|
Packit |
5c3484 |
dnl AMD64 mpn_gcd_1 optimised for Intel C2, NHM, SBR and AMD K10, BD.
|
|
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 8.50
|
|
Packit |
5c3484 |
C AMD K10 4.30
|
|
Packit |
5c3484 |
C AMD bd1 5.00
|
|
Packit |
5c3484 |
C AMD bobcat 10.0
|
|
Packit |
5c3484 |
C Intel P4 18.6
|
|
Packit |
5c3484 |
C Intel core2 3.83
|
|
Packit |
5c3484 |
C Intel NHM 5.17
|
|
Packit |
5c3484 |
C Intel SBR 4.69
|
|
Packit |
5c3484 |
C Intel atom 17.0
|
|
Packit |
5c3484 |
C VIA nano 5.44
|
|
Packit |
5c3484 |
C Numbers measured with: speed -CD -s16-64 -t48 mpn_gcd_1
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
C TODO
|
|
Packit |
5c3484 |
C * Optimise inner-loop for specific CPUs.
|
|
Packit |
5c3484 |
C * Use DIV for 1-by-1 reductions, at least for some CPUs.
|
|
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', 6)
|
|
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 |
C Undo some configure cleverness.
|
|
Packit |
5c3484 |
C The problem is that C only defines the '1c' variant, and that configure
|
|
Packit |
5c3484 |
C therefore considers modexact_1c to be the base function. It then adds a
|
|
Packit |
5c3484 |
C special fat rule for mpn_modexact_1_odd, messing up things when a cpudep
|
|
Packit |
5c3484 |
C gcd_1 exists without a corresponding cpudep mode1o.
|
|
Packit |
5c3484 |
ifdef(`WANT_FAT_BINARY', `
|
|
Packit |
5c3484 |
define(`mpn_modexact_1_odd', `MPN_PREFIX`modexact_1_odd_x86_64'')')
|
|
Packit |
5c3484 |
|
|
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 |
or v0, %rax
|
|
Packit |
5c3484 |
bsf %rax, %rax C min(ctz(u0),ctz(v0))
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
bsf v0, %rcx
|
|
Packit |
5c3484 |
shr R8(%rcx), v0
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
push %rax C preserve common twos over call
|
|
Packit |
5c3484 |
push v0 C preserve v0 argument over call
|
|
Packit |
5c3484 |
sub $STACK_ALLOC, %rsp C maintain ABI required rsp alignment
|
|
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(reduced)
|
|
Packit |
5c3484 |
jmp L(bmod)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
L(reduce_nby1):
|
|
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 |
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 |
L(reduced):
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
add $STACK_ALLOC, %rsp
|
|
Packit |
5c3484 |
pop %rdx
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
bsf %rax, %rcx
|
|
Packit |
5c3484 |
C test %rax, %rax C FIXME: does this lower latency?
|
|
Packit |
5c3484 |
jnz L(mid)
|
|
Packit |
5c3484 |
jmp L(end)
|
|
Packit |
5c3484 |
|
|
Packit |
5c3484 |
ALIGN(16) C K10 BD C2 NHM SBR
|
|
Packit |
5c3484 |
L(top): cmovc %r10, %rax C if x-y < 0 0,3 0,3 0,6 0,5 0,5
|
|
Packit |
5c3484 |
cmovc %r9, %rdx C use x,y-x 0,3 0,3 2,8 1,7 1,7
|
|
Packit |
5c3484 |
L(mid): shr R8(%rcx), %rax C 1,7 1,6 2,8 2,8 2,8
|
|
Packit |
5c3484 |
mov %rdx, %r10 C 1 1 4 3 3
|
|
Packit |
5c3484 |
sub %rax, %r10 C 2 2 5 4 4
|
|
Packit |
5c3484 |
bsf %r10, %rcx C 3 3 6 5 5
|
|
Packit |
5c3484 |
mov %rax, %r9 C 2 2 3 3 4
|
|
Packit |
5c3484 |
sub %rdx, %rax C 2 2 4 3 4
|
|
Packit |
5c3484 |
jnz L(top) C
|
|
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 |
EPILOGUE()
|
|
rpm-build |
c3cd4f |
CF_PROT
|