|
Packit |
5c3484 |
dnl AMD64 mpn_gcd_1 -- mpn by 1 gcd.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Based on the K7 gcd_1.asm, by Kevin Ryde. Rehacked for AMD64 by Torbjorn
|
|
Packit |
5c3484 |
dnl Granlund.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
dnl Copyright 2000-2002, 2005, 2009, 2011, 2012 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 cycles/bit (approx)
|
|
Packit |
5c3484 |
C AMD K8,K9 5.21 (4.95)
|
|
Packit |
5c3484 |
C AMD K10 5.15 (5.00)
|
|
Packit |
5c3484 |
C AMD bd1 5.42 (5.14)
|
|
Packit |
5c3484 |
C AMD bobcat 6.71 (6.56)
|
|
Packit |
5c3484 |
C Intel P4 13.5 (12.75)
|
|
Packit |
5c3484 |
C Intel core2 6.20 (6.16)
|
|
Packit |
5c3484 |
C Intel NHM 6.49 (6.25)
|
|
Packit |
5c3484 |
C Intel SBR 7.75 (7.57)
|
|
Packit |
5c3484 |
C Intel atom 8.77 (8.54)
|
|
Packit |
5c3484 |
C VIA nano 6.60 (6.20)
|
|
Packit |
5c3484 |
C Numbers measured with: speed -CD -s16-64 -t48 mpn_gcd_1
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C ctz_table[n] is the number of trailing zeros on n, or MAXSHIFT if n==0.
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
deflit(MAXSHIFT, 7)
|
|
Packit |
5c3484 |
deflit(MASK, eval((m4_lshift(1,MAXSHIFT))-1))
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
DEF_OBJECT(ctz_table,64)
|
|
Packit |
5c3484 |
.byte MAXSHIFT
|
|
Packit |
5c3484 |
forloop(i,1,MASK,
|
|
Packit |
5c3484 |
` .byte m4_count_trailing_zeros(i)
|
|
Packit |
5c3484 |
')
|
|
Packit |
5c3484 |
END_OBJECT(ctz_table)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C Threshold of when to call bmod when U is one limb. Should be about
|
|
Packit |
5c3484 |
C (time_in_cycles(bmod_1,1) + call_overhead) / (cycles/bit).
|
|
Packit |
5c3484 |
define(`BMOD_THRES_LOG2', 8)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C INPUT PARAMETERS
|
|
Packit |
5c3484 |
define(`up', `%rdi')
|
|
Packit |
5c3484 |
define(`n', `%rsi')
|
|
Packit |
5c3484 |
define(`v0', `%rdx')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ABI_SUPPORT(DOS64)
|
|
Packit |
5c3484 |
ABI_SUPPORT(STD64)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
IFDOS(`define(`STACK_ALLOC', 40)')
|
|
Packit |
5c3484 |
IFSTD(`define(`STACK_ALLOC', 8)')
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ASM_START()
|
|
Packit |
5c3484 |
TEXT
|
|
Packit |
5c3484 |
ALIGN(16)
|
|
Packit |
5c3484 |
PROLOGUE(mpn_gcd_1)
|
|
Packit |
5c3484 |
FUNC_ENTRY(3)
|
|
Packit |
5c3484 |
mov (up), %rax C U low limb
|
|
Packit |
5c3484 |
mov $-1, R32(%rcx)
|
|
Packit |
5c3484 |
or v0, %rax C x | y
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(twos):
|
|
Packit |
5c3484 |
inc R32(%rcx)
|
|
Packit |
5c3484 |
shr %rax
|
|
Packit |
5c3484 |
jnc L(twos)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
shr R8(%rcx), v0
|
|
Packit |
5c3484 |
push %rcx C common twos
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(divide_strip_y):
|
|
Packit |
5c3484 |
shr v0
|
|
Packit |
5c3484 |
jnc L(divide_strip_y)
|
|
Packit |
5c3484 |
adc v0, v0
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
cmp $1, n
|
|
Packit |
5c3484 |
jnz L(reduce_nby1)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C Both U and V are single limbs, reduce with bmod if u0 >> v0.
|
|
Packit |
5c3484 |
mov (up), %r8
|
|
Packit |
5c3484 |
mov %r8, %rax
|
|
Packit |
5c3484 |
shr $BMOD_THRES_LOG2, %r8
|
|
Packit |
5c3484 |
cmp %r8, v0
|
|
Packit |
5c3484 |
ja L(noreduce)
|
|
Packit |
5c3484 |
push v0
|
|
Packit |
5c3484 |
sub $STACK_ALLOC, %rsp C maintain ABI required rsp alignment
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(bmod):
|
|
Packit |
5c3484 |
IFDOS(` mov %rdx, %r8 ')
|
|
Packit |
5c3484 |
IFDOS(` mov %rsi, %rdx ')
|
|
Packit |
5c3484 |
IFDOS(` mov %rdi, %rcx ')
|
|
Packit |
5c3484 |
ASSERT(nz, `test $15, %rsp')
|
|
Packit |
5c3484 |
CALL( mpn_modexact_1_odd)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(reduced):
|
|
Packit |
5c3484 |
add $STACK_ALLOC, %rsp
|
|
Packit |
5c3484 |
pop %rdx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(noreduce):
|
|
Packit |
5c3484 |
LEA( ctz_table, %rsi)
|
|
Packit |
5c3484 |
test %rax, %rax
|
|
Packit |
5c3484 |
mov %rax, %rcx
|
|
Packit |
5c3484 |
jnz L(mid)
|
|
Packit |
5c3484 |
jmp L(end)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(reduce_nby1):
|
|
Packit |
5c3484 |
push v0
|
|
Packit |
5c3484 |
sub $STACK_ALLOC, %rsp C maintain ABI required rsp alignment
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
cmp $BMOD_1_TO_MOD_1_THRESHOLD, n
|
|
Packit |
5c3484 |
jl L(bmod)
|
|
Packit |
5c3484 |
IFDOS(` mov %rdx, %r8 ')
|
|
Packit |
5c3484 |
IFDOS(` mov %rsi, %rdx ')
|
|
Packit |
5c3484 |
IFDOS(` mov %rdi, %rcx ')
|
|
Packit |
5c3484 |
ASSERT(nz, `test $15, %rsp')
|
|
Packit |
5c3484 |
CALL( mpn_mod_1)
|
|
Packit |
5c3484 |
jmp L(reduced)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16) C K8 BC P4 NHM SBR
|
|
Packit |
5c3484 |
L(top): cmovc %rcx, %rax C if x-y < 0 0
|
|
Packit |
5c3484 |
cmovc %rdi, %rdx C use x,y-x 0
|
|
Packit |
5c3484 |
L(mid): and $MASK, R32(%rcx) C 0
|
|
Packit |
5c3484 |
movzbl (%rsi,%rcx), R32(%rcx) C 1
|
|
Packit |
5c3484 |
jz L(shift_alot) C 1
|
|
Packit |
5c3484 |
shr R8(%rcx), %rax C 3
|
|
Packit |
5c3484 |
mov %rax, %rdi C 4
|
|
Packit |
5c3484 |
mov %rdx, %rcx C 3
|
|
Packit |
5c3484 |
sub %rax, %rcx C 4
|
|
Packit |
5c3484 |
sub %rdx, %rax C 4
|
|
Packit |
5c3484 |
jnz L(top) C 5
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(end): pop %rcx
|
|
Packit |
5c3484 |
mov %rdx, %rax
|
|
Packit |
5c3484 |
shl R8(%rcx), %rax
|
|
Packit |
5c3484 |
FUNC_EXIT()
|
|
Packit |
5c3484 |
ret
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(shift_alot):
|
|
Packit |
5c3484 |
shr $MAXSHIFT, %rax
|
|
Packit |
5c3484 |
mov %rax, %rcx
|
|
Packit |
5c3484 |
jmp L(mid)
|
|
Packit |
5c3484 |
EPILOGUE()
|
|
rpm-build |
c3cd4f |
CF_PROT
|