| 1 |
dnl AMD64 mpn_addmul_2 optimised for Intel Haswell. |
| 2 |
|
| 3 |
dnl Contributed to the GNU project by Torbjörn Granlund. |
| 4 |
|
| 5 |
dnl Copyright 2013 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 AMD K8,K9 n/a |
| 37 |
C AMD K10 n/a |
| 38 |
C AMD bull n/a |
| 39 |
C AMD pile n/a |
| 40 |
C AMD steam ? |
| 41 |
C AMD bobcat n/a |
| 42 |
C AMD jaguar ? |
| 43 |
C Intel P4 n/a |
| 44 |
C Intel core n/a |
| 45 |
C Intel NHM n/a |
| 46 |
C Intel SBR n/a |
| 47 |
C Intel IBR n/a |
| 48 |
C Intel HWL 2.15 |
| 49 |
C Intel BWL ? |
| 50 |
C Intel atom n/a |
| 51 |
C VIA nano n/a |
| 52 |
|
| 53 |
C The loop of this code is the result of running a code generation and |
| 54 |
C optimisation tool suite written by David Harvey and Torbjörn Granlund. |
| 55 |
|
| 56 |
define(`rp', `%rdi') |
| 57 |
define(`up', `%rsi') |
| 58 |
define(`n_param',`%rdx') |
| 59 |
define(`vp', `%rcx') |
| 60 |
|
| 61 |
define(`v0', `%r8') |
| 62 |
define(`v1', `%r9') |
| 63 |
define(`w0', `%rbx') |
| 64 |
define(`w1', `%rcx') |
| 65 |
define(`w2', `%rbp') |
| 66 |
define(`w3', `%r10') |
| 67 |
define(`n', `%r11') |
| 68 |
define(`X0', `%r12') |
| 69 |
define(`X1', `%r13') |
| 70 |
|
| 71 |
ABI_SUPPORT(DOS64) |
| 72 |
ABI_SUPPORT(STD64) |
| 73 |
|
| 74 |
ASM_START() |
| 75 |
TEXT |
| 76 |
ALIGN(32) |
| 77 |
PROLOGUE(mpn_addmul_2) |
| 78 |
FUNC_ENTRY(4) |
| 79 |
push %rbx |
| 80 |
push %rbp |
| 81 |
push %r12 |
| 82 |
push %r13 |
| 83 |
|
| 84 |
mov (vp), v0 |
| 85 |
mov 8(vp), v1 |
| 86 |
|
| 87 |
mov n_param, n |
| 88 |
shr $2, n |
| 89 |
|
| 90 |
test $1, R8(n_param) |
| 91 |
jnz L(bx1) |
| 92 |
|
| 93 |
L(bx0): mov (rp), X0 |
| 94 |
mov 8(rp), X1 |
| 95 |
test $2, R8(n_param) |
| 96 |
jnz L(b10) |
| 97 |
|
| 98 |
L(b00): mov (up), %rdx |
| 99 |
lea 16(up), up |
| 100 |
mulx( v0, %rax, w1) |
| 101 |
add %rax, X0 |
| 102 |
mulx( v1, %rax, w2) |
| 103 |
adc $0, w1 |
| 104 |
mov X0, (rp) |
| 105 |
add %rax, X1 |
| 106 |
adc $0, w2 |
| 107 |
mov -8(up), %rdx |
| 108 |
lea 16(rp), rp |
| 109 |
jmp L(lo0) |
| 110 |
|
| 111 |
L(b10): mov (up), %rdx |
| 112 |
inc n |
| 113 |
mulx( v0, %rax, w1) |
| 114 |
add %rax, X0 |
| 115 |
adc $0, w1 |
| 116 |
mulx( v1, %rax, w2) |
| 117 |
mov X0, (rp) |
| 118 |
mov 16(rp), X0 |
| 119 |
add %rax, X1 |
| 120 |
adc $0, w2 |
| 121 |
xor w0, w0 |
| 122 |
jmp L(lo2) |
| 123 |
|
| 124 |
L(bx1): mov (rp), X1 |
| 125 |
mov 8(rp), X0 |
| 126 |
test $2, R8(n_param) |
| 127 |
jnz L(b11) |
| 128 |
|
| 129 |
L(b01): mov (up), %rdx |
| 130 |
mulx( v0, %rax, w3) |
| 131 |
add %rax, X1 |
| 132 |
adc $0, w3 |
| 133 |
mulx( v1, %rax, w0) |
| 134 |
add %rax, X0 |
| 135 |
adc $0, w0 |
| 136 |
mov 8(up), %rdx |
| 137 |
mov X1, (rp) |
| 138 |
mov 16(rp), X1 |
| 139 |
mulx( v0, %rax, w1) |
| 140 |
lea 24(rp), rp |
| 141 |
lea 24(up), up |
| 142 |
jmp L(lo1) |
| 143 |
|
| 144 |
L(b11): mov (up), %rdx |
| 145 |
inc n |
| 146 |
mulx( v0, %rax, w3) |
| 147 |
add %rax, X1 |
| 148 |
adc $0, w3 |
| 149 |
mulx( v1, %rax, w0) |
| 150 |
add %rax, X0 |
| 151 |
adc $0, w0 |
| 152 |
mov X1, (rp) |
| 153 |
mov 8(up), %rdx |
| 154 |
mulx( v0, %rax, w1) |
| 155 |
lea 8(rp), rp |
| 156 |
lea 8(up), up |
| 157 |
jmp L(lo3) |
| 158 |
|
| 159 |
ALIGN(16) |
| 160 |
L(top): mulx( v0, %rax, w3) |
| 161 |
add w0, X1 |
| 162 |
adc $0, w2 |
| 163 |
add %rax, X1 |
| 164 |
adc $0, w3 |
| 165 |
mulx( v1, %rax, w0) |
| 166 |
add %rax, X0 |
| 167 |
adc $0, w0 |
| 168 |
lea 32(rp), rp |
| 169 |
add w1, X1 |
| 170 |
mov -16(up), %rdx |
| 171 |
mov X1, -24(rp) |
| 172 |
adc $0, w3 |
| 173 |
add w2, X0 |
| 174 |
mov -8(rp), X1 |
| 175 |
mulx( v0, %rax, w1) |
| 176 |
adc $0, w0 |
| 177 |
L(lo1): add %rax, X0 |
| 178 |
mulx( v1, %rax, w2) |
| 179 |
adc $0, w1 |
| 180 |
add w3, X0 |
| 181 |
mov X0, -16(rp) |
| 182 |
adc $0, w1 |
| 183 |
add %rax, X1 |
| 184 |
adc $0, w2 |
| 185 |
add w0, X1 |
| 186 |
mov -8(up), %rdx |
| 187 |
adc $0, w2 |
| 188 |
L(lo0): mulx( v0, %rax, w3) |
| 189 |
add %rax, X1 |
| 190 |
adc $0, w3 |
| 191 |
mov (rp), X0 |
| 192 |
mulx( v1, %rax, w0) |
| 193 |
add %rax, X0 |
| 194 |
adc $0, w0 |
| 195 |
add w1, X1 |
| 196 |
mov X1, -8(rp) |
| 197 |
adc $0, w3 |
| 198 |
mov (up), %rdx |
| 199 |
add w2, X0 |
| 200 |
mulx( v0, %rax, w1) |
| 201 |
adc $0, w0 |
| 202 |
L(lo3): add %rax, X0 |
| 203 |
adc $0, w1 |
| 204 |
mulx( v1, %rax, w2) |
| 205 |
add w3, X0 |
| 206 |
mov 8(rp), X1 |
| 207 |
mov X0, (rp) |
| 208 |
mov 16(rp), X0 |
| 209 |
adc $0, w1 |
| 210 |
add %rax, X1 |
| 211 |
adc $0, w2 |
| 212 |
L(lo2): mov 8(up), %rdx |
| 213 |
lea 32(up), up |
| 214 |
dec n |
| 215 |
jnz L(top) |
| 216 |
|
| 217 |
L(end): mulx( v0, %rax, w3) |
| 218 |
add w0, X1 |
| 219 |
adc $0, w2 |
| 220 |
add %rax, X1 |
| 221 |
adc $0, w3 |
| 222 |
mulx( v1, %rdx, %rax) |
| 223 |
add w1, X1 |
| 224 |
mov X1, 8(rp) |
| 225 |
adc $0, w3 |
| 226 |
add w2, %rdx |
| 227 |
adc $0, %rax |
| 228 |
add w3, %rdx |
| 229 |
mov %rdx, 16(rp) |
| 230 |
adc $0, %rax |
| 231 |
|
| 232 |
pop %r13 |
| 233 |
pop %r12 |
| 234 |
pop %rbp |
| 235 |
pop %rbx |
| 236 |
FUNC_EXIT() |
| 237 |
ret |
| 238 |
EPILOGUE() |