dnl Contributed to the GNU project by Torbjörn Granlund.
dnl Copyright 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 n/a
C AMD K10 n/a
C AMD bull n/a
C AMD pile n/a
C AMD steam ?
C AMD bobcat n/a
C AMD jaguar ?
C Intel P4 n/a
C Intel core n/a
C Intel NHM n/a
C Intel SBR n/a
C Intel IBR n/a
C Intel HWL 2.32
C Intel BWL ?
C Intel atom n/a
C VIA nano n/a
C The inner loops of thiscode 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.
C * Consider inlining mpn_add_n. Tests indicate that this saves just 1-2
C cycles, though.
define(`rp', `%rdi') C rcx
define(`up', `%rsi') C rdx
define(`mp_param', `%rdx') C r8
define(`n', `%rcx') C r9
define(`u0inv_param', `%r8') C stack
define(`i', `%r14')
define(`j', `%r15')
define(`mp', `%rdi')
define(`u0inv', `(%rsp)') C stack
ABI_SUPPORT(DOS64) C FIXME: needs verification
ABI_SUPPORT(STD64)
ASM_START()
TEXT ALIGN(16)
PROLOGUE(mpn_redc_1)
FUNC_ENTRY(4)
IFDOS(` mov56(%rsp), %r8 ') push %rbx push %rbp push %r12 push %r13 push %r14 push %r15 push rp mov mp_param, mp C note that rp and mp shares register mov (up), %rdx
neg n push %r8 C put u0inv on stack
imul u0inv_param, %rdx C first iteration q0 mov n, j C outer loop induction var
test $1, R8(n) jnz L(bx1)
L(bx0): test $2, R8(n) jz L(o0b)
cmp $-2, R32(n) jnz L(o2)
C Special codefor n = 2 since general code cannot handle it mov8(%rsp), %rbx C rp lea16(%rsp), %rsp C deallocate two slots
mulx( (mp), %r9, %r12)
mulx( 8,(mp), %r11, %r10) add %r12, %r11 adc $0, %r10 add (up), %r9 C = 0 adc8(up), %r11 C r11 = up[1] adc $0, %r10 C -> up[0] mov %r11, %rdx
imul u0inv_param, %rdx
mulx( (mp), %r13, %r12)
mulx( 8,(mp), %r14, %r15) xor R32(%rax), R32(%rax) add %r12, %r14 adc $0, %r15 add %r11, %r13 C = 0 adc16(up), %r14 C rp[2] adc $0, %r15 C -> up[1] add %r14, %r10 adc24(up), %r15 mov %r10, (%rbx) mov %r15, 8(%rbx)
setc R8(%rax) jmp L(ret)
L(o2): lea2(n), i C inner loop induction var
mulx( (mp), %r9, %r8)
mulx( 8,(mp), %r11, %r10)
sar $2, i add %r8, %r11 jmp L(lo2)
L(ed2): mov56(up,n,8), %rdx C next iteration up[0] lea16(mp,n,8), mp C mp = (last starting mp) adc %rax, %r9 adc %r8, %r11 mov32(up), %r8 adc $0, %r10
imul u0inv, %rdx C next iteration q0 mov40(up), %rax add %r9, %r8 mov %r8, 32(up) adc %r11, %rax mov %rax, 40(up) lea56(up,n,8), up C up = (last starting up) + 1 adc $0, %r10 mov %r10, -8(up) inc j jnz L(o2)
jmp L(cj)
L(bx1): test $2, R8(n) jz L(o3a)
L(o1a): cmp $-1, R32(n) jnz L(o1b)
C Special codefor n = 1 since general code cannot handle it mov8(%rsp), %rbx C rp lea16(%rsp), %rsp C deallocate two slots
mulx( (mp), %r11, %r10) add (up), %r11 adc8(up), %r10 mov %r10, (%rbx) mov $0, R32(%rax)
setc R8(%rax) jmp L(ret)
L(o1b): lea24(mp), mp
L(o1): lea1(n), i C inner loop induction var
mulx( -24,(mp), %r11, %r10)
mulx( -16,(mp), %r13, %r12)
mulx( -8,(mp), %rbx, %rax)
sar $2, i add %r10, %r13 adc %r12, %rbx adc $0, %rax mov (up), %r10 mov8(up), %r12 mov16(up), %rbp add %r11, %r10 jmp L(lo1)
L(ed1): mov48(up,n,8), %rdx C next iteration up[0] lea40(mp,n,8), mp C mp = (last starting mp) adc %rax, %r9 adc %r8, %r11 mov24(up), %r8 adc $0, %r10
imul u0inv, %rdx C next iteration q0 mov32(up), %rax add %r9, %r8 mov %r8, 24(up) adc %r11, %rax mov %rax, 32(up) lea48(up,n,8), up C up = (last starting up) + 1 adc $0, %r10 mov %r10, -8(up) inc j jnz L(o1)
jmp L(cj)
L(o3a): cmp $-3, R32(n) jnz L(o3b)
C Special codefor n = 3 since general code cannot handle it
L(n3): mulx( (mp), %rbx, %rax)
mulx( 8,(mp), %r9, %r14) add (up), %rbx
mulx( 16,(mp), %r11, %r10) adc %rax, %r9 C W 1 adc %r14, %r11 C W 2 mov8(up), %r14 mov u0inv_param, %rdx adc $0, %r10 C W 3 mov16(up), %rax add %r9, %r14 C W 1 mov %r14, 8(up)
mulx( %r14, %rdx, %r13) C next iteration q0 adc %r11, %rax C W 2 mov %rax, 16(up) adc $0, %r10 C W 3 mov %r10, (up) lea8(up), up C up = (last starting up) + 1 inc j jnz L(n3)
jmp L(cj)
L(o3b): lea8(mp), mp
L(o3): lea4(n), i C inner loop induction var
mulx( -8,(mp), %rbx, %rax)
mulx( (mp), %r9, %r8) mov (up), %rbp
mulx( 8,(mp), %r11, %r10)
sar $2, i add %rbx, %rbp nop adc %rax, %r9 jmp L(lo3)
L(ed0): mov40(up,n,8), %rdx C next iteration up[0] lea32(mp,n,8), mp C mp = (last starting mp) adc %rax, %r9 adc %r8, %r11 mov16(up), %r8 adc $0, %r10
imul u0inv, %rdx C next iteration q0 mov24(up), %rax add %r9, %r8 mov %r8, 16(up) adc %r11, %rax mov %rax, 24(up) lea40(up,n,8), up C up = (last starting up) + 1 adc $0, %r10 mov %r10, -8(up) inc j jnz L(o0)
L(cj):
IFSTD(` mov8(%rsp), %rdi C param 1: rp lea16-8(%rsp), %rsp C deallocate 2, add back for alignment lea (up,n,8), %rdx C param 3: up - n neg R32(n) ') C param 4: n
IFDOS(` mov up, %rdx C param 2: up lea (up,n,8), %r8 C param 3: up - n neg R32(n) mov n, %r9 C param 4: n mov8(%rsp), %rcx C param 1: rp lea16-32-8(%rsp), %rsp') C deallocate 2, allocate shadow, align
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.