Home | History | Annotate | Line # | Download | only in powerpc32
      1      1.1  mrg dnl  PowerPC-32 mpn_add_n and mpn_sub_n.
      2      1.1  mrg 
      3      1.1  mrg dnl  Copyright 2002, 2005, 2007 Free Software Foundation, Inc.
      4      1.1  mrg 
      5      1.1  mrg dnl  This file is part of the GNU MP Library.
      6  1.1.1.3  mrg dnl
      7      1.1  mrg dnl  The GNU MP Library is free software; you can redistribute it and/or modify
      8  1.1.1.3  mrg dnl  it under the terms of either:
      9  1.1.1.3  mrg dnl
     10  1.1.1.3  mrg dnl    * the GNU Lesser General Public License as published by the Free
     11  1.1.1.3  mrg dnl      Software Foundation; either version 3 of the License, or (at your
     12  1.1.1.3  mrg dnl      option) any later version.
     13  1.1.1.3  mrg dnl
     14  1.1.1.3  mrg dnl  or
     15  1.1.1.3  mrg dnl
     16  1.1.1.3  mrg dnl    * the GNU General Public License as published by the Free Software
     17  1.1.1.3  mrg dnl      Foundation; either version 2 of the License, or (at your option) any
     18  1.1.1.3  mrg dnl      later version.
     19  1.1.1.3  mrg dnl
     20  1.1.1.3  mrg dnl  or both in parallel, as here.
     21  1.1.1.3  mrg dnl
     22      1.1  mrg dnl  The GNU MP Library is distributed in the hope that it will be useful, but
     23      1.1  mrg dnl  WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
     24  1.1.1.3  mrg dnl  or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
     25  1.1.1.3  mrg dnl  for more details.
     26  1.1.1.3  mrg dnl
     27  1.1.1.3  mrg dnl  You should have received copies of the GNU General Public License and the
     28  1.1.1.3  mrg dnl  GNU Lesser General Public License along with the GNU MP Library.  If not,
     29  1.1.1.3  mrg dnl  see https://www.gnu.org/licenses/.
     30      1.1  mrg 
     31      1.1  mrg include(`../config.m4')
     32      1.1  mrg 
     33  1.1.1.2  mrg C                   cycles/limb
     34  1.1.1.2  mrg C 603e:                  ?
     35  1.1.1.2  mrg C 604e:                  ?		old: 3.25
     36  1.1.1.2  mrg C 75x (G3):              ?		old: 3.5
     37  1.1.1.2  mrg C 7400,7410 (G4):        3.25
     38  1.1.1.2  mrg C 744x,745x (G4+):       4
     39  1.1.1.2  mrg C POWER3/PPC630          2
     40  1.1.1.2  mrg C POWER4/PPC970          2.4
     41  1.1.1.2  mrg C POWER5                 2.75
     42  1.1.1.2  mrg C POWER6               40-140
     43  1.1.1.2  mrg C POWER7                 3
     44      1.1  mrg 
     45      1.1  mrg C INPUT PARAMETERS
     46      1.1  mrg define(`rp',	`r3')
     47      1.1  mrg define(`up',	`r4')
     48      1.1  mrg define(`vp',	`r5')
     49      1.1  mrg define(`n',	`r6')
     50      1.1  mrg define(`cy',	`r7')
     51      1.1  mrg 
     52      1.1  mrg ifdef(`OPERATION_add_n', `
     53      1.1  mrg 	define(ADCSBC,	adde)
     54      1.1  mrg 	define(func,	mpn_add_n)
     55      1.1  mrg 	define(func_nc,	mpn_add_nc)
     56      1.1  mrg 	define(IFADD,	`$1')
     57      1.1  mrg 	define(IFSUB,	`')')
     58      1.1  mrg ifdef(`OPERATION_sub_n', `
     59      1.1  mrg 	define(ADCSBC,	subfe)
     60      1.1  mrg 	define(func,	mpn_sub_n)
     61      1.1  mrg 	define(func_nc,	mpn_sub_nc)
     62      1.1  mrg 	define(IFADD,	`')
     63      1.1  mrg 	define(IFSUB,	`$1')')
     64      1.1  mrg 
     65      1.1  mrg MULFUNC_PROLOGUE(mpn_add_n mpn_add_nc mpn_sub_n mpn_sub_nc)
     66      1.1  mrg 
     67      1.1  mrg ASM_START()
     68      1.1  mrg 
     69      1.1  mrg PROLOGUE(func_nc)
     70      1.1  mrg IFADD(`	addic	r0, cy, -1')		C set carry from argument
     71      1.1  mrg IFSUB(`	subfic	r0, cy, 0')		C set carry from argument
     72      1.1  mrg 	b	L(ent)
     73      1.1  mrg EPILOGUE()
     74      1.1  mrg 
     75      1.1  mrg PROLOGUE(func)
     76      1.1  mrg IFADD(`	addic	r0, n, 0')		C clear carry
     77      1.1  mrg IFSUB(`	addic	r0, n, -1')		C set carry
     78      1.1  mrg L(ent):	andi.	r0, n, 3
     79      1.1  mrg 	addi	r3, r3, -12
     80      1.1  mrg 	addi	n, n, 1
     81      1.1  mrg 	cmpwi	cr7, r0, 2
     82      1.1  mrg 	srwi	r0, n, 2
     83      1.1  mrg 	sub	r4, r4, r3
     84      1.1  mrg 	sub	r5, r5, r3
     85      1.1  mrg 	mtctr	r0
     86      1.1  mrg 	bne	cr0, L(n00)
     87      1.1  mrg 
     88      1.1  mrg 	lwzx	r7, r4, r3		C n = 4, 8, 12, ...
     89      1.1  mrg 	lwzx	r8, r5, r3
     90      1.1  mrg 	addi	r3, r3, 4
     91      1.1  mrg 	lwzx	r9, r4, r3
     92      1.1  mrg 	ADCSBC	r7, r8, r7
     93      1.1  mrg 	lwzx	r10, r5, r3
     94      1.1  mrg 	addi	r3, r3, 4
     95      1.1  mrg 	b	L(00)
     96      1.1  mrg 
     97      1.1  mrg L(n00):	bge	cr7, L(n01)
     98      1.1  mrg 	cmpwi	cr0, r0, 0		C n = 1, 5, 9, 13, ...
     99      1.1  mrg 	lwzx	r0, r4, r3
    100      1.1  mrg 	lwzx	r6, r5, r3
    101      1.1  mrg 	addi	r3, r3, 4
    102      1.1  mrg 	ADCSBC	r0, r6, r0
    103      1.1  mrg 	ble	L(ret)
    104      1.1  mrg L(gt1):	lwzx	r7, r4, r3
    105      1.1  mrg 	lwzx	r8, r5, r3
    106      1.1  mrg 	addi	r3, r3, 4
    107      1.1  mrg 	b	L(01)
    108      1.1  mrg 
    109      1.1  mrg L(n10):
    110      1.1  mrg 	lwzx	r9, r4, r3		C n = 3, 7, 11, 15, ...
    111      1.1  mrg 	lwzx	r10, r5, r3
    112      1.1  mrg 	addi	r3, r3, 4
    113      1.1  mrg 	lwzx	r11, r4, r3
    114      1.1  mrg 	ADCSBC	r9, r10, r9
    115      1.1  mrg 	lwzx	r12, r5, r3
    116      1.1  mrg 	addi	r3, r3, 4
    117      1.1  mrg 	b	L(11)
    118      1.1  mrg 
    119      1.1  mrg L(n01):	bne	cr7, L(n10)
    120      1.1  mrg 	cmpwi	cr0, r0, 0		C n = 2, 6, 10, 14, ...
    121      1.1  mrg 	lwzx	r11, r4, r3
    122      1.1  mrg 	lwzx	r12, r5, r3
    123      1.1  mrg 	addi	r3, r3, 4
    124      1.1  mrg 	lwzx	r0, r4, r3
    125      1.1  mrg 	ADCSBC	r11, r12, r11
    126      1.1  mrg 	lwzx	r6, r5, r3
    127      1.1  mrg 	addi	r3, r3, 4
    128      1.1  mrg 	ble	cr0, L(end)
    129      1.1  mrg 
    130      1.1  mrg 
    131      1.1  mrg L(lp):	lwzx	r7, r4, r3
    132      1.1  mrg 	ADCSBC	r0, r6, r0
    133      1.1  mrg 	lwzx	r8, r5, r3
    134      1.1  mrg 	stwu	r11, 4(r3)
    135      1.1  mrg L(01):	lwzx	r9, r4, r3
    136      1.1  mrg 	ADCSBC	r7, r8, r7
    137      1.1  mrg 	lwzx	r10, r5, r3
    138      1.1  mrg 	stwu	r0, 4(r3)
    139      1.1  mrg L(00):	lwzx	r11, r4, r3
    140      1.1  mrg 	ADCSBC	r9, r10, r9
    141      1.1  mrg 	lwzx	r12, r5, r3
    142      1.1  mrg 	stwu	r7, 4(r3)
    143      1.1  mrg L(11):	lwzx	r0, r4, r3
    144      1.1  mrg 	ADCSBC	r11, r12, r11
    145      1.1  mrg 	lwzx	r6, r5, r3
    146      1.1  mrg 	stwu	r9, 4(r3)
    147      1.1  mrg 	bdnz	L(lp)
    148      1.1  mrg 
    149      1.1  mrg L(end):	ADCSBC	r0, r6, r0
    150      1.1  mrg 	stw	r11, 4(r3)
    151      1.1  mrg L(ret):	stw	r0, 8(r3)
    152      1.1  mrg IFADD(`	li	r3, 0	')
    153      1.1  mrg IFADD(`	addze	r3, r3	')
    154      1.1  mrg IFSUB(`	subfe	r3, r0, r0')
    155      1.1  mrg IFSUB(`	neg	r3, r3')
    156      1.1  mrg 	blr
    157      1.1  mrg EPILOGUE()
    158