| Austin Schuh | dace2a6 | 2020-08-18 10:56:48 -0700 | [diff] [blame^] | 1 | dnl x86 mpn_copyi -- copy limb vector, incrementing. |
| 2 | |
| 3 | dnl Copyright 1999-2002 Free Software Foundation, Inc. |
| 4 | |
| 5 | dnl This file is part of the GNU MP Library. |
| 6 | dnl |
| 7 | dnl The GNU MP Library is free software; you can redistribute it and/or modify |
| 8 | dnl it under the terms of either: |
| 9 | dnl |
| 10 | dnl * the GNU Lesser General Public License as published by the Free |
| 11 | dnl Software Foundation; either version 3 of the License, or (at your |
| 12 | dnl option) any later version. |
| 13 | dnl |
| 14 | dnl or |
| 15 | dnl |
| 16 | dnl * the GNU General Public License as published by the Free Software |
| 17 | dnl Foundation; either version 2 of the License, or (at your option) any |
| 18 | dnl later version. |
| 19 | dnl |
| 20 | dnl or both in parallel, as here. |
| 21 | dnl |
| 22 | dnl The GNU MP Library is distributed in the hope that it will be useful, but |
| 23 | dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY |
| 24 | dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 25 | dnl for more details. |
| 26 | dnl |
| 27 | dnl You should have received copies of the GNU General Public License and the |
| 28 | dnl GNU Lesser General Public License along with the GNU MP Library. If not, |
| 29 | dnl see https://www.gnu.org/licenses/. |
| 30 | |
| 31 | include(`../config.m4') |
| 32 | |
| 33 | |
| 34 | C cycles/limb startup (approx) |
| 35 | C P5 1.0 35 |
| 36 | C P6 0.75 45 |
| 37 | C K6 1.0 30 |
| 38 | C K7 1.3 65 |
| 39 | C P4 1.0 120 |
| 40 | C |
| 41 | C (Startup time includes some function call overheads.) |
| 42 | |
| 43 | |
| 44 | C void mpn_copyi (mp_ptr dst, mp_srcptr src, mp_size_t size); |
| 45 | C |
| 46 | C Copy src,size to dst,size, working from low to high addresses. |
| 47 | C |
| 48 | C The code here is very generic and can be expected to be reasonable on all |
| 49 | C the x86 family. |
| 50 | C |
| 51 | C P6 - An MMX based copy was tried, but was found to be slower than a rep |
| 52 | C movs in all cases. The fastest MMX found was 0.8 cycles/limb (when |
| 53 | C fully aligned). A rep movs seems to have a startup time of about 15 |
| 54 | C cycles, but doing something special for small sizes could lead to a |
| 55 | C branch misprediction that would destroy any saving. For now a plain |
| 56 | C rep movs seems ok. |
| 57 | C |
| 58 | C K62 - We used to have a big chunk of code doing an MMX copy at 0.56 c/l if |
| 59 | C aligned or a 1.0 rep movs if not. But that seemed excessive since |
| 60 | C it only got an advantage half the time, and even then only showed it |
| 61 | C above 50 limbs or so. |
| 62 | |
| 63 | defframe(PARAM_SIZE,12) |
| 64 | defframe(PARAM_SRC, 8) |
| 65 | defframe(PARAM_DST, 4) |
| 66 | deflit(`FRAME',0) |
| 67 | |
| 68 | TEXT |
| 69 | ALIGN(32) |
| 70 | |
| 71 | C eax saved esi |
| 72 | C ebx |
| 73 | C ecx counter |
| 74 | C edx saved edi |
| 75 | C esi src |
| 76 | C edi dst |
| 77 | C ebp |
| 78 | |
| 79 | PROLOGUE(mpn_copyi) |
| 80 | |
| 81 | movl PARAM_SIZE, %ecx |
| 82 | movl %esi, %eax |
| 83 | |
| 84 | movl PARAM_SRC, %esi |
| 85 | movl %edi, %edx |
| 86 | |
| 87 | movl PARAM_DST, %edi |
| 88 | |
| 89 | cld C better safe than sorry, see mpn/x86/README |
| 90 | |
| 91 | rep |
| 92 | movsl |
| 93 | |
| 94 | movl %eax, %esi |
| 95 | movl %edx, %edi |
| 96 | |
| 97 | ret |
| 98 | |
| 99 | EPILOGUE() |