/*******************************************************************************
 * 
 * Copyright (c) 1993 Intel Corporation
 * 
 * Intel hereby grants you permission to copy, modify, and distribute this
 * software and its documentation.  Intel grants this permission provided
 * that the above copyright notice appears in all copies and that both the
 * copyright notice and this permission notice appear in supporting
 * documentation.  In addition, Intel grants this permission provided that
 * you prominently mark as "not part of the original" any modifications
 * made to this software or documentation, and that the name of Intel
 * Corporation not be used in advertising or publicity pertaining to
 * distribution of the software or the documentation without specific,
 * written prior permission.
 * 
 * Intel Corporation provides this AS IS, WITHOUT ANY WARRANTY, EXPRESS OR
 * IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTY OF MERCHANTABILITY
 * OR FITNESS FOR A PARTICULAR PURPOSE.  Intel makes no guarantee or
 * representations regarding the use of, or the results of the use of,
 * the software and documentation in terms of correctness, accuracy,
 * reliability, currentness, or otherwise; and you rely on the software,
 * documentation and results solely at your own risk.
 *
 * IN NO EVENT SHALL INTEL BE LIABLE FOR ANY LOSS OF USE, LOSS OF BUSINESS,
 * LOSS OF PROFITS, INDIRECT, INCIDENTAL, SPECIAL OR CONSEQUENTIAL DAMAGES
 * OF ANY KIND.  IN NO EVENT SHALL INTEL'S TOTAL LIABILITY EXCEED THE SUM
 * PAID TO INTEL FOR THE PRODUCT LICENSED HEREUNDER.
 * 
 ******************************************************************************/

	.file "memcm_ca.s"
#ifdef	__PIC
	.pic
#endif
#ifdef	__PID
	.pid
#endif
/*
 * (c) copyright 1988,1992,1993 Intel Corp., all rights reserved
 */

/*
	procedure memcmp  (optimized assembler version for the CA)     

	result = memcmp (src1_addr, src2_addr, max_bytes)

	compare the byte array pointed to by src1_addr to the byte array
	pointed to by src2_addr.  Return 0 iff the arrays are equal, -1 if 
	src1_addr is lexicly less than src2_addr, and 1 if it is lexicly 
	greater.  Do not compare more than max_bytes bytes.

	Undefined behavior will occur if the end of either source array
	is in the last word of the program's allocated memory space.  This 
	is so because, in several cases, memcmp will fetch ahead one word.
	Disallowing the fetch ahead would impose a severe performance penalty.

	This program handles five cases:

	1) both arguments start on a word boundary
	2) neither are word aligned, but they are offset by the same amount
	3) source1 is word aligned, source2 is not
	4) source2 is word aligned, source1 is not
	5) neither is word aligned, and they are offset by differing amounts

	At the time of this writing, only g0 thru g7 and g14 are available 
	for use in this leafproc;  other registers would have to be saved and
	restored.  These nine registers are sufficient to implement the routine.
	The registers are used as follows:

	g0  original src1 ptr;  extracted word;  return result
	g1  src2 ptr; byt extraction mask
	g2  maximum number of bytes to compare
	g3  src2 word ptr 
	Little endian
		g4  lsw of src1
		g5  msw of src1
		g6  src2 word
		g7  src1 word ptr
	Big endian
		g4  msw of src1
		g5  lsw of src1
		g6  src1 word ptr
		g7  src2 word
	g13 return address
	g14 shift count
*/

#if __i960_BIG_ENDIAN__
#define MSW g4
#define LSW g5
#define SRC1 g6
#define SRC2 g7
#else
#define LSW g4
#define MSW g5
#define SRC2 g6
#define SRC1 g7
#endif

	.globl	_memcmp
	.globl	__memcmp
	.leafproc	_memcmp, __memcmp
	.align	2
_memcmp:
#ifndef __PIC
	lda 	Lrett,g14
#else
	lda 	Lrett-(.+8)(ip),g14
#endif
__memcmp:
Lrestart:
#if __i960_BIG_ENDIAN__
	subo	1,g0,SRC1
	notand	SRC1,3,SRC1	# extract word addr of start of src1
#else
	notand	g0,3,SRC1	# extract word addr of start of src1
#endif
	 lda	(g14),g13	# preserve return address
	cmpibge.f 0,g2,Lequal_exit	# return equality if number bytes 0
	notand	g1,3,g3		# extract word addr of start of src2
	 ld	(SRC1),LSW	# fetch word with at least first byte of src1
	cmpo	g3,g1		# check alignment of src2
	 ld	4(SRC1),MSW	# fetch second word of src1
	shlo	3,g0,g14	# compute shift count for src1
#if __i960_BIG_ENDIAN__
	subo	g14,0,g14	# adjust shift count for big endian.
#endif
	 ld	(g3),SRC2	# fetch word with at least first byte of src2
	eshro	g14,g4,LSW	# extract word of src1
	 lda	8(SRC1),SRC1	# advance src1 word addr
	 bne.f	Lsrc2_unaligned	# branch if src2 is NOT word aligned

	mov	LSW,g0		# at least src2 is word aligned

	 lda	0xff,g1

Lwloop:				# word comparing loop
	cmpo	SRC2,g0		# compare src1 and src2 words
	 lda	4(g3),g3	# pre-increment src2 addr
	mov	MSW,LSW		# move msw of src1 to lsw
	 ld	(SRC1),MSW	# pre-fetch next msw of src1
	subi	4,g2,g2		# decrement maximum byte count
	 bne.f	Lcloop		# branch if src1 and src2 unequal
	cmpi	0,g2
	 ld	(g3),SRC2	# pre-fetch next word of src2
	eshro	g14,g4,g0	# extract word of src1
	 lda	4(SRC1),SRC1	# post-increment src1 addr
	bl.t	Lwloop		# branch if max_bytes not reached yet

	b	Lequal_exit	# strings were equal up through max_bytes

Lcloop_setup:			# setup for coming from Lsrc2_unaligned
	mov	LSW,g0		# restore extracted src1 word
	subo	4,g2,g2		# make up for later re-incrementing
	 lda	0xff,g1		# byte extraction mask

Lcloop:				# character comparing loop
#if __i960_BIG_ENDIAN__
	rotate	24,g1,g1	# shift mask for next byte
#endif
	and	SRC2,g1,g3	# extract next char of src2
	and	g0,g1,LSW	# extract next char of src1
	cmpobne.f LSW,g3,.diff	# check for equality
#if ! __i960_BIG_ENDIAN__
	shlo	8,g1,g1		# shift mask for next byte
#endif
	subi	1,g2,g2		# decrement character counter
	 b	Lcloop		# branch if null not reached


Lequal_exit: 			# words are equal up thru null byte
	mov	0,g14		# conform to register conventions
	 lda	0,g0		# return zero, indicating equality
	bx	(g13)		# return
Lrett:
	ret

.diff:
	addo	4,g2,g2		# to make up for extra decrement in loop
	 lda	0,g14
	 bl	Lless_than_exit
Lgreater_than_exit:
	cmpibge.f 0,g2,Lequal_exit  # branch if difference is beyond max_bytes
	mov	1,g0
	 bx	(g13)		# g0 = 1 (src1 > src2)
Lless_than_exit:
	cmpibge.f 0,g2,Lequal_exit  # branch if difference is beyond max_bytes
	subi	1,0,g0
	 bx	(g13)		# g0 = -1 (src1 < src2)

Lsrc2_unaligned:
	notor	g1,3,g14	# first step in computing new src1 ptr
	 ld	4(g3),SRC1	# fetch second word of src2
	shlo	3,g1,MSW	# compute shift count for src2
#if __i960_BIG_ENDIAN__
	subo	MSW,0,MSW
#endif
	eshro	MSW,g6,SRC2	# extract word of src2
	cmpo	LSW,SRC2	# compare src1 and src2 words
	 lda	4(g3),g1	# set new src2 ptr
	 bne.f	Lcloop_setup	# first four bytes differ
	subo	g14,g0,g0	# second (final) step in computing new src1 ptr
	addi	g14,g2,g2	# compute new max_bytes too
	 lda	(g13),g14	# prepare return pointer for Lrestart
	 b	Lrestart		# continue with both string fetches shifted