Blame mpn/sparc64/ultrasparct3/invert_limb.asm
|
Packit |
5c3484 |
dnl SPARC T3/T4/T5 mpn_invert_limb.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Contributed to the GNU project by Torbjörn Granlund.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2013 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 UltraSPARC T3: ?
|
|
Packit |
5c3484 |
C UltraSPARC T4/T5: ?
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C INPUT PARAMETERS
|
|
Packit |
5c3484 |
define(`d', `%o0')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
REGISTER(%g2,#scratch)
|
|
Packit |
5c3484 |
REGISTER(%g3,#scratch)
|
|
Packit |
5c3484 |
PROLOGUE(mpn_invert_limb)
|
|
Packit |
5c3484 |
srlx d, 54, %g1
|
|
Packit |
5c3484 |
LEA64(approx_tab, g2, g3)
|
|
Packit |
5c3484 |
and %g1, 0x1fe, %g1
|
|
Packit |
5c3484 |
srlx d, 24, %g4
|
|
Packit |
5c3484 |
lduh [%g2+%g1], %g3
|
|
Packit |
5c3484 |
add %g4, 1, %g4
|
|
Packit |
5c3484 |
sllx %g3, 11, %g2
|
|
Packit |
5c3484 |
add %g2, -1, %g2
|
|
Packit |
5c3484 |
mulx %g3, %g3, %g3
|
|
Packit |
5c3484 |
mulx %g3, %g4, %g3
|
|
Packit |
5c3484 |
srlx %g3, 40, %g3
|
|
Packit |
5c3484 |
sub %g2, %g3, %g2
|
|
Packit |
5c3484 |
sllx %g2, 60, %g1
|
|
Packit |
5c3484 |
mulx %g2, %g2, %g3
|
|
Packit |
5c3484 |
mulx %g3, %g4, %g4
|
|
Packit |
5c3484 |
sub %g1, %g4, %g1
|
|
Packit |
5c3484 |
srlx %g1, 47, %g1
|
|
Packit |
5c3484 |
sllx %g2, 13, %g2
|
|
Packit |
5c3484 |
add %g1, %g2, %g1
|
|
Packit |
5c3484 |
and d, 1, %g2
|
|
Packit |
5c3484 |
srlx %g1, 1, %g4
|
|
Packit |
5c3484 |
sub %g0, %g2, %g3
|
|
Packit |
5c3484 |
and %g4, %g3, %g3
|
|
Packit |
5c3484 |
srlx d, 1, %g4
|
|
Packit |
5c3484 |
add %g4, %g2, %g2
|
|
Packit |
5c3484 |
mulx %g1, %g2, %g2
|
|
Packit |
5c3484 |
sub %g3, %g2, %g2
|
|
Packit |
5c3484 |
umulxhi(%g1, %g2, %g2)
|
|
Packit |
5c3484 |
srlx %g2, 1, %g2
|
|
Packit |
5c3484 |
sllx %g1, 31, %g1
|
|
Packit |
5c3484 |
add %g2, %g1, %g1
|
|
Packit |
5c3484 |
mulx %g1, d, %g3
|
|
Packit |
5c3484 |
umulxhi(d, %g1, %g4)
|
|
Packit |
5c3484 |
addcc %g3, d, %g0
|
|
Packit |
5c3484 |
addxc( %g4, d, %o0)
|
|
Packit |
5c3484 |
jmp %o7+8
|
|
Packit |
5c3484 |
sub %g1, %o0, %o0
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
RODATA
|
|
Packit |
5c3484 |
ALIGN(2)
|
|
Packit |
5c3484 |
TYPE( approx_tab, object)
|
|
Packit |
5c3484 |
SIZE( approx_tab, 512)
|
|
Packit |
5c3484 |
approx_tab:
|
|
Packit |
5c3484 |
forloop(i,256,512-1,dnl
|
|
Packit |
5c3484 |
` .half eval(0x7fd00/i)
|
|
Packit |
5c3484 |
')dnl
|