXME: This code is commented out since it's apparently not actually faster than the generic code. */
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, mmx_composite_over_x888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_over_x888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8r8g8, PIXMAN_a8,       PIXMAN_x8b8g8r8, mmx_composite_over_x888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8r8g8, PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_over_x888_8_8888    },
+#endif
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_a8r8g8b8, mmx_composite_over_n_8888        },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_over_n_8888        },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_r5g6b5,   mmx_composite_over_n_0565        },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_copy_area          },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, mmx_composite_copy_area          },
+
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, mmx_composite_over_8888_8888     },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_over_8888_8888     },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   mmx_composite_over_8888_0565     },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, mmx_composite_over_8888_8888     },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, mmx_composite_over_8888_8888     },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   mmx_composite_over_8888_0565     },
+
+    { PIXMAN_OP_ADD, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_a8r8g8b8, mmx_composite_add_8888_8888   },
+    { PIXMAN_OP_ADD, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_a8b8g8r8, mmx_composite_add_8888_8888   },
+    { PIXMAN_OP_ADD, PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       mmx_composite_add_8000_8000   },
+    { PIXMAN_OP_ADD, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       mmx_composite_add_n_8_8    },
+
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8r8g8b8, mmx_composite_src_n_8_8888 },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_x8r8g8b8, mmx_composite_src_n_8_8888 },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8b8g8r8, mmx_composite_src_n_8_8888 },
+    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_x8b8g8r8, mmx_composite_src_n_8_8888 },
+    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_a8r8g8b8, mmx_composite_copy_area },
+    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_a8b8g8r8, mmx_composite_copy_area },
+    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_copy_area },
+    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, mmx_composite_copy_area },
+    { PIXMAN_OP_SRC, PIXMAN_x8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, mmx_composite_copy_area },
+    { PIXMAN_OP_SRC, PIXMAN_x8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, mmx_composite_copy_area },
+    { PIXMAN_OP_SRC, PIXMAN_r5g6b5,    PIXMAN_null,     PIXMAN_r5g6b5,   mmx_composite_copy_area },
+    { PIXMAN_OP_SRC, PIXMAN_b5g6r5,    PIXMAN_null,     PIXMAN_b5g6r5,   mmx_composite_copy_area },
+
+    { PIXMAN_OP_IN,  PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       mmx_composite_in_8_8    },
+    { PIXMAN_OP_IN,  PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       mmx_composite_in_n_8_8  },
+
+    { PIXMAN_OP_NONE },
+};
+
+static void
+mmx_composite (pixman_implementation_t *imp,
+               pixman_op_t              op,
+               pixman_image_t *         src,
+               pixman_image_t *         mask,
+               pixman_image_t *         dest,
+               int32_t                  src_x,
+               int32_t                  src_y,
+               int32_t                  mask_x,
+               int32_t                  mask_y,
+               int32_t                  dest_x,
+               int32_t                  dest_y,
+               int32_t                  width,
+               int32_t                  height)
+{
+    if (_pixman_run_fast_path (mmx_fast_paths, imp,
+                               op, src, mask, dest,
+                               src_x, src_y,
+                               mask_x, mask_y,
+                               dest_x, dest_y,
+                               width, height))
+    {
+	return;
+    }
+
+    _pixman_implementation_composite (imp->delegate,
+                                      op, src, mask, dest, src_x, src_y,
+                                      mask_x, mask_y, dest_x, dest_y,
+                                      width, height);
+}
+
+static pixman_bool_t
+mmx_blt (pixman_implementation_t *imp,
+         uint32_t *               src_bits,
+         uint32_t *               dst_bits,
+         int                      src_stride,
+         int                      dst_stride,
+         int                      src_bpp,
+         int                      dst_bpp,
+         int                      src_x,
+         int                      src_y,
+         int                      dst_x,
+         int                      dst_y,
+         int                      width,
+         int                      height)
+{
+    if (!pixman_blt_mmx (
+            src_bits, dst_bits, src_stride, dst_stride, src_bpp, dst_bpp,
+            src_x, src_y, dst_x, dst_y, width, height))
+
+    {
+	return _pixman_implementation_blt (
+	           imp->delegate,
+	           src_bits, dst_bits, src_stride, dst_stride, src_bpp, dst_bpp,
+	           src_x, src_y, dst_x, dst_y, width, height);
+    }
+
+    return TRUE;
+}
+
+static pixman_bool_t
+mmx_fill (pixman_implementation_t *imp,
+          uint32_t *               bits,
+          int                      stride,
+          int                      bpp,
+          int                      x,
+          int                      y,
+          int                      width,
+          int                      height,
+          uint32_t xor)
+{
+    if (!pixman_fill_mmx (bits, stride, bpp, x, y, width, height, xor))
+    {
+	return _pixman_implementation_fill (
+	           imp->delegate, bits, stride, bpp, x, y, width, height, xor);
+    }
+
+    return TRUE;
+}
+
+pixman_implementation_t *
+_pixman_implementation_create_mmx (void)
+{
+    pixman_implementation_t *general = _pixman_implementation_create_fast_path ();
+    pixman_implementation_t *imp = _pixman_implementation_create (general);
+
+    imp->combine_32[PIXMAN_OP_OVER] = mmx_combine_over_u;
+    imp->combine_32[PIXMAN_OP_OVER_REVERSE] = mmx_combine_over_reverse_u;
+    imp->combine_32[PIXMAN_OP_IN] = mmx_combine_in_u;
+    imp->combine_32[PIXMAN_OP_IN_REVERSE] = mmx_combine_in_reverse_u;
+    imp->combine_32[PIXMAN_OP_OUT] = mmx_combine_out_u;
+    imp->combine_32[PIXMAN_OP_OUT_REVERSE] = mmx_combine_out_reverse_u;
+    imp->combine_32[PIXMAN_OP_ATOP] = mmx_combine_atop_u;
+    imp->combine_32[PIXMAN_OP_ATOP_REVERSE] = mmx_combine_atop_reverse_u;
+    imp->combine_32[PIXMAN_OP_XOR] = mmx_combine_xor_u;
+    imp->combine_32[PIXMAN_OP_ADD] = mmx_combine_add_u;
+    imp->combine_32[PIXMAN_OP_SATURATE] = mmx_combine_saturate_u;
+
+    imp->combine_32_ca[PIXMAN_OP_SRC] = mmx_combine_src_ca;
+    imp->combine_32_ca[PIXMAN_OP_OVER] = mmx_combine_over_ca;
+    imp->combine_32_ca[PIXMAN_OP_OVER_REVERSE] = mmx_combine_over_reverse_ca;
+    imp->combine_32_ca[PIXMAN_OP_IN] = mmx_combine_in_ca;
+    imp->combine_32_ca[PIXMAN_OP_IN_REVERSE] = mmx_combine_in_reverse_ca;
+    imp->combine_32_ca[PIXMAN_OP_OUT] = mmx_combine_out_ca;
+    imp->combine_32_ca[PIXMAN_OP_OUT_REVERSE] = mmx_combine_out_reverse_ca;
+    imp->combine_32_ca[PIXMAN_OP_ATOP] = mmx_combine_atop_ca;
+    imp->combine_32_ca[PIXMAN_OP_ATOP_REVERSE] = mmx_combine_atop_reverse_ca;
+    imp->combine_32_ca[PIXMAN_OP_XOR] = mmx_combine_xor_ca;
+    imp->combine_32_ca[PIXMAN_OP_ADD] = mmx_combine_add_ca;
+
+    imp->composite = mmx_composite;
+    imp->blt = mmx_blt;
+    imp->fill = mmx_fill;
+
+    return imp;
+}
+
+#endif /* USE_MMX */
diff --git a/pixman/pixman/pixman-private.h b/pixman/pixman/pixman-private.h
index ff7a65f88..5000f910d 100644
--- a/pixman/pixman/pixman-private.h
+++ b/pixman/pixman/pixman-private.h
@@ -5,6 +5,9 @@
 #ifndef PIXMAN_PRIVATE_H
 #define PIXMAN_PRIVATE_H
 
+#define PIXMAN_DISABLE_DEPRECATED
+#define PIXMAN_USE_INTERNAL_API
+
 #include "pixman.h"
 #include <time.h>
 #include <assert.h>
@@ -120,8 +123,6 @@ struct gradient
     int                     n_stops;
     pixman_gradient_stop_t *stops;
     int                     stop_range;
-    uint32_t *              color_table;
-    int                     color_table_size;
 };
 
 struct linear_gradient
@@ -437,10 +438,10 @@ struct pixman_implementation_t
     pixman_blt_func_t        blt;
     pixman_fill_func_t       fill;
 
-    pixman_combine_32_func_t combine_32[PIXMAN_OP_LAST];
-    pixman_combine_32_func_t combine_32_ca[PIXMAN_OP_LAST];
-    pixman_combine_64_func_t combine_64[PIXMAN_OP_LAST];
-    pixman_combine_64_func_t combine_64_ca[PIXMAN_OP_LAST];
+    pixman_combine_32_func_t combine_32[PIXMAN_N_OPERATORS];
+    pixman_combine_32_func_t combine_32_ca[PIXMAN_N_OPERATORS];
+    pixman_combine_64_func_t combine_64[PIXMAN_N_OPERATORS];
+    pixman_combine_64_func_t combine_64_ca[PIXMAN_N_OPERATORS];
 };
 
 pixman_implementation_t *
@@ -556,15 +557,15 @@ _pixman_choose_implementation (void);
  * Utilities
  */
 
-/* These "formats" both have depth 0, so they
+/* These "formats" all have depth 0, so they
  * will never clash with any real ones
  */
 #define PIXMAN_null             PIXMAN_FORMAT (0, 0, 0, 0, 0, 0)
 #define PIXMAN_solid            PIXMAN_FORMAT (0, 1, 0, 0, 0, 0)
-
-#define NEED_COMPONENT_ALPHA            (1 << 0)
-#define NEED_PIXBUF                     (1 << 1)
-#define NEED_SOLID_MASK                 (1 << 2)
+#define PIXMAN_a8r8g8b8_ca	PIXMAN_FORMAT (0, 2, 0, 0, 0, 0)
+#define PIXMAN_a8b8g8r8_ca	PIXMAN_FORMAT (0, 3, 0, 0, 0, 0)
+#define PIXMAN_pixbuf		PIXMAN_FORMAT (0, 4, 0, 0, 0, 0)
+#define PIXMAN_rpixbuf		PIXMAN_FORMAT (0, 5, 0, 0, 0, 0)
 
 typedef struct
 {
@@ -573,7 +574,6 @@ typedef struct
     pixman_format_code_t    mask_format;
     pixman_format_code_t    dest_format;
     pixman_composite_func_t func;
-    uint32_t                flags;
 } pixman_fast_path_t;
 
 /* Memory allocation helpers */
@@ -612,14 +612,14 @@ _pixman_walk_composite_region (pixman_implementation_t *imp,
                                pixman_image_t *         src_image,
                                pixman_image_t *         mask_image,
                                pixman_image_t *         dst_image,
-                               int16_t                  src_x,
-                               int16_t                  src_y,
-                               int16_t                  mask_x,
-                               int16_t                  mask_y,
-                               int16_t                  dest_x,
-                               int16_t                  dest_y,
-                               uint16_t                 width,
-                               uint16_t                 height,
+                               int32_t                  src_x,
+                               int32_t                  src_y,
+                               int32_t                  mask_x,
+                               int32_t                  mask_y,
+                               int32_t                  dest_x,
+                               int32_t                  dest_y,
+                               int32_t                  width,
+                               int32_t                  height,
                                pixman_composite_func_t  composite_rect);
 
 void
diff --git a/pixman/pixman/pixman-sse2.c b/pixman/pixman/pixman-sse2.c
index bb74882b2..78b0ad185 100644
--- a/pixman/pixman/pixman-sse2.c
+++ b/pixman/pixman/pixman-sse2.c
@@ -368,6 +368,22 @@ cache_prefetch_next (__m128i* addr)
     _mm_prefetch ((void const *)(addr + 4), _MM_HINT_T0); /* 64 bytes ahead */
 }
 
+/* prefetching NULL is very slow on some systems. don't do that. */
+
+static force_inline void
+maybe_prefetch (__m128i* addr)
+{
+    if (addr)
+	cache_prefetch (addr);
+}
+
+static force_inline void
+maybe_prefetch_next (__m128i* addr)
+{
+    if (addr)
+	cache_prefetch_next (addr);
+}
+
 /* load 4 pixels from a 16-byte boundary aligned address */
 static force_inline __m128i
 load_128_aligned (__m128i* src)
@@ -413,9 +429,15 @@ save_128_unaligned (__m128i* dst,
  */
 
 static force_inline __m64
+load_32_1x64 (uint32_t data)
+{
+    return _mm_cvtsi32_si64 (data);
+}
+
+static force_inline __m64
 unpack_32_1x64 (uint32_t data)
 {
-    return _mm_unpacklo_pi8 (_mm_cvtsi32_si64 (data), _mm_setzero_si64 ());
+    return _mm_unpacklo_pi8 (load_32_1x64 (data), _mm_setzero_si64 ());
 }
 
 static force_inline __m64
@@ -629,7 +651,7 @@ core_combine_over_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     /* Align dst on a 16-byte boundary */
     while (w && ((unsigned long)pd & 15))
@@ -647,14 +669,14 @@ core_combine_over_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	/* I'm loading unaligned because I'm not sure about
 	 * the address alignment.
@@ -720,7 +742,7 @@ core_combine_over_reverse_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     /* Align dst on a 16-byte boundary */
     while (w &&
@@ -739,14 +761,14 @@ core_combine_over_reverse_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	/* I'm loading unaligned because I'm not sure
 	 * about the address alignment.
@@ -822,7 +844,7 @@ core_combine_in_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && ((unsigned long) pd & 15))
     {
@@ -839,14 +861,14 @@ core_combine_in_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	xmm_dst_hi = load_128_aligned ((__m128i*) pd);
 	xmm_src_hi = combine4 ((__m128i*) ps, (__m128i*) pm);
@@ -896,7 +918,7 @@ core_combine_reverse_in_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && ((unsigned long) pd & 15))
     {
@@ -913,14 +935,14 @@ core_combine_reverse_in_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	xmm_dst_hi = load_128_aligned ((__m128i*) pd);
 	xmm_src_hi = combine4 ((__m128i*) ps, (__m128i*)pm);
@@ -965,7 +987,7 @@ core_combine_reverse_out_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && ((unsigned long) pd & 15))
     {
@@ -986,7 +1008,7 @@ core_combine_reverse_out_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
@@ -996,7 +1018,7 @@ core_combine_reverse_out_u_sse2 (uint32_t*       pd,
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	xmm_src_hi = combine4 ((__m128i*)ps, (__m128i*)pm);
 	xmm_dst_hi = load_128_aligned ((__m128i*) pd);
@@ -1047,7 +1069,7 @@ core_combine_out_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && ((unsigned long) pd & 15))
     {
@@ -1067,7 +1089,7 @@ core_combine_out_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
@@ -1077,7 +1099,7 @@ core_combine_out_u_sse2 (uint32_t*       pd,
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	xmm_src_hi = combine4 ((__m128i*) ps, (__m128i*)pm);
 	xmm_dst_hi = load_128_aligned ((__m128i*) pd);
@@ -1147,7 +1169,7 @@ core_combine_atop_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && ((unsigned long) pd & 15))
     {
@@ -1164,14 +1186,14 @@ core_combine_atop_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	xmm_src_hi = combine4 ((__m128i*)ps, (__m128i*)pm);
 	xmm_dst_hi = load_128_aligned ((__m128i*) pd);
@@ -1244,7 +1266,7 @@ core_combine_reverse_atop_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && ((unsigned long) pd & 15))
     {
@@ -1261,14 +1283,14 @@ core_combine_reverse_atop_u_sse2 (uint32_t*       pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	xmm_src_hi = combine4 ((__m128i*)ps, (__m128i*)pm);
 	xmm_dst_hi = load_128_aligned ((__m128i*) pd);
@@ -1345,7 +1367,7 @@ core_combine_xor_u_sse2 (uint32_t*       dst,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && ((unsigned long) pd & 15))
     {
@@ -1362,14 +1384,14 @@ core_combine_xor_u_sse2 (uint32_t*       dst,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	xmm_src = combine4 ((__m128i*) ps, (__m128i*) pm);
 	xmm_dst = load_128_aligned ((__m128i*) pd);
@@ -1430,7 +1452,7 @@ core_combine_add_u_sse2 (uint32_t*       dst,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && (unsigned long)pd & 15)
     {
@@ -1448,7 +1470,7 @@ core_combine_add_u_sse2 (uint32_t*       dst,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
@@ -1457,7 +1479,7 @@ core_combine_add_u_sse2 (uint32_t*       dst,
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	s = combine4 ((__m128i*)ps, (__m128i*)pm);
 
@@ -1516,7 +1538,7 @@ core_combine_saturate_u_sse2 (uint32_t *      pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w && (unsigned long)pd & 15)
     {
@@ -1533,14 +1555,14 @@ core_combine_saturate_u_sse2 (uint32_t *      pd,
     /* call prefetch hint to optimize cache load*/
     cache_prefetch ((__m128i*)ps);
     cache_prefetch ((__m128i*)pd);
-    cache_prefetch ((__m128i*)pm);
+    maybe_prefetch ((__m128i*)pm);
 
     while (w >= 4)
     {
 	/* fill cache line with next memory */
 	cache_prefetch_next ((__m128i*)ps);
 	cache_prefetch_next ((__m128i*)pd);
-	cache_prefetch_next ((__m128i*)pm);
+	maybe_prefetch_next ((__m128i*)pm);
 
 	xmm_dst = load_128_aligned  ((__m128i*)pd);
 	xmm_src = combine4 ((__m128i*)ps, (__m128i*)pm);
@@ -3221,7 +3243,7 @@ sse2_composite_add_n_8888_8888_ca (pixman_implementation_t *imp,
 	    if (m)
 	    {
 		d = *pd;
-		
+
 		mmx_mask = unpack_32_1x64 (m);
 		mmx_dest = unpack_32_1x64 (d);
 
@@ -3412,7 +3434,7 @@ sse2_composite_over_8888_n_8888 (pixman_implementation_t *imp,
     PIXMAN_IMAGE_GET_LINE (
 	src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
 
-    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);
+    mask = _pixman_image_get_solid (mask_image, PIXMAN_a8r8g8b8);
 
     xmm_mask = create_mask_16_128 (mask >> 24);
 
@@ -3528,7 +3550,7 @@ sse2_composite_over_x888_n_8888 (pixman_implementation_t *imp,
     PIXMAN_IMAGE_GET_LINE (
 	src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
 
-    mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);
+    mask = _pixman_image_get_solid (mask_image, PIXMAN_a8r8g8b8);
 
     xmm_mask = create_mask_16_128 (mask >> 24);
     xmm_alpha = mask_00ff;
@@ -5033,23 +5055,23 @@ sse2_composite_in_8_8 (pixman_implementation_t *imp,
 }
 
 /* -------------------------------------------------------------------------
- * composite_add_8888_8_8
+ * composite_add_n_8_8
  */
 
 static void
-sse2_composite_add_8888_8_8 (pixman_implementation_t *imp,
-                             pixman_op_t              op,
-                             pixman_image_t *         src_image,
-                             pixman_image_t *         mask_image,
-                             pixman_image_t *         dst_image,
-                             int32_t                  src_x,
-                             int32_t                  src_y,
-                             int32_t                  mask_x,
-                             int32_t                  mask_y,
-                             int32_t                  dest_x,
-                             int32_t                  dest_y,
-                             int32_t                  width,
-                             int32_t                  height)
+sse2_composite_add_n_8_8 (pixman_implementation_t *imp,
+			  pixman_op_t              op,
+			  pixman_image_t *         src_image,
+			  pixman_image_t *         mask_image,
+			  pixman_image_t *         dst_image,
+			  int32_t                  src_x,
+			  int32_t                  src_y,
+			  int32_t                  mask_x,
+			  int32_t                  mask_y,
+			  int32_t                  dest_x,
+			  int32_t                  dest_y,
+			  int32_t                  width,
+			  int32_t                  height)
 {
     uint8_t     *dst_line, *dst;
     uint8_t     *mask_line, *mask;
@@ -5428,9 +5450,7 @@ sse2_composite_copy_area (pixman_implementation_t *imp,
                      src_x, src_y, dest_x, dest_y, width, height);
 }
 
-#if 0
-/* This code are buggy in MMX version, now the bug was translated to SSE2 version */
-void
+static void
 sse2_composite_over_x888_8_8888 (pixman_implementation_t *imp,
                                  pixman_op_t              op,
                                  pixman_image_t *         src_image,
@@ -5451,6 +5471,7 @@ sse2_composite_over_x888_8_8888 (pixman_implementation_t *imp,
     uint32_t m;
     int src_stride, mask_stride, dst_stride;
     uint16_t w;
+    __m64 ms;
 
     __m128i xmm_src, xmm_src_lo, xmm_src_hi;
     __m128i xmm_dst, xmm_dst_lo, xmm_dst_hi;
@@ -5465,199 +5486,355 @@ sse2_composite_over_x888_8_8888 (pixman_implementation_t *imp,
 
     while (height--)
     {
-	src = src_line;
-	src_line += src_stride;
-	dst = dst_line;
-	dst_line += dst_stride;
-	mask = mask_line;
-	mask_line += mask_stride;
+        src = src_line;
+        src_line += src_stride;
+        dst = dst_line;
+        dst_line += dst_stride;
+        mask = mask_line;
+        mask_line += mask_stride;
+
+        w = width;
+
+        /* call prefetch hint to optimize cache load*/
+        cache_prefetch ((__m128i*)src);
+        cache_prefetch ((__m128i*)dst);
+        cache_prefetch ((__m128i*)mask);
+
+        while (w && (unsigned long)dst & 15)
+        {
+            s = 0xff000000 | *src++;
+            m = (uint32_t) *mask++;
+            d = *dst;
+            ms = unpack_32_1x64 (s);
+
+            if (m != 0xff)
+            {
+		__m64 ma = expand_alpha_rev_1x64 (unpack_32_1x64 (m));
+		__m64 md = unpack_32_1x64 (d);
+
+                ms = in_over_1x64 (&ms, &mask_x00ff, &ma, &md);
+            }
+
+            *dst++ = pack_1x64_32 (ms);
+            w--;
+        }
+
+        /* call prefetch hint to optimize cache load*/
+        cache_prefetch ((__m128i*)src);
+        cache_prefetch ((__m128i*)dst);
+        cache_prefetch ((__m128i*)mask);
+
+        while (w >= 4)
+        {
+            /* fill cache line with next memory */
+            cache_prefetch_next ((__m128i*)src);
+            cache_prefetch_next ((__m128i*)dst);
+            cache_prefetch_next ((__m128i*)mask);
+
+            m = *(uint32_t*) mask;
+            xmm_src = _mm_or_si128 (load_128_unaligned ((__m128i*)src), mask_ff000000);
+
+            if (m == 0xffffffff)
+            {
+                save_128_aligned ((__m128i*)dst, xmm_src);
+            }
+            else
+            {
+                xmm_dst = load_128_aligned ((__m128i*)dst);
+
+                xmm_mask = _mm_unpacklo_epi16 (unpack_32_1x128 (m), _mm_setzero_si128());
+
+                unpack_128_2x128 (xmm_src, &xmm_src_lo, &xmm_src_hi);
+                unpack_128_2x128 (xmm_mask, &xmm_mask_lo, &xmm_mask_hi);
+                unpack_128_2x128 (xmm_dst, &xmm_dst_lo, &xmm_dst_hi);
+
+                expand_alpha_rev_2x128 (xmm_mask_lo, xmm_mask_hi, &xmm_mask_lo, &xmm_mask_hi);
+
+                in_over_2x128 (&xmm_src_lo, &xmm_src_hi, &mask_00ff, &mask_00ff, &xmm_mask_lo, &xmm_mask_hi, &xmm_dst_lo, &xmm_dst_hi);
+
+                save_128_aligned ((__m128i*)dst, pack_2x128_128 (xmm_dst_lo, xmm_dst_hi));
+            }
+
+            src += 4;
+            dst += 4;
+            mask += 4;
+            w -= 4;
+        }
+
+        while (w)
+        {
+            m = (uint32_t) *mask++;
+
+            if (m)
+            {
+                s = 0xff000000 | *src;
+
+                if (m == 0xff)
+                {
+                    *dst = s;
+                }
+                else
+                {
+		    __m64 ma, md, ms;
+
+                    d = *dst;
+
+		    ma = expand_alpha_rev_1x64 (unpack_32_1x64 (m));
+		    md = unpack_32_1x64 (d);
+		    ms = unpack_32_1x64 (s);
+
+                    *dst = pack_1x64_32 (in_over_1x64 (&ms, &mask_x00ff, &ma, &md));
+                }
+
+            }
+
+            src++;
+            dst++;
+            w--;
+        }
+    }
 
-	w = width;
+    _mm_empty ();
+}
 
-	/* call prefetch hint to optimize cache load*/
-	cache_prefetch ((__m128i*)src);
-	cache_prefetch ((__m128i*)dst);
-	cache_prefetch ((__m128i*)mask);
+static void
+sse2_composite_over_8888_8_8888 (pixman_implementation_t *imp,
+                                 pixman_op_t              op,
+                                 pixman_image_t *         src_image,
+                                 pixman_image_t *         mask_image,
+                                 pixman_image_t *         dst_image,
+                                 int32_t                  src_x,
+                                 int32_t                  src_y,
+                                 int32_t                  mask_x,
+                                 int32_t                  mask_y,
+                                 int32_t                  dest_x,
+                                 int32_t                  dest_y,
+                                 int32_t                  width,
+                                 int32_t                  height)
+{
+    uint32_t    *src, *src_line, s;
+    uint32_t    *dst, *dst_line, d;
+    uint8_t         *mask, *mask_line;
+    uint32_t m;
+    int src_stride, mask_stride, dst_stride;
+    uint16_t w;
 
-	while (w && (unsigned long)dst & 15)
-	{
-	    s = 0xff000000 | *src++;
-	    m = (uint32_t) *mask++;
-	    d = *dst;
+    __m128i xmm_src, xmm_src_lo, xmm_src_hi, xmm_srca_lo, xmm_srca_hi;
+    __m128i xmm_dst, xmm_dst_lo, xmm_dst_hi;
+    __m128i xmm_mask, xmm_mask_lo, xmm_mask_hi;
 
-	    __m64 ms = unpack_32_1x64 (s);
+    PIXMAN_IMAGE_GET_LINE (
+	dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+    PIXMAN_IMAGE_GET_LINE (
+	mask_image, mask_x, mask_y, uint8_t, mask_stride, mask_line, 1);
+    PIXMAN_IMAGE_GET_LINE (
+	src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
 
-	    if (m != 0xff)
-	    {
-		ms = in_over_1x64 (ms,
-		                   mask_x00ff,
-		                   expand_alpha_rev_1x64 (unpack_32_1x64 (m)),
-		                   unpack_32_1x64 (d));
-	    }
+    while (height--)
+    {
+        src = src_line;
+        src_line += src_stride;
+        dst = dst_line;
+        dst_line += dst_stride;
+        mask = mask_line;
+        mask_line += mask_stride;
 
-	    *dst++ = pack_1x64_32 (ms);
-	    w--;
-	}
+        w = width;
 
-	/* call prefetch hint to optimize cache load*/
-	cache_prefetch ((__m128i*)src);
-	cache_prefetch ((__m128i*)dst);
-	cache_prefetch ((__m128i*)mask);
+        /* call prefetch hint to optimize cache load*/
+        cache_prefetch ((__m128i *)src);
+        cache_prefetch ((__m128i *)dst);
+        cache_prefetch ((__m128i *)mask);
 
-	while (w >= 4)
-	{
-	    /* fill cache line with next memory */
-	    cache_prefetch_next ((__m128i*)src);
-	    cache_prefetch_next ((__m128i*)dst);
-	    cache_prefetch_next ((__m128i*)mask);
+        while (w && (unsigned long)dst & 15)
+        {
+	    uint32_t sa;
 
-	    m = *(uint32_t*) mask;
-	    xmm_src = _mm_or_si128 (load_128_unaligned ((__m128i*)src), mask_ff000000);
+            s = *src++;
+            m = (uint32_t) *mask++;
+            d = *dst;
 
-	    if (m == 0xffffffff)
+	    sa = s >> 24;
+
+	    if (m)
 	    {
-		save_128_aligned ((__m128i*)dst, xmm_src);
+		if (sa == 0xff && m == 0xff)
+		{
+		    *dst = s;
+		}
+		else
+		{
+		    __m64 ms, md, ma, msa;
+
+		    ma = expand_alpha_rev_1x64 (load_32_1x64 (m));
+		    ms = unpack_32_1x64 (s);
+		    md = unpack_32_1x64 (d);
+
+		    msa = expand_alpha_rev_1x64 (load_32_1x64 (sa));
+
+		    *dst = pack_1x64_32 (in_over_1x64 (&ms, &msa, &ma, &md));
+		}
 	    }
-	    else
+
+	    dst++;
+            w--;
+        }
+
+        /* call prefetch hint to optimize cache load*/
+        cache_prefetch ((__m128i *)src);
+        cache_prefetch ((__m128i *)dst);
+        cache_prefetch ((__m128i *)mask);
+
+        while (w >= 4)
+        {
+            /* fill cache line with next memory */
+            cache_prefetch_next ((__m128i *)src);
+            cache_prefetch_next ((__m128i *)dst);
+            cache_prefetch_next ((__m128i *)mask);
+
+            m = *(uint32_t *) mask;
+
+	    if (m)
 	    {
-		xmm_dst = load_128_aligned ((__m128i*)dst);
+		xmm_src = load_128_unaligned ((__m128i*)src);
 
-		xmm_mask = _mm_unpacklo_epi16 (
-		    unpack_32_1x128 (m), _mm_setzero_si128 ());
+		if (m == 0xffffffff && is_opaque (xmm_src))
+		{
+		    save_128_aligned ((__m128i *)dst, xmm_src);
+		}
+		else
+		{
+		    xmm_dst = load_128_aligned ((__m128i *)dst);
 
-		unpack_128_2x128 (xmm_src, &xmm_src_lo, &xmm_src_hi);
-		unpack_128_2x128 (xmm_mask, &xmm_mask_lo, &xmm_mask_hi);
-		unpack_128_2x128 (xmm_dst, &xmm_dst_lo, &xmm_dst_hi);
+		    xmm_mask = _mm_unpacklo_epi16 (unpack_32_1x128 (m), _mm_setzero_si128());
 
-		expand_alpha_rev_2x128 (xmm_mask_lo, xmm_mask_hi,
-					&xmm_mask_lo, &xmm_mask_hi);
+		    unpack_128_2x128 (xmm_src, &xmm_src_lo, &xmm_src_hi);
+		    unpack_128_2x128 (xmm_mask, &xmm_mask_lo, &xmm_mask_hi);
+		    unpack_128_2x128 (xmm_dst, &xmm_dst_lo, &xmm_dst_hi);
 
-		in_over_2x128 (xmm_src_lo, xmm_src_hi,
-			       mask_00ff, mask_00ff,
-			       xmm_mask_lo, xmm_mask_hi,
-			       &xmm_dst_lo, &xmm_dst_hi);
+		    expand_alpha_2x128 (xmm_src_lo, xmm_src_hi, &xmm_srca_lo, &xmm_srca_hi);
+		    expand_alpha_rev_2x128 (xmm_mask_lo, xmm_mask_hi, &xmm_mask_lo, &xmm_mask_hi);
+		    
+		    in_over_2x128 (&xmm_src_lo, &xmm_src_hi, &xmm_srca_lo, &xmm_srca_hi,
+				   &xmm_mask_lo, &xmm_mask_hi, &xmm_dst_lo, &xmm_dst_hi);
 
-		save_128_aligned (
-		    (__m128i*)dst, pack_2x128_128 (xmm_dst_lo, xmm_dst_hi));
+		    save_128_aligned ((__m128i*)dst, pack_2x128_128 (xmm_dst_lo, xmm_dst_hi));
+		}
 	    }
 
-	    src += 4;
-	    dst += 4;
-	    mask += 4;
-	    w -= 4;
-	}
+            src += 4;
+            dst += 4;
+            mask += 4;
+            w -= 4;
+        }
 
-	while (w)
-	{
-	    m = (uint32_t) *mask++;
+        while (w)
+        {
+	    uint32_t sa;
+
+            s = *src++;
+            m = (uint32_t) *mask++;
+            d = *dst;
+
+	    sa = s >> 24;
 
 	    if (m)
 	    {
-		s = 0xff000000 | *src;
-
-		if (m == 0xff)
+		if (sa == 0xff && m == 0xff)
 		{
 		    *dst = s;
 		}
 		else
 		{
-		    d = *dst;
+		    __m64 ms, md, ma, msa;
 
-		    *dst = pack_1x64_32 (
-			in_over_1x64 (
-			    unpack_32_1x64 (s),
-			    mask_x00ff,
-			    expand_alpha_rev_1x64 (unpack_32_1x64 (m)),
-			    unpack_32_1x64 (d)));
-		}
+		    ma = expand_alpha_rev_1x64 (load_32_1x64 (m));
+		    ms = unpack_32_1x64 (s);
+		    md = unpack_32_1x64 (d);
 
+		    msa = expand_alpha_rev_1x64 (load_32_1x64 (sa));
+
+		    *dst = pack_1x64_32 (in_over_1x64 (&ms, &msa, &ma, &md));
+		}
 	    }
 
-	    src++;
 	    dst++;
-	    w--;
-	}
+            w--;
+        }
     }
 
     _mm_empty ();
 }
 
-#endif
-
 static const pixman_fast_path_t sse2_fast_paths[] =
 {
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r5g6b5,   sse2_composite_over_n_8_0565,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b5g6r5,   sse2_composite_over_n_8_0565,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_a8r8g8b8, sse2_composite_over_n_8888,         0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_over_n_8888,         0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_r5g6b5,   sse2_composite_over_n_0565,         0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, sse2_composite_over_8888_8888,      0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_over_8888_8888,      0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, sse2_composite_over_8888_8888,      0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, sse2_composite_over_8888_8888,      0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   sse2_composite_over_8888_0565,      0 },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   sse2_composite_over_8888_0565,      0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_over_n_8_8888,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_over_n_8_8888,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_over_n_8_8888,       0 },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_over_n_8_8888,       0 },
-#if 0
-    /* FIXME: This code are buggy in MMX version, now the bug was translated to SSE2 version */
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_over_x888_8_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_over_x888_8_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_over_x888_8_8888,    0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_over_x888_8_8888,    0 },
-#endif
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_over_x888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_over_x888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_over_x888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_over_x888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_over_8888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_over_8888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_over_8888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_over_8888_n_8888,    NEED_SOLID_MASK },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_a8r8g8b8, sse2_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_x8r8g8b8, sse2_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_a8b8g8r8, sse2_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_x8b8g8r8, sse2_composite_over_n_8888_8888_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_r5g6b5,   sse2_composite_over_n_8888_0565_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8b8g8r8, PIXMAN_b5g6r5,   sse2_composite_over_n_8888_0565_ca, NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8r8g8b8, PIXMAN_a8r8g8b8, sse2_composite_over_pixbuf_8888,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8b8g8r8, PIXMAN_a8r8g8b8, sse2_composite_over_pixbuf_8888,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8r8g8b8, PIXMAN_x8r8g8b8, sse2_composite_over_pixbuf_8888,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8b8g8r8, PIXMAN_x8r8g8b8, sse2_composite_over_pixbuf_8888,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8r8g8b8, PIXMAN_a8b8g8r8, sse2_composite_over_pixbuf_8888,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8b8g8r8, PIXMAN_a8b8g8r8, sse2_composite_over_pixbuf_8888,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8r8g8b8, PIXMAN_x8b8g8r8, sse2_composite_over_pixbuf_8888,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8b8g8r8, PIXMAN_x8b8g8r8, sse2_composite_over_pixbuf_8888,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8r8g8b8, PIXMAN_r5g6b5,   sse2_composite_over_pixbuf_0565,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8b8g8r8, PIXMAN_r5g6b5,   sse2_composite_over_pixbuf_0565,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8r8g8b8, PIXMAN_b5g6r5,   sse2_composite_over_pixbuf_0565,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8b8g8r8, PIXMAN_b5g6r5,   sse2_composite_over_pixbuf_0565,    NEED_PIXBUF },
-    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_copy_area,           0 },
-    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, sse2_composite_copy_area,           0 },
-
-    { PIXMAN_OP_ADD,  PIXMAN_solid,    PIXMAN_a8r8g8b8, PIXMAN_a8r8g8b8, sse2_composite_add_n_8888_8888_ca,  NEED_COMPONENT_ALPHA },
-    { PIXMAN_OP_ADD,  PIXMAN_a8,       PIXMAN_null,     PIXMAN_a8,       sse2_composite_add_8000_8000,       0 },
-    { PIXMAN_OP_ADD,  PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, sse2_composite_add_8888_8888,       0 },
-    { PIXMAN_OP_ADD,  PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, sse2_composite_add_8888_8888,       0 },
-    { PIXMAN_OP_ADD,  PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8,       sse2_composite_add_8888_8_8,        0 },
-
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_src_n_8_8888,        0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_src_n_8_8888,        0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_src_n_8_8888,        0 },
-    { PIXMAN_OP_SRC, PIXMAN_solid,     PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_src_n_8_8888,        0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_a8r8g8b8, sse2_composite_copy_area,           0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_a8b8g8r8, sse2_composite_copy_area,           0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_copy_area,           0 },
-    { PIXMAN_OP_SRC, PIXMAN_a8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, sse2_composite_copy_area,           0 },
-    { PIXMAN_OP_SRC, PIXMAN_x8r8g8b8,  PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_copy_area,           0 },
-    { PIXMAN_OP_SRC, PIXMAN_x8b8g8r8,  PIXMAN_null,     PIXMAN_x8b8g8r8, sse2_composite_copy_area,           0 },
-    { PIXMAN_OP_SRC, PIXMAN_r5g6b5,    PIXMAN_null,     PIXMAN_r5g6b5,   sse2_composite_copy_area,           0 },
-    { PIXMAN_OP_SRC, PIXMAN_b5g6r5,    PIXMAN_null,     PIXMAN_b5g6r5,   sse2_composite_copy_area,           0 },
-
-    { PIXMAN_OP_IN,  PIXMAN_a8,        PIXMAN_null,     PIXMAN_a8,       sse2_composite_in_8_8,              0 },
-    { PIXMAN_OP_IN,  PIXMAN_solid,     PIXMAN_a8,       PIXMAN_a8,       sse2_composite_in_n_8_8,            0 },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_r5g6b5,   sse2_composite_over_n_8_0565       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_b5g6r5,   sse2_composite_over_n_8_0565       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_a8r8g8b8, sse2_composite_over_n_8888         },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_over_n_8888         },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_null,     PIXMAN_r5g6b5,   sse2_composite_over_n_0565         },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, sse2_composite_over_8888_8888      },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_over_8888_8888      },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, sse2_composite_over_8888_8888      },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, sse2_composite_over_8888_8888      },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_r5g6b5,   sse2_composite_over_8888_0565      },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_b5g6r5,   sse2_composite_over_8888_0565      },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_over_n_8_8888       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_over_n_8_8888       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_over_n_8_8888       },
+    { PIXMAN_OP_OVER, PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_over_n_8_8888       },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_over_8888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_over_8888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_over_8888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_over_8888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_over_x888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_over_x888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_over_x888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_over_x888_8_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_solid,    PIXMAN_a8r8g8b8, sse2_composite_over_x888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_solid,    PIXMAN_x8r8g8b8, sse2_composite_over_x888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_solid,    PIXMAN_a8b8g8r8, sse2_composite_over_x888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_solid,    PIXMAN_x8b8g8r8, sse2_composite_over_x888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_solid,    PIXMAN_a8r8g8b8, sse2_composite_over_8888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8r8g8b8, PIXMAN_solid,    PIXMAN_x8r8g8b8, sse2_composite_over_8888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_solid,    PIXMAN_a8b8g8r8, sse2_composite_over_8888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_a8b8g8r8, PIXMAN_solid,    PIXMAN_x8b8g8r8, sse2_composite_over_8888_n_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_a8r8g8b8, sse2_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_x8r8g8b8, sse2_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_a8b8g8r8, sse2_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_x8b8g8r8, sse2_composite_over_n_8888_8888_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_r5g6b5,   sse2_composite_over_n_8888_0565_ca },
+    { PIXMAN_OP_OVER, PIXMAN_solid, PIXMAN_a8b8g8r8_ca, PIXMAN_b5g6r5,   sse2_composite_over_n_8888_0565_ca },
+    { PIXMAN_OP_OVER, PIXMAN_pixbuf,   PIXMAN_pixbuf,   PIXMAN_a8r8g8b8, sse2_composite_over_pixbuf_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_pixbuf,   PIXMAN_pixbuf,   PIXMAN_x8r8g8b8, sse2_composite_over_pixbuf_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_rpixbuf,  PIXMAN_rpixbuf,  PIXMAN_a8b8g8r8, sse2_composite_over_pixbuf_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_rpixbuf,  PIXMAN_rpixbuf,  PIXMAN_x8b8g8r8, sse2_composite_over_pixbuf_8888    },
+    { PIXMAN_OP_OVER, PIXMAN_pixbuf,   PIXMAN_pixbuf,   PIXMAN_r5g6b5,   sse2_composite_over_pixbuf_0565    },
+    { PIXMAN_OP_OVER, PIXMAN_rpixbuf,  PIXMAN_rpixbuf,  PIXMAN_b5g6r5,   sse2_composite_over_pixbuf_0565    },
+    { PIXMAN_OP_OVER, PIXMAN_x8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_copy_area           },
+    { PIXMAN_OP_OVER, PIXMAN_x8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, sse2_composite_copy_area           },
+
+    { PIXMAN_OP_ADD,  PIXMAN_solid, PIXMAN_a8r8g8b8_ca, PIXMAN_a8r8g8b8, sse2_composite_add_n_8888_8888_ca  },
+    { PIXMAN_OP_ADD,  PIXMAN_a8,       PIXMAN_null,     PIXMAN_a8,       sse2_composite_add_8000_8000       },
+    { PIXMAN_OP_ADD,  PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, sse2_composite_add_8888_8888       },
+    { PIXMAN_OP_ADD,  PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, sse2_composite_add_8888_8888       },
+    { PIXMAN_OP_ADD,  PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8,       sse2_composite_add_n_8_8           },
+
+    { PIXMAN_OP_SRC,  PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8r8g8b8, sse2_composite_src_n_8_8888        },
+    { PIXMAN_OP_SRC,  PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8r8g8b8, sse2_composite_src_n_8_8888        },
+    { PIXMAN_OP_SRC,  PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8b8g8r8, sse2_composite_src_n_8_8888        },
+    { PIXMAN_OP_SRC,  PIXMAN_solid,    PIXMAN_a8,       PIXMAN_x8b8g8r8, sse2_composite_src_n_8_8888        },
+    { PIXMAN_OP_SRC,  PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_a8r8g8b8, sse2_composite_copy_area           },
+    { PIXMAN_OP_SRC,  PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_a8b8g8r8, sse2_composite_copy_area           },
+    { PIXMAN_OP_SRC,  PIXMAN_a8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_copy_area           },
+    { PIXMAN_OP_SRC,  PIXMAN_a8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, sse2_composite_copy_area           },
+    { PIXMAN_OP_SRC,  PIXMAN_x8r8g8b8, PIXMAN_null,     PIXMAN_x8r8g8b8, sse2_composite_copy_area           },
+    { PIXMAN_OP_SRC,  PIXMAN_x8b8g8r8, PIXMAN_null,     PIXMAN_x8b8g8r8, sse2_composite_copy_area           },
+    { PIXMAN_OP_SRC,  PIXMAN_r5g6b5,   PIXMAN_null,     PIXMAN_r5g6b5,   sse2_composite_copy_area           },
+    { PIXMAN_OP_SRC,  PIXMAN_b5g6r5,   PIXMAN_null,     PIXMAN_b5g6r5,   sse2_composite_copy_area           },
+
+    { PIXMAN_OP_IN,   PIXMAN_a8,       PIXMAN_null,     PIXMAN_a8,       sse2_composite_in_8_8              },
+    { PIXMAN_OP_IN,   PIXMAN_solid,    PIXMAN_a8,       PIXMAN_a8,       sse2_composite_in_n_8_8            },
 
     { PIXMAN_OP_NONE },
 };
diff --git a/pixman/pixman/pixman-utils.c b/pixman/pixman/pixman-utils.c
index 71282062c..6d03f3c07 100644
--- a/pixman/pixman/pixman-utils.c
+++ b/pixman/pixman/pixman-utils.c
@@ -33,8 +33,6 @@
 /*
  * Computing composite region
  */
-#define BOUND(v)        (int16_t) ((v) < INT16_MIN ? INT16_MIN : (v) > INT16_MAX ? INT16_MAX : (v))
-
 static inline pixman_bool_t
 clip_general_image (pixman_region32_t * region,
                     pixman_region32_t * clip,
@@ -49,17 +47,17 @@ clip_general_image (pixman_region32_t * region,
 	int v;
 
 	if (rbox->x1 < (v = cbox->x1 + dx))
-	    rbox->x1 = BOUND (v);
+	    rbox->x1 = v;
 	if (rbox->x2 > (v = cbox->x2 + dx))
-	    rbox->x2 = BOUND (v);
+	    rbox->x2 = v;
 	if (rbox->y1 < (v = cbox->y1 + dy))
-	    rbox->y1 = BOUND (v);
+	    rbox->y1 = v;
 	if (rbox->y2 > (v = cbox->y2 + dy))
-	    rbox->y2 = BOUND (v);
-	if (rbox->x1 >= rbox->x2 ||
-	    rbox->y1 >= rbox->y2)
+	    rbox->y2 = v;
+	if (rbox->x1 >= rbox->x2 || rbox->y1 >= rbox->y2)
 	{
 	    pixman_region32_init (region);
+	    return FALSE;
 	}
     }
     else if (!pixman_region32_not_empty (clip))
@@ -70,11 +68,14 @@ clip_general_image (pixman_region32_t * region,
     {
 	if (dx || dy)
 	    pixman_region32_translate (region, -dx, -dy);
+
 	if (!pixman_region32_intersect (region, region, clip))
 	    return FALSE;
+
 	if (dx || dy)
 	    pixman_region32_translate (region, dx, dy);
     }
+
     return pixman_region32_not_empty (region);
 }
 
@@ -106,23 +107,19 @@ pixman_compute_composite_region32 (pixman_region32_t * region,
                                    pixman_image_t *    src_image,
                                    pixman_image_t *    mask_image,
                                    pixman_image_t *    dst_image,
-                                   int16_t             src_x,
-                                   int16_t             src_y,
-                                   int16_t             mask_x,
-                                   int16_t             mask_y,
-                                   int16_t             dest_x,
-                                   int16_t             dest_y,
-                                   uint16_t            width,
-                                   uint16_t            height)
+                                   int32_t             src_x,
+                                   int32_t             src_y,
+                                   int32_t             mask_x,
+                                   int32_t             mask_y,
+                                   int32_t             dest_x,
+                                   int32_t             dest_y,
+                                   int32_t             width,
+                                   int32_t             height)
 {
-    int v;
-
     region->extents.x1 = dest_x;
-    v = dest_x + width;
-    region->extents.x2 = BOUND (v);
+    region->extents.x2 = dest_x + width;
     region->extents.y1 = dest_y;
-    v = dest_y + height;
-    region->extents.y2 = BOUND (v);
+    region->extents.y2 = dest_y + height;
 
     region->extents.x1 = MAX (region->extents.x1, 0);
     region->extents.y1 = MAX (region->extents.y1, 0);
@@ -378,14 +375,14 @@ walk_region_internal (pixman_implementation_t *imp,
                       pixman_image_t *         src_image,
                       pixman_image_t *         mask_image,
                       pixman_image_t *         dst_image,
-                      int16_t                  src_x,
-                      int16_t                  src_y,
-                      int16_t                  mask_x,
-                      int16_t                  mask_y,
-                      int16_t                  dest_x,
-                      int16_t                  dest_y,
-                      uint16_t                 width,
-                      uint16_t                 height,
+                      int32_t                  src_x,
+                      int32_t                  src_y,
+                      int32_t                  mask_x,
+                      int32_t                  mask_y,
+                      int32_t                  dest_x,
+                      int32_t                  dest_y,
+                      int32_t                  width,
+                      int32_t                  height,
                       pixman_bool_t            src_repeat,
                       pixman_bool_t            mask_repeat,
                       pixman_region32_t *      region,
@@ -471,14 +468,14 @@ _pixman_walk_composite_region (pixman_implementation_t *imp,
                                pixman_image_t *         src_image,
                                pixman_image_t *         mask_image,
                                pixman_image_t *         dst_image,
-                               int16_t                  src_x,
-                               int16_t                  src_y,
-                               int16_t                  mask_x,
-                               int16_t                  mask_y,
-                               int16_t                  dest_x,
-                               int16_t                  dest_y,
-                               uint16_t                 width,
-                               uint16_t                 height,
+                               int32_t                  src_x,
+                               int32_t                  src_y,
+                               int32_t                  mask_x,
+                               int32_t                  mask_y,
+                               int32_t                  dest_x,
+                               int32_t                  dest_y,
+                               int32_t                  width,
+                               int32_t                  height,
                                pixman_composite_func_t  composite_rect)
 {
     pixman_region32_t region;
@@ -501,78 +498,97 @@ _pixman_walk_composite_region (pixman_implementation_t *imp,
     }
 }
 
-static pixman_bool_t
-mask_is_solid (pixman_image_t *mask)
-{
-    if (mask->type == SOLID)
-	return TRUE;
-
-    if (mask->type == BITS &&
-        mask->common.repeat == PIXMAN_REPEAT_NORMAL &&
-        mask->bits.width == 1 &&
-        mask->bits.height == 1)
-    {
-	return TRUE;
-    }
-
-    return FALSE;
-}
-
 static const pixman_fast_path_t *
 get_fast_path (const pixman_fast_path_t *fast_paths,
                pixman_op_t               op,
                pixman_image_t *          src_image,
                pixman_image_t *          mask_image,
                pixman_image_t *          dst_image,
-               pixman_bool_t             is_pixbuf)
+	       int			 src_x,
+	       int			 src_y,
+	       int			 mask_x,
+	       int			 mask_y)
 {
+    pixman_format_code_t src_format, mask_format;
     const pixman_fast_path_t *info;
 
-    for (info = fast_paths; info->op != PIXMAN_OP_NONE; info++)
+    /* Check for pixbufs */
+    if (mask_image && mask_image->type == BITS								&&
+	(mask_image->bits.format == PIXMAN_a8r8g8b8 || mask_image->bits.format == PIXMAN_a8b8g8r8)	&&
+	(src_image->type == BITS && src_image->bits.bits == mask_image->bits.bits)			&&
+	(src_image->common.repeat == mask_image->common.repeat)						&&
+	(src_x == mask_x && src_y == mask_y))
     {
-	pixman_bool_t valid_src = FALSE;
-	pixman_bool_t valid_mask = FALSE;
-
-	if (info->op != op)
-	    continue;
-
-	if ((info->src_format == PIXMAN_solid &&
-	     _pixman_image_is_solid (src_image)) ||
-	    (src_image->type == BITS &&
-	     info->src_format == src_image->bits.format))
+	if (src_image->bits.format == PIXMAN_x8b8g8r8)
+	    src_format = mask_format = PIXMAN_pixbuf;
+	else if (src_image->bits.format == PIXMAN_x8r8g8b8)
+	    src_format = mask_format = PIXMAN_rpixbuf;
+	else
+	    return NULL;
+    }
+    else
+    {
+	if (_pixman_image_is_solid (src_image))
 	{
-	    valid_src = TRUE;
+	    src_format = PIXMAN_solid;
 	}
-
-	if (!valid_src)
-	    continue;
-
-	if ((info->mask_format == PIXMAN_null && !mask_image) ||
-	    (mask_image && mask_image->type == BITS &&
-	     info->mask_format == mask_image->bits.format))
+	else if (src_image->type == BITS)
 	{
-	    valid_mask = TRUE;
-
-	    if (info->flags & NEED_SOLID_MASK)
+	    src_format = src_image->bits.format;
+	}
+	else
+	{
+	    return NULL;
+	}
+	
+	if (!mask_image)
+	{
+	    mask_format = PIXMAN_null;
+	}
+	else if (mask_image->common.component_alpha)
+	{
+	    if (mask_image->type == BITS)
 	    {
-		if (!mask_image || !mask_is_solid (mask_image))
-		    valid_mask = FALSE;
+		/* These are the *only* component_alpha formats
+		 * we support for fast paths
+		 */
+		if (mask_image->bits.format == PIXMAN_a8r8g8b8)
+		    mask_format = PIXMAN_a8r8g8b8_ca;
+		else if (mask_image->bits.format == PIXMAN_a8b8g8r8)
+		    mask_format = PIXMAN_a8b8g8r8_ca;
+		else
+		    return NULL;
 	    }
-
-	    if (info->flags & NEED_COMPONENT_ALPHA)
+	    else
 	    {
-		if (!mask_image || !mask_image->common.component_alpha)
-		    valid_mask = FALSE;
+		return NULL;
 	    }
 	}
-
-	if (!valid_mask)
+	else if (_pixman_image_is_solid (mask_image))
+	{
+	    mask_format = PIXMAN_solid;
+	}
+	else if (mask_image->common.type == BITS)
+	{
+	    mask_format = mask_image->bits.format;
+	}
+	else
+	{
+	    return NULL;
+	}
+    }
+    
+    for (info = fast_paths; info->op != PIXMAN_OP_NONE; info++)
+    {
+	if (info->op != op)
 	    continue;
 
-	if (info->dest_format != dst_image->bits.format)
+	if (info->src_format != src_format)
+	    continue;
+	if (info->mask_format != mask_format)
 	    continue;
 
-	if ((info->flags & NEED_PIXBUF) && !is_pixbuf)
+	if (info->dest_format != dst_image->bits.format)
 	    continue;
 
 	return info;
@@ -617,7 +633,7 @@ sources_cover (pixman_image_t *src,
 
     if (!mask)
 	return TRUE;
-    
+
     if (!image_covers (mask, extents, dest_x - mask_x, dest_y - mask_y))
 	return FALSE;
 
@@ -654,8 +670,7 @@ _pixman_run_fast_path (const pixman_fast_path_t *paths,
 
     if (has_fast_path)
     {
-	has_fast_path = (src->type == BITS || _pixman_image_is_solid (src)) &&
-	                !src->common.transform &&
+	has_fast_path = !src->common.transform &&
 	                !src->common.alpha_map &&
 			src->common.filter != PIXMAN_FILTER_CONVOLUTION &&
 			src->common.repeat != PIXMAN_REPEAT_PAD &&
@@ -671,7 +686,6 @@ _pixman_run_fast_path (const pixman_fast_path_t *paths,
     if (mask && has_fast_path)
     {
 	has_fast_path =
-	    mask->type == BITS &&
 	    !mask->common.transform &&
 	    !mask->common.alpha_map &&
 	    !mask->bits.read_func &&
@@ -685,28 +699,15 @@ _pixman_run_fast_path (const pixman_fast_path_t *paths,
     if (has_fast_path)
     {
 	const pixman_fast_path_t *info;
-	pixman_bool_t pixbuf;
-
-	pixbuf =
-	    src && src->type == BITS            &&
-	    mask && mask->type == BITS          &&
-	    src->bits.bits == mask->bits.bits   &&
-	    src_x == mask_x                     &&
-	    src_y == mask_y                     &&
-	    !mask->common.component_alpha       &&
-	    !mask_repeat;
-
-	info = get_fast_path (paths, op, src, mask, dest, pixbuf);
 
-	if (info)
+	if ((info = get_fast_path (paths, op, src, mask, dest, src_x, src_y, mask_x, mask_y)))
 	{
 	    func = info->func;
 
 	    if (info->src_format == PIXMAN_solid)
 		src_repeat = FALSE;
 
-	    if (info->mask_format == PIXMAN_solid ||
-		info->flags & NEED_SOLID_MASK)
+	    if (info->mask_format == PIXMAN_solid)
 	    {
 		mask_repeat = FALSE;
 	    }
diff --git a/pixman/pixman/pixman-version.h b/pixman/pixman/pixman-version.h
index 79d74fed4..6b93790e1 100644
--- a/pixman/pixman/pixman-version.h
+++ b/pixman/pixman/pixman-version.h
@@ -32,10 +32,10 @@
 #endif
 
 #define PIXMAN_VERSION_MAJOR 0
-#define PIXMAN_VERSION_MINOR 16
+#define PIXMAN_VERSION_MINOR 17
 #define PIXMAN_VERSION_MICRO 2
 
-#define PIXMAN_VERSION_STRING "0.16.2"
+#define PIXMAN_VERSION_STRING "0.17.2"
 
 #define PIXMAN_VERSION_ENCODE(major, minor, micro) (	\
 	  ((major) * 10000)				\
diff --git a/pixman/pixman/pixman.h b/pixman/pixman/pixman.h
index 5b90a0c8d..ba2b242ad 100644
--- a/pixman/pixman/pixman.h
+++ b/pixman/pixman/pixman.h
@@ -74,7 +74,7 @@ SOFTWARE.
 /*
  * Standard integers
  */
-#if defined (_SVR4) || defined (SVR4) || defined (__OpenBSD__) || defined (_sgi) || defined (__sun) || defined (sun) || defined (__digital__)
+#if defined (_SVR4) || defined (SVR4) || defined (__OpenBSD__) || defined (_sgi) || defined (__sun) || defined (sun) || defined (__digital__) || defined (__HP_cc)
 #  include <inttypes.h>
 #elif defined (_MSC_VER)
 typedef __int8 int8_t;
@@ -109,6 +109,7 @@ typedef pixman_fixed_16_16_t	pixman_fixed_t;
 #define pixman_fixed_e			((pixman_fixed_t) 1)
 #define pixman_fixed_1			(pixman_int_to_fixed(1))
 #define pixman_fixed_1_minus_e		(pixman_fixed_1 - pixman_fixed_e)
+#define pixman_fixed_minus_1		(pixman_int_to_fixed(-1))
 #define pixman_fixed_to_int(f)		((int) ((f) >> 16))
 #define pixman_int_to_fixed(i)		((pixman_fixed_t) ((i) << 16))
 #define pixman_fixed_to_double(f)	(double) ((f) / (double) pixman_fixed_1)
@@ -333,8 +334,10 @@ typedef enum
     PIXMAN_OP_HSL_COLOR			= 0x3d,
     PIXMAN_OP_HSL_LUMINOSITY		= 0x3e,
 
-    PIXMAN_OP_NONE,
-    PIXMAN_OP_LAST = PIXMAN_OP_NONE
+#ifdef PIXMAN_USE_INTERNAL_API
+    PIXMAN_N_OPERATORS,
+    PIXMAN_OP_NONE = PIXMAN_N_OPERATORS
+#endif
 } pixman_op_t;
 
 /*
@@ -637,11 +640,11 @@ typedef enum {
 /* 24bpp formats */
     PIXMAN_r8g8b8 =	 PIXMAN_FORMAT(24,PIXMAN_TYPE_ARGB,0,8,8,8),
     PIXMAN_b8g8r8 =	 PIXMAN_FORMAT(24,PIXMAN_TYPE_ABGR,0,8,8,8),
-    
+
 /* 16bpp formats */
     PIXMAN_r5g6b5 =	 PIXMAN_FORMAT(16,PIXMAN_TYPE_ARGB,0,5,6,5),
     PIXMAN_b5g6r5 =	 PIXMAN_FORMAT(16,PIXMAN_TYPE_ABGR,0,5,6,5),
-    
+
     PIXMAN_a1r5g5b5 =	 PIXMAN_FORMAT(16,PIXMAN_TYPE_ARGB,1,5,5,5),
     PIXMAN_x1r5g5b5 =	 PIXMAN_FORMAT(16,PIXMAN_TYPE_ARGB,0,5,5,5),
     PIXMAN_a1b5g5r5 =	 PIXMAN_FORMAT(16,PIXMAN_TYPE_ABGR,1,5,5,5),
@@ -650,35 +653,35 @@ typedef enum {
     PIXMAN_x4r4g4b4 =	 PIXMAN_FORMAT(16,PIXMAN_TYPE_ARGB,0,4,4,4),
     PIXMAN_a4b4g4r4 =	 PIXMAN_FORMAT(16,PIXMAN_TYPE_ABGR,4,4,4,4),
     PIXMAN_x4b4g4r4 =	 PIXMAN_FORMAT(16,PIXMAN_TYPE_ABGR,0,4,4,4),
-    
+
 /* 8bpp formats */
     PIXMAN_a8 =		 PIXMAN_FORMAT(8,PIXMAN_TYPE_A,8,0,0,0),
     PIXMAN_r3g3b2 =	 PIXMAN_FORMAT(8,PIXMAN_TYPE_ARGB,0,3,3,2),
     PIXMAN_b2g3r3 =	 PIXMAN_FORMAT(8,PIXMAN_TYPE_ABGR,0,3,3,2),
     PIXMAN_a2r2g2b2 =	 PIXMAN_FORMAT(8,PIXMAN_TYPE_ARGB,2,2,2,2),
     PIXMAN_a2b2g2r2 =	 PIXMAN_FORMAT(8,PIXMAN_TYPE_ABGR,2,2,2,2),
-    
+
     PIXMAN_c8 =		 PIXMAN_FORMAT(8,PIXMAN_TYPE_COLOR,0,0,0,0),
     PIXMAN_g8 =		 PIXMAN_FORMAT(8,PIXMAN_TYPE_GRAY,0,0,0,0),
-    
+
     PIXMAN_x4a4 =	 PIXMAN_FORMAT(8,PIXMAN_TYPE_A,4,0,0,0),
-    
+
     PIXMAN_x4c4 =	 PIXMAN_FORMAT(8,PIXMAN_TYPE_COLOR,0,0,0,0),
     PIXMAN_x4g4 =	 PIXMAN_FORMAT(8,PIXMAN_TYPE_GRAY,0,0,0,0),
-    
+
 /* 4bpp formats */
     PIXMAN_a4 =		 PIXMAN_FORMAT(4,PIXMAN_TYPE_A,4,0,0,0),
     PIXMAN_r1g2b1 =	 PIXMAN_FORMAT(4,PIXMAN_TYPE_ARGB,0,1,2,1),
     PIXMAN_b1g2r1 =	 PIXMAN_FORMAT(4,PIXMAN_TYPE_ABGR,0,1,2,1),
     PIXMAN_a1r1g1b1 =	 PIXMAN_FORMAT(4,PIXMAN_TYPE_ARGB,1,1,1,1),
     PIXMAN_a1b1g1r1 =	 PIXMAN_FORMAT(4,PIXMAN_TYPE_ABGR,1,1,1,1),
-    
+
     PIXMAN_c4 =		 PIXMAN_FORMAT(4,PIXMAN_TYPE_COLOR,0,0,0,0),
     PIXMAN_g4 =		 PIXMAN_FORMAT(4,PIXMAN_TYPE_GRAY,0,0,0,0),
-    
+
 /* 1bpp formats */
     PIXMAN_a1 =		 PIXMAN_FORMAT(1,PIXMAN_TYPE_A,1,0,0,0),
-    
+
     PIXMAN_g1 =		 PIXMAN_FORMAT(1,PIXMAN_TYPE_GRAY,0,0,0,0),
 
 /* YUV formats */
author	marha <marha@users.sourceforge.net>	2009-11-20 14:53:20 +0000
committer	marha <marha@users.sourceforge.net>	2009-11-20 14:53:20 +0000
commit	9e11b82359e90c0963f56c4695a4dcc8731e45c3 (patch)
tree	ffa3a53246352ca44ccfc4d90e7cca993ad0b02e /pixman/pixman
parent	607c11e20aca5c9f13d5e6ab80153b10c249162b (diff)
parent	4ac4a5b7ce8cc8f195d69a42da10d386eaa5c056 (diff)
download	vcxsrv-9e11b82359e90c0963f56c4695a4dcc8731e45c3.tar.gz vcxsrv-9e11b82359e90c0963f56c4695a4dcc8731e45c3.tar.bz2 vcxsrv-9e11b82359e90c0963f56c4695a4dcc8731e45c3.zip