|
Packit |
5c3484 |
dnl PowerPC-32 mpn_divexact_by3 -- mpn by 3 exact division
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2002, 2003, 2005, 2006 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: 5
|
|
Packit |
5c3484 |
C 75x (G3): ?
|
|
Packit |
5c3484 |
C 7400,7410 (G4): 8
|
|
Packit |
5c3484 |
C 744x,745x (G4+): 6
|
|
Packit |
5c3484 |
C power4/ppc970: 12
|
|
Packit |
5c3484 |
C power5: ?
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C void mpn_divexact_by3 (mp_ptr dst, mp_srcptr src, mp_size_t size);
|
|
Packit |
5c3484 |
C
|
|
Packit |
5c3484 |
C We avoid the slow subfe instruction and instead rely on an extremely unlikely
|
|
Packit |
5c3484 |
C branch.
|
|
Packit |
5c3484 |
C
|
|
Packit |
5c3484 |
C The mullw has the inverse in the first operand, since 0xAA..AB won't allow
|
|
Packit |
5c3484 |
C any early-out. The src[] data normally won't either, but there's at least
|
|
Packit |
5c3484 |
C a chance, whereas 0xAA..AB never will. If, for instance, src[] is all
|
|
Packit |
5c3484 |
C zeros (not a sensible input of course) we run at 7.0 c/l on ppc750.
|
|
Packit |
5c3484 |
C
|
|
Packit |
5c3484 |
C The mulhwu has the "3" multiplier in the second operand, which lets 750 and
|
|
Packit |
5c3484 |
C 7400 use an early-out.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C INPUT PARAMETERS
|
|
Packit |
5c3484 |
define(`rp', `r3')
|
|
Packit |
5c3484 |
define(`up', `r4')
|
|
Packit |
5c3484 |
define(`n', `r5')
|
|
Packit |
5c3484 |
define(`cy', `r6')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
PROLOGUE(mpn_divexact_by3c)
|
|
Packit |
5c3484 |
lwz r11, 0(up)
|
|
Packit |
5c3484 |
mtctr n
|
|
Packit |
5c3484 |
lis r12, 0xAAAA
|
|
Packit |
5c3484 |
ori r12, r12, 0xAAAB
|
|
Packit |
5c3484 |
li r10, 3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
cmplw cr7, cy, r11
|
|
Packit |
5c3484 |
subf r11, cy, r11
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
mullw r0, r11, r12
|
|
Packit |
5c3484 |
stw r0, 0(rp)
|
|
Packit |
5c3484 |
bdz L(one)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(top): lwzu r9, 4(up)
|
|
Packit |
5c3484 |
mulhwu r7, r0, r10
|
|
Packit |
5c3484 |
bgt- cr7, L(adj) C very unlikely branch
|
|
Packit |
5c3484 |
L(bko): cmplw cr7, r7, r9
|
|
Packit |
5c3484 |
subf r0, r7, r9
|
|
Packit |
5c3484 |
mullw r0, r12, r0
|
|
Packit |
5c3484 |
stwu r0, 4(rp)
|
|
Packit |
5c3484 |
bdnz L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(one): mulhwu r3, r0, r10
|
|
Packit |
5c3484 |
blelr+ cr7
|
|
Packit |
5c3484 |
addi r3, r3, 1
|
|
Packit |
5c3484 |
blr
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(adj): addi r7, r7, 1
|
|
Packit |
5c3484 |
b L(bko)
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
Packit |
5c3484 |
ASM_END()
|