|
Packit |
5c3484 |
dnl Intel P5 mpn_mod_34lsub1 -- mpn remainder modulo 2**24-1.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2000-2002 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 |
|
|
Packit |
5c3484 |
C P5: 1.66 cycles/limb
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C mp_limb_t mpn_mod_34lsub1 (mp_srcptr src, mp_size_t size)
|
|
Packit |
5c3484 |
C
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
defframe(PARAM_SIZE, 8)
|
|
Packit |
5c3484 |
defframe(PARAM_SRC, 4)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
TEXT
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
PROLOGUE(mpn_mod_34lsub1)
|
|
Packit |
5c3484 |
deflit(`FRAME',0)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl PARAM_SIZE, %ecx
|
|
Packit |
5c3484 |
movl PARAM_SRC, %edx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
subl $2, %ecx
|
|
Packit |
5c3484 |
ja L(three_or_more)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl (%edx), %eax
|
|
Packit |
5c3484 |
jne L(one)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl 4(%edx), %ecx
|
|
Packit |
5c3484 |
movl %eax, %edx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shrl $24, %edx
|
|
Packit |
5c3484 |
andl $0xFFFFFF, %eax
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addl %edx, %eax
|
|
Packit |
5c3484 |
movl %ecx, %edx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shrl $16, %ecx
|
|
Packit |
5c3484 |
andl $0xFFFF, %edx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shll $8, %edx
|
|
Packit |
5c3484 |
addl %ecx, %eax
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addl %edx, %eax
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(one):
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(three_or_more):
|
|
Packit |
5c3484 |
C eax
|
|
Packit |
5c3484 |
C ebx
|
|
Packit |
5c3484 |
C ecx size-2
|
|
Packit |
5c3484 |
C edx src
|
|
Packit |
5c3484 |
C esi
|
|
Packit |
5c3484 |
C edi
|
|
Packit |
5c3484 |
C ebp
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
pushl %ebx FRAME_pushl()
|
|
Packit |
5c3484 |
pushl %esi FRAME_pushl()
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
pushl %edi FRAME_pushl()
|
|
Packit |
5c3484 |
pushl %ebp FRAME_pushl()
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
xorl %esi, %esi C 0mod3
|
|
Packit |
5c3484 |
xorl %edi, %edi C 1mod3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
xorl %ebp, %ebp C 2mod3, and clear carry
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(top):
|
|
Packit |
5c3484 |
C eax scratch
|
|
Packit |
5c3484 |
C ebx scratch
|
|
Packit |
5c3484 |
C ecx counter, limbs
|
|
Packit |
5c3484 |
C edx src
|
|
Packit |
5c3484 |
C esi 0mod3
|
|
Packit |
5c3484 |
C edi 1mod3
|
|
Packit |
5c3484 |
C ebp 2mod3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl (%edx), %eax
|
|
Packit |
5c3484 |
movl 4(%edx), %ebx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
adcl %eax, %esi
|
|
Packit |
5c3484 |
movl 8(%edx), %eax
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
adcl %ebx, %edi
|
|
Packit |
5c3484 |
leal 12(%edx), %edx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
adcl %eax, %ebp
|
|
Packit |
5c3484 |
leal -2(%ecx), %ecx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
decl %ecx
|
|
Packit |
5c3484 |
jg L(top)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C ecx is -2, -1 or 0, representing 0, 1 or 2 more limbs, respectively
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl $0xFFFFFFFF, %ebx C mask
|
|
Packit |
5c3484 |
incl %ecx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
js L(combine) C 0 more
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl (%edx), %eax
|
|
Packit |
5c3484 |
movl $0xFFFFFF00, %ebx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
adcl %eax, %esi
|
|
Packit |
5c3484 |
decl %ecx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
js L(combine) C 1 more
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
movl 4(%edx), %eax
|
|
Packit |
5c3484 |
movl $0xFFFF0000, %ebx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
adcl %eax, %edi
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(combine):
|
|
Packit |
5c3484 |
C eax
|
|
Packit |
5c3484 |
C ebx mask
|
|
Packit |
5c3484 |
C ecx
|
|
Packit |
5c3484 |
C edx
|
|
Packit |
5c3484 |
C esi 0mod3
|
|
Packit |
5c3484 |
C edi 1mod3
|
|
Packit |
5c3484 |
C ebp 2mod3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
sbbl %ecx, %ecx C carry
|
|
Packit |
5c3484 |
movl %esi, %eax C 0mod3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
andl %ebx, %ecx C masked for position
|
|
Packit |
5c3484 |
andl $0xFFFFFF, %eax C 0mod3 low
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shrl $24, %esi C 0mod3 high
|
|
Packit |
5c3484 |
subl %ecx, %eax C apply carry
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addl %esi, %eax C apply 0mod3
|
|
Packit |
5c3484 |
movl %edi, %ebx C 1mod3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shrl $16, %edi C 1mod3 high
|
|
Packit |
5c3484 |
andl $0x0000FFFF, %ebx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shll $8, %ebx C 1mod3 low
|
|
Packit |
5c3484 |
addl %edi, %eax C apply 1mod3 high
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addl %ebx, %eax C apply 1mod3 low
|
|
Packit |
5c3484 |
movl %ebp, %ebx C 2mod3
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shrl $8, %ebp C 2mod3 high
|
|
Packit |
5c3484 |
andl $0xFF, %ebx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shll $16, %ebx C 2mod3 low
|
|
Packit |
5c3484 |
addl %ebp, %eax C apply 2mod3 high
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
addl %ebx, %eax C apply 2mod3 low
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
popl %ebp
|
|
Packit |
5c3484 |
popl %edi
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
popl %esi
|
|
Packit |
5c3484 |
popl %ebx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
EPILOGUE()
|