SSE LEAF EE works

10 files changed, 720 insertions, 137 deletions

diff --git a/src/Makefile.am b/src/Makefile.am
index 84265d6..0b0791e 100644
--- a/src/Makefile.am
+++ b/src/Makefile.am
@@ -1,20 +1,12 @@
 
-OBJLIBS = libffts.a
-HDRS = cp_sse.h patterns.h macros.h neon_float.h codegen.h neon.h
-FILES = cp_sse.c patterns.c codegen.c neon.s
-OBJS = cp_sse.o patterns.o codegen.o neon.o
 
-all: $(OBJLIBS)
+lib_LTLIBRARIES = libffts.la
 
-%.o: %.c $(HDRS)
-	$(CC) $(CFLAGS) -c -o $@ $< -I../include -g
-	$(CC) $(CFLAGS) -S $< -I../include
+libffts_la_SOURCES = cp_sse.c patterns.c codegen.c
 
-%.o: %.s 
-	$(CC) $(CFLAGS) -c -o $@ $< -I../include -g
-	
-$(OBJLIBS): $(OBJS) 
-	$(AR) rcs libffts.a $(OBJS)
-
-clean:
-	$(RM) -f *.o $(OBJLIBS)
+if HAVE_NEON
+libffts_la_SOURCES += neon.s
+endif
+if HAVE_SSE
+libffts_la_SOURCES += sse.s
+endif
diff --git a/src/Makefile.in b/src/Makefile.in
index 52c6551..3d87aa0 100644
--- a/src/Makefile.in
+++ b/src/Makefile.in
@@ -13,6 +13,7 @@
 # PARTICULAR PURPOSE.
 
 @SET_MAKE@
+
 VPATH = @srcdir@
 am__make_dryrun = \
   { \
@@ -47,8 +48,13 @@ POST_INSTALL = :
 NORMAL_UNINSTALL = :
 PRE_UNINSTALL = :
 POST_UNINSTALL = :
+build_triplet = @build@
+host_triplet = @host@
+@HAVE_NEON_TRUE@am__append_1 = neon.s
+@HAVE_SSE_TRUE@am__append_2 = sse.s
 subdir = src
-DIST_COMMON = $(srcdir)/Makefile.am $(srcdir)/Makefile.in
+DIST_COMMON = $(srcdir)/Makefile.am $(srcdir)/Makefile.in \
+	$(top_srcdir)/depcomp
 ACLOCAL_M4 = $(top_srcdir)/aclocal.m4
 am__aclocal_m4_deps = $(top_srcdir)/configure.ac
 am__configure_deps = $(am__aclocal_m4_deps) $(CONFIGURE_DEPENDENCIES) \
@@ -57,49 +63,123 @@ mkinstalldirs = $(install_sh) -d
 CONFIG_HEADER = $(top_builddir)/config.h
 CONFIG_CLEAN_FILES =
 CONFIG_CLEAN_VPATH_FILES =
-SOURCES =
-DIST_SOURCES =
+am__vpath_adj_setup = srcdirstrip=`echo "$(srcdir)" | sed 's|.|.|g'`;
+am__vpath_adj = case $$p in \
+    $(srcdir)/*) f=`echo "$$p" | sed "s|^$$srcdirstrip/||"`;; \
+    *) f=$$p;; \
+  esac;
+am__strip_dir = f=`echo $$p | sed -e 's|^.*/||'`;
+am__install_max = 40
+am__nobase_strip_setup = \
+  srcdirstrip=`echo "$(srcdir)" | sed 's/[].[^$$\\*|]/\\\\&/g'`
+am__nobase_strip = \
+  for p in $$list; do echo "$$p"; done | sed -e "s|$$srcdirstrip/||"
+am__nobase_list = $(am__nobase_strip_setup); \
+  for p in $$list; do echo "$$p $$p"; done | \
+  sed "s| $$srcdirstrip/| |;"' / .*\//!s/ .*/ ./; s,\( .*\)/[^/]*$$,\1,' | \
+  $(AWK) 'BEGIN { files["."] = "" } { files[$$2] = files[$$2] " " $$1; \
+    if (++n[$$2] == $(am__install_max)) \
+      { print $$2, files[$$2]; n[$$2] = 0; files[$$2] = "" } } \
+    END { for (dir in files) print dir, files[dir] }'
+am__base_list = \
+  sed '$$!N;$$!N;$$!N;$$!N;$$!N;$$!N;$$!N;s/\n/ /g' | \
+  sed '$$!N;$$!N;$$!N;$$!N;s/\n/ /g'
+am__uninstall_files_from_dir = { \
+  test -z "$$files" \
+    || { test ! -d "$$dir" && test ! -f "$$dir" && test ! -r "$$dir"; } \
+    || { echo " ( cd '$$dir' && rm -f" $$files ")"; \
+         $(am__cd) "$$dir" && rm -f $$files; }; \
+  }
+am__installdirs = "$(DESTDIR)$(libdir)"
+LTLIBRARIES = $(lib_LTLIBRARIES)
+libffts_la_LIBADD =
+am__libffts_la_SOURCES_DIST = cp_sse.c patterns.c codegen.c neon.s \
+	sse.s
+@HAVE_NEON_TRUE@am__objects_1 = neon.lo
+@HAVE_SSE_TRUE@am__objects_2 = sse.lo
+am_libffts_la_OBJECTS = cp_sse.lo patterns.lo codegen.lo \
+	$(am__objects_1) $(am__objects_2)
+libffts_la_OBJECTS = $(am_libffts_la_OBJECTS)
+DEFAULT_INCLUDES = -I.@am__isrc@ -I$(top_builddir)
+depcomp = $(SHELL) $(top_srcdir)/depcomp
+am__depfiles_maybe = depfiles
+am__mv = mv -f
+COMPILE = $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) \
+	$(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS)
+LTCOMPILE = $(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) \
+	--mode=compile $(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) \
+	$(AM_CPPFLAGS) $(CPPFLAGS) $(AM_CFLAGS) $(CFLAGS)
+CCLD = $(CC)
+LINK = $(LIBTOOL) --tag=CC $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) \
+	--mode=link $(CCLD) $(AM_CFLAGS) $(CFLAGS) $(AM_LDFLAGS) \
+	$(LDFLAGS) -o $@
+CCASCOMPILE = $(CCAS) $(AM_CCASFLAGS) $(CCASFLAGS)
+LTCCASCOMPILE = $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) \
+	--mode=compile $(CCAS) $(AM_CCASFLAGS) $(CCASFLAGS)
+SOURCES = $(libffts_la_SOURCES)
+DIST_SOURCES = $(am__libffts_la_SOURCES_DIST)
 am__can_run_installinfo = \
   case $$AM_UPDATE_INFO_DIR in \
     n|no|NO) false;; \
     *) (install-info --version) >/dev/null 2>&1;; \
   esac
+ETAGS = etags
+CTAGS = ctags
 DISTFILES = $(DIST_COMMON) $(DIST_SOURCES) $(TEXINFOS) $(EXTRA_DIST)
 ACLOCAL = @ACLOCAL@
 AMTAR = @AMTAR@
+AR = @AR@
 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@
+DLLTOOL = @DLLTOOL@
+DSYMUTIL = @DSYMUTIL@
+DUMPBIN = @DUMPBIN@
 ECHO_C = @ECHO_C@
 ECHO_N = @ECHO_N@
 ECHO_T = @ECHO_T@
 EGREP = @EGREP@
 EXEEXT = @EXEEXT@
+FGREP = @FGREP@
 GREP = @GREP@
 INSTALL = @INSTALL@
 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@
 MAKEINFO = @MAKEINFO@
+MANIFEST_TOOL = @MANIFEST_TOOL@
 MKDIR_P = @MKDIR_P@
+NM = @NM@
+NMEDIT = @NMEDIT@
+OBJDUMP = @OBJDUMP@
 OBJEXT = @OBJEXT@
+OTOOL = @OTOOL@
+OTOOL64 = @OTOOL64@
 PACKAGE = @PACKAGE@
 PACKAGE_BUGREPORT = @PACKAGE_BUGREPORT@
 PACKAGE_NAME = @PACKAGE_NAME@
@@ -108,6 +188,8 @@ PACKAGE_TARNAME = @PACKAGE_TARNAME@
 PACKAGE_URL = @PACKAGE_URL@
 PACKAGE_VERSION = @PACKAGE_VERSION@
 PATH_SEPARATOR = @PATH_SEPARATOR@
+RANLIB = @RANLIB@
+SED = @SED@
 SET_MAKE = @SET_MAKE@
 SHELL = @SHELL@
 STRIP = @STRIP@
@@ -116,22 +198,32 @@ abs_builddir = @abs_builddir@
 abs_srcdir = @abs_srcdir@
 abs_top_builddir = @abs_top_builddir@
 abs_top_srcdir = @abs_top_srcdir@
+ac_ct_AR = @ac_ct_AR@
 ac_ct_CC = @ac_ct_CC@
 ac_ct_CXX = @ac_ct_CXX@
+ac_ct_DUMPBIN = @ac_ct_DUMPBIN@
 am__include = @am__include@
 am__leading_dot = @am__leading_dot@
 am__quote = @am__quote@
 am__tar = @am__tar@
 am__untar = @am__untar@
 bindir = @bindir@
+build = @build@
 build_alias = @build_alias@
+build_cpu = @build_cpu@
+build_os = @build_os@
+build_vendor = @build_vendor@
 builddir = @builddir@
 datadir = @datadir@
 datarootdir = @datarootdir@
 docdir = @docdir@
 dvidir = @dvidir@
 exec_prefix = @exec_prefix@
+host = @host@
 host_alias = @host_alias@
+host_cpu = @host_cpu@
+host_os = @host_os@
+host_vendor = @host_vendor@
 htmldir = @htmldir@
 includedir = @includedir@
 infodir = @infodir@
@@ -155,13 +247,13 @@ target_alias = @target_alias@
 top_build_prefix = @top_build_prefix@
 top_builddir = @top_builddir@
 top_srcdir = @top_srcdir@
-OBJLIBS = libffts.a
-HDRS = cp_sse.h patterns.h macros.h neon_float.h codegen.h neon.h
-FILES = cp_sse.c patterns.c codegen.c neon.s
-OBJS = cp_sse.o patterns.o codegen.o neon.o
+lib_LTLIBRARIES = libffts.la
+libffts_la_SOURCES = cp_sse.c patterns.c codegen.c $(am__append_1) \
+	$(am__append_2)
 all: all-am
 
 .SUFFIXES:
+.SUFFIXES: .c .lo .o .obj .s
 $(srcdir)/Makefile.in:  $(srcdir)/Makefile.am  $(am__configure_deps)
 	@for dep in $?; do \
 	  case '$(am__configure_deps)' in \
@@ -192,14 +284,154 @@ $(top_srcdir)/configure:  $(am__configure_deps)
 $(ACLOCAL_M4):  $(am__aclocal_m4_deps)
 	cd $(top_builddir) && $(MAKE) $(AM_MAKEFLAGS) am--refresh
 $(am__aclocal_m4_deps):
+install-libLTLIBRARIES: $(lib_LTLIBRARIES)
+	@$(NORMAL_INSTALL)
+	@list='$(lib_LTLIBRARIES)'; test -n "$(libdir)" || list=; \
+	list2=; for p in $$list; do \
+	  if test -f $$p; then \
+	    list2="$$list2 $$p"; \
+	  else :; fi; \
+	done; \
+	test -z "$$list2" || { \
+	  echo " $(MKDIR_P) '$(DESTDIR)$(libdir)'"; \
+	  $(MKDIR_P) "$(DESTDIR)$(libdir)" || exit 1; \
+	  echo " $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=install $(INSTALL) $(INSTALL_STRIP_FLAG) $$list2 '$(DESTDIR)$(libdir)'"; \
+	  $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=install $(INSTALL) $(INSTALL_STRIP_FLAG) $$list2 "$(DESTDIR)$(libdir)"; \
+	}
+
+uninstall-libLTLIBRARIES:
+	@$(NORMAL_UNINSTALL)
+	@list='$(lib_LTLIBRARIES)'; test -n "$(libdir)" || list=; \
+	for p in $$list; do \
+	  $(am__strip_dir) \
+	  echo " $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=uninstall rm -f '$(DESTDIR)$(libdir)/$$f'"; \
+	  $(LIBTOOL) $(AM_LIBTOOLFLAGS) $(LIBTOOLFLAGS) --mode=uninstall rm -f "$(DESTDIR)$(libdir)/$$f"; \
+	done
+
+clean-libLTLIBRARIES:
+	-test -z "$(lib_LTLIBRARIES)" || rm -f $(lib_LTLIBRARIES)
+	@list='$(lib_LTLIBRARIES)'; \
+	locs=`for p in $$list; do echo $$p; done | \
+	      sed 's|^[^/]*$$|.|; s|/[^/]*$$||; s|$$|/so_locations|' | \
+	      sort -u`; \
+	test -z "$$locs" || { \
+	  echo rm -f $${locs}; \
+	  rm -f $${locs}; \
+	}
+libffts.la: $(libffts_la_OBJECTS) $(libffts_la_DEPENDENCIES) $(EXTRA_libffts_la_DEPENDENCIES) 
+	$(LINK) -rpath $(libdir) $(libffts_la_OBJECTS) $(libffts_la_LIBADD) $(LIBS)
+
+mostlyclean-compile:
+	-rm -f *.$(OBJEXT)
+
+distclean-compile:
+	-rm -f *.tab.c
+
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/codegen.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/cp_sse.Plo@am__quote@
+@AMDEP_TRUE@@am__include@ @am__quote@./$(DEPDIR)/patterns.Plo@am__quote@
+
+.c.o:
+@am__fastdepCC_TRUE@	$(COMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
+@am__fastdepCC_TRUE@	$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(COMPILE) -c $<
+
+.c.obj:
+@am__fastdepCC_TRUE@	$(COMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ `$(CYGPATH_W) '$<'`
+@am__fastdepCC_TRUE@	$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Po
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='$<' object='$@' libtool=no @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(COMPILE) -c `$(CYGPATH_W) '$<'`
+
+.c.lo:
+@am__fastdepCC_TRUE@	$(LTCOMPILE) -MT $@ -MD -MP -MF $(DEPDIR)/$*.Tpo -c -o $@ $<
+@am__fastdepCC_TRUE@	$(am__mv) $(DEPDIR)/$*.Tpo $(DEPDIR)/$*.Plo
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	source='$<' object='$@' libtool=yes @AMDEPBACKSLASH@
+@AMDEP_TRUE@@am__fastdepCC_FALSE@	DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@
+@am__fastdepCC_FALSE@	$(LTCOMPILE) -c -o $@ $<
+
+.s.o:
+	$(CCASCOMPILE) -c -o $@ $<
+
+.s.obj:
+	$(CCASCOMPILE) -c -o $@ `$(CYGPATH_W) '$<'`
+
+.s.lo:
+	$(LTCCASCOMPILE) -c -o $@ $<
+
+mostlyclean-libtool:
+	-rm -f *.lo
+
+clean-libtool:
+	-rm -rf .libs _libs
+
+ID: $(HEADERS) $(SOURCES) $(LISP) $(TAGS_FILES)
+	list='$(SOURCES) $(HEADERS) $(LISP) $(TAGS_FILES)'; \
+	unique=`for i in $$list; do \
+	    if test -f "$$i"; then echo $$i; else echo $(srcdir)/$$i; fi; \
+	  done | \
+	  $(AWK) '{ files[$$0] = 1; nonempty = 1; } \
+	      END { if (nonempty) { for (i in files) print i; }; }'`; \
+	mkid -fID $$unique
 tags: TAGS
-TAGS:
 
+TAGS:  $(HEADERS) $(SOURCES)  $(TAGS_DEPENDENCIES) \
+		$(TAGS_FILES) $(LISP)
+	set x; \
+	here=`pwd`; \
+	list='$(SOURCES) $(HEADERS)  $(LISP) $(TAGS_FILES)'; \
+	unique=`for i in $$list; do \
+	    if test -f "$$i"; then echo $$i; else echo $(srcdir)/$$i; fi; \
+	  done | \
+	  $(AWK) '{ files[$$0] = 1; nonempty = 1; } \
+	      END { if (nonempty) { for (i in files) print i; }; }'`; \
+	shift; \
+	if test -z "$(ETAGS_ARGS)$$*$$unique"; then :; else \
+	  test -n "$$unique" || unique=$$empty_fix; \
+	  if test $$# -gt 0; then \
+	    $(ETAGS) $(ETAGSFLAGS) $(AM_ETAGSFLAGS) $(ETAGS_ARGS) \
+	      "$$@" $$unique; \
+	  else \
+	    $(ETAGS) $(ETAGSFLAGS) $(AM_ETAGSFLAGS) $(ETAGS_ARGS) \
+	      $$unique; \
+	  fi; \
+	fi
 ctags: CTAGS
-CTAGS:
-
-cscope cscopelist:
+CTAGS:  $(HEADERS) $(SOURCES)  $(TAGS_DEPENDENCIES) \
+		$(TAGS_FILES) $(LISP)
+	list='$(SOURCES) $(HEADERS)  $(LISP) $(TAGS_FILES)'; \
+	unique=`for i in $$list; do \
+	    if test -f "$$i"; then echo $$i; else echo $(srcdir)/$$i; fi; \
+	  done | \
+	  $(AWK) '{ files[$$0] = 1; nonempty = 1; } \
+	      END { if (nonempty) { for (i in files) print i; }; }'`; \
+	test -z "$(CTAGS_ARGS)$$unique" \
+	  || $(CTAGS) $(CTAGSFLAGS) $(AM_CTAGSFLAGS) $(CTAGS_ARGS) \
+	     $$unique
+
+GTAGS:
+	here=`$(am__cd) $(top_builddir) && pwd` \
+	  && $(am__cd) $(top_srcdir) \
+	  && gtags -i $(GTAGS_ARGS) "$$here"
+
+cscopelist:  $(HEADERS) $(SOURCES) $(LISP)
+	list='$(SOURCES) $(HEADERS) $(LISP)'; \
+	case "$(srcdir)" in \
+	  [\\/]* | ?:[\\/]*) sdir="$(srcdir)" ;; \
+	  *) sdir=$(subdir)/$(srcdir) ;; \
+	esac; \
+	for i in $$list; do \
+	  if test -f "$$i"; then \
+	    echo "$(subdir)/$$i"; \
+	  else \
+	    echo "$$sdir/$$i"; \
+	  fi; \
+	done >> $(top_builddir)/cscope.files
 
+distclean-tags:
+	-rm -f TAGS ID GTAGS GRTAGS GSYMS GPATH tags
 
 distdir: $(DISTFILES)
 	@srcdirstrip=`echo "$(srcdir)" | sed 's/[].[^$$\\*]/\\\\&/g'`; \
@@ -233,8 +465,11 @@ distdir: $(DISTFILES)
 	done
 check-am: all-am
 check: check-am
-all-am: Makefile
+all-am: Makefile $(LTLIBRARIES)
 installdirs:
+	for dir in "$(DESTDIR)$(libdir)"; do \
+	  test -z "$$dir" || $(MKDIR_P) "$$dir"; \
+	done
 install: install-am
 install-exec: install-exec-am
 install-data: install-data-am
@@ -265,11 +500,16 @@ distclean-generic:
 maintainer-clean-generic:
 	@echo "This command is intended for maintainers to use"
 	@echo "it deletes files that may require special tools to rebuild."
-clean-am: clean-generic mostlyclean-am
+clean: clean-am
+
+clean-am: clean-generic clean-libLTLIBRARIES clean-libtool \
+	mostlyclean-am
 
 distclean: distclean-am
+	-rm -rf ./$(DEPDIR)
 	-rm -f Makefile
-distclean-am: clean-am distclean-generic
+distclean-am: clean-am distclean-compile distclean-generic \
+	distclean-tags
 
 dvi: dvi-am
 
@@ -289,7 +529,7 @@ install-dvi: install-dvi-am
 
 install-dvi-am:
 
-install-exec-am:
+install-exec-am: install-libLTLIBRARIES
 
 install-html: install-html-am
 
@@ -312,12 +552,14 @@ install-ps-am:
 installcheck-am:
 
 maintainer-clean: maintainer-clean-am
+	-rm -rf ./$(DEPDIR)
 	-rm -f Makefile
 maintainer-clean-am: distclean-am maintainer-clean-generic
 
 mostlyclean: mostlyclean-am
 
-mostlyclean-am: mostlyclean-generic
+mostlyclean-am: mostlyclean-compile mostlyclean-generic \
+	mostlyclean-libtool
 
 pdf: pdf-am
 
@@ -327,35 +569,24 @@ ps: ps-am
 
 ps-am:
 
-uninstall-am:
+uninstall-am: uninstall-libLTLIBRARIES
 
 .MAKE: install-am install-strip
 
-.PHONY: all all-am check check-am clean clean-generic distclean \
-	distclean-generic distdir dvi dvi-am html html-am info info-am \
+.PHONY: CTAGS GTAGS all all-am check check-am clean clean-generic \
+	clean-libLTLIBRARIES clean-libtool cscopelist ctags distclean \
+	distclean-compile distclean-generic distclean-libtool \
+	distclean-tags distdir dvi dvi-am html html-am info info-am \
 	install install-am install-data install-data-am install-dvi \
 	install-dvi-am install-exec install-exec-am install-html \
-	install-html-am install-info install-info-am install-man \
-	install-pdf install-pdf-am install-ps install-ps-am \
-	install-strip installcheck installcheck-am installdirs \
-	maintainer-clean maintainer-clean-generic mostlyclean \
-	mostlyclean-generic pdf pdf-am ps ps-am uninstall uninstall-am
-
-
-all: $(OBJLIBS)
-
-%.o: %.c $(HDRS)
-	$(CC) $(CFLAGS) -c -o $@ $< -I../include -g
-	$(CC) $(CFLAGS) -S $< -I../include
-
-%.o: %.s 
-	$(CC) $(CFLAGS) -c -o $@ $< -I../include -g
-
-$(OBJLIBS): $(OBJS) 
-	$(AR) rcs libffts.a $(OBJS)
+	install-html-am install-info install-info-am \
+	install-libLTLIBRARIES install-man install-pdf install-pdf-am \
+	install-ps install-ps-am install-strip installcheck \
+	installcheck-am installdirs maintainer-clean \
+	maintainer-clean-generic mostlyclean mostlyclean-compile \
+	mostlyclean-generic mostlyclean-libtool pdf pdf-am ps ps-am \
+	tags uninstall uninstall-am uninstall-libLTLIBRARIES
 
-clean:
-	$(RM) -f *.o $(OBJLIBS)
 
 # 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.
diff --git a/src/codegen.c b/src/codegen.c
index 681414f..68d3b3d 100644
--- a/src/codegen.c
+++ b/src/codegen.c
@@ -1,10 +1,19 @@
 #include "codegen.h"
 #include "macros.h"
-#include "neon_float.h"
 //#include "mini_macros.h"
-#include "neon.h"
 #include "cp_sse.h"
 #include <libkern/OSCacheControl.h>
+#include <sys/types.h>
+#include <sys/mman.h>
+
+#ifdef __ARM_NEON__
+	#include "codegen_neon.h"
+	#include "neon_float.h"
+	#include "neon.h"
+#else
+	#include "codegen_sse.h"
+	#include "sse_float.h"
+#endif
 
 int tree_count(int N, int leafN, int offset) {
 	
@@ -35,69 +44,7 @@ void elaborate_tree(size_t **p, int N, int leafN, int offset) {
 }
 
 
-void neon_x4_x(data_t * restrict data, const data_t * restrict LUT) {
-	X_8_SPLIT(data, 64, LUT);
-}
-
-
-uint32_t BL(void *pos, void *target) {
-	return 0xeb000000 | (((target - pos) / 4) & 0xffffff);
-}
-
-uint32_t B(uint8_t r) {
-	return 0xe12fff10 | r;
-}
-
-uint32_t MOV(uint8_t dst, uint8_t src) {
-		return 0xe1a00000 | (src & 0xf) | ((dst & 0xf) << 12);
-}
-
-void ADDI(uint32_t **p, uint8_t dst, uint8_t src, int32_t imm) {
-	int32_t oimm = imm;
-	if(imm < 0) {
-		imm = -imm;
-		uint32_t shamt = (__builtin_ctzl(imm)>15)?15:__builtin_ctzl(imm);
-		if(shamt & 1) shamt -= 1;
-		imm >>= shamt;
-		shamt = (32 - shamt)/2;
-		
-	//	if(imm > 255) fprintf(stderr, "imm>255: %d\n", oimm);
-		*(*p)++ = 0xe2400000 | ((src & 0xf) << 16) | ((dst & 0xf) << 12) | ((shamt & 0xf) << 8) | (imm & 0xff);
-	
-		if(imm > 255) ADDI(p, dst, src, (oimm + ((imm & 0xff) << (32-shamt*2))));
-
-	}else{
-		uint32_t shamt = (__builtin_ctzl(imm)>15)?15:__builtin_ctzl(imm);
-		if(shamt & 1) shamt -= 1;
-		imm >>= shamt;
-		shamt = (32 - shamt)/2;
-
-//		if(imm > 255) fprintf(stderr, "imm>255: %d\n", oimm);
-
-		*(*p)++ = 0xe2800000 | ((src & 0xf) << 16) | ((dst & 0xf) << 12) | ((shamt & 0xf) << 8) | (imm & 0xff);
-		
-		if(imm > 255) ADDI(p, dst, src, (oimm - ((imm & 0xff) << (32-shamt*2))));
-	}
-}
-
-uint32_t LDRI(uint8_t dst, uint8_t base, uint32_t offset) {
-	return 0xe5900000 | ((dst & 0xf) << 12)
-	                  | ((base & 0xf) << 16) | (offset & 0xfff) ;
-}
 
-uint32_t MOVI(uint32_t **p, uint8_t dst, uint32_t imm) {
-	uint32_t oimm = imm;
-	
-		uint32_t shamt = (__builtin_ctzl(imm)>15)?15:__builtin_ctzl(imm);
-		if(shamt & 1) shamt -= 1;
-		imm >>= shamt;
-		shamt = (32 - shamt)/2;
-	*(*p)++ = 0xe3a00000 | ((dst & 0xf) << 12) | ((shamt & 0xf) << 8) | (imm & 0xff) ;
-		if(imm > 255) ADDI(p, dst, dst, (oimm - ((imm & 0xff) << (32-shamt*2))));
-}
-
-uint32_t PUSH_LR() { return 0xe92d4ff0; } //0xe92d4000; }
-uint32_t POP_LR() { return 0xe8bd8ff0; } //0xe8bd8000; }
 
 uint32_t LUT_offset(size_t N, size_t leafN) {
 		int i;
@@ -130,7 +77,13 @@ uint32_t LUT_offset(size_t N, size_t leafN) {
 	return lut_size;
 }
 
-transform_func_t ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN) {
+#ifdef __ARM_NEON__
+	typedef uint32_t insns_t;
+#else
+	typedef uint8_t insns_t;
+#endif
+
+void ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN) {
 	int count = tree_count(N, leafN, 0) + 1;
 	size_t *ps = malloc(count * 2 * sizeof(size_t));
 	size_t *pps = ps;
@@ -144,29 +97,37 @@ transform_func_t ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN)
 	if(N < 8192) p->transform_size = 8192;
 	else p->transform_size = N;
 
-	p->transform_base = valloc(p->transform_size);//(void *)func;
-	uint32_t *func = p->transform_base;//valloc(8192);
-	uint32_t *fp = func;
+	p->transform_base = valloc(p->transform_size);//mmap(NULL, p->transform_size, PROT_WRITE | PROT_READ, MAP_ANON | MAP_SHARED, -1, 0);
+	/*
+	if(p->transform_base == MAP_FAILED) {
+		fprintf(stderr, "MAP FAILED\n");
+		exit(1);
+	}*/
+
+	insns_t *func = p->transform_base;//valloc(8192);
+	insns_t *fp = func;
 
 	fprintf(stderr, "Allocating %d bytes \n", p->transform_size);
+	fprintf(stderr, "Base address = %016p\n", func);
 
 	if(!func) { 
 		fprintf(stderr, "NOMEM\n");
 		exit(1);
 	}
 
-	uint32_t *x_8_addr = fp;
+	insns_t *x_8_addr = fp;
 	memcpy(fp, neon_x8, neon_x8_t - neon_x8);
 	fp += (neon_x8_t - neon_x8) / 4;
 //uint32_t *x_8_t_addr = fp;
 //memcpy(fp, neon_x8_t, neon_end - neon_x8_t);
 //fp += (neon_end - neon_x8_t) / 4;
-	uint32_t *x_4_addr = fp;
+	insns_t *x_4_addr = fp;
 	memcpy(fp, neon_x4, neon_x8 - neon_x4);
 	fp += (neon_x8 - neon_x4) / 4;
 
-	uint32_t *start = fp;
+	insns_t *start = fp;
 
+#ifdef __ARM_NEON__
 	*fp++ = PUSH_LR();
 	
 	ADDI(&fp, 3, 1, 0);
@@ -183,20 +144,40 @@ transform_func_t ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN)
 	ADDI(&fp, 1, 0, 0);
 
 	ADDI(&fp, 0, 2, 0), // mov out into r0
+#endif
 
 	p->oe_ws = oe_w_data;
 	p->ee_ws = ee_w_data;
 	p->eo_ws = eo_w_data;
 
 
-
+#ifdef __ARM_NEON__
 	*fp++ = LDRI(2, 1, ((uint32_t)&p->ee_ws) - ((uint32_t)p)); 
 	MOVI(&fp, 11, p->i0);
+#else
 	
+	*fp++ = 0x4c;
+	*fp++ = 0x8b;
+	*fp++ = 0x07;
+	MOVI(&fp, RCX, p->i0 * 4);
+	//LEA(&fp, R8, RDI, ((uint32_t)&p->offsets) - ((uint32_t)p)); 
+#endif
 	//fp++;
-	memcpy(fp, neon_ee, neon_oo - neon_ee);
-	fp += (neon_oo - neon_ee) / 4;
+	memcpy(fp, leaf_ee, neon_oo - leaf_ee);
+#ifdef __ARM_NEON__
+	fp += (neon_oo - leaf_ee) / 4;
+#else
+	int i;
+
+	IMM32_NI(fp + 3, READ_IMM32(fp + 3) + ((void *)leaf_ee - (void *)fp )); 
+
+	uint32_t offsets[8] = {0, N, N/2, 3*N/2, N/4, 5*N/4, 7*N/4, 3*N/4};
 	
+	for(i=0;i<8;i++) IMM32_NI(fp + sse_leaf_ee_offsets[i], offsets[i]*4); 
+	
+	fp += (neon_oo - leaf_ee);
+#endif
+#ifdef __ARM_NEON__
 	if(__builtin_ctzl(N) & 1){
 		ADDI(&fp, 2, 7, 0);
 		ADDI(&fp, 7, 9, 0);
@@ -316,6 +297,23 @@ transform_func_t ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN)
 	}
 	
 	*fp++ = POP_LR(); count++;
+#else
+	RET(&fp);	
+
+
+	uint8_t *pp = func;
+	int counter = 0;
+	do{ 
+		printf("%02x ", *pp);
+		if(counter++ % 16 == 15) printf("\n");
+	} while(++pp < fp);
+
+	printf("\n");
+
+
+#endif
+
+
 //	*fp++ = B(14); count++;
 
 //for(int i=0;i<(neon_x8 - neon_x4)/4;i++) 
@@ -327,7 +325,7 @@ transform_func_t ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN)
 	
 	if (mprotect(func, p->transform_size, PROT_READ | PROT_EXEC)) {
 		perror("Couldn't mprotect");
-		return NULL;
+		exit(1);
 	}
 
 	sys_icache_invalidate(func, p->transform_size);
@@ -335,5 +333,5 @@ transform_func_t ffts_generate_func_code(ffts_plan_t *p, size_t N, size_t leafN)
 
 	fprintf(stderr, "size of transform %zu = %d\n", N, (fp-func)*4);
 
-	return (transform_func_t)start;
+	p->transform = start;
 }
diff --git a/src/codegen.h b/src/codegen.h
index f887ca0..a41735b 100644
--- a/src/codegen.h
+++ b/src/codegen.h
@@ -13,7 +13,7 @@ typedef struct _ffts_plan_t ffts_plan_t;
 
 typedef void (*transform_func_t)(float *data, size_t N, float *LUT);
 
-transform_func_t ffts_generate_func_code(ffts_plan_t *, size_t N, size_t leafN); 
+void ffts_generate_func_code(ffts_plan_t *, size_t N, size_t leafN); 
 
 //static const __attribute__ ((aligned(16))) float ee_w_data[4] = 
 //{0.70710678118654757273731092936941,
diff --git a/src/codegen_neon.h b/src/codegen_neon.h
new file mode 100644
index 0000000..b14cb1c
--- /dev/null
+++ b/src/codegen_neon.h
@@ -0,0 +1,68 @@
+#ifndef __CODEGEN_NEON_H__
+#define __CODEGEN_NEON_H__
+
+
+
+uint32_t BL(void *pos, void *target) {
+	return 0xeb000000 | (((target - pos) / 4) & 0xffffff);
+}
+
+uint32_t B(uint8_t r) {
+	return 0xe12fff10 | r;
+}
+
+uint32_t MOV(uint8_t dst, uint8_t src) {
+		return 0xe1a00000 | (src & 0xf) | ((dst & 0xf) << 12);
+}
+
+void ADDI(uint32_t **p, uint8_t dst, uint8_t src, int32_t imm) {
+	int32_t oimm = imm;
+	if(imm < 0) {
+		imm = -imm;
+		uint32_t shamt = (__builtin_ctzl(imm)>15)?15:__builtin_ctzl(imm);
+		if(shamt & 1) shamt -= 1;
+		imm >>= shamt;
+		shamt = (32 - shamt)/2;
+		
+	//	if(imm > 255) fprintf(stderr, "imm>255: %d\n", oimm);
+		*(*p)++ = 0xe2400000 | ((src & 0xf) << 16) | ((dst & 0xf) << 12) | ((shamt & 0xf) << 8) | (imm & 0xff);
+	
+		if(imm > 255) ADDI(p, dst, src, (oimm + ((imm & 0xff) << (32-shamt*2))));
+
+	}else{
+		uint32_t shamt = (__builtin_ctzl(imm)>15)?15:__builtin_ctzl(imm);
+		if(shamt & 1) shamt -= 1;
+		imm >>= shamt;
+		shamt = (32 - shamt)/2;
+
+//		if(imm > 255) fprintf(stderr, "imm>255: %d\n", oimm);
+
+		*(*p)++ = 0xe2800000 | ((src & 0xf) << 16) | ((dst & 0xf) << 12) | ((shamt & 0xf) << 8) | (imm & 0xff);
+		
+		if(imm > 255) ADDI(p, dst, src, (oimm - ((imm & 0xff) << (32-shamt*2))));
+	}
+}
+
+uint32_t LDRI(uint8_t dst, uint8_t base, uint32_t offset) {
+	return 0xe5900000 | ((dst & 0xf) << 12)
+	                  | ((base & 0xf) << 16) | (offset & 0xfff) ;
+}
+
+uint32_t MOVI(uint32_t **p, uint8_t dst, uint32_t imm) {
+	uint32_t oimm = imm;
+	
+		uint32_t shamt = (__builtin_ctzl(imm)>15)?15:__builtin_ctzl(imm);
+		if(shamt & 1) shamt -= 1;
+		imm >>= shamt;
+		shamt = (32 - shamt)/2;
+	*(*p)++ = 0xe3a00000 | ((dst & 0xf) << 12) | ((shamt & 0xf) << 8) | (imm & 0xff) ;
+		if(imm > 255) ADDI(p, dst, dst, (oimm - ((imm & 0xff) << (32-shamt*2))));
+}
+
+uint32_t PUSH_LR() { return 0xe92d4ff0; } //0xe92d4000; }
+uint32_t POP_LR() { return 0xe8bd8ff0; } //0xe8bd8000; }
+
+
+
+
+#endif
diff --git a/src/codegen_sse.h b/src/codegen_sse.h
new file mode 100644
index 0000000..d3b136a
--- /dev/null
+++ b/src/codegen_sse.h
@@ -0,0 +1,104 @@
+#ifndef __CODEGEN_SSE_H__
+#define __CODEGEN_SSE_H__
+
+static const __attribute__ ((aligned(16))) float ee_w_data[4] = {0.70710678118654757273731092936941,0.70710678118654746171500846685376,
+    -0.70710678118654757273731092936941,-0.70710678118654746171500846685376};
+static const __attribute__ ((aligned(16))) data_t oe_w_data[4] = {1.0f,0.70710678118654757273731092936941f, 0.0f,-0.70710678118654746171500846685376};
+static const __attribute__ ((aligned(16))) data_t eo_w_data[4] = {1.0f,0.70710678118654757273731092936941f, 0.0f,-0.70710678118654746171500846685376};
+void neon_x4(float *, size_t, float *);
+void neon_x8(float *, size_t, float *);
+void neon_x8_t(float *, size_t, float *);
+void leaf_ee();
+void neon_oo();
+void neon_eo();
+void neon_oe();
+void neon_end();
+
+
+extern const uint32_t sse_leaf_ee_offsets[8];
+
+#define EAX 0
+#define ECX 1
+#define EDX 2
+#define EBX 3
+#define ESI 6
+#define EDI 7
+#define EBP 5
+
+#define RAX 0
+#define RCX 1
+#define RDX 2
+#define RBX 3
+#define RSI 6
+#define RDI 7
+#define RBP 5
+#define R8 8 
+#define R9 9 
+#define R10 10 
+#define R11 11 
+#define R12 12 
+#define R13 13 
+#define R14 14 
+#define R15 15 
+
+void IMM8(uint8_t **p, uint32_t imm) {
+		*(*p)++ = (imm & 0xff);
+}
+
+void IMM32(uint8_t **p, uint32_t imm) {
+	int i;
+	for(i=0;i<4;i++) {
+		*(*p)++ = (imm & (0xff << (i*8))) >> (i*8);
+	} 
+}
+void IMM32_NI(uint8_t *p, uint32_t imm) {
+	int i;
+	for(i=0;i<4;i++) {
+		*(p+i) = (imm & (0xff << (i*8))) >> (i*8);
+	} 
+}
+
+uint32_t READ_IMM32(uint8_t *p) {
+	uint32_t rval = 0;
+	int i;
+	for(i=0;i<4;i++) {
+		rval |= *(p+i) << (i*8);
+	} 
+	return rval;
+}
+
+void MOVI(uint8_t **p, uint8_t dst, uint32_t imm) {
+	if(dst < 8) {
+		*(*p)++ = 0xb8 + dst; 
+	}else{
+		*(*p)++ = 0x49;
+		*(*p)++ = 0xc7;
+		*(*p)++ = 0xc0 | (dst - 8);
+	}
+	IMM32(p, imm);
+}
+
+void ADDRMODE(uint8_t **p, uint8_t reg, uint8_t rm, int32_t disp) {
+	if(disp == 0) {
+		*(*p)++ = (rm & 7) | ((reg & 7) << 3);
+	}else if(disp <= 127 || disp >= -128) {
+		*(*p)++ = 0x40 | (rm & 7) | ((reg & 7) << 3);
+		IMM8(p, disp);
+	}else{
+		*(*p)++ = 0x80 | (rm & 7) | ((reg & 7) << 3);
+		IMM32(p, disp);
+	}
+}
+
+void LEA(uint8_t **p, uint8_t dst, uint8_t base, int32_t disp) {
+
+	*(*p)++ = 0x48 | ((base & 0x8) >> 3) | ((dst & 0x8) >> 1);
+	*(*p)++ = 0x8d;
+	ADDRMODE(p, dst, base, disp);		
+}
+
+void RET(uint8_t **p) {
+	*(*p)++ = 0xc3;
+}
+
+#endif
diff --git a/src/cp_sse.c b/src/cp_sse.c
index 088798b..f9a5c8f 100644
--- a/src/cp_sse.c
+++ b/src/cp_sse.c
@@ -308,8 +308,7 @@ ffts_plan_t *ffts_init(size_t N, int sign) {
 	p->N = N;
 	p->lastlut = w;
 	p->n_luts = n_luts;
-	if(N>=32)  
-		p->transform = ffts_generate_func_code(p, N, leafN);
+	if(N>=32)  ffts_generate_func_code(p, N, leafN);
 //	fprintf(stderr, "sizeof(size_t) == %lu\n", sizeof(size_t));
 
 	return p;
diff --git a/src/neon.h b/src/neon.h
new file mode 100644
index 0000000..16cb38d
--- /dev/null
+++ b/src/neon.h
@@ -0,0 +1,17 @@
+#ifndef __NEON_H__
+#define __NEON_H__
+
+static const __attribute__ ((aligned(16))) float ee_w_data[4] = {0.70710678118654757273731092936941,0.70710678118654746171500846685376,
+    -0.70710678118654757273731092936941,-0.70710678118654746171500846685376};
+static const __attribute__ ((aligned(16))) data_t oe_w_data[4] = {1.0f,0.70710678118654757273731092936941f, 0.0f,-0.70710678118654746171500846685376};
+static const __attribute__ ((aligned(16))) data_t eo_w_data[4] = {1.0f,0.70710678118654757273731092936941f, 0.0f,-0.70710678118654746171500846685376};
+void neon_x4(float *, size_t, float *);
+void neon_x8(float *, size_t, float *);
+void neon_x8_t(float *, size_t, float *);
+void neon_ee();
+void neon_oo();
+void neon_eo();
+void neon_oe();
+void neon_end();
+
+#endif
diff --git a/src/patterns.c b/src/patterns.c
index 664f20e..29fa5ae 100644
--- a/src/patterns.c
+++ b/src/patterns.c
@@ -114,9 +114,9 @@ void ffts_init_offsets(ffts_plan_t *p, int N, int leafN) {
 	for(i=0;i<N/leafN;i++) {
 		p->offsets[i] = offsets[i*2+1]*2;
 	}
-//for(i=0;i<N/leafN;i++) {
-//	printf("%4d %4d\n", p->offsets[i], reverse_bits(p->offsets[i], __builtin_ctzl(2*N)));
-//}
+  for(i=0;i<N/leafN;i++) {
+  	printf("%4d %4d\n", p->offsets[i], reverse_bits(p->offsets[i], __builtin_ctzl(2*N)));
+  }
 	free(offsets);
 
 
diff --git a/src/sse.s b/src/sse.s
new file mode 100644
index 0000000..c8e509f
--- /dev/null
+++ b/src/sse.s
@@ -0,0 +1,174 @@
+
+	.globl	_neon_x4
+	.align	4
+_neon_x4:
+
+	.globl _neon_x8
+	.align 4
+_neon_x8:
+
+	.globl _neon_x8_t
+	.align 4
+_neon_x8_t:
+
+
+
+# eax is loop counter (init to 0)
+# rcx is loop max count
+# rsi is 'in' base pointer
+# rdx is 'out' base pointer
+# r8 is offsets pointer
+# r9 is constants pointer
+# scratch: rax r11 r12
+	.globl _leaf_ee
+	.align	4, 0x90
+_leaf_ee:
+ 		lea L_sse_constants(%rip), %r9 
+		movaps    32(%r9), %xmm0            #83.5
+ 		movaps    0x0(%r9), %xmm8            #83.5
+		xorl	%eax, %eax
+	.align	4, 0x90
+LEAF_EE_1:
+LEAF_EE_const_0:
+				movaps    0xBEBAFECA(%rsi,%rax,4), %xmm7                           #83.5
+LEAF_EE_const_2:
+        movaps    0xBEBAFECA(%rsi,%rax,4), %xmm12                         #83.5
+        movaps    %xmm7, %xmm6                                  #83.5
+LEAF_EE_const_3:
+        movaps    0xBEBAFECA(%rsi,%rax,4), %xmm10                         #83.5
+        movaps    %xmm12, %xmm11                                #83.5
+        subps     %xmm10, %xmm12                                #83.5
+        addps     %xmm10, %xmm11                                #83.5
+        xorps     %xmm8, %xmm12                                 #83.5
+LEAF_EE_const_1:
+        movaps    0xBEBAFECA(%rsi,%rax,4), %xmm9                          #83.5
+LEAF_EE_const_4:
+        movaps    0xBEBAFECA(%rsi,%rax,4), %xmm10                         #83.5
+        addps     %xmm9, %xmm6                                  #83.5
+        subps     %xmm9, %xmm7                                  #83.5
+LEAF_EE_const_5:
+        movaps    0xBEBAFECA(%rsi,%rax,4), %xmm13                         #83.5
+        movaps    %xmm10, %xmm9                                 #83.5
+LEAF_EE_const_6:
+        movaps    0xBEBAFECA(%rsi,%rax,4), %xmm3                          #83.5
+        movaps    %xmm6, %xmm5                                  #83.5
+LEAF_EE_const_7:
+        movaps    0xBEBAFECA(%rsi,%rax,4), %xmm14                          #83.5
+        movaps    %xmm3, %xmm15                                 #83.5
+        shufps    $177, %xmm12, %xmm12                          #83.5
+        movaps    %xmm7, %xmm4                                  #83.5
+        movslq    (%r8, %rax, 8), %r11                                   #83.44
+        subps     %xmm13, %xmm10                                #83.5
+        subps     %xmm14, %xmm3                                 #83.5
+        addps     %xmm11, %xmm5                                 #83.5
+        subps     %xmm11, %xmm6                                 #83.5
+        subps     %xmm12, %xmm4                                 #83.5
+        addps     %xmm12, %xmm7                                 #83.5
+        addps     %xmm13, %xmm9                                 #83.5
+        addps     %xmm14, %xmm15                                #83.5
+        movaps    16(%r9), %xmm12           #83.5
+        movaps    %xmm9, %xmm1                                  #83.5
+        movaps    16(%r9), %xmm11           #83.5
+        movaps    %xmm5, %xmm2                                  #83.5
+        mulps     %xmm10, %xmm12                                #83.5
+        subps     %xmm15, %xmm9                                 #83.5
+        addps     %xmm15, %xmm1                                 #83.5
+        mulps     %xmm3, %xmm11                                 #83.5
+        addps     %xmm1, %xmm2                                  #83.5
+        subps     %xmm1, %xmm5                                  #83.5
+        shufps    $177, %xmm10, %xmm10                          #83.5
+        xorps     %xmm8, %xmm9                                  #83.5
+        shufps    $177, %xmm3, %xmm3                            #83.5
+        movaps    %xmm6, %xmm1                                  #83.5
+        mulps     %xmm0, %xmm10                                 #83.5
+        movaps    %xmm4, %xmm13                                 #83.5
+        mulps     %xmm0, %xmm3                                  #83.5
+        subps     %xmm10, %xmm12                                #83.5
+        addps     %xmm3, %xmm11                                 #83.5
+        movaps    %xmm12, %xmm3                                 #83.5
+        movaps    %xmm7, %xmm14                                 #83.5
+        shufps    $177, %xmm9, %xmm9                            #83.5
+        subps     %xmm11, %xmm12                                #83.5
+        addps     %xmm11, %xmm3                                 #83.5
+        subps     %xmm9, %xmm1                                  #83.5
+        addps     %xmm9, %xmm6                                  #83.5
+        addps     %xmm3, %xmm4                                  #83.5
+        subps     %xmm3, %xmm13                                 #83.5
+        xorps     %xmm8, %xmm12                                 #83.5
+        movaps    %xmm2, %xmm3                                  #83.5
+        shufps    $177, %xmm12, %xmm12                          #83.5
+        movaps    %xmm6, %xmm9                                  #83.5
+        movslq    4(%r8, %rax, 8), %r12                                  #83.59
+        movlhps   %xmm4, %xmm3                                  #83.5
+				addq	    $4, %rax
+        shufps    $238, %xmm4, %xmm2                            #83.5
+        movaps    %xmm1, %xmm4                                  #83.5
+        movaps    %xmm3, (%rdx,%r11,4)                          #83.5
+        subps     %xmm12, %xmm7                                 #83.5
+        addps     %xmm12, %xmm14                                #83.5
+        movlhps   %xmm7, %xmm4                                  #83.5
+        shufps    $238, %xmm7, %xmm1                            #83.5
+        movaps    %xmm5, %xmm7                                  #83.5
+        movlhps   %xmm13, %xmm7                                 #83.5
+        movlhps   %xmm14, %xmm9                                 #83.5
+        shufps    $238, %xmm13, %xmm5                           #83.5
+        shufps    $238, %xmm14, %xmm6                           #83.5
+        movaps    %xmm4, 16(%rdx,%r11,4)                        #83.5
+        movaps    %xmm7, 32(%rdx,%r11,4)                        #83.5
+        movaps    %xmm9, 48(%rdx,%r11,4)                        #83.5
+        movaps    %xmm2, (%rdx,%r12,4)                          #83.5
+        movaps    %xmm1, 16(%rdx,%r12,4)                        #83.5
+        movaps    %xmm5, 32(%rdx,%r12,4)                        #83.5
+        movaps    %xmm6, 48(%rdx,%r12,4)                        #83.5
+				cmpq	%rcx, %rax
+        jne      LEAF_EE_1 
+
+	.globl _neon_oo
+	.align 4
+_neon_oo:
+
+
+	.globl _neon_eo
+	.align 4
+_neon_eo:
+	
+
+	.globl _neon_oe
+	.align 4
+_neon_oe:
+	
+	
+	.globl	_neon_end
+	.align	4
+_neon_end:
+	
+	
+	.globl _sse_leaf_ee_offsets
+	.align 4
+_sse_leaf_ee_offsets:
+	.long LEAF_EE_const_0-_leaf_ee+0x4
+	.long LEAF_EE_const_1-_leaf_ee+0x5
+	.long LEAF_EE_const_2-_leaf_ee+0x5
+	.long LEAF_EE_const_3-_leaf_ee+0x5
+	.long LEAF_EE_const_4-_leaf_ee+0x5
+	.long LEAF_EE_const_5-_leaf_ee+0x5
+	.long LEAF_EE_const_6-_leaf_ee+0x4
+	.long LEAF_EE_const_7-_leaf_ee+0x5
+	
+	.section	__TEXT, __const
+	.align 4
+L_sse_constants:
+L_2il0floatpacket.719:
+	.long	0x00000000,0x80000000,0x00000000,0x80000000
+	.align 4
+L_2il0floatpacket.720:
+	.long	0x3f3504f3,0x3f3504f3,0x3f3504f3,0x3f3504f3
+	.align 4
+L_2il0floatpacket.721:
+	.long	0xbf3504f3,0x3f3504f3,0xbf3504f3,0x3f3504f3
+	.align 4
+L_2il0floatpacket.722:
+	.long	0x3f800000,0x3f800000,0x3f3504f3,0x3f3504f3
+	.align 4
+L_2il0floatpacket.723:
+	.long	0x00000000,0x00000000,0xbf3504f3,0x3f3504f3