Home | History | Annotate | Line # | Download | only in gen
divrem.m4 revision 1.3 
      1 /*	$NetBSD: divrem.m4,v 1.3 1995/08/14 02:20:07 cgd Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1994, 1995 Carnegie-Mellon University.
      5  * All rights reserved.
      6  *
      7  * Author: Chris G. Demetriou
      8  *
      9  * Permission to use, copy, modify and distribute this software and
     10  * its documentation is hereby granted, provided that both the copyright
     11  * notice and this permission notice appear in all copies of the
     12  * software, derivative works or modified versions, and any portions
     13  * thereof, and that both notices appear in supporting documentation.
     14  *
     15  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
     16  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
     17  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
     18  *
     19  * Carnegie Mellon requests users of this software to return to
     20  *
     21  *  Software Distribution Coordinator  or  Software.Distribution (at) CS.CMU.EDU
     22  *  School of Computer Science
     23  *  Carnegie Mellon University
     24  *  Pittsburgh PA 15213-3890
     25  *
     26  * any improvements or extensions that they make and grant Carnegie the
     27  * rights to redistribute these changes.
     28  */
     29 
     30 /*
     31  * Division and remainder.
     32  *
     33  * The use of m4 is modeled after the sparc code, but the algorithm is
     34  * simple binary long division.
     35  *
     36  * Note that the loops could probably benefit from unrolling.
     37  */
     38 
     39 /*
     40  * M4 Parameters
     41  * NAME		name of function to generate
     42  * OP		OP=div: t10 / t11 -> t12; OP=rem: t10 % t11 -> t12
     43  * S		S=true: signed; S=false: unsigned [XXX NOT YET]
     44  * WORDSIZE	total number of bits
     45  */
     46 
     47 define(A, `t10')
     48 define(B, `t11')
     49 define(RESULT, `t12')
     50 
     51 define(BIT, `t0')
     52 define(I, `t1')
     53 define(CC, `t2')
     54 define(T_0, `t3')
     55 ifelse(S, `true', `define(SIGN, `t4')')
     56 
     57 #include "DEFS.h"
     58 
     59 LEAF(NAME, 0)					/* XXX */
     60 	lda	sp, -64(sp)
     61 	stq	BIT, 0(sp)
     62 	stq	I, 8(sp)
     63 	stq	CC, 16(sp)
     64 	stq	T_0, 24(sp)
     65 ifelse(S, `true',
     66 `	stq	SIGN, 32(sp)')
     67 	stq	A, 40(sp)
     68 	stq	B, 48(sp)
     69 	mov	zero, RESULT			/* Initialize result to zero */
     70 
     71 ifelse(S, `true',
     72 `
     73 	/* Compute sign of result.  If either is negative, this is easy.  */
     74 	or	A, B, SIGN			/* not the sign, but... */
     75 	bgt	SIGN, Ldoit			/* neither negative? do it! */
     76 
     77 ifelse(OP, `div',
     78 `	xor	A, B, SIGN			/* THIS is the sign! */
     79 ', `	mov	A, SIGN				/* sign follows A. */
     80 ')
     81 	bge	A, LnegB			/* see if A is negative */
     82 	/* A is negative; flip it. */
     83 	subq	zero, A, A
     84 	bge	B, Ldoit			/* see if B is negative */
     85 LnegB:
     86 	/* B is definitely negative, no matter how we got here. */
     87 	subq	zero, B, B
     88 Ldoit:
     89 ', `
     90 ifelse(WORDSIZE, `32', `
     91 	/*
     92 	 * Clear the top 32 bits of each operand, as the compiler may
     93 	 * have sign extended them, if the 31st bit was set.
     94 	 */
     95 	zap	A, 0xf0, A
     96 	zap	B, 0xf0, B
     97 ')' )
     98 
     99 	/* kill the special cases. */
    100 	beq	B, Ldotrap			/* division by zero! XXX */
    101 
    102 1:	cmpult	A, B, CC			/* A < B? */
    103 	/* RESULT is already zero, from above.  A is untouched. */
    104 	bne	CC, Lret_result
    105 
    106 	cmpeq	A, B, CC			/* A == B? */
    107 	cmovne	CC, 1, RESULT
    108 	cmovne	CC, zero, A
    109 	bne	CC, Lret_result
    110 
    111 	/*
    112 	 * Find out how many bits of zeros are at the beginning of the divisor.
    113 	 */
    114 LBbits:
    115 	CONST(1, T_0)				/* I = 0; BIT = 1<<WORDSIZE-1 */
    116 	mov	zero, I
    117 	sll	T_0, WORDSIZE-1, BIT
    118 LBloop:
    119 	and	B, BIT, CC			/* if bit in B is set, done. */
    120 	bne	CC, LAbits
    121 	addq	I, 1, I				/* increment I, shift bit */
    122 	srl	BIT, 1, BIT
    123 	cmplt	I, WORDSIZE-1, CC		/* if I leaves one bit, done. */
    124 	bne	CC, LBloop
    125 
    126 LAbits:
    127 	beq	I, Ldodiv			/* If I = 0, divide now.  */
    128 	CONST(1, T_0)				/* BIT = 1<<WORDSIZE-1 */
    129 	sll	T_0, WORDSIZE-1, BIT
    130 
    131 LAloop:
    132 	and	A, BIT, CC			/* if bit in A is set, done. */
    133 	bne	CC, Ldodiv
    134 	subq	I, 1, I				/* decrement I, shift bit */
    135 	srl     BIT, 1, BIT
    136 	bne	I, LAloop			/* If I != 0, loop again */
    137 
    138 Ldodiv:
    139 	sll	B, I, B				/* B <<= i */
    140 	CONST(1, T_0)
    141 	sll	T_0, I, BIT
    142 
    143 Ldivloop:
    144 	cmpult	A, B, CC
    145 	or	RESULT, BIT, T_0
    146 	cmoveq	CC, T_0, RESULT
    147 	subq	A, B, T_0
    148 	cmoveq	CC, T_0, A
    149 	srl	BIT, 1, BIT
    150 	srl	B, 1, B
    151 	beq	A, Lret_result
    152 	bne	BIT, Ldivloop
    153 
    154 Lret_result:
    155 ifelse(OP, `div',
    156 `', `	mov	A, RESULT
    157 ')
    158 ifelse(S, `true',
    159 `
    160 	/* Check to see if we should negate it. */
    161 	subqv	zero, RESULT, T_0
    162 	cmovlt	SIGN, T_0, RESULT
    163 ')
    164 
    165 	ldq	BIT, 0(sp)
    166 	ldq	I, 8(sp)
    167 	ldq	CC, 16(sp)
    168 	ldq	T_0, 24(sp)
    169 ifelse(S, `true',
    170 `	ldq	SIGN, 32(sp)')
    171 	ldq	A, 40(sp)
    172 	ldq	B, 48(sp)
    173 	lda	sp, 64(sp)
    174 	ret	zero, (t9), 1
    175 
    176 Ldotrap:
    177 	CONST(-2, a0)			/* This is the signal to SIGFPE! */
    178 	call_pal PAL_gentrap
    179 ifelse(OP, `div',
    180 `', `	mov	zero, A			/* so that zero will be returned */
    181 ')
    182 	br	zero, Lret_result
    183 
    184 END(NAME)
    185