sqr_diag_addlsh1.asm revision 1.1
1 dnl ARM64 mpn_sqr_diag_addlsh1. 2 3 dnl Contributed to the GNU project by Torbjrn Granlund. 4 5 dnl Copyright 2016, 2017 Free Software Foundation, Inc. 6 7 dnl This file is part of the GNU MP Library. 8 dnl 9 dnl The GNU MP Library is free software; you can redistribute it and/or modify 10 dnl it under the terms of either: 11 dnl 12 dnl * the GNU Lesser General Public License as published by the Free 13 dnl Software Foundation; either version 3 of the License, or (at your 14 dnl option) any later version. 15 dnl 16 dnl or 17 dnl 18 dnl * the GNU General Public License as published by the Free Software 19 dnl Foundation; either version 2 of the License, or (at your option) any 20 dnl later version. 21 dnl 22 dnl or both in parallel, as here. 23 dnl 24 dnl The GNU MP Library is distributed in the hope that it will be useful, but 25 dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 26 dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 27 dnl for more details. 28 dnl 29 dnl You should have received copies of the GNU General Public License and the 30 dnl GNU Lesser General Public License along with the GNU MP Library. If not, 31 dnl see https://www.gnu.org/licenses/. 32 33 include(`../config.m4') 34 35 C cycles/limb 36 C Cortex-A53 5.65 37 C Cortex-A57 3.5 38 C X-Gene 3.38 39 40 changecom(blah) 41 42 define(`rp', `x0') 43 define(`tp', `x1') 44 define(`up', `x2') 45 define(`n', `x3') 46 47 ASM_START() 48 PROLOGUE(mpn_sqr_diag_addlsh1) 49 ldr x15, [up],#8 50 lsr x18, n, #1 51 tbz n, #0, L(bx0) 52 53 L(bx1): adds x7, xzr, xzr 54 mul x12, x15, x15 55 ldr x16, [up],#8 56 ldp x4, x5, [tp],#16 57 umulh x11, x15, x15 58 b L(mid) 59 60 L(bx0): adds x5, xzr, xzr 61 mul x12, x15, x15 62 ldr x17, [up],#16 63 ldp x6, x7, [tp],#32 64 umulh x11, x15, x15 65 sub x18, x18, #1 66 cbz x18, L(end) 67 68 ALIGN(16) 69 L(top): extr x9, x6, x5, #63 70 mul x10, x17, x17 71 ldr x16, [up,#-8] 72 adcs x13, x9, x11 73 ldp x4, x5, [tp,#-16] 74 umulh x11, x17, x17 75 extr x8, x7, x6, #63 76 stp x12, x13, [rp],#16 77 adcs x12, x8, x10 78 L(mid): extr x9, x4, x7, #63 79 mul x10, x16, x16 80 ldr x17, [up],#16 81 adcs x13, x9, x11 82 ldp x6, x7, [tp],#32 83 umulh x11, x16, x16 84 extr x8, x5, x4, #63 85 stp x12, x13, [rp],#16 86 adcs x12, x8, x10 87 sub x18, x18, #1 88 cbnz x18, L(top) 89 90 L(end): extr x9, x6, x5, #63 91 mul x10, x17, x17 92 adcs x13, x9, x11 93 umulh x11, x17, x17 94 extr x8, x7, x6, #63 95 stp x12, x13, [rp] 96 adcs x12, x8, x10 97 extr x9, xzr, x7, #63 98 adcs x13, x9, x11 99 stp x12, x13, [rp,#16] 100 101 ret 102 EPILOGUE() 103
Indexes created Wed Aug 26 00:26:28 UTC 2026