| 1 |
dnl ARM mpn_addmul_1 optimised for A15. |
| 2 |
|
| 3 |
dnl Copyright 2012, 2013 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 |
C cycles/limb best |
| 34 |
C StrongARM: - |
| 35 |
C XScale ? |
| 36 |
C Cortex-A7 ? |
| 37 |
C Cortex-A8 ? |
| 38 |
C Cortex-A9 6 3.25 |
| 39 |
C Cortex-A15 2 this |
| 40 |
|
| 41 |
C This code uses umlal for adding in the rp[] data, keeping the recurrency path |
| 42 |
C separate from any multiply instructions. It performs well on A15, at umlal's |
| 43 |
C bandwidth. |
| 44 |
C |
| 45 |
C An A9 variant should perhaps stick to 3-way unrolling, and use ldm and stm |
| 46 |
C for all loads and stores. Alternatively, it could do 2-way or 4-way, but |
| 47 |
C then alignment aware code will be necessary (adding O(1) bookkeeping |
| 48 |
C overhead). |
| 49 |
C |
| 50 |
C We don't use r12 due to ldrd and strd limitations. |
| 51 |
|
| 52 |
C Architecture requirements: |
| 53 |
C v5 - |
| 54 |
C v5t - |
| 55 |
C v5te ldrd strd |
| 56 |
C v6 - |
| 57 |
C v6t2 - |
| 58 |
C v7a - |
| 59 |
|
| 60 |
define(`rp', `r0') |
| 61 |
define(`up', `r1') |
| 62 |
define(`n', `r2') |
| 63 |
define(`v0', `r3') |
| 64 |
|
| 65 |
define(`w0', `r10') define(`w1', `r11') |
| 66 |
define(`u0', `r8') define(`u1', `r9') |
| 67 |
|
| 68 |
ASM_START() |
| 69 |
PROLOGUE(mpn_addmul_1) |
| 70 |
push { r4-r11 } |
| 71 |
|
| 72 |
ands r6, n, #3 |
| 73 |
sub n, n, #3 |
| 74 |
beq L(b00) |
| 75 |
cmp r6, #2 |
| 76 |
bcc L(b01) |
| 77 |
beq L(b10) |
| 78 |
|
| 79 |
L(b11): mov r6, #0 |
| 80 |
cmn r13, #0 C carry clear |
| 81 |
ldr u1, [up], #-4 |
| 82 |
ldr w1, [rp], #-4 |
| 83 |
mov r7, #0 |
| 84 |
b L(mid) |
| 85 |
|
| 86 |
L(b00): ldrd u0, u1, [up] |
| 87 |
ldrd w0, w1, [rp] |
| 88 |
mov r6, #0 |
| 89 |
umlal w0, r6, u0, v0 |
| 90 |
cmn r13, #0 C carry clear |
| 91 |
mov r7, #0 |
| 92 |
str w0, [rp] |
| 93 |
b L(mid) |
| 94 |
|
| 95 |
L(b10): ldrd u0, u1, [up], #8 |
| 96 |
ldrd w0, w1, [rp] |
| 97 |
mov r4, #0 |
| 98 |
umlal w0, r4, u0, v0 |
| 99 |
cmn r13, #0 C carry clear |
| 100 |
mov r5, #0 |
| 101 |
str w0, [rp], #8 |
| 102 |
umlal w1, r5, u1, v0 |
| 103 |
tst n, n |
| 104 |
bmi L(end) |
| 105 |
b L(top) |
| 106 |
|
| 107 |
L(b01): mov r4, #0 |
| 108 |
ldr u1, [up], #4 |
| 109 |
ldr w1, [rp], #4 |
| 110 |
mov r5, #0 |
| 111 |
umlal w1, r5, u1, v0 |
| 112 |
tst n, n |
| 113 |
bmi L(end) |
| 114 |
|
| 115 |
ALIGN(16) |
| 116 |
L(top): ldrd u0, u1, [up, #0] |
| 117 |
adcs r4, r4, w1 |
| 118 |
ldrd w0, w1, [rp, #0] |
| 119 |
mov r6, #0 |
| 120 |
umlal w0, r6, u0, v0 C 1 2 |
| 121 |
adcs r5, r5, w0 |
| 122 |
mov r7, #0 |
| 123 |
strd r4, r5, [rp, #-4] |
| 124 |
L(mid): umlal w1, r7, u1, v0 C 2 3 |
| 125 |
ldrd u0, u1, [up, #8] |
| 126 |
adcs r6, r6, w1 |
| 127 |
ldrd w0, w1, [rp, #8] |
| 128 |
mov r4, #0 |
| 129 |
umlal w0, r4, u0, v0 C 3 4 |
| 130 |
adcs r7, r7, w0 |
| 131 |
mov r5, #0 |
| 132 |
strd r6, r7, [rp, #4] |
| 133 |
umlal w1, r5, u1, v0 C 0 1 |
| 134 |
sub n, n, #4 |
| 135 |
add up, up, #16 |
| 136 |
add rp, rp, #16 |
| 137 |
tst n, n |
| 138 |
bpl L(top) |
| 139 |
|
| 140 |
L(end): adcs r4, r4, w1 |
| 141 |
str r4, [rp, #-4] |
| 142 |
adc r0, r5, #0 |
| 143 |
pop { r4-r11 } |
| 144 |
bx r14 |
| 145 |
EPILOGUE() |