| 1 |
dnl x86-64 mpn_rsh1add_n/mpn_rsh1sub_n. |
| 2 |
|
| 3 |
dnl Contributed to the GNU project by Torbjorn Granlund. |
| 4 |
|
| 5 |
dnl Copyright 2011, 2012 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 TODO |
| 36 |
C * Schedule loop less. It is now almost surely overscheduled, resulting in |
| 37 |
C large feed-in and wind-down code. |
| 38 |
|
| 39 |
C cycles/limb |
| 40 |
C AMD K8,K9 ? |
| 41 |
C AMD K10 ? |
| 42 |
C Intel P4 ? |
| 43 |
C Intel core2 ? |
| 44 |
C Intel NMH ? |
| 45 |
C Intel SBR ? |
| 46 |
C Intel atom 5.25 |
| 47 |
C VIA nano ? |
| 48 |
|
| 49 |
C INPUT PARAMETERS |
| 50 |
define(`rp',`%rdi') |
| 51 |
define(`up',`%rsi') |
| 52 |
define(`vp',`%rdx') |
| 53 |
define(`n',`%rcx') |
| 54 |
|
| 55 |
ifdef(`OPERATION_rsh1add_n', ` |
| 56 |
define(ADDSUB, add) |
| 57 |
define(ADCSBB, adc) |
| 58 |
define(func_n, mpn_rsh1add_n) |
| 59 |
define(func_nc, mpn_rsh1add_nc)') |
| 60 |
ifdef(`OPERATION_rsh1sub_n', ` |
| 61 |
define(ADDSUB, sub) |
| 62 |
define(ADCSBB, sbb) |
| 63 |
define(func_n, mpn_rsh1sub_n) |
| 64 |
define(func_nc, mpn_rsh1sub_nc)') |
| 65 |
|
| 66 |
ABI_SUPPORT(DOS64) |
| 67 |
ABI_SUPPORT(STD64) |
| 68 |
|
| 69 |
MULFUNC_PROLOGUE(mpn_rsh1add_n mpn_rsh1sub_n) |
| 70 |
|
| 71 |
ASM_START() |
| 72 |
TEXT |
| 73 |
ALIGN(16) |
| 74 |
PROLOGUE(func_n) |
| 75 |
FUNC_ENTRY(4) |
| 76 |
push %rbx |
| 77 |
push %rbp |
| 78 |
push %r12 |
| 79 |
push %r13 |
| 80 |
push %r14 |
| 81 |
push %r15 |
| 82 |
|
| 83 |
mov (up), %r15 |
| 84 |
ADDSUB (vp), %r15 |
| 85 |
sbb R32(%rbx), R32(%rbx) |
| 86 |
xor R32(%rax), R32(%rax) |
| 87 |
shr %r15 |
| 88 |
adc R32(%rax), R32(%rax) C return value |
| 89 |
|
| 90 |
mov R32(n), R32(%rbp) |
| 91 |
and $3, R32(%rbp) |
| 92 |
jz L(b0) |
| 93 |
cmp $2, R32(%rbp) |
| 94 |
jae L(b23) |
| 95 |
|
| 96 |
L(b1): dec n |
| 97 |
jnz L(gt1) |
| 98 |
shl $63, %rbx |
| 99 |
add %rbx, %r15 |
| 100 |
mov %r15, (rp) |
| 101 |
jmp L(cj1) |
| 102 |
L(gt1): lea 24(up), up |
| 103 |
lea 24(vp), vp |
| 104 |
mov -16(up), %r9 |
| 105 |
add R32(%rbx), R32(%rbx) |
| 106 |
mov -8(up), %r10 |
| 107 |
lea 24(rp), rp |
| 108 |
mov (up), %r11 |
| 109 |
ADCSBB -16(vp), %r9 |
| 110 |
ADCSBB -8(vp), %r10 |
| 111 |
mov %r15, %r12 |
| 112 |
ADCSBB (vp), %r11 |
| 113 |
mov %r9, %r13 |
| 114 |
sbb R32(%rbx), R32(%rbx) |
| 115 |
mov %r11, %r15 |
| 116 |
mov %r10, %r14 |
| 117 |
shl $63, %r11 |
| 118 |
shl $63, %r10 |
| 119 |
shl $63, %r9 |
| 120 |
or %r9, %r12 |
| 121 |
shr %r13 |
| 122 |
mov 8(up), %r8 |
| 123 |
shr %r14 |
| 124 |
or %r10, %r13 |
| 125 |
shr %r15 |
| 126 |
or %r11, %r14 |
| 127 |
sub $4, n |
| 128 |
jz L(cj5) |
| 129 |
L(gt5): mov 16(up), %r9 |
| 130 |
add R32(%rbx), R32(%rbx) |
| 131 |
mov 24(up), %r10 |
| 132 |
ADCSBB 8(vp), %r8 |
| 133 |
mov %r15, %rbp |
| 134 |
mov 32(up), %r11 |
| 135 |
jmp L(lo1) |
| 136 |
|
| 137 |
L(b23): jnz L(b3) |
| 138 |
mov 8(up), %r8 |
| 139 |
sub $2, n |
| 140 |
jnz L(gt2) |
| 141 |
add R32(%rbx), R32(%rbx) |
| 142 |
ADCSBB 8(vp), %r8 |
| 143 |
mov %r8, %r12 |
| 144 |
jmp L(cj2) |
| 145 |
L(gt2): mov 16(up), %r9 |
| 146 |
add R32(%rbx), R32(%rbx) |
| 147 |
mov 24(up), %r10 |
| 148 |
ADCSBB 8(vp), %r8 |
| 149 |
mov %r15, %rbp |
| 150 |
mov 32(up), %r11 |
| 151 |
ADCSBB 16(vp), %r9 |
| 152 |
lea 32(up), up |
| 153 |
ADCSBB 24(vp), %r10 |
| 154 |
mov %r9, %r13 |
| 155 |
ADCSBB 32(vp), %r11 |
| 156 |
mov %r8, %r12 |
| 157 |
jmp L(lo2) |
| 158 |
|
| 159 |
L(b3): lea 40(up), up |
| 160 |
lea 8(vp), vp |
| 161 |
mov %r15, %r14 |
| 162 |
add R32(%rbx), R32(%rbx) |
| 163 |
mov -32(up), %r11 |
| 164 |
ADCSBB 0(vp), %r11 |
| 165 |
lea 8(rp), rp |
| 166 |
sbb R32(%rbx), R32(%rbx) |
| 167 |
mov %r11, %r15 |
| 168 |
shl $63, %r11 |
| 169 |
mov -24(up), %r8 |
| 170 |
shr %r15 |
| 171 |
or %r11, %r14 |
| 172 |
sub $3, n |
| 173 |
jnz L(gt3) |
| 174 |
add R32(%rbx), R32(%rbx) |
| 175 |
ADCSBB 8(vp), %r8 |
| 176 |
jmp L(cj3) |
| 177 |
L(gt3): mov -16(up), %r9 |
| 178 |
add R32(%rbx), R32(%rbx) |
| 179 |
mov -8(up), %r10 |
| 180 |
ADCSBB 8(vp), %r8 |
| 181 |
mov %r15, %rbp |
| 182 |
mov (up), %r11 |
| 183 |
ADCSBB 16(vp), %r9 |
| 184 |
ADCSBB 24(vp), %r10 |
| 185 |
mov %r8, %r12 |
| 186 |
jmp L(lo3) |
| 187 |
|
| 188 |
L(b0): lea 48(up), up |
| 189 |
lea 16(vp), vp |
| 190 |
add R32(%rbx), R32(%rbx) |
| 191 |
mov -40(up), %r10 |
| 192 |
lea 16(rp), rp |
| 193 |
mov -32(up), %r11 |
| 194 |
ADCSBB -8(vp), %r10 |
| 195 |
mov %r15, %r13 |
| 196 |
ADCSBB (vp), %r11 |
| 197 |
sbb R32(%rbx), R32(%rbx) |
| 198 |
mov %r11, %r15 |
| 199 |
mov %r10, %r14 |
| 200 |
shl $63, %r11 |
| 201 |
shl $63, %r10 |
| 202 |
mov -24(up), %r8 |
| 203 |
shr %r14 |
| 204 |
or %r10, %r13 |
| 205 |
shr %r15 |
| 206 |
or %r11, %r14 |
| 207 |
sub $4, n |
| 208 |
jnz L(gt4) |
| 209 |
add R32(%rbx), R32(%rbx) |
| 210 |
ADCSBB 8(vp), %r8 |
| 211 |
jmp L(cj4) |
| 212 |
L(gt4): mov -16(up), %r9 |
| 213 |
add R32(%rbx), R32(%rbx) |
| 214 |
mov -8(up), %r10 |
| 215 |
ADCSBB 8(vp), %r8 |
| 216 |
mov %r15, %rbp |
| 217 |
mov (up), %r11 |
| 218 |
ADCSBB 16(vp), %r9 |
| 219 |
jmp L(lo0) |
| 220 |
|
| 221 |
ALIGN(8) |
| 222 |
L(top): mov 16(up), %r9 |
| 223 |
shr %r14 |
| 224 |
or %r10, %r13 |
| 225 |
shr %r15 |
| 226 |
or %r11, %r14 |
| 227 |
add R32(%rbx), R32(%rbx) |
| 228 |
mov 24(up), %r10 |
| 229 |
mov %rbp, (rp) |
| 230 |
ADCSBB 8(vp), %r8 |
| 231 |
mov %r15, %rbp |
| 232 |
lea 32(rp), rp |
| 233 |
mov 32(up), %r11 |
| 234 |
L(lo1): ADCSBB 16(vp), %r9 |
| 235 |
lea 32(up), up |
| 236 |
mov %r12, -24(rp) |
| 237 |
L(lo0): ADCSBB 24(vp), %r10 |
| 238 |
mov %r8, %r12 |
| 239 |
mov %r13, -16(rp) |
| 240 |
L(lo3): ADCSBB 32(vp), %r11 |
| 241 |
mov %r9, %r13 |
| 242 |
mov %r14, -8(rp) |
| 243 |
L(lo2): sbb R32(%rbx), R32(%rbx) |
| 244 |
shl $63, %r8 |
| 245 |
mov %r11, %r15 |
| 246 |
shr %r12 |
| 247 |
mov %r10, %r14 |
| 248 |
shl $63, %r9 |
| 249 |
lea 32(vp), vp |
| 250 |
shl $63, %r10 |
| 251 |
or %r8, %rbp |
| 252 |
shl $63, %r11 |
| 253 |
or %r9, %r12 |
| 254 |
shr %r13 |
| 255 |
mov 8(up), %r8 |
| 256 |
sub $4, n |
| 257 |
jg L(top) |
| 258 |
|
| 259 |
L(end): shr %r14 |
| 260 |
or %r10, %r13 |
| 261 |
shr %r15 |
| 262 |
or %r11, %r14 |
| 263 |
mov %rbp, (rp) |
| 264 |
lea 32(rp), rp |
| 265 |
L(cj5): add R32(%rbx), R32(%rbx) |
| 266 |
ADCSBB 8(vp), %r8 |
| 267 |
mov %r12, -24(rp) |
| 268 |
L(cj4): mov %r13, -16(rp) |
| 269 |
L(cj3): mov %r8, %r12 |
| 270 |
mov %r14, -8(rp) |
| 271 |
L(cj2): sbb R32(%rbx), R32(%rbx) |
| 272 |
shl $63, %r8 |
| 273 |
shr %r12 |
| 274 |
or %r8, %r15 |
| 275 |
shl $63, %rbx |
| 276 |
add %rbx, %r12 |
| 277 |
mov %r15, (rp) |
| 278 |
mov %r12, 8(rp) |
| 279 |
L(cj1): pop %r15 |
| 280 |
pop %r14 |
| 281 |
pop %r13 |
| 282 |
pop %r12 |
| 283 |
pop %rbp |
| 284 |
pop %rbx |
| 285 |
FUNC_EXIT() |
| 286 |
ret |
| 287 |
EPILOGUE() |