1dnl AMD64 mpn_copyi -- copy limb vector, incrementing. 2 3dnl Copyright 2003, 2005, 2007, 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 22C cycles/limb 23C AMD K8,K9 1 24C AMD K10 1 25C AMD bd1 1.36 26C AMD bobcat 1.71 27C Intel P4 2-3 28C Intel core2 1 29C Intel NHM 1 30C Intel SBR 1 31C Intel atom 2 32C VIA nano 2 33 34 35IFSTD(`define(`rp',`%rdi')') 36IFSTD(`define(`up',`%rsi')') 37IFSTD(`define(`n', `%rdx')') 38 39IFDOS(`define(`rp',`%rcx')') 40IFDOS(`define(`up',`%rdx')') 41IFDOS(`define(`n', `%r8')') 42 43ABI_SUPPORT(DOS64) 44ABI_SUPPORT(STD64) 45 46ASM_START() 47 TEXT 48 ALIGN(64) 49 .byte 0,0,0,0,0,0 50PROLOGUE(mpn_copyi) 51 lea -8(rp), rp 52 sub $4, n 53 jc L(end) 54 55L(top): mov (up), %rax 56 mov 8(up), %r9 57 lea 32(rp), rp 58 mov 16(up), %r10 59 mov 24(up), %r11 60 lea 32(up), up 61 mov %rax, -24(rp) 62 mov %r9, -16(rp) 63 sub $4, n 64 mov %r10, -8(rp) 65 mov %r11, (rp) 66 jnc L(top) 67 68L(end): shr R32(n) 69 jnc 1f 70 mov (up), %rax 71 mov %rax, 8(rp) 72 lea 8(rp), rp 73 lea 8(up), up 741: shr R32(n) 75 jnc 1f 76 mov (up), %rax 77 mov 8(up), %r9 78 mov %rax, 8(rp) 79 mov %r9, 16(rp) 801: ret 81EPILOGUE() 82