| 1 |
dnl Alpha ev6 nails mpn_mul_1. |
| 2 |
|
| 3 |
dnl Copyright 2002, 2005, 2006 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 |
| 34 |
C EV4: 42 |
| 35 |
C EV5: 18 |
| 36 |
C EV6: 3.25 |
| 37 |
|
| 38 |
C TODO |
| 39 |
C * Reroll loop for 3.0 c/l with current 4-way unrolling. |
| 40 |
C * The loop is overscheduled wrt loads and wrt multiplies, in particular |
| 41 |
C umulh. |
| 42 |
C * Use FP loop count and multiple exit points, that would simplify feed-in lp0 |
| 43 |
C and would work since the loop structure is really regular. |
| 44 |
|
| 45 |
C INPUT PARAMETERS |
| 46 |
define(`rp',`r16') |
| 47 |
define(`up',`r17') |
| 48 |
define(`n', `r18') |
| 49 |
define(`vl0',`r19') |
| 50 |
|
| 51 |
define(`numb_mask',`r6') |
| 52 |
|
| 53 |
define(`m0a',`r0') |
| 54 |
define(`m0b',`r1') |
| 55 |
define(`m1a',`r2') |
| 56 |
define(`m1b',`r3') |
| 57 |
define(`m2a',`r20') |
| 58 |
define(`m2b',`r21') |
| 59 |
define(`m3a',`r22') |
| 60 |
define(`m3b',`r23') |
| 61 |
|
| 62 |
define(`acc0',`r25') |
| 63 |
define(`acc1',`r27') |
| 64 |
|
| 65 |
define(`ul0',`r4') |
| 66 |
define(`ul1',`r5') |
| 67 |
define(`ul2',`r4') |
| 68 |
define(`ul3',`r5') |
| 69 |
|
| 70 |
define(`rl0',`r24') |
| 71 |
define(`rl1',`r24') |
| 72 |
define(`rl2',`r24') |
| 73 |
define(`rl3',`r24') |
| 74 |
|
| 75 |
define(`t0',`r7') |
| 76 |
define(`t1',`r8') |
| 77 |
|
| 78 |
define(`NAIL_BITS',`GMP_NAIL_BITS') |
| 79 |
define(`NUMB_BITS',`GMP_NUMB_BITS') |
| 80 |
|
| 81 |
dnl This declaration is munged by configure |
| 82 |
NAILS_SUPPORT(1-63) |
| 83 |
|
| 84 |
ASM_START() |
| 85 |
PROLOGUE(mpn_mul_1) |
| 86 |
sll vl0, NAIL_BITS, vl0 |
| 87 |
lda numb_mask, -1(r31) |
| 88 |
srl numb_mask, NAIL_BITS, numb_mask |
| 89 |
|
| 90 |
and n, 3, r25 |
| 91 |
cmpeq r25, 1, r21 |
| 92 |
bne r21, L(1m4) |
| 93 |
cmpeq r25, 2, r21 |
| 94 |
bne r21, L(2m4) |
| 95 |
beq r25, L(0m4) |
| 96 |
|
| 97 |
L(3m4): ldq ul3, 0(up) |
| 98 |
lda n, -4(n) |
| 99 |
ldq ul0, 8(up) |
| 100 |
mulq vl0, ul3, m3a |
| 101 |
umulh vl0, ul3, m3b |
| 102 |
ldq ul1, 16(up) |
| 103 |
lda up, 24(up) |
| 104 |
lda rp, -8(rp) |
| 105 |
mulq vl0, ul0, m0a |
| 106 |
umulh vl0, ul0, m0b |
| 107 |
bge n, L(ge3) |
| 108 |
|
| 109 |
mulq vl0, ul1, m1a |
| 110 |
umulh vl0, ul1, m1b |
| 111 |
srl m3a,NAIL_BITS, t0 |
| 112 |
addq t0, r31, acc1 |
| 113 |
srl m0a,NAIL_BITS, t0 |
| 114 |
addq t0, m3b, acc0 |
| 115 |
srl acc1,NUMB_BITS, t1 |
| 116 |
br r31, L(ta3) |
| 117 |
|
| 118 |
L(ge3): ldq ul2, 0(up) |
| 119 |
mulq vl0, ul1, m1a |
| 120 |
umulh vl0, ul1, m1b |
| 121 |
srl m3a,NAIL_BITS, t0 |
| 122 |
ldq ul3, 8(up) |
| 123 |
lda n, -4(n) |
| 124 |
mulq vl0, ul2, m2a |
| 125 |
addq t0, r31, acc1 |
| 126 |
umulh vl0, ul2, m2b |
| 127 |
srl m0a,NAIL_BITS, t0 |
| 128 |
ldq ul0, 16(up) |
| 129 |
mulq vl0, ul3, m3a |
| 130 |
addq t0, m3b, acc0 |
| 131 |
srl acc1,NUMB_BITS, t1 |
| 132 |
br r31, L(el3) |
| 133 |
|
| 134 |
L(0m4): lda n, -8(n) |
| 135 |
ldq ul2, 0(up) |
| 136 |
ldq ul3, 8(up) |
| 137 |
mulq vl0, ul2, m2a |
| 138 |
umulh vl0, ul2, m2b |
| 139 |
ldq ul0, 16(up) |
| 140 |
mulq vl0, ul3, m3a |
| 141 |
umulh vl0, ul3, m3b |
| 142 |
ldq ul1, 24(up) |
| 143 |
lda up, 32(up) |
| 144 |
mulq vl0, ul0, m0a |
| 145 |
umulh vl0, ul0, m0b |
| 146 |
bge n, L(ge4) |
| 147 |
|
| 148 |
srl m2a,NAIL_BITS, t0 |
| 149 |
mulq vl0, ul1, m1a |
| 150 |
addq t0, r31, acc0 |
| 151 |
umulh vl0, ul1, m1b |
| 152 |
srl m3a,NAIL_BITS, t0 |
| 153 |
addq t0, m2b, acc1 |
| 154 |
srl acc0,NUMB_BITS, t1 |
| 155 |
br r31, L(ta4) |
| 156 |
|
| 157 |
L(ge4): srl m2a,NAIL_BITS, t0 |
| 158 |
ldq ul2, 0(up) |
| 159 |
mulq vl0, ul1, m1a |
| 160 |
addq t0, r31, acc0 |
| 161 |
umulh vl0, ul1, m1b |
| 162 |
srl m3a,NAIL_BITS, t0 |
| 163 |
ldq ul3, 8(up) |
| 164 |
lda n, -4(n) |
| 165 |
mulq vl0, ul2, m2a |
| 166 |
addq t0, m2b, acc1 |
| 167 |
srl acc0,NUMB_BITS, t1 |
| 168 |
br r31, L(el0) |
| 169 |
|
| 170 |
L(2m4): lda n, -4(n) |
| 171 |
ldq ul0, 0(up) |
| 172 |
ldq ul1, 8(up) |
| 173 |
lda up, 16(up) |
| 174 |
lda rp, -16(rp) |
| 175 |
mulq vl0, ul0, m0a |
| 176 |
umulh vl0, ul0, m0b |
| 177 |
bge n, L(ge2) |
| 178 |
|
| 179 |
mulq vl0, ul1, m1a |
| 180 |
umulh vl0, ul1, m1b |
| 181 |
srl m0a,NAIL_BITS, t0 |
| 182 |
addq t0, r31, acc0 |
| 183 |
srl m1a,NAIL_BITS, t0 |
| 184 |
addq t0, m0b, acc1 |
| 185 |
srl acc0,NUMB_BITS, t1 |
| 186 |
br r31, L(ta2) |
| 187 |
|
| 188 |
L(ge2): ldq ul2, 0(up) |
| 189 |
mulq vl0, ul1, m1a |
| 190 |
umulh vl0, ul1, m1b |
| 191 |
ldq ul3, 8(up) |
| 192 |
lda n, -4(n) |
| 193 |
mulq vl0, ul2, m2a |
| 194 |
umulh vl0, ul2, m2b |
| 195 |
srl m0a,NAIL_BITS, t0 |
| 196 |
ldq ul0, 16(up) |
| 197 |
mulq vl0, ul3, m3a |
| 198 |
addq t0, r31, acc0 |
| 199 |
umulh vl0, ul3, m3b |
| 200 |
srl m1a,NAIL_BITS, t0 |
| 201 |
ldq ul1, 24(up) |
| 202 |
lda up, 32(up) |
| 203 |
lda rp, 32(rp) |
| 204 |
mulq vl0, ul0, m0a |
| 205 |
addq t0, m0b, acc1 |
| 206 |
srl acc0,NUMB_BITS, t1 |
| 207 |
bge n, L(el2) |
| 208 |
|
| 209 |
br r31, L(ta6) |
| 210 |
|
| 211 |
L(1m4): lda n, -4(n) |
| 212 |
ldq ul1, 0(up) |
| 213 |
lda up, 8(up) |
| 214 |
lda rp, -24(rp) |
| 215 |
bge n, L(ge1) |
| 216 |
|
| 217 |
mulq vl0, ul1, m1a |
| 218 |
umulh vl0, ul1, m1b |
| 219 |
srl m1a,NAIL_BITS, t0 |
| 220 |
addq t0, r31, acc1 |
| 221 |
and acc1,numb_mask, r28 |
| 222 |
srl acc1,NUMB_BITS, t1 |
| 223 |
stq r28, 24(rp) |
| 224 |
addq t1, m1b, r0 |
| 225 |
ret r31, (r26), 1 |
| 226 |
|
| 227 |
L(ge1): ldq ul2, 0(up) |
| 228 |
mulq vl0, ul1, m1a |
| 229 |
umulh vl0, ul1, m1b |
| 230 |
ldq ul3, 8(up) |
| 231 |
lda n, -4(n) |
| 232 |
mulq vl0, ul2, m2a |
| 233 |
umulh vl0, ul2, m2b |
| 234 |
ldq ul0, 16(up) |
| 235 |
mulq vl0, ul3, m3a |
| 236 |
umulh vl0, ul3, m3b |
| 237 |
srl m1a,NAIL_BITS, t0 |
| 238 |
ldq ul1, 24(up) |
| 239 |
lda up, 32(up) |
| 240 |
lda rp, 32(rp) |
| 241 |
mulq vl0, ul0, m0a |
| 242 |
addq t0, r31, acc1 |
| 243 |
umulh vl0, ul0, m0b |
| 244 |
srl m2a,NAIL_BITS, t0 |
| 245 |
mulq vl0, ul1, m1a |
| 246 |
addq t0, m1b, acc0 |
| 247 |
srl acc1,NUMB_BITS, t1 |
| 248 |
blt n, L(ta5) |
| 249 |
|
| 250 |
L(ge5): ldq ul2, 0(up) |
| 251 |
br r31, L(el1) |
| 252 |
|
| 253 |
ALIGN(16) |
| 254 |
L(top): mulq vl0, ul0, m0a C U1 |
| 255 |
addq t0, m0b, acc1 C L0 |
| 256 |
srl acc0,NUMB_BITS, t1 C U0 |
| 257 |
stq r28, -24(rp) C L1 |
| 258 |
C |
| 259 |
L(el2): umulh vl0, ul0, m0b C U1 |
| 260 |
and acc0,numb_mask, r28 C L0 |
| 261 |
unop C U0 |
| 262 |
unop C L1 |
| 263 |
C |
| 264 |
unop C U1 |
| 265 |
addq t1, acc1, acc1 C L0 |
| 266 |
srl m2a,NAIL_BITS, t0 C U0 |
| 267 |
ldq ul2, 0(up) C L1 |
| 268 |
C |
| 269 |
mulq vl0, ul1, m1a C U1 |
| 270 |
addq t0, m1b, acc0 C L0 |
| 271 |
srl acc1,NUMB_BITS, t1 C U0 |
| 272 |
stq r28, -16(rp) C L1 |
| 273 |
C |
| 274 |
L(el1): umulh vl0, ul1, m1b C U1 |
| 275 |
and acc1,numb_mask, r28 C L0 |
| 276 |
unop C U0 |
| 277 |
lda n, -4(n) C L1 |
| 278 |
C |
| 279 |
unop C U1 |
| 280 |
addq t1, acc0, acc0 C L0 |
| 281 |
srl m3a,NAIL_BITS, t0 C U0 |
| 282 |
ldq ul3, 8(up) C L1 |
| 283 |
C |
| 284 |
mulq vl0, ul2, m2a C U1 |
| 285 |
addq t0, m2b, acc1 C L0 |
| 286 |
srl acc0,NUMB_BITS, t1 C U0 |
| 287 |
stq r28, -8(rp) C L1 |
| 288 |
C |
| 289 |
L(el0): umulh vl0, ul2, m2b C U1 |
| 290 |
and acc0,numb_mask, r28 C L0 |
| 291 |
unop C U0 |
| 292 |
unop C L1 |
| 293 |
C |
| 294 |
unop C U1 |
| 295 |
addq t1, acc1, acc1 C L0 |
| 296 |
srl m0a,NAIL_BITS, t0 C U0 |
| 297 |
ldq ul0, 16(up) C L1 |
| 298 |
C |
| 299 |
mulq vl0, ul3, m3a C U1 |
| 300 |
addq t0, m3b, acc0 C L0 |
| 301 |
srl acc1,NUMB_BITS, t1 C U0 |
| 302 |
stq r28, 0(rp) C L1 |
| 303 |
C |
| 304 |
L(el3): umulh vl0, ul3, m3b C U1 |
| 305 |
and acc1,numb_mask, r28 C L0 |
| 306 |
unop C U0 |
| 307 |
unop C L1 |
| 308 |
C |
| 309 |
unop C U1 |
| 310 |
addq t1, acc0, acc0 C L0 |
| 311 |
srl m1a,NAIL_BITS, t0 C U0 |
| 312 |
ldq ul1, 24(up) C L1 |
| 313 |
C |
| 314 |
lda up, 32(up) C L0 |
| 315 |
unop C U1 |
| 316 |
lda rp, 32(rp) C L1 |
| 317 |
bge n, L(top) C U0 |
| 318 |
|
| 319 |
L(end): mulq vl0, ul0, m0a |
| 320 |
addq t0, m0b, acc1 |
| 321 |
srl acc0,NUMB_BITS, t1 |
| 322 |
stq r28, -24(rp) |
| 323 |
L(ta6): umulh vl0, ul0, m0b |
| 324 |
and acc0,numb_mask, r28 |
| 325 |
addq t1, acc1, acc1 |
| 326 |
srl m2a,NAIL_BITS, t0 |
| 327 |
mulq vl0, ul1, m1a |
| 328 |
addq t0, m1b, acc0 |
| 329 |
srl acc1,NUMB_BITS, t1 |
| 330 |
stq r28, -16(rp) |
| 331 |
L(ta5): umulh vl0, ul1, m1b |
| 332 |
and acc1,numb_mask, r28 |
| 333 |
addq t1, acc0, acc0 |
| 334 |
srl m3a,NAIL_BITS, t0 |
| 335 |
addq t0, m2b, acc1 |
| 336 |
srl acc0,NUMB_BITS, t1 |
| 337 |
stq r28, -8(rp) |
| 338 |
ALIGN(16) |
| 339 |
L(ta4): and acc0,numb_mask, r28 |
| 340 |
addq t1, acc1, acc1 |
| 341 |
srl m0a,NAIL_BITS, t0 |
| 342 |
addq t0, m3b, acc0 |
| 343 |
srl acc1,NUMB_BITS, t1 |
| 344 |
stq r28, 0(rp) |
| 345 |
unop |
| 346 |
ALIGN(16) |
| 347 |
L(ta3): and acc1,numb_mask, r28 |
| 348 |
addq t1, acc0, acc0 |
| 349 |
srl m1a,NAIL_BITS, t0 |
| 350 |
addq t0, m0b, acc1 |
| 351 |
srl acc0,NUMB_BITS, t1 |
| 352 |
stq r28, 8(rp) |
| 353 |
unop |
| 354 |
ALIGN(16) |
| 355 |
L(ta2): and acc0,numb_mask, r28 |
| 356 |
addq t1, acc1, acc1 |
| 357 |
srl acc1,NUMB_BITS, t1 |
| 358 |
stq r28, 16(rp) |
| 359 |
and acc1,numb_mask, r28 |
| 360 |
addq t1, m1b, r0 |
| 361 |
stq r28, 24(rp) |
| 362 |
ret r31, (r26), 1 |
| 363 |
EPILOGUE() |
| 364 |
ASM_END() |