|
Packit |
5c3484 |
dnl PowerPC-32 mpn_add_n and mpn_sub_n.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2002, 2005, 2007 Free Software Foundation, Inc.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl This file is part of the GNU MP Library.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl The GNU MP Library is free software; you can redistribute it and/or modify
|
|
Packit |
5c3484 |
dnl it under the terms of either:
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl * the GNU Lesser General Public License as published by the Free
|
|
Packit |
5c3484 |
dnl Software Foundation; either version 3 of the License, or (at your
|
|
Packit |
5c3484 |
dnl option) any later version.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl or
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl * the GNU General Public License as published by the Free Software
|
|
Packit |
5c3484 |
dnl Foundation; either version 2 of the License, or (at your option) any
|
|
Packit |
5c3484 |
dnl later version.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl or both in parallel, as here.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl The GNU MP Library is distributed in the hope that it will be useful, but
|
|
Packit |
5c3484 |
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
|
|
Packit |
5c3484 |
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
|
|
Packit |
5c3484 |
dnl for more details.
|
|
Packit |
5c3484 |
dnl
|
|
Packit |
5c3484 |
dnl You should have received copies of the GNU General Public License and the
|
|
Packit |
5c3484 |
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
|
|
Packit |
5c3484 |
dnl see https://www.gnu.org/licenses/.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
include(`../config.m4')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C cycles/limb
|
|
Packit |
5c3484 |
C 603e: ?
|
|
Packit |
5c3484 |
C 604e: ? old: 3.25
|
|
Packit |
5c3484 |
C 75x (G3): ? old: 3.5
|
|
Packit |
5c3484 |
C 7400,7410 (G4): 3.25
|
|
Packit |
5c3484 |
C 744x,745x (G4+): 4
|
|
Packit |
5c3484 |
C POWER3/PPC630 2
|
|
Packit |
5c3484 |
C POWER4/PPC970 2.4
|
|
Packit |
5c3484 |
C POWER5 2.75
|
|
Packit |
5c3484 |
C POWER6 40-140
|
|
Packit |
5c3484 |
C POWER7 3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C INPUT PARAMETERS
|
|
Packit |
5c3484 |
define(`rp', `r3')
|
|
Packit |
5c3484 |
define(`up', `r4')
|
|
Packit |
5c3484 |
define(`vp', `r5')
|
|
Packit |
5c3484 |
define(`n', `r6')
|
|
Packit |
5c3484 |
define(`cy', `r7')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ifdef(`OPERATION_add_n', `
|
|
Packit |
5c3484 |
define(ADCSBC, adde)
|
|
Packit |
5c3484 |
define(func, mpn_add_n)
|
|
Packit |
5c3484 |
define(func_nc, mpn_add_nc)
|
|
Packit |
5c3484 |
define(IFADD, `$1')
|
|
Packit |
5c3484 |
define(IFSUB, `')')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_sub_n', `
|
|
Packit |
5c3484 |
define(ADCSBC, subfe)
|
|
Packit |
5c3484 |
define(func, mpn_sub_n)
|
|
Packit |
5c3484 |
define(func_nc, mpn_sub_nc)
|
|
Packit |
5c3484 |
define(IFADD, `')
|
|
Packit |
5c3484 |
define(IFSUB, `$1')')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
MULFUNC_PROLOGUE(mpn_add_n mpn_add_nc mpn_sub_n mpn_sub_nc)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
PROLOGUE(func_nc)
|
|
Packit |
5c3484 |
IFADD(` addic r0, cy, -1') C set carry from argument
|
|
Packit |
5c3484 |
IFSUB(` subfic r0, cy, 0') C set carry from argument
|
|
Packit |
5c3484 |
b L(ent)
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
PROLOGUE(func)
|
|
Packit |
5c3484 |
IFADD(` addic r0, n, 0') C clear carry
|
|
Packit |
5c3484 |
IFSUB(` addic r0, n, -1') C set carry
|
|
Packit |
5c3484 |
L(ent): andi. r0, n, 3
|
|
Packit |
5c3484 |
addi r3, r3, -12
|
|
Packit |
5c3484 |
addi n, n, 1
|
|
Packit |
5c3484 |
cmpwi cr7, r0, 2
|
|
Packit |
5c3484 |
srwi r0, n, 2
|
|
Packit |
5c3484 |
sub r4, r4, r3
|
|
Packit |
5c3484 |
sub r5, r5, r3
|
|
Packit |
5c3484 |
mtctr r0
|
|
Packit |
5c3484 |
bne cr0, L(n00)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
lwzx r7, r4, r3 C n = 4, 8, 12, ...
|
|
Packit |
5c3484 |
lwzx r8, r5, r3
|
|
Packit |
5c3484 |
addi r3, r3, 4
|
|
Packit |
5c3484 |
lwzx r9, r4, r3
|
|
Packit |
5c3484 |
ADCSBC r7, r8, r7
|
|
Packit |
5c3484 |
lwzx r10, r5, r3
|
|
Packit |
5c3484 |
addi r3, r3, 4
|
|
Packit |
5c3484 |
b L(00)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(n00): bge cr7, L(n01)
|
|
Packit |
5c3484 |
cmpwi cr0, r0, 0 C n = 1, 5, 9, 13, ...
|
|
Packit |
5c3484 |
lwzx r0, r4, r3
|
|
Packit |
5c3484 |
lwzx r6, r5, r3
|
|
Packit |
5c3484 |
addi r3, r3, 4
|
|
Packit |
5c3484 |
ADCSBC r0, r6, r0
|
|
Packit |
5c3484 |
ble L(ret)
|
|
Packit |
5c3484 |
L(gt1): lwzx r7, r4, r3
|
|
Packit |
5c3484 |
lwzx r8, r5, r3
|
|
Packit |
5c3484 |
addi r3, r3, 4
|
|
Packit |
5c3484 |
b L(01)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(n10):
|
|
Packit |
5c3484 |
lwzx r9, r4, r3 C n = 3, 7, 11, 15, ...
|
|
Packit |
5c3484 |
lwzx r10, r5, r3
|
|
Packit |
5c3484 |
addi r3, r3, 4
|
|
Packit |
5c3484 |
lwzx r11, r4, r3
|
|
Packit |
5c3484 |
ADCSBC r9, r10, r9
|
|
Packit |
5c3484 |
lwzx r12, r5, r3
|
|
Packit |
5c3484 |
addi r3, r3, 4
|
|
Packit |
5c3484 |
b L(11)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(n01): bne cr7, L(n10)
|
|
Packit |
5c3484 |
cmpwi cr0, r0, 0 C n = 2, 6, 10, 14, ...
|
|
Packit |
5c3484 |
lwzx r11, r4, r3
|
|
Packit |
5c3484 |
lwzx r12, r5, r3
|
|
Packit |
5c3484 |
addi r3, r3, 4
|
|
Packit |
5c3484 |
lwzx r0, r4, r3
|
|
Packit |
5c3484 |
ADCSBC r11, r12, r11
|
|
Packit |
5c3484 |
lwzx r6, r5, r3
|
|
Packit |
5c3484 |
addi r3, r3, 4
|
|
Packit |
5c3484 |
ble cr0, L(end)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(lp): lwzx r7, r4, r3
|
|
Packit |
5c3484 |
ADCSBC r0, r6, r0
|
|
Packit |
5c3484 |
lwzx r8, r5, r3
|
|
Packit |
5c3484 |
stwu r11, 4(r3)
|
|
Packit |
5c3484 |
L(01): lwzx r9, r4, r3
|
|
Packit |
5c3484 |
ADCSBC r7, r8, r7
|
|
Packit |
5c3484 |
lwzx r10, r5, r3
|
|
Packit |
5c3484 |
stwu r0, 4(r3)
|
|
Packit |
5c3484 |
L(00): lwzx r11, r4, r3
|
|
Packit |
5c3484 |
ADCSBC r9, r10, r9
|
|
Packit |
5c3484 |
lwzx r12, r5, r3
|
|
Packit |
5c3484 |
stwu r7, 4(r3)
|
|
Packit |
5c3484 |
L(11): lwzx r0, r4, r3
|
|
Packit |
5c3484 |
ADCSBC r11, r12, r11
|
|
Packit |
5c3484 |
lwzx r6, r5, r3
|
|
Packit |
5c3484 |
stwu r9, 4(r3)
|
|
Packit |
5c3484 |
bdnz L(lp)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(end): ADCSBC r0, r6, r0
|
|
Packit |
5c3484 |
stw r11, 4(r3)
|
|
Packit |
5c3484 |
L(ret): stw r0, 8(r3)
|
|
Packit |
5c3484 |
IFADD(` li r3, 0 ')
|
|
Packit |
5c3484 |
IFADD(` addze r3, r3 ')
|
|
Packit |
5c3484 |
IFSUB(` subfe r3, r0, r0')
|
|
Packit |
5c3484 |
IFSUB(` neg r3, r3')
|
|
Packit |
5c3484 |
blr
|
|
Packit |
5c3484 |
EPILOGUE()
|