1dnl AMD64 mpn_com. 2 3dnl Copyright 2004, 2005, 2006, 2011, 2012 Free Software Foundation, Inc. 4 5dnl This file is part of the GNU MP Library. 6 7dnl The GNU MP Library is free software; you can redistribute it and/or modify 8dnl it under the terms of the GNU Lesser General Public License as published 9dnl by the Free Software Foundation; either version 3 of the License, or (at 10dnl your option) any later version. 11 12dnl The GNU MP Library is distributed in the hope that it will be useful, but 13dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 14dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public 15dnl License for more details. 16 17dnl You should have received a copy of the GNU Lesser General Public License 18dnl along with the GNU MP Library. If not, see http://www.gnu.org/licenses/. 19 20include(`../config.m4') 21 22 23C cycles/limb 24C AMD K8,K9 1.25 25C AMD K10 1.25 26C Intel P4 2.78 27C Intel core2 1.1 28C Intel corei 1.5 29C Intel atom ? 30C VIA nano 2 31 32C INPUT PARAMETERS 33define(`rp',`%rdi') 34define(`up',`%rsi') 35define(`n',`%rdx') 36 37ABI_SUPPORT(DOS64) 38ABI_SUPPORT(STD64) 39 40ASM_START() 41 TEXT 42 ALIGN(32) 43PROLOGUE(mpn_com) 44 FUNC_ENTRY(3) 45 movq (up), %r8 46 movl R32(%rdx), R32(%rax) 47 leaq (up,n,8), up 48 leaq (rp,n,8), rp 49 negq n 50 andl $3, R32(%rax) 51 je L(b00) 52 cmpl $2, R32(%rax) 53 jc L(b01) 54 je L(b10) 55 56L(b11): notq %r8 57 movq %r8, (rp,n,8) 58 decq n 59 jmp L(e11) 60L(b10): addq $-2, n 61 jmp L(e10) 62 .byte 0x90,0x90,0x90,0x90,0x90,0x90 63L(b01): notq %r8 64 movq %r8, (rp,n,8) 65 incq n 66 jz L(ret) 67 68L(oop): movq (up,n,8), %r8 69L(b00): movq 8(up,n,8), %r9 70 notq %r8 71 notq %r9 72 movq %r8, (rp,n,8) 73 movq %r9, 8(rp,n,8) 74L(e11): movq 16(up,n,8), %r8 75L(e10): movq 24(up,n,8), %r9 76 notq %r8 77 notq %r9 78 movq %r8, 16(rp,n,8) 79 movq %r9, 24(rp,n,8) 80 addq $4, n 81 jnc L(oop) 82L(ret): FUNC_EXIT() 83 ret 84EPILOGUE() 85