dnl X86-64 mpn_redc_1 optimised for AMD K8-K10.
dnl Contributed to the GNU project by Torbjörn Granlund.
dnl Copyright 2004, 2008, 2013 Free Software Foundation, Inc.
dnl This file is part of the GNU MP Library.
dnl
dnl The GNU MP Library is free software; you can redistribute it and/or modify
dnl it under the terms of either:
dnl
dnl * the GNU Lesser General Public License as published by the Free
dnl Software Foundation; either version 3 of the License, or (at your
dnl option) any later version.
dnl
dnl or
dnl
dnl * the GNU General Public License as published by the Free Software
dnl Foundation; either version 2 of the License, or (at your option) any
dnl later version.
dnl
dnl or both in parallel, as here.
dnl
dnl The GNU MP Library is distributed in the hope that it will be useful, but
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
dnl for more details.
dnl
dnl You should have received copies of the GNU General Public License and the
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
dnl see https://www.gnu.org/licenses/.
include(`../config.m4')
C cycles/limb
C AMD K8,K9 ?
C AMD K10 ?
C AMD bull ?
C AMD pile ?
C AMD steam ?
C AMD bobcat ?
C AMD jaguar ?
C Intel P4 ?
C Intel core ?
C Intel NHM ?
C Intel SBR ?
C Intel IBR ?
C Intel HWL ?
C Intel BWL ?
C Intel atom ?
C VIA nano ?
C The inner loops of this code are the result of running a code generation and
C optimisation tool suite written by David Harvey and Torbjörn Granlund.
C TODO
C * Micro-optimise, none performed thus far.
C * This looks different from other current redc_1.asm variants. Consider
C adapting this to the mainstream style.
C * Is this code really faster than more approaches which compute q0 later?
C Is the use of a jump jump table faster? Or is the edge of this due to the
C inlined add_n code?
C * Put initial m[0] x q0 computation in header.
C * Put basecases at the file's end, single them out before the pushes.
define(`rp', `%rdi') C rcx
define(`up', `%rsi') C rdx
define(`mp_param', `%rdx') C r8
define(`n', `%rcx') C r9
define(`u0inv', `%r8') C stack
define(`i', `%r11')
define(`nneg', `%r12')
define(`mp', `%r13')
define(`q0', `%rbp')
define(`vp', `%rdx')
ABI_SUPPORT(DOS64)
ABI_SUPPORT(STD64)
ASM_START()
TEXT
ALIGN(32)
PROLOGUE(mpn_redc_1)
FUNC_ENTRY(4)
IFDOS(` mov 56(%rsp), %r8 ')
push %rbp
mov (up), q0 C up[0]
push %rbx
imul u0inv, q0 C first q0, for all execution paths
push %r12
push %r13
push %r14
push %r15
mov n, nneg
neg nneg
lea (mp_param,n,8), mp C mp += n
lea -16(up,n,8), up C up += n
mov R32(n), R32(%rax)
and $3, R32(%rax)
lea 4(%rax), %r9
cmp $4, R32(n)
cmovg %r9, %rax
lea L(tab)(%rip), %r9
ifdef(`PIC',`
movslq (%r9,%rax,4), %rax
add %r9, %rax
jmp *%rax
',`
jmp *(%r9,%rax,8)
')
JUMPTABSECT
ALIGN(8)
L(tab): JMPENT( L(0), L(tab))
JMPENT( L(1), L(tab))
JMPENT( L(2), L(tab))
JMPENT( L(3), L(tab))
JMPENT( L(0m4), L(tab))
JMPENT( L(1m4), L(tab))
JMPENT( L(2m4), L(tab))
JMPENT( L(3m4), L(tab))
TEXT
ALIGN(16)
L(1): mov (mp_param), %rax
mul q0
add 8(up), %rax
adc 16(up), %rdx
mov %rdx, (rp)
mov $0, R32(%rax)
adc R32(%rax), R32(%rax)
jmp L(ret)
ALIGN(16)
L(2): mov (mp_param), %rax
mul q0
xor R32(%r14), R32(%r14)
mov %rax, %r10
mov -8(mp), %rax
mov %rdx, %r9
mul q0
add (up), %r10
adc %rax, %r9
adc %rdx, %r14
add 8(up), %r9
adc $0, %r14
mov %r9, q0
imul u0inv, q0
mov -16(mp), %rax
mul q0
xor R32(%rbx), R32(%rbx)
mov %rax, %r10
mov -8(mp), %rax
mov %rdx, %r11
mul q0
add %r9, %r10
adc %rax, %r11
adc %rdx, %rbx
add 16(up), %r11
adc $0, %rbx
xor R32(%rax), R32(%rax)
add %r11, %r14
adc 24(up), %rbx
mov %r14, (rp)
mov %rbx, 8(rp)
adc R32(%rax), R32(%rax)
jmp L(ret)
L(3): mov (mp_param), %rax
mul q0
mov %rax, %rbx
mov %rdx, %r10
mov -16(mp), %rax
mul q0
xor R32(%r9), R32(%r9)
xor R32(%r14), R32(%r14)
add -8(up), %rbx
adc %rax, %r10
mov -8(mp), %rax
adc %rdx, %r9
mul q0
add (up), %r10
mov %r10, (up)
adc %rax, %r9
adc %rdx, %r14
mov %r10, q0
imul u0inv, q0
add %r9, 8(up)
adc $0, %r14
mov %r14, -8(up)
mov -24(mp), %rax
mul q0
mov %rax, %rbx
mov %rdx, %r10
mov -16(mp), %rax
mul q0
xor R32(%r9), R32(%r9)
xor R32(%r14), R32(%r14)
add (up), %rbx
adc %rax, %r10
mov -8(mp), %rax
adc %rdx, %r9
mul q0
add 8(up), %r10
mov %r10, 8(up)
adc %rax, %r9
adc %rdx, %r14
mov %r10, q0
imul u0inv, q0
add %r9, 16(up)
adc $0, %r14
mov %r14, (up)
mov -24(mp), %rax
mul q0
mov %rax, %rbx
mov %rdx, %r10
mov -16(mp), %rax
mul q0
xor R32(%r9), R32(%r9)
xor R32(%r14), R32(%r14)
add 8(up), %rbx
adc %rax, %r10
mov -8(mp), %rax
adc %rdx, %r9
mul q0
add 16(up), %r10
adc %rax, %r9
adc %rdx, %r14
add 24(up), %r9
adc $0, %r14
xor R32(%rax), R32(%rax)
add -8(up), %r10
adc (up), %r9
adc 32(up), %r14
mov %r10, (rp)
mov %r9, 8(rp)
mov %r14, 16(rp)
adc R32(%rax), R32(%rax)
jmp L(ret)
ALIGN(16)
L(2m4):
L(lo2): mov (mp,nneg,8), %rax
mul q0
xor R32(%r14), R32(%r14)
xor R32(%rbx), R32(%rbx)
mov %rax, %r10
mov 8(mp,nneg,8), %rax
mov 24(up,nneg,8), %r15
mov %rdx, %r9
mul q0
add 16(up,nneg,8), %r10
adc %rax, %r9
mov 16(mp,nneg,8), %rax
adc %rdx, %r14
mul q0
mov $0, R32(%r10) C xor?
lea 2(nneg), i
add %r9, %r15
imul u0inv, %r15
jmp L(e2)
ALIGN(16)
L(li2): add %r10, (up,i,8)
adc %rax, %r9
mov (mp,i,8), %rax
adc %rdx, %r14
xor R32(%r10), R32(%r10)
mul q0
L(e2): add %r9, 8(up,i,8)
adc %rax, %r14
adc %rdx, %rbx
mov 8(mp,i,8), %rax
mul q0
add %r14, 16(up,i,8)
adc %rax, %rbx
adc %rdx, %r10
mov 16(mp,i,8), %rax
mul q0
add %rbx, 24(up,i,8)
mov $0, R32(%r14) C zero
mov %r14, %rbx C zero
adc %rax, %r10
mov 24(mp,i,8), %rax
mov %r14, %r9 C zero
adc %rdx, %r9
mul q0
add $4, i
js L(li2)
L(le2): add %r10, (up)
adc %rax, %r9
adc %r14, %rdx
add %r9, 8(up)
adc $0, %rdx
mov %rdx, 16(up,nneg,8) C up[0]
add $8, up
mov %r15, q0
dec n
jnz L(lo2)
mov nneg, n
sar $2, n
lea 32(up,nneg,8), up
lea (up,nneg,8), vp
mov -16(up), %r8
mov -8(up), %r9
add -16(vp), %r8
adc -8(vp), %r9
mov %r8, (rp)
mov %r9, 8(rp)
lea 16(rp), rp
jmp L(addx)
ALIGN(16)
L(1m4):
L(lo1): mov (mp,nneg,8), %rax
xor %r9, %r9
xor R32(%rbx), R32(%rbx)
mul q0
mov %rax, %r9
mov 8(mp,nneg,8), %rax
mov 24(up,nneg,8), %r15
mov %rdx, %r14
mov $0, R32(%r10) C xor?
mul q0
add 16(up,nneg,8), %r9
adc %rax, %r14
adc %rdx, %rbx
mov 16(mp,nneg,8), %rax
mul q0
lea 1(nneg), i
add %r14, %r15
imul u0inv, %r15
jmp L(e1)
ALIGN(16)
L(li1): add %r10, (up,i,8)
adc %rax, %r9
mov (mp,i,8), %rax
adc %rdx, %r14
xor R32(%r10), R32(%r10)
mul q0
add %r9, 8(up,i,8)
adc %rax, %r14
adc %rdx, %rbx
mov 8(mp,i,8), %rax
mul q0
L(e1): add %r14, 16(up,i,8)
adc %rax, %rbx
adc %rdx, %r10
mov 16(mp,i,8), %rax
mul q0
add %rbx, 24(up,i,8)
mov $0, R32(%r14) C zero
mov %r14, %rbx C zero
adc %rax, %r10
mov 24(mp,i,8), %rax
mov %r14, %r9 C zero
adc %rdx, %r9
mul q0
add $4, i
js L(li1)
L(le1): add %r10, (up)
adc %rax, %r9
adc %r14, %rdx
add %r9, 8(up)
adc $0, %rdx
mov %rdx, 16(up,nneg,8) C up[0]
add $8, up
mov %r15, q0
dec n
jnz L(lo1)
mov nneg, n
sar $2, n
lea 24(up,nneg,8), up
lea (up,nneg,8), vp
mov -8(up), %r8
add -8(vp), %r8
mov %r8, (rp)
lea 8(rp), rp
jmp L(addx)
ALIGN(16)
L(0):
L(0m4):
L(lo0): mov (mp,nneg,8), %rax
mov nneg, i
mul q0
xor R32(%r10), R32(%r10)
mov %rax, %r14
mov %rdx, %rbx
mov 8(mp,nneg,8), %rax
mov 24(up,nneg,8), %r15
mul q0
add 16(up,nneg,8), %r14
adc %rax, %rbx
adc %rdx, %r10
add %rbx, %r15
imul u0inv, %r15
jmp L(e0)
ALIGN(16)
L(li0): add %r10, (up,i,8)
adc %rax, %r9
mov (mp,i,8), %rax
adc %rdx, %r14
xor R32(%r10), R32(%r10)
mul q0
add %r9, 8(up,i,8)
adc %rax, %r14
adc %rdx, %rbx
mov 8(mp,i,8), %rax
mul q0
add %r14, 16(up,i,8)
adc %rax, %rbx
adc %rdx, %r10
L(e0): mov 16(mp,i,8), %rax
mul q0
add %rbx, 24(up,i,8)
mov $0, R32(%r14) C zero
mov %r14, %rbx C zero
adc %rax, %r10
mov 24(mp,i,8), %rax
mov %r14, %r9 C zero
adc %rdx, %r9
mul q0
add $4, i
js L(li0)
L(le0): add %r10, (up)
adc %rax, %r9
adc %r14, %rdx
add %r9, 8(up)
adc $0, %rdx
mov %rdx, 16(up,nneg,8) C up[0]
add $8, up
mov %r15, q0
dec n
jnz L(lo0)
mov nneg, n
sar $2, n
clc
lea 16(up,nneg,8), up
lea (up,nneg,8), vp
jmp L(addy)
ALIGN(16)
L(3m4):
L(lo3): mov (mp,nneg,8), %rax
mul q0
mov %rax, %rbx
mov %rdx, %r10
mov 8(mp,nneg,8), %rax
mov 24(up,nneg,8), %r15
mul q0
add 16(up,nneg,8), %rbx C result is zero, might carry
mov $0, R32(%rbx) C zero
mov %rbx, %r14 C zero
adc %rax, %r10
mov 16(mp,nneg,8), %rax
mov %r14, %r9 C zero
adc %rdx, %r9
add %r10, %r15
mul q0
lea 3(nneg), i
imul u0inv, %r15
C jmp L(li3)
ALIGN(16)
L(li3): add %r10, (up,i,8)
adc %rax, %r9
mov (mp,i,8), %rax
adc %rdx, %r14
xor R32(%r10), R32(%r10)
mul q0
add %r9, 8(up,i,8)
adc %rax, %r14
adc %rdx, %rbx
mov 8(mp,i,8), %rax
mul q0
add %r14, 16(up,i,8)
adc %rax, %rbx
adc %rdx, %r10
mov 16(mp,i,8), %rax
mul q0
add %rbx, 24(up,i,8)
mov $0, R32(%r14) C zero
mov %r14, %rbx C zero
adc %rax, %r10
mov 24(mp,i,8), %rax
mov %r14, %r9 C zero
adc %rdx, %r9
mul q0
add $4, i
js L(li3)
L(le3): add %r10, (up)
adc %rax, %r9
adc %r14, %rdx
add %r9, 8(up)
adc $0, %rdx
mov %rdx, 16(up,nneg,8) C up[0]
mov %r15, q0
lea 8(up), up
dec n
jnz L(lo3)
C ==== Addition code ====
mov nneg, n
sar $2, n
lea 40(up,nneg,8), up
lea (up,nneg,8), vp
mov -24(up), %r8
mov -16(up), %r9
mov -8(up), %r10
add -24(vp), %r8
adc -16(vp), %r9
adc -8(vp), %r10
mov %r8, (rp)
mov %r9, 8(rp)
mov %r10, 16(rp)
lea 24(rp), rp
L(addx):inc n
jz L(ad3)
L(addy):mov (up), %r8
mov 8(up), %r9
inc n
jmp L(mid)
C ALIGN(16)
L(al3): adc (vp), %r8
adc 8(vp), %r9
adc 16(vp), %r10
adc 24(vp), %r11
mov %r8, (rp)
lea 32(up), up
mov %r9, 8(rp)
mov %r10, 16(rp)
inc n
mov %r11, 24(rp)
lea 32(vp), vp
mov (up), %r8
mov 8(up), %r9
lea 32(rp), rp
L(mid): mov 16(up), %r10
mov 24(up), %r11
jnz L(al3)
L(ae3): adc (vp), %r8
adc 8(vp), %r9
adc 16(vp), %r10
adc 24(vp), %r11
mov %r8, (rp)
mov %r9, 8(rp)
mov %r10, 16(rp)
mov %r11, 24(rp)
L(ad3): mov R32(n), R32(%rax) C zero
adc R32(%rax), R32(%rax)
L(ret): pop %r15
pop %r14
pop %r13
pop %r12
pop %rbx
pop %rbp
FUNC_EXIT()
ret
EPILOGUE()