|
Packit |
5c3484 |
dnl Alpha mpn_addlsh1_n/mpn_sublsh1_n -- rp[] = up[] +- (vp[] << 1).
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2003, 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 EV4: ?
|
|
Packit |
5c3484 |
C EV5: 6.25
|
|
Packit |
5c3484 |
C EV6: 4.5
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`rp',`r16')
|
|
Packit |
5c3484 |
define(`up',`r17')
|
|
Packit |
5c3484 |
define(`vp',`r18')
|
|
Packit |
5c3484 |
define(`n', `r19')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`u0', `r8')
|
|
Packit |
5c3484 |
define(`u1', `r1')
|
|
Packit |
5c3484 |
define(`v0', `r4')
|
|
Packit |
5c3484 |
define(`v1', `r5')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
define(`cy0', `r0')
|
|
Packit |
5c3484 |
define(`cy1', `r20')
|
|
Packit |
5c3484 |
define(`cy', `r22')
|
|
Packit |
5c3484 |
define(`rr', `r24')
|
|
Packit |
5c3484 |
define(`ps', `r25')
|
|
Packit |
5c3484 |
define(`sl', `r28')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ifdef(`OPERATION_addlsh1_n',`
|
|
Packit |
5c3484 |
define(ADDSUB, addq)
|
|
Packit |
5c3484 |
define(CARRY, `cmpult $1,$2,$3')
|
|
Packit |
5c3484 |
define(func, mpn_addlsh1_n)
|
|
Packit |
5c3484 |
')
|
|
Packit |
5c3484 |
ifdef(`OPERATION_sublsh1_n',`
|
|
Packit |
5c3484 |
define(ADDSUB, subq)
|
|
Packit |
5c3484 |
define(CARRY, `cmpult $2,$1,$3')
|
|
Packit |
5c3484 |
define(func, mpn_sublsh1_n)
|
|
Packit |
5c3484 |
')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
MULFUNC_PROLOGUE(mpn_addlsh1_n mpn_sublsh1_n)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
PROLOGUE(func)
|
|
Packit |
5c3484 |
and n, 2, cy0
|
|
Packit |
5c3484 |
blbs n, L(bx1)
|
|
Packit |
5c3484 |
L(bx0): ldq v1, 0(vp)
|
|
Packit |
5c3484 |
ldq u1, 0(up)
|
|
Packit |
5c3484 |
nop
|
|
Packit |
5c3484 |
bne cy0, L(b10)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b00): lda vp, 48(vp)
|
|
Packit |
5c3484 |
lda up, -16(up)
|
|
Packit |
5c3484 |
lda rp, -8(rp)
|
|
Packit |
5c3484 |
br r31, L(lo0)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b10): lda vp, 32(vp)
|
|
Packit |
5c3484 |
lda rp, 8(rp)
|
|
Packit |
5c3484 |
lda cy0, 0(r31)
|
|
Packit |
5c3484 |
br r31, L(lo2)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(bx1): ldq v0, 0(vp)
|
|
Packit |
5c3484 |
ldq u0, 0(up)
|
|
Packit |
5c3484 |
lda cy1, 0(r31)
|
|
Packit |
5c3484 |
beq cy0, L(b01)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b11): lda vp, 40(vp)
|
|
Packit |
5c3484 |
lda up, -24(up)
|
|
Packit |
5c3484 |
lda rp, 16(rp)
|
|
Packit |
5c3484 |
br r31, L(lo3)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(b01): lda n, -4(n)
|
|
Packit |
5c3484 |
ble n, L(end)
|
|
Packit |
5c3484 |
lda vp, 24(vp)
|
|
Packit |
5c3484 |
lda up, -8(up)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
L(top): addq v0, v0, sl C left shift vlimb
|
|
Packit |
5c3484 |
ldq v1, -16(vp)
|
|
Packit |
5c3484 |
ADDSUB u0, sl, ps C ulimb + (vlimb << 1)
|
|
Packit |
5c3484 |
cmplt v0, r31, cy0 C carry out #1
|
|
Packit |
5c3484 |
ldq u1, 16(up)
|
|
Packit |
5c3484 |
ADDSUB ps, cy1, rr C consume carry from previous operation
|
|
Packit |
5c3484 |
CARRY( ps, u0, cy) C carry out #2
|
|
Packit |
5c3484 |
stq rr, 0(rp)
|
|
Packit |
5c3484 |
addq cy, cy0, cy0 C combine carry out #1 and #2
|
|
Packit |
5c3484 |
CARRY( rr, ps, cy) C carry out #3
|
|
Packit |
5c3484 |
addq cy, cy0, cy0 C final carry out
|
|
Packit |
5c3484 |
lda vp, 32(vp) C bookkeeping
|
|
Packit |
5c3484 |
L(lo0): addq v1, v1, sl
|
|
Packit |
5c3484 |
ldq v0, -40(vp)
|
|
Packit |
5c3484 |
ADDSUB u1, sl, ps
|
|
Packit |
5c3484 |
cmplt v1, r31, cy1
|
|
Packit |
5c3484 |
ldq u0, 24(up)
|
|
Packit |
5c3484 |
ADDSUB ps, cy0, rr
|
|
Packit |
5c3484 |
CARRY( ps, u1, cy)
|
|
Packit |
5c3484 |
stq rr, 8(rp)
|
|
Packit |
5c3484 |
addq cy, cy1, cy1
|
|
Packit |
5c3484 |
CARRY( rr, ps, cy)
|
|
Packit |
5c3484 |
addq cy, cy1, cy1
|
|
Packit |
5c3484 |
lda rp, 32(rp) C bookkeeping
|
|
Packit |
5c3484 |
L(lo3): addq v0, v0, sl
|
|
Packit |
5c3484 |
ldq v1, -32(vp)
|
|
Packit |
5c3484 |
ADDSUB u0, sl, ps
|
|
Packit |
5c3484 |
cmplt v0, r31, cy0
|
|
Packit |
5c3484 |
ldq u1, 32(up)
|
|
Packit |
5c3484 |
ADDSUB ps, cy1, rr
|
|
Packit |
5c3484 |
CARRY( ps, u0, cy)
|
|
Packit |
5c3484 |
stq rr, -16(rp)
|
|
Packit |
5c3484 |
addq cy, cy0, cy0
|
|
Packit |
5c3484 |
CARRY( rr, ps, cy)
|
|
Packit |
5c3484 |
addq cy, cy0, cy0
|
|
Packit |
5c3484 |
lda up, 32(up) C bookkeeping
|
|
Packit |
5c3484 |
L(lo2): addq v1, v1, sl
|
|
Packit |
5c3484 |
ldq v0, -24(vp)
|
|
Packit |
5c3484 |
ADDSUB u1, sl, ps
|
|
Packit |
5c3484 |
cmplt v1, r31, cy1
|
|
Packit |
5c3484 |
ldq u0, 8(up)
|
|
Packit |
5c3484 |
ADDSUB ps, cy0, rr
|
|
Packit |
5c3484 |
CARRY( ps, u1, cy)
|
|
Packit |
5c3484 |
stq rr, -8(rp)
|
|
Packit |
5c3484 |
addq cy, cy1, cy1
|
|
Packit |
5c3484 |
CARRY( rr, ps, cy)
|
|
Packit |
5c3484 |
addq cy, cy1, cy1
|
|
Packit |
5c3484 |
lda n, -4(n) C bookkeeping
|
|
Packit |
5c3484 |
bgt n, L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(end): addq v0, v0, sl
|
|
Packit |
5c3484 |
ADDSUB u0, sl, ps
|
|
Packit |
5c3484 |
ADDSUB ps, cy1, rr
|
|
Packit |
5c3484 |
cmplt v0, r31, cy0
|
|
Packit |
5c3484 |
CARRY( ps, u0, cy)
|
|
Packit |
5c3484 |
stq rr, 0(rp)
|
|
Packit |
5c3484 |
addq cy, cy0, cy0
|
|
Packit |
5c3484 |
CARRY( rr, ps, cy)
|
|
Packit |
5c3484 |
addq cy, cy0, r0
|
|
Packit |
5c3484 |
ret r31,(r26),1
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
Packit |
5c3484 |
ASM_END()
|