|
| 1 | +/*********************************************************************/ |
| 2 | +/* Copyright 2009, 2010 The University of Texas at Austin. */ |
| 3 | +/* All rights reserved. */ |
| 4 | +/* */ |
| 5 | +/* Redistribution and use in source and binary forms, with or */ |
| 6 | +/* without modification, are permitted provided that the following */ |
| 7 | +/* conditions are met: */ |
| 8 | +/* */ |
| 9 | +/* 1. Redistributions of source code must retain the above */ |
| 10 | +/* copyright notice, this list of conditions and the following */ |
| 11 | +/* disclaimer. */ |
| 12 | +/* */ |
| 13 | +/* 2. Redistributions in binary form must reproduce the above */ |
| 14 | +/* copyright notice, this list of conditions and the following */ |
| 15 | +/* disclaimer in the documentation and/or other materials */ |
| 16 | +/* provided with the distribution. */ |
| 17 | +/* */ |
| 18 | +/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */ |
| 19 | +/* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */ |
| 20 | +/* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */ |
| 21 | +/* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */ |
| 22 | +/* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */ |
| 23 | +/* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */ |
| 24 | +/* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */ |
| 25 | +/* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */ |
| 26 | +/* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */ |
| 27 | +/* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */ |
| 28 | +/* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */ |
| 29 | +/* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */ |
| 30 | +/* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */ |
| 31 | +/* POSSIBILITY OF SUCH DAMAGE. */ |
| 32 | +/* */ |
| 33 | +/* The views and conclusions contained in the software and */ |
| 34 | +/* documentation are those of the authors and should not be */ |
| 35 | +/* interpreted as representing official policies, either expressed */ |
| 36 | +/* or implied, of The University of Texas at Austin. */ |
| 37 | +/*********************************************************************/ |
| 38 | + |
| 39 | +#define ASSEMBLER |
| 40 | +#include "common.h" |
| 41 | + |
| 42 | +#define N $4 |
| 43 | +#define X $5 |
| 44 | +#define INCX $6 |
| 45 | + |
| 46 | +#define I $2 |
| 47 | +#define TEMP $3 |
| 48 | + |
| 49 | +#define a1 $f2 |
| 50 | +#define a2 $f3 |
| 51 | +#define a3 $f4 |
| 52 | +#define a4 $f5 |
| 53 | +#define a5 $f6 |
| 54 | +#define a6 $f7 |
| 55 | +#define a7 $f8 |
| 56 | +#define a8 $f9 |
| 57 | + |
| 58 | +#define t1 $f10 |
| 59 | +#define t2 $f11 |
| 60 | +#define t3 $f12 |
| 61 | +#define t4 $f13 |
| 62 | + |
| 63 | +#define s1 $f0 |
| 64 | +#define s2 $f1 |
| 65 | + |
| 66 | + PROLOGUE |
| 67 | + |
| 68 | +#ifdef F_INTERFACE |
| 69 | + LDINT N, 0(N) |
| 70 | + LDINT INCX, 0(INCX) |
| 71 | +#endif |
| 72 | + |
| 73 | + MTC $0, s1 |
| 74 | + |
| 75 | + MTC $0, s2 |
| 76 | + dsll INCX, INCX, BASE_SHIFT |
| 77 | + |
| 78 | + blez N, .L999 |
| 79 | + li TEMP, SIZE |
| 80 | + |
| 81 | + bne INCX, TEMP, .L20 |
| 82 | + dsra I, N, 3 |
| 83 | + |
| 84 | + blez I, .L15 |
| 85 | + NOP |
| 86 | + |
| 87 | + LD a1, 0 * SIZE(X) |
| 88 | + LD a2, 1 * SIZE(X) |
| 89 | + LD a3, 2 * SIZE(X) |
| 90 | + LD a4, 3 * SIZE(X) |
| 91 | + |
| 92 | + LD a5, 4 * SIZE(X) |
| 93 | + MOV t1, a1 |
| 94 | + LD a6, 5 * SIZE(X) |
| 95 | + MOV t2, a2 |
| 96 | + LD a7, 6 * SIZE(X) |
| 97 | + MOV t3, a3 |
| 98 | + |
| 99 | + MOV t4, a4 |
| 100 | + daddiu I, I, -1 |
| 101 | + |
| 102 | + blez I, .L13 |
| 103 | + LD a8, 7 * SIZE(X) |
| 104 | + .align 3 |
| 105 | + |
| 106 | +.L12: |
| 107 | + ADD s1, s1, t1 |
| 108 | + LD a1, 8 * SIZE(X) |
| 109 | + |
| 110 | + MOV t1, a5 |
| 111 | + daddiu I, I, -1 |
| 112 | + |
| 113 | + ADD s2, s2, t2 |
| 114 | + LD a2, 9 * SIZE(X) |
| 115 | + |
| 116 | + MOV t2, a6 |
| 117 | + NOP |
| 118 | + |
| 119 | + ADD s1, s1, t3 |
| 120 | + LD a3, 10 * SIZE(X) |
| 121 | + |
| 122 | + MOV t3, a7 |
| 123 | + NOP |
| 124 | + |
| 125 | + ADD s2, s2, t4 |
| 126 | + LD a4, 11 * SIZE(X) |
| 127 | + |
| 128 | + MOV t4, a8 |
| 129 | + daddiu X, X, 8 * SIZE |
| 130 | + |
| 131 | + ADD s1, s1, t1 |
| 132 | + LD a5, 4 * SIZE(X) |
| 133 | + |
| 134 | + MOV t1, a1 |
| 135 | + NOP |
| 136 | + |
| 137 | + ADD s2, s2, t2 |
| 138 | + LD a6, 5 * SIZE(X) |
| 139 | + |
| 140 | + MOV t2, a2 |
| 141 | + NOP |
| 142 | + |
| 143 | + ADD s1, s1, t3 |
| 144 | + LD a7, 6 * SIZE(X) |
| 145 | + |
| 146 | + MOV t3, a3 |
| 147 | + NOP |
| 148 | + |
| 149 | + ADD s2, s2, t4 |
| 150 | + LD a8, 7 * SIZE(X) |
| 151 | + |
| 152 | + bgtz I, .L12 |
| 153 | + MOV t4, a4 |
| 154 | + .align 3 |
| 155 | + |
| 156 | +.L13: |
| 157 | + ADD s1, s1, t1 |
| 158 | + daddiu X, X, 8 * SIZE |
| 159 | + |
| 160 | + MOV t1, a5 |
| 161 | + NOP |
| 162 | + |
| 163 | + ADD s2, s2, t2 |
| 164 | + MOV t2, a6 |
| 165 | + |
| 166 | + ADD s1, s1, t3 |
| 167 | + MOV t3, a7 |
| 168 | + |
| 169 | + ADD s2, s2, t4 |
| 170 | + MOV t4, a8 |
| 171 | + |
| 172 | + ADD s1, s1, t1 |
| 173 | + ADD s2, s2, t2 |
| 174 | + ADD s1, s1, t3 |
| 175 | + ADD s2, s2, t4 |
| 176 | + .align 3 |
| 177 | + |
| 178 | +.L15: |
| 179 | + andi I, N, 7 |
| 180 | + |
| 181 | + blez I, .L999 |
| 182 | + NOP |
| 183 | + .align 3 |
| 184 | + |
| 185 | +.L16: |
| 186 | + LD a1, 0 * SIZE(X) |
| 187 | + daddiu I, I, -1 |
| 188 | + |
| 189 | + MOV t1, a1 |
| 190 | + |
| 191 | + ADD s1, s1, t1 |
| 192 | + |
| 193 | + bgtz I, .L16 |
| 194 | + daddiu X, X, SIZE |
| 195 | + |
| 196 | + j .L999 |
| 197 | + NOP |
| 198 | + .align 3 |
| 199 | + |
| 200 | +.L20: |
| 201 | + blez I, .L25 |
| 202 | + NOP |
| 203 | + |
| 204 | + LD a1, 0 * SIZE(X) |
| 205 | + daddu X, X, INCX |
| 206 | + |
| 207 | + LD a2, 0 * SIZE(X) |
| 208 | + daddu X, X, INCX |
| 209 | + |
| 210 | + LD a3, 0 * SIZE(X) |
| 211 | + daddu X, X, INCX |
| 212 | + |
| 213 | + LD a4, 0 * SIZE(X) |
| 214 | + daddu X, X, INCX |
| 215 | + |
| 216 | + LD a5, 0 * SIZE(X) |
| 217 | + daddu X, X, INCX |
| 218 | + |
| 219 | + LD a6, 0 * SIZE(X) |
| 220 | + daddu X, X, INCX |
| 221 | + |
| 222 | + MOV t1, a1 |
| 223 | + LD a7, 0 * SIZE(X) |
| 224 | + |
| 225 | + MOV t2, a2 |
| 226 | + daddu X, X, INCX |
| 227 | + |
| 228 | + MOV t3, a3 |
| 229 | + LD a8, 0 * SIZE(X) |
| 230 | + |
| 231 | + MOV t4, a4 |
| 232 | + daddiu I, I, -1 |
| 233 | + |
| 234 | + blez I, .L24 |
| 235 | + daddu X, X, INCX |
| 236 | + .align 3 |
| 237 | + |
| 238 | +.L23: |
| 239 | + ADD s1, s1, t1 |
| 240 | + LD a1, 0 * SIZE(X) |
| 241 | + |
| 242 | + MOV t1, a5 |
| 243 | + daddu X, X, INCX |
| 244 | + |
| 245 | + ADD s2, s2, t2 |
| 246 | + LD a2, 0 * SIZE(X) |
| 247 | + |
| 248 | + MOV t2, a6 |
| 249 | + daddu X, X, INCX |
| 250 | + |
| 251 | + ADD s1, s1, t3 |
| 252 | + LD a3, 0 * SIZE(X) |
| 253 | + |
| 254 | + MOV t3, a7 |
| 255 | + daddu X, X, INCX |
| 256 | + |
| 257 | + ADD s2, s2, t4 |
| 258 | + LD a4, 0 * SIZE(X) |
| 259 | + |
| 260 | + MOV t4, a8 |
| 261 | + daddu X, X, INCX |
| 262 | + |
| 263 | + ADD s1, s1, t1 |
| 264 | + LD a5, 0 * SIZE(X) |
| 265 | + |
| 266 | + MOV t1, a1 |
| 267 | + daddu X, X, INCX |
| 268 | + |
| 269 | + ADD s2, s2, t2 |
| 270 | + LD a6, 0 * SIZE(X) |
| 271 | + |
| 272 | + MOV t2, a2 |
| 273 | + daddu X, X, INCX |
| 274 | + |
| 275 | + ADD s1, s1, t3 |
| 276 | + LD a7, 0 * SIZE(X) |
| 277 | + |
| 278 | + MOV t3, a3 |
| 279 | + daddu X, X, INCX |
| 280 | + |
| 281 | + ADD s2, s2, t4 |
| 282 | + LD a8, 0 * SIZE(X) |
| 283 | + |
| 284 | + MOV t4, a4 |
| 285 | + daddiu I, I, -1 |
| 286 | + |
| 287 | + bgtz I, .L23 |
| 288 | + daddu X, X, INCX |
| 289 | + .align 3 |
| 290 | + |
| 291 | +.L24: |
| 292 | + ADD s1, s1, t1 |
| 293 | + MOV t1, a5 |
| 294 | + |
| 295 | + ADD s2, s2, t2 |
| 296 | + MOV t2, a6 |
| 297 | + |
| 298 | + ADD s1, s1, t3 |
| 299 | + MOV t3, a7 |
| 300 | + |
| 301 | + ADD s2, s2, t4 |
| 302 | + MOV t4, a8 |
| 303 | + |
| 304 | + ADD s1, s1, t1 |
| 305 | + ADD s2, s2, t2 |
| 306 | + ADD s1, s1, t3 |
| 307 | + ADD s2, s2, t4 |
| 308 | + .align 3 |
| 309 | + |
| 310 | +.L25: |
| 311 | + andi I, N, 7 |
| 312 | + |
| 313 | + blez I, .L999 |
| 314 | + NOP |
| 315 | + .align 3 |
| 316 | + |
| 317 | +.L26: |
| 318 | + LD a1, 0 * SIZE(X) |
| 319 | + daddiu I, I, -1 |
| 320 | + |
| 321 | + MOV t1, a1 |
| 322 | + daddu X, X, INCX |
| 323 | + |
| 324 | + bgtz I, .L26 |
| 325 | + ADD s1, s1, t1 |
| 326 | + .align 3 |
| 327 | + |
| 328 | +.L999: |
| 329 | + j $31 |
| 330 | + ADD s1, s1, s2 |
| 331 | + |
| 332 | + EPILOGUE |
0 commit comments