import files from glibc for an ia64/static port

51 files changed, 5209 insertions, 0 deletions

diff --git a/libc/string/ia64/Makefile b/libc/string/ia64/Makefile
new file mode 100644
index 000000000..0a95346fd
--- /dev/null
+++ b/libc/string/ia64/Makefile
@@ -0,0 +1,13 @@
+# Makefile for uClibc
+#
+# Copyright (C) 2000-2005 Erik Andersen <andersen@uclibc.org>
+#
+# Licensed under the LGPL v2.1, see the file COPYING.LIB in this tarball.
+#
+
+top_srcdir:=../../../
+top_builddir:=../../../
+all: objs
+include $(top_builddir)Rules.mak
+include ../Makefile.in
+include $(top_srcdir)Makerules
diff --git a/libc/string/ia64/bcopy.S b/libc/string/ia64/bcopy.S
new file mode 100644
index 000000000..c4eb22b1f
--- /dev/null
+++ b/libc/string/ia64/bcopy.S
@@ -0,0 +1,10 @@
+#include "sysdep.h"
+
+ENTRY(bcopy)
+	.regstk 3, 0, 0, 0
+	mov r8 = in0
+	mov in0 = in1
+	;;
+	mov in1 = r8
+	br.cond.sptk.many HIDDEN_JUMPTARGET(memmove)
+END(bcopy)
diff --git a/libc/string/ia64/bzero.S b/libc/string/ia64/bzero.S
new file mode 100644
index 000000000..bcca41d5e
--- /dev/null
+++ b/libc/string/ia64/bzero.S
@@ -0,0 +1,315 @@
+/* Optimized version of the standard bzero() function.
+   This file is part of the GNU C Library.
+   Copyright (C) 2000, 2001, 2002 Free Software Foundation, Inc.
+   Contributed by Dan Pop for Itanium <Dan.Pop@cern.ch>.
+   Rewritten for McKinley by Sverre Jarp, HP Labs/CERN <Sverre.Jarp@cern.ch>
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+/* Return: dest
+
+   Inputs:
+        in0:    dest
+        in1:    count
+
+   The algorithm is fairly straightforward: set byte by byte until we
+   we get to a 16B-aligned address, then loop on 128 B chunks using an
+   early store as prefetching, then loop on 32B chucks, then clear remaining
+   words, finally clear remaining bytes.
+   Since a stf.spill f0 can store 16B in one go, we use this instruction
+   to get peak speed.  */
+
+#include <sysdep.h>
+#undef ret
+
+#define dest		in0
+#define	cnt		in1
+
+#define tmp		r31
+#define save_lc		r30
+#define ptr0		r29
+#define ptr1		r28
+#define ptr2		r27
+#define ptr3		r26
+#define ptr9 		r24
+#define	loopcnt		r23
+#define linecnt		r22
+#define bytecnt		r21
+
+// This routine uses only scratch predicate registers (p6 - p15)
+#define p_scr		p6	// default register for same-cycle branches
+#define p_unalgn	p9
+#define p_y		p11
+#define p_n		p12
+#define p_yy		p13
+#define p_nn		p14
+
+#define movi0		mov
+
+#define MIN1		15
+#define MIN1P1HALF	8
+#define LINE_SIZE	128
+#define LSIZE_SH        7			// shift amount
+#define PREF_AHEAD	8
+
+#define USE_FLP
+#if defined(USE_INT)
+#define store		st8
+#define myval		r0
+#elif defined(USE_FLP)
+#define store		stf8
+#define myval		f0
+#endif
+
+.align	64
+ENTRY(bzero)
+{ .mmi
+	.prologue
+	alloc	tmp = ar.pfs, 2, 0, 0, 0
+	lfetch.nt1 [dest]
+	.save   ar.lc, save_lc
+	movi0	save_lc = ar.lc
+} { .mmi
+	.body
+	mov	ret0 = dest		// return value
+	nop.m	0
+	cmp.eq	p_scr, p0 = cnt, r0
+;; }
+{ .mmi
+	and	ptr2 = -(MIN1+1), dest	// aligned address
+	and	tmp = MIN1, dest	// prepare to check for alignment
+	tbit.nz p_y, p_n = dest, 0	// Do we have an odd address? (M_B_U)
+} { .mib
+	mov	ptr1 = dest
+	nop.i	0
+(p_scr)	br.ret.dpnt.many rp		// return immediately if count = 0
+;; }
+{ .mib
+	cmp.ne	p_unalgn, p0 = tmp, r0
+} { .mib					// NB: # of bytes to move is 1
+	sub	bytecnt = (MIN1+1), tmp		//     higher than loopcnt
+	cmp.gt	p_scr, p0 = 16, cnt		// is it a minimalistic task?
+(p_scr)	br.cond.dptk.many .move_bytes_unaligned	// go move just a few (M_B_U)
+;; }
+{ .mmi
+(p_unalgn) add	ptr1 = (MIN1+1), ptr2		// after alignment
+(p_unalgn) add	ptr2 = MIN1P1HALF, ptr2		// after alignment
+(p_unalgn) tbit.nz.unc p_y, p_n = bytecnt, 3	// should we do a st8 ?
+;; }
+{ .mib
+(p_y)	add	cnt = -8, cnt
+(p_unalgn) tbit.nz.unc p_yy, p_nn = bytecnt, 2	// should we do a st4 ?
+} { .mib
+(p_y)	st8	[ptr2] = r0,-4
+(p_n)	add	ptr2 = 4, ptr2
+;; }
+{ .mib
+(p_yy)	add	cnt = -4, cnt
+(p_unalgn) tbit.nz.unc p_y, p_n = bytecnt, 1	// should we do a st2 ?
+} { .mib
+(p_yy)	st4	[ptr2] = r0,-2
+(p_nn)	add	ptr2 = 2, ptr2
+;; }
+{ .mmi
+	mov	tmp = LINE_SIZE+1		// for compare
+(p_y)	add	cnt = -2, cnt
+(p_unalgn) tbit.nz.unc p_yy, p_nn = bytecnt, 0	// should we do a st1 ?
+} { .mmi
+	nop.m	0
+(p_y)	st2	[ptr2] = r0,-1
+(p_n)	add	ptr2 = 1, ptr2
+;; }
+
+{ .mmi
+(p_yy)	st1	[ptr2] = r0
+  	cmp.gt	p_scr, p0 = tmp, cnt		// is it a minimalistic task?
+} { .mbb
+(p_yy)	add	cnt = -1, cnt
+(p_scr)	br.cond.dpnt.many .fraction_of_line	// go move just a few
+;; }
+{ .mib
+	nop.m 	0
+	shr.u	linecnt = cnt, LSIZE_SH
+	nop.b	0
+;; }
+
+	.align 32
+.l1b:	// ------------------//  L1B: store ahead into cache lines; fill later
+{ .mmi
+	and	tmp = -(LINE_SIZE), cnt		// compute end of range
+	mov	ptr9 = ptr1			// used for prefetching
+	and	cnt = (LINE_SIZE-1), cnt	// remainder
+} { .mmi
+	mov	loopcnt = PREF_AHEAD-1		// default prefetch loop
+	cmp.gt	p_scr, p0 = PREF_AHEAD, linecnt	// check against actual value
+;; }
+{ .mmi
+(p_scr)	add	loopcnt = -1, linecnt
+	add	ptr2 = 16, ptr1	// start of stores (beyond prefetch stores)
+	add	ptr1 = tmp, ptr1	// first address beyond total range
+;; }
+{ .mmi
+	add	tmp = -1, linecnt	// next loop count
+	movi0	ar.lc = loopcnt
+;; }
+.pref_l1b:
+{ .mib
+	stf.spill [ptr9] = f0, 128	// Do stores one cache line apart
+	nop.i   0
+	br.cloop.dptk.few .pref_l1b
+;; }
+{ .mmi
+	add	ptr0 = 16, ptr2		// Two stores in parallel
+	movi0	ar.lc = tmp
+;; }
+.l1bx:
+ { .mmi
+	stf.spill [ptr2] = f0, 32
+	stf.spill [ptr0] = f0, 32
+ ;; }
+ { .mmi
+	stf.spill [ptr2] = f0, 32
+	stf.spill [ptr0] = f0, 32
+ ;; }
+ { .mmi
+	stf.spill [ptr2] = f0, 32
+	stf.spill [ptr0] = f0, 64
+ 	cmp.lt	p_scr, p0 = ptr9, ptr1	// do we need more prefetching?
+ ;; }
+{ .mmb
+	stf.spill [ptr2] = f0, 32
+(p_scr)	stf.spill [ptr9] = f0, 128
+	br.cloop.dptk.few .l1bx
+;; }
+{ .mib
+	cmp.gt  p_scr, p0 = 8, cnt	// just a few bytes left ?
+(p_scr)	br.cond.dpnt.many  .move_bytes_from_alignment
+;; }
+
+.fraction_of_line:
+{ .mib
+	add	ptr2 = 16, ptr1
+	shr.u	loopcnt = cnt, 5   	// loopcnt = cnt / 32
+;; }
+{ .mib
+	cmp.eq	p_scr, p0 = loopcnt, r0
+	add	loopcnt = -1, loopcnt
+(p_scr)	br.cond.dpnt.many .store_words
+;; }
+{ .mib
+	and	cnt = 0x1f, cnt		// compute the remaining cnt
+	movi0   ar.lc = loopcnt
+;; }
+	.align 32
+.l2:	// -----------------------------//  L2A:  store 32B in 2 cycles
+{ .mmb
+	store	[ptr1] = myval, 8
+	store	[ptr2] = myval, 8
+;; } { .mmb
+	store	[ptr1] = myval, 24
+	store	[ptr2] = myval, 24
+	br.cloop.dptk.many .l2
+;; }
+.store_words:
+{ .mib
+	cmp.gt	p_scr, p0 = 8, cnt	// just a few bytes left ?
+(p_scr)	br.cond.dpnt.many .move_bytes_from_alignment	// Branch
+;; }
+
+{ .mmi
+	store	[ptr1] = myval, 8	// store
+	cmp.le	p_y, p_n = 16, cnt	//
+	add	cnt = -8, cnt		// subtract
+;; }
+{ .mmi
+(p_y)	store	[ptr1] = myval, 8	// store
+(p_y)	cmp.le.unc p_yy, p_nn = 16, cnt
+(p_y)	add	cnt = -8, cnt		// subtract
+;; }
+{ .mmi					// store
+(p_yy)	store	[ptr1] = myval, 8
+(p_yy)	add	cnt = -8, cnt		// subtract
+;; }
+
+.move_bytes_from_alignment:
+{ .mib
+	cmp.eq	p_scr, p0 = cnt, r0
+	tbit.nz.unc p_y, p0 = cnt, 2	// should we terminate with a st4 ?
+(p_scr)	br.cond.dpnt.few .restore_and_exit
+;; }
+{ .mib
+(p_y)	st4	[ptr1] = r0,4
+	tbit.nz.unc p_yy, p0 = cnt, 1	// should we terminate with a st2 ?
+;; }
+{ .mib
+(p_yy)	st2	[ptr1] = r0,2
+	tbit.nz.unc p_y, p0 = cnt, 0	// should we terminate with a st1 ?
+;; }
+
+{ .mib
+(p_y)	st1	[ptr1] = r0
+;; }
+.restore_and_exit:
+{ .mib
+	nop.m	0
+	movi0	ar.lc = save_lc
+	br.ret.sptk.many rp
+;; }
+
+.move_bytes_unaligned:
+{ .mmi
+       .pred.rel "mutex",p_y, p_n
+       .pred.rel "mutex",p_yy, p_nn
+(p_n)	cmp.le  p_yy, p_nn = 4, cnt
+(p_y)	cmp.le  p_yy, p_nn = 5, cnt
+(p_n)	add	ptr2 = 2, ptr1
+} { .mmi
+(p_y)	add	ptr2 = 3, ptr1
+(p_y)	st1	[ptr1] = r0, 1		// fill 1 (odd-aligned) byte
+(p_y)	add	cnt = -1, cnt		// [15, 14 (or less) left]
+;; }
+{ .mmi
+(p_yy)	cmp.le.unc p_y, p0 = 8, cnt
+	add	ptr3 = ptr1, cnt	// prepare last store
+	movi0	ar.lc = save_lc
+} { .mmi
+(p_yy)	st2	[ptr1] = r0, 4		// fill 2 (aligned) bytes
+(p_yy)	st2	[ptr2] = r0, 4		// fill 2 (aligned) bytes
+(p_yy)	add	cnt = -4, cnt		// [11, 10 (o less) left]
+;; }
+{ .mmi
+(p_y)	cmp.le.unc p_yy, p0 = 8, cnt
+	add	ptr3 = -1, ptr3		// last store
+	tbit.nz p_scr, p0 = cnt, 1	// will there be a st2 at the end ?
+} { .mmi
+(p_y)	st2	[ptr1] = r0, 4		// fill 2 (aligned) bytes
+(p_y)	st2	[ptr2] = r0, 4		// fill 2 (aligned) bytes
+(p_y)	add	cnt = -4, cnt		// [7, 6 (or less) left]
+;; }
+{ .mmi
+(p_yy)	st2	[ptr1] = r0, 4		// fill 2 (aligned) bytes
+(p_yy)	st2	[ptr2] = r0, 4		// fill 2 (aligned) bytes
+					// [3, 2 (or less) left]
+	tbit.nz p_y, p0 = cnt, 0	// will there be a st1 at the end ?
+} { .mmi
+(p_yy)	add	cnt = -4, cnt
+;; }
+{ .mmb
+(p_scr)	st2	[ptr1] = r0		// fill 2 (aligned) bytes
+(p_y)	st1	[ptr3] = r0		// fill last byte (using ptr3)
+	br.ret.sptk.many rp
+;; }
+END(bzero)
diff --git a/libc/string/ia64/memccpy.S b/libc/string/ia64/memccpy.S
new file mode 100644
index 000000000..53c43c512
--- /dev/null
+++ b/libc/string/ia64/memccpy.S
@@ -0,0 +1,213 @@
+/* Optimized version of the memccpy() function.
+   This file is part of the GNU C Library.
+   Copyright (C) 2000, 2001, 2003 Free Software Foundation, Inc.
+   Contributed by Dan Pop <Dan.Pop@cern.ch>.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+/* Return: a pointer to the next byte after char in dest or NULL
+
+   Inputs:
+        in0:    dest
+        in1:    src
+  	in2:	char
+        in3:    byte count
+
+   This implementation assumes little endian mode (UM.be = 0).
+
+   This implementation assumes that it is safe to do read ahead
+   in the src block, without getting beyond its limit.  */
+
+#include <sysdep.h>
+#undef ret
+
+#define OP_T_THRES 	16
+#define OPSIZ 		8
+
+#define saved_pr	r17
+#define saved_lc	r18
+#define dest		r19
+#define src		r20
+#define len		r21
+#define asrc		r22
+#define tmp		r23
+#define char		r24
+#define charx8		r25
+#define saved_ec	r26
+#define sh2		r28
+#define	sh1		r29
+#define loopcnt		r30
+#define	value		r31
+
+#ifdef GAS_ALIGN_BREAKS_UNWIND_INFO
+/* Manually force proper loop-alignment.  Note: be sure to
+   double-check the code-layout after making any changes to
+   this routine! */
+# define ALIGN(n)	{ nop 0 }
+#else
+# define ALIGN(n)	.align n
+#endif
+
+ENTRY(memccpy)
+	.prologue
+	alloc 	r2 = ar.pfs, 4, 40 - 4, 0, 40
+
+#include "softpipe.h"
+	.rotr	r[MEMLAT + 7], tmp1[4], tmp2[4], val[4], tmp3[2], pos0[2]
+	.rotp	p[MEMLAT + 6 + 1]
+
+	mov	ret0 = r0		// return NULL if no match
+	.save pr, saved_pr
+	mov	saved_pr = pr		// save the predicate registers
+	mov 	dest = in0		// dest
+	.save ar.lc, saved_lc
+        mov 	saved_lc = ar.lc	// save the loop counter
+        mov 	saved_ec = ar.ec	// save the loop counter
+	.body
+	mov 	src = in1		// src
+	extr.u	char = in2, 0, 8	// char
+	mov	len = in3		// len
+	sub	tmp = r0, in0		// tmp = -dest
+	cmp.ne	p7, p0 = r0, r0		// clear p7
+	;;
+	and	loopcnt = 7, tmp	// loopcnt = -dest % 8
+	cmp.ge	p6, p0 = OP_T_THRES, len	// is len <= OP_T_THRES
+	mov	ar.ec = 0		// ec not guaranteed zero on entry
+(p6)	br.cond.spnt	.cpyfew		// copy byte by byte
+	;;
+	cmp.eq	p6, p0 = loopcnt, r0
+	mux1	charx8 = char, @brcst
+(p6)	br.cond.sptk .dest_aligned
+	sub	len = len, loopcnt	// len -= -dest % 8
+	adds	loopcnt = -1, loopcnt	// --loopcnt
+	;;
+	mov	ar.lc = loopcnt
+.l1:					// copy -dest % 8 bytes
+	ld1	value = [src], 1	// value = *src++
+	;;
+	st1	[dest] = value, 1	// *dest++ = value
+	cmp.eq	p6, p0 = value, char
+(p6)	br.cond.spnt .foundit
+	br.cloop.dptk .l1
+.dest_aligned:
+	and	sh1 = 7, src 		// sh1 = src % 8
+	and	tmp = -8, len   	// tmp = len & -OPSIZ
+	and	asrc = -8, src		// asrc = src & -OPSIZ  -- align src
+	shr.u	loopcnt = len, 3	// loopcnt = len / 8
+	and	len = 7, len ;;		// len = len % 8
+	shl	sh1 = sh1, 3		// sh1 = 8 * (src % 8)
+	adds	loopcnt = -1, loopcnt	// --loopcnt
+	mov     pr.rot = 1 << 16 ;;	// set rotating predicates
+	sub	sh2 = 64, sh1		// sh2 = 64 - sh1
+	mov	ar.lc = loopcnt		// set LC
+	cmp.eq  p6, p0 = sh1, r0 	// is the src aligned?
+(p6)    br.cond.sptk .src_aligned ;;
+	add	src = src, tmp		// src += len & -OPSIZ
+	mov	ar.ec = MEMLAT + 6 + 1 	// six more passes needed
+	ld8	r[1] = [asrc], 8 	// r[1] = w0
+	cmp.ne	p6, p0 = r0, r0	;;	// clear p6
+	ALIGN(32)
+.l2:
+(p[0])		ld8.s	r[0] = [asrc], 8		// r[0] = w1
+(p[MEMLAT])	shr.u	tmp1[0] = r[1 + MEMLAT], sh1	// tmp1 = w0 >> sh1
+(p[MEMLAT])	shl	tmp2[0] = r[0 + MEMLAT], sh2  	// tmp2 = w1 << sh2
+(p[MEMLAT+4])	xor	tmp3[0] = val[1], charx8
+(p[MEMLAT+5])	czx1.r	pos0[0] = tmp3[1]
+(p[MEMLAT+6])	chk.s	r[6 + MEMLAT], .recovery1	// our data isn't
+							// valid - rollback!
+(p[MEMLAT+6])	cmp.ne	p6, p0 = 8, pos0[1]
+(p6)		br.cond.spnt	.gotit
+(p[MEMLAT+6])	st8	[dest] = val[3], 8		// store val to dest
+(p[MEMLAT+3])	or	val[0] = tmp1[3], tmp2[3] 	// val = tmp1 | tmp2
+		br.ctop.sptk    .l2
+		br.cond.sptk .cpyfew
+
+.src_aligned:
+		cmp.ne  p6, p0 = r0, r0			// clear p6
+		mov     ar.ec = MEMLAT + 2 + 1 ;;	// set EC
+.l3:
+(p[0])		ld8.s	r[0] = [src], 8
+(p[MEMLAT])	xor	tmp3[0] = r[MEMLAT], charx8
+(p[MEMLAT+1])	czx1.r	pos0[0] = tmp3[1]
+(p[MEMLAT+2])	cmp.ne	p7, p0 = 8, pos0[1]
+(p[MEMLAT+2])	chk.s	r[MEMLAT+2], .recovery2
+(p7)		br.cond.spnt	.gotit
+.back2:
+(p[MEMLAT+2])	st8	[dest] = r[MEMLAT+2], 8
+		br.ctop.dptk .l3
+.cpyfew:
+	cmp.eq	p6, p0 = len, r0	// is len == 0 ?
+	adds	len = -1, len		// --len;
+(p6)	br.cond.spnt	.restore_and_exit ;;
+	mov	ar.lc = len
+.l4:
+	ld1	value = [src], 1
+	;;
+	st1	[dest] = value, 1
+	cmp.eq	p6, p0 = value, char
+(p6)	br.cond.spnt .foundit
+	br.cloop.dptk	.l4 ;;
+.foundit:
+(p6)	mov	ret0 = dest
+.restore_and_exit:
+	mov     pr = saved_pr, -1    	// restore the predicate registers
+	mov 	ar.lc = saved_lc	// restore the loop counter
+	mov 	ar.ec = saved_ec ;;	// restore the epilog counter
+	br.ret.sptk.many b0
+.gotit:
+	.pred.rel "mutex" p6, p7
+(p6)	mov	value = val[3]		// if coming from l2
+(p7)	mov	value = r[MEMLAT+2]	// if coming from l3
+	mov	ar.lc = pos0[1] ;;
+.l5:
+	extr.u	tmp = value, 0, 8 ;;
+	st1	[dest] = tmp, 1
+	shr.u	value = value, 8
+	br.cloop.sptk .l5 ;;
+	mov 	ret0 = dest
+	mov	pr = saved_pr, -1
+	mov	ar.lc = saved_lc
+	br.ret.sptk.many b0
+
+.recovery1:
+	adds	src = -(MEMLAT + 6 + 1) * 8, asrc
+	mov	loopcnt = ar.lc
+	mov	tmp = ar.ec ;;
+	sub	sh1 = (MEMLAT + 6 + 1), tmp
+	shr.u	sh2 = sh2, 3
+	;; 
+	shl	loopcnt = loopcnt, 3
+	sub	src = src, sh2
+	shl	sh1 = sh1, 3
+	shl	tmp = tmp, 3
+	;;
+	add	len = len, loopcnt
+	add	src = sh1, src ;;
+	add	len = tmp, len
+.back1:
+	br.cond.sptk .cpyfew
+
+.recovery2:
+	add	tmp = -(MEMLAT + 3) * 8, src
+(p7)	br.cond.spnt .gotit
+	;;
+	ld8	r[MEMLAT+2] = [tmp] ;;
+	xor	pos0[1] = r[MEMLAT+2], charx8 ;;
+	czx1.r	pos0[1] = pos0[1] ;;
+	cmp.ne	p7, p6 = 8, pos0[1]
+(p7)	br.cond.spnt .gotit
+	br.cond.sptk .back2
+END(memccpy)
diff --git a/libc/string/ia64/memchr.S b/libc/string/ia64/memchr.S
new file mode 100644
index 000000000..d7742fe8a
--- /dev/null
+++ b/libc/string/ia64/memchr.S
@@ -0,0 +1,133 @@
+/* Optimized version of the standard memchr() function.
+   This file is part of the GNU C Library.
+   Copyright (C) 2000, 2001, 2003 Free Software Foundation, Inc.
+   Contributed by Dan Pop <Dan.Pop@cern.ch>.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+/* Return: the address of the first occurence of chr in str or NULL
+
+   Inputs:
+  	in0:	str
+  	in1:	chr
+  	in2:	byte count
+
+   This implementation assumes little endian mode.  For big endian mode,
+   the instruction czx1.r should be replaced by czx1.l.
+
+   The algorithm is fairly straightforward: search byte by byte until we
+   we get to a word aligned address, then search word by word as much as
+   possible; the remaining few bytes are searched one at a time.
+
+   The word by word search is performed by xor-ing the word with a word
+   containing chr in every byte.  If there is a hit, the result will
+   contain a zero byte in the corresponding position.  The presence and
+   position of that zero byte is detected with a czx instruction.
+
+   All the loops in this function could have had the internal branch removed
+   if br.ctop and br.cloop could be predicated :-(.  */
+
+#include <sysdep.h>
+#undef ret
+
+#define saved_pr	r15
+#define saved_lc	r16
+#define	chr		r17
+#define len		r18
+#define pos0		r20
+#define val		r21
+#define tmp		r24
+#define chrx8		r25
+#define loopcnt		r30
+
+#define str		in0
+
+ENTRY(__memchr)
+	.prologue
+	alloc r2 = ar.pfs, 3, 0, 29, 32
+#include "softpipe.h"
+	.rotr	value[MEMLAT+1], addr[MEMLAT+3], aux[2], poschr[2]
+	.rotp	p[MEMLAT+3]
+	.save ar.lc, saved_lc
+        mov 	saved_lc = ar.lc 	// save the loop counter
+	.save pr, saved_pr
+	mov	saved_pr = pr		// save the predicates
+	.body
+	mov 	ret0 = str
+	and 	tmp = 7, str		// tmp = str % 8
+	cmp.ne	p7, p0 = r0, r0		// clear p7
+	extr.u	chr = in1, 0, 8		// chr = (unsigned char) in1
+	mov	len = in2
+	cmp.gtu	p6, p0 = 16, in2	// use a simple loop for short
+(p6)	br.cond.spnt .srchfew ;;	// searches
+	sub	loopcnt = 8, tmp	// loopcnt = 8 - tmp
+	cmp.eq	p6, p0 = tmp, r0
+(p6)	br.cond.sptk	.str_aligned;;
+	sub	len = len, loopcnt
+	adds	loopcnt = -1, loopcnt;;
+	mov	ar.lc = loopcnt
+.l1:
+	ld1	val = [ret0], 1
+	;;
+	cmp.eq	p6, p0 = val, chr
+(p6)	br.cond.spnt	.foundit
+	br.cloop.sptk	.l1 ;;
+.str_aligned:
+	cmp.ne	p6, p0 = r0, r0		// clear p6
+	shr.u	loopcnt = len, 3	// loopcnt = len / 8
+	and 	len = 7, len ;;		// remaining len = len & 7
+	adds	loopcnt = -1, loopcnt
+	mov	ar.ec = MEMLAT + 3
+	mux1	chrx8 = chr, @brcst ;;	// get a word full of chr
+	mov	ar.lc = loopcnt
+	mov	pr.rot = 1 << 16 ;;
+.l2:
+(p[0])		mov	addr[0] = ret0
+(p[0])		ld8	value[0] = [ret0], 8
+(p[MEMLAT])	xor	aux[0] = value[MEMLAT], chrx8
+(p[MEMLAT+1])	czx1.r	poschr[0] = aux[1]
+(p[MEMLAT+2])	cmp.ne	p7, p0 = 8, poschr[1]
+(p7)		br.cond.dpnt .foundit
+		br.ctop.dptk .l2
+.srchfew:
+	adds	loopcnt = -1, len
+	cmp.eq	p6, p0 = len, r0
+(p6)	br.cond.spnt .notfound ;;
+	mov	ar.lc = loopcnt
+.l3:
+	ld1	val = [ret0], 1
+	;;
+	cmp.eq	p6, p0 = val, chr
+(p6)	br.cond.dpnt	.foundit
+	br.cloop.sptk	.l3 ;;
+.notfound:
+	cmp.ne	p6, p0 = r0, r0	// clear p6 (p7 was already 0 when we got here)
+	mov	ret0 = r0 ;;	// return NULL
+.foundit:
+	.pred.rel "mutex" p6, p7
+(p6)	adds	ret0 = -1, ret0 		   // if we got here from l1 or l3
+(p7)	add	ret0 = addr[MEMLAT+2], poschr[1]   // if we got here from l2
+	mov	pr = saved_pr, -1
+	mov	ar.lc = saved_lc