|
| 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 | +#include <stdio.h> |
| 40 | +#include "common.h" |
| 41 | + |
| 42 | +int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, FLOAT *b){ |
| 43 | + BLASLONG i, j; |
| 44 | + |
| 45 | + IFLOAT *aoffset; |
| 46 | + IFLOAT *aoffset1, *aoffset2, *aoffset3, *aoffset4; |
| 47 | + IFLOAT *aoffset5, *aoffset6, *aoffset7, *aoffset8; |
| 48 | + IFLOAT *aoffset9, *aoffset10, *aoffset11, *aoffset12; |
| 49 | + IFLOAT *aoffset13, *aoffset14, *aoffset15, *aoffset16; |
| 50 | + |
| 51 | + IFLOAT *boffset; |
| 52 | + IFLOAT ctemp01, ctemp02, ctemp03, ctemp04; |
| 53 | + IFLOAT ctemp05, ctemp06, ctemp07, ctemp08; |
| 54 | + IFLOAT ctemp09, ctemp10, ctemp11, ctemp12; |
| 55 | + IFLOAT ctemp13, ctemp14, ctemp15, ctemp16; |
| 56 | + IFLOAT ctemp17, ctemp18, ctemp19, ctemp20; |
| 57 | + IFLOAT ctemp21, ctemp22, ctemp23, ctemp24; |
| 58 | + IFLOAT ctemp25, ctemp26, ctemp27, ctemp28; |
| 59 | + IFLOAT ctemp29, ctemp30, ctemp31, ctemp32; |
| 60 | + |
| 61 | + aoffset = a; |
| 62 | + boffset = b; |
| 63 | + lda *= 2; |
| 64 | + |
| 65 | + j = (n >> 4); |
| 66 | + if (j > 0){ |
| 67 | + do{ |
| 68 | + aoffset1 = aoffset; |
| 69 | + aoffset2 = aoffset1 + lda; |
| 70 | + aoffset3 = aoffset2 + lda; |
| 71 | + aoffset4 = aoffset3 + lda; |
| 72 | + aoffset5 = aoffset4 + lda; |
| 73 | + aoffset6 = aoffset5 + lda; |
| 74 | + aoffset7 = aoffset6 + lda; |
| 75 | + aoffset8 = aoffset7 + lda; |
| 76 | + aoffset9 = aoffset8 + lda; |
| 77 | + aoffset10 = aoffset9 + lda; |
| 78 | + aoffset11 = aoffset10 + lda; |
| 79 | + aoffset12 = aoffset11 + lda; |
| 80 | + aoffset13 = aoffset12 + lda; |
| 81 | + aoffset14 = aoffset13 + lda; |
| 82 | + aoffset15 = aoffset14 + lda; |
| 83 | + aoffset16 = aoffset15 + lda; |
| 84 | + aoffset += 16 * lda; |
| 85 | + |
| 86 | + i = m; |
| 87 | + if (i > 0){ |
| 88 | + do{ |
| 89 | + ctemp01 = *(aoffset1 + 0); |
| 90 | + ctemp02 = *(aoffset1 + 1); |
| 91 | + ctemp03 = *(aoffset2 + 0); |
| 92 | + ctemp04 = *(aoffset2 + 1); |
| 93 | + ctemp05 = *(aoffset3 + 0); |
| 94 | + ctemp06 = *(aoffset3 + 1); |
| 95 | + ctemp07 = *(aoffset4 + 0); |
| 96 | + ctemp08 = *(aoffset4 + 1); |
| 97 | + ctemp09 = *(aoffset5 + 0); |
| 98 | + ctemp10 = *(aoffset5 + 1); |
| 99 | + ctemp11 = *(aoffset6 + 0); |
| 100 | + ctemp12 = *(aoffset6 + 1); |
| 101 | + ctemp13 = *(aoffset7 + 0); |
| 102 | + ctemp14 = *(aoffset7 + 1); |
| 103 | + ctemp15 = *(aoffset8 + 0); |
| 104 | + ctemp16 = *(aoffset8 + 1); |
| 105 | + |
| 106 | + ctemp17 = *(aoffset9 + 0); |
| 107 | + ctemp18 = *(aoffset9 + 1); |
| 108 | + ctemp19 = *(aoffset10 + 0); |
| 109 | + ctemp20 = *(aoffset10 + 1); |
| 110 | + ctemp21 = *(aoffset11 + 0); |
| 111 | + ctemp22 = *(aoffset11 + 1); |
| 112 | + ctemp23 = *(aoffset12 + 0); |
| 113 | + ctemp24 = *(aoffset12 + 1); |
| 114 | + ctemp25 = *(aoffset13 + 0); |
| 115 | + ctemp26 = *(aoffset13 + 1); |
| 116 | + ctemp27 = *(aoffset14 + 0); |
| 117 | + ctemp28 = *(aoffset14 + 1); |
| 118 | + ctemp29 = *(aoffset15 + 0); |
| 119 | + ctemp30 = *(aoffset15 + 1); |
| 120 | + ctemp31 = *(aoffset16 + 0); |
| 121 | + ctemp32 = *(aoffset16 + 1); |
| 122 | + |
| 123 | + *(boffset + 0) = ctemp01; |
| 124 | + *(boffset + 1) = ctemp02; |
| 125 | + *(boffset + 2) = ctemp03; |
| 126 | + *(boffset + 3) = ctemp04; |
| 127 | + *(boffset + 4) = ctemp05; |
| 128 | + *(boffset + 5) = ctemp06; |
| 129 | + *(boffset + 6) = ctemp07; |
| 130 | + *(boffset + 7) = ctemp08; |
| 131 | + *(boffset + 8) = ctemp09; |
| 132 | + *(boffset + 9) = ctemp10; |
| 133 | + *(boffset + 10) = ctemp11; |
| 134 | + *(boffset + 11) = ctemp12; |
| 135 | + *(boffset + 12) = ctemp13; |
| 136 | + *(boffset + 13) = ctemp14; |
| 137 | + *(boffset + 14) = ctemp15; |
| 138 | + *(boffset + 15) = ctemp16; |
| 139 | + |
| 140 | + *(boffset + 16) = ctemp17; |
| 141 | + *(boffset + 17) = ctemp18; |
| 142 | + *(boffset + 18) = ctemp19; |
| 143 | + *(boffset + 19) = ctemp20; |
| 144 | + *(boffset + 20) = ctemp21; |
| 145 | + *(boffset + 21) = ctemp22; |
| 146 | + *(boffset + 22) = ctemp23; |
| 147 | + *(boffset + 23) = ctemp24; |
| 148 | + *(boffset + 24) = ctemp25; |
| 149 | + *(boffset + 25) = ctemp26; |
| 150 | + *(boffset + 26) = ctemp27; |
| 151 | + *(boffset + 27) = ctemp28; |
| 152 | + *(boffset + 28) = ctemp29; |
| 153 | + *(boffset + 29) = ctemp30; |
| 154 | + *(boffset + 30) = ctemp31; |
| 155 | + *(boffset + 31) = ctemp32; |
| 156 | + |
| 157 | + aoffset1 += 2; |
| 158 | + aoffset2 += 2; |
| 159 | + aoffset3 += 2; |
| 160 | + aoffset4 += 2; |
| 161 | + aoffset5 += 2; |
| 162 | + aoffset6 += 2; |
| 163 | + aoffset7 += 2; |
| 164 | + aoffset8 += 2; |
| 165 | + aoffset9 += 2; |
| 166 | + aoffset10 += 2; |
| 167 | + aoffset11 += 2; |
| 168 | + aoffset12 += 2; |
| 169 | + aoffset13 += 2; |
| 170 | + aoffset14 += 2; |
| 171 | + aoffset15 += 2; |
| 172 | + aoffset16 += 2; |
| 173 | + |
| 174 | + boffset += 32; |
| 175 | + i --; |
| 176 | + }while(i > 0); |
| 177 | + } |
| 178 | + j--; |
| 179 | + }while(j > 0); |
| 180 | + } /* end of if(j > 0) */ |
| 181 | + |
| 182 | + if (n & 8){ |
| 183 | + aoffset1 = aoffset; |
| 184 | + aoffset2 = aoffset1 + lda; |
| 185 | + aoffset3 = aoffset2 + lda; |
| 186 | + aoffset4 = aoffset3 + lda; |
| 187 | + aoffset5 = aoffset4 + lda; |
| 188 | + aoffset6 = aoffset5 + lda; |
| 189 | + aoffset7 = aoffset6 + lda; |
| 190 | + aoffset8 = aoffset7 + lda; |
| 191 | + aoffset += 8 * lda; |
| 192 | + |
| 193 | + i = m; |
| 194 | + if (i > 0){ |
| 195 | + do{ |
| 196 | + ctemp01 = *(aoffset1 + 0); |
| 197 | + ctemp02 = *(aoffset1 + 1); |
| 198 | + ctemp03 = *(aoffset2 + 0); |
| 199 | + ctemp04 = *(aoffset2 + 1); |
| 200 | + ctemp05 = *(aoffset3 + 0); |
| 201 | + ctemp06 = *(aoffset3 + 1); |
| 202 | + ctemp07 = *(aoffset4 + 0); |
| 203 | + ctemp08 = *(aoffset4 + 1); |
| 204 | + ctemp09 = *(aoffset5 + 0); |
| 205 | + ctemp10 = *(aoffset5 + 1); |
| 206 | + ctemp11 = *(aoffset6 + 0); |
| 207 | + ctemp12 = *(aoffset6 + 1); |
| 208 | + ctemp13 = *(aoffset7 + 0); |
| 209 | + ctemp14 = *(aoffset7 + 1); |
| 210 | + ctemp15 = *(aoffset8 + 0); |
| 211 | + ctemp16 = *(aoffset8 + 1); |
| 212 | + |
| 213 | + *(boffset + 0) = ctemp01; |
| 214 | + *(boffset + 1) = ctemp02; |
| 215 | + *(boffset + 2) = ctemp03; |
| 216 | + *(boffset + 3) = ctemp04; |
| 217 | + *(boffset + 4) = ctemp05; |
| 218 | + *(boffset + 5) = ctemp06; |
| 219 | + *(boffset + 6) = ctemp07; |
| 220 | + *(boffset + 7) = ctemp08; |
| 221 | + *(boffset + 8) = ctemp09; |
| 222 | + *(boffset + 9) = ctemp10; |
| 223 | + *(boffset + 10) = ctemp11; |
| 224 | + *(boffset + 11) = ctemp12; |
| 225 | + *(boffset + 12) = ctemp13; |
| 226 | + *(boffset + 13) = ctemp14; |
| 227 | + *(boffset + 14) = ctemp15; |
| 228 | + *(boffset + 15) = ctemp16; |
| 229 | + |
| 230 | + aoffset1 += 2; |
| 231 | + aoffset2 += 2; |
| 232 | + aoffset3 += 2; |
| 233 | + aoffset4 += 2; |
| 234 | + aoffset5 += 2; |
| 235 | + aoffset6 += 2; |
| 236 | + aoffset7 += 2; |
| 237 | + aoffset8 += 2; |
| 238 | + |
| 239 | + boffset += 16; |
| 240 | + i --; |
| 241 | + }while(i > 0); |
| 242 | + } |
| 243 | + } |
| 244 | + |
| 245 | + if (n & 4){ |
| 246 | + aoffset1 = aoffset; |
| 247 | + aoffset2 = aoffset1 + lda; |
| 248 | + aoffset3 = aoffset2 + lda; |
| 249 | + aoffset4 = aoffset3 + lda; |
| 250 | + aoffset += 4 * lda; |
| 251 | + |
| 252 | + i = m; |
| 253 | + if (i > 0){ |
| 254 | + do{ |
| 255 | + ctemp01 = *(aoffset1 + 0); |
| 256 | + ctemp02 = *(aoffset1 + 1); |
| 257 | + ctemp03 = *(aoffset2 + 0); |
| 258 | + ctemp04 = *(aoffset2 + 1); |
| 259 | + ctemp05 = *(aoffset3 + 0); |
| 260 | + ctemp06 = *(aoffset3 + 1); |
| 261 | + ctemp07 = *(aoffset4 + 0); |
| 262 | + ctemp08 = *(aoffset4 + 1); |
| 263 | + |
| 264 | + *(boffset + 0) = ctemp01; |
| 265 | + *(boffset + 1) = ctemp02; |
| 266 | + *(boffset + 2) = ctemp03; |
| 267 | + *(boffset + 3) = ctemp04; |
| 268 | + *(boffset + 4) = ctemp05; |
| 269 | + *(boffset + 5) = ctemp06; |
| 270 | + *(boffset + 6) = ctemp07; |
| 271 | + *(boffset + 7) = ctemp08; |
| 272 | + |
| 273 | + aoffset1 += 2; |
| 274 | + aoffset2 += 2; |
| 275 | + aoffset3 += 2; |
| 276 | + aoffset4 += 2; |
| 277 | + |
| 278 | + boffset += 8; |
| 279 | + i --; |
| 280 | + }while(i > 0); |
| 281 | + } |
| 282 | + } /* end of if(j > 0) */ |
| 283 | + |
| 284 | + if (n & 2){ |
| 285 | + aoffset1 = aoffset; |
| 286 | + aoffset2 = aoffset1 + lda; |
| 287 | + aoffset += 2 * lda; |
| 288 | + |
| 289 | + i = m; |
| 290 | + if (i > 0){ |
| 291 | + do{ |
| 292 | + ctemp01 = *(aoffset1 + 0); |
| 293 | + ctemp02 = *(aoffset1 + 1); |
| 294 | + ctemp03 = *(aoffset2 + 0); |
| 295 | + ctemp04 = *(aoffset2 + 1); |
| 296 | + |
| 297 | + *(boffset + 0) = ctemp01; |
| 298 | + *(boffset + 1) = ctemp02; |
| 299 | + *(boffset + 2) = ctemp03; |
| 300 | + *(boffset + 3) = ctemp04; |
| 301 | + |
| 302 | + aoffset1 += 2; |
| 303 | + aoffset2 += 2; |
| 304 | + boffset += 4; |
| 305 | + i --; |
| 306 | + }while(i > 0); |
| 307 | + } |
| 308 | + |
| 309 | + } /* end of if(j > 0) */ |
| 310 | + |
| 311 | + if (n & 1){ |
| 312 | + aoffset1 = aoffset; |
| 313 | + |
| 314 | + i = m; |
| 315 | + if (i > 0){ |
| 316 | + do{ |
| 317 | + ctemp01 = *(aoffset1 + 0); |
| 318 | + ctemp02 = *(aoffset1 + 1); |
| 319 | + |
| 320 | + *(boffset + 0) = ctemp01; |
| 321 | + *(boffset + 1) = ctemp02; |
| 322 | + |
| 323 | + aoffset1 += 2; |
| 324 | + boffset += 2; |
| 325 | + i --; |
| 326 | + }while(i > 0); |
| 327 | + } |
| 328 | + |
| 329 | + } /* end of if(j > 0) */ |
| 330 | + |
| 331 | + return 0; |
| 332 | +} |
0 commit comments