|
| 1 | +/*************************************************************************** |
| 2 | +Copyright (c) 2013-2016, The OpenBLAS Project |
| 3 | +All rights reserved. |
| 4 | +Redistribution and use in source and binary forms, with or without |
| 5 | +modification, are permitted provided that the following conditions are |
| 6 | +met: |
| 7 | +1. Redistributions of source code must retain the above copyright |
| 8 | +notice, this list of conditions and the following disclaimer. |
| 9 | +2. Redistributions in binary form must reproduce the above copyright |
| 10 | +notice, this list of conditions and the following disclaimer in |
| 11 | +the documentation and/or other materials provided with the |
| 12 | +distribution. |
| 13 | +3. Neither the name of the OpenBLAS project nor the names of |
| 14 | +its contributors may be used to endorse or promote products |
| 15 | +derived from this software without specific prior written permission. |
| 16 | +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 17 | +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 18 | +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 19 | +ARE DISCLAIMED. IN NO EVENT SHALL THE OPENBLAS PROJECT OR CONTRIBUTORS BE |
| 20 | +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
| 21 | +DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR |
| 22 | +SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
| 23 | +CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 24 | +OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE |
| 25 | +USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 26 | +*****************************************************************************/ |
| 27 | + |
| 28 | +/************************************************************************************** |
| 29 | +* 2016/04/21 Werner Saar (wernsaar@googlemail.com) |
| 30 | +* BLASTEST : OK |
| 31 | +* CTEST : OK |
| 32 | +* TEST : OK |
| 33 | +* LAPACK-TEST : OK |
| 34 | +**************************************************************************************/ |
| 35 | + |
| 36 | + |
| 37 | +/********************************************************************************************** |
| 38 | +* Macros for N=4 and M=16 |
| 39 | +**********************************************************************************************/ |
| 40 | + |
| 41 | +#if defined(_AIX) |
| 42 | +define(`COPY_4x16', ` |
| 43 | +#else |
| 44 | +.macro COPY_4x16 |
| 45 | +#endif |
| 46 | + |
| 47 | + lxvpx vs32, o0, A0 |
| 48 | + lxvpx vs34, o32, A0 |
| 49 | + |
| 50 | + lxvpx vs36, o0, A1 |
| 51 | + lxvpx vs38, o32, A1 |
| 52 | + |
| 53 | + lxvpx vs40, o0, A2 |
| 54 | + lxvpx vs42, o32, A2 |
| 55 | + |
| 56 | + lxvpx vs44, o0, A3 |
| 57 | + lxvpx vs46, o32, A3 |
| 58 | + |
| 59 | + mr T1, BO |
| 60 | + |
| 61 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 62 | + stxvx vs32, o0, T1 |
| 63 | + stxvx vs33, o16, T1 |
| 64 | + stxvx vs34, o32, T1 |
| 65 | + stxvx vs35, o48, T1 |
| 66 | +#else |
| 67 | + stxvx vs33, o0, T1 |
| 68 | + stxvx vs32, o16, T1 |
| 69 | + stxvx vs35, o32, T1 |
| 70 | + stxvx vs34, o48, T1 |
| 71 | +#endif |
| 72 | + |
| 73 | + addi T1, T1, 64 |
| 74 | + |
| 75 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 76 | + stxvx vs36, o0, T1 |
| 77 | + stxvx vs37, o16, T1 |
| 78 | + stxvx vs38, o32, T1 |
| 79 | + stxvx vs39, o48, T1 |
| 80 | +#else |
| 81 | + stxvx vs37, o0, T1 |
| 82 | + stxvx vs36, o16, T1 |
| 83 | + stxvx vs39, o32, T1 |
| 84 | + stxvx vs38, o48, T1 |
| 85 | +#endif |
| 86 | + |
| 87 | + addi T1, T1, 64 |
| 88 | + |
| 89 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 90 | + stxvx vs40, o0, T1 |
| 91 | + stxvx vs41, o16, T1 |
| 92 | + stxvx vs42, o32, T1 |
| 93 | + stxvx vs43, o48, T1 |
| 94 | +#else |
| 95 | + stxvx vs41, o0, T1 |
| 96 | + stxvx vs40, o16, T1 |
| 97 | + stxvx vs43, o32, T1 |
| 98 | + stxvx vs42, o48, T1 |
| 99 | +#endif |
| 100 | + |
| 101 | + addi T1, T1, 64 |
| 102 | + |
| 103 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 104 | + stxvx vs44, o0, T1 |
| 105 | + stxvx vs45, o16, T1 |
| 106 | + stxvx vs46, o32, T1 |
| 107 | + stxvx vs47, o48, T1 |
| 108 | +#else |
| 109 | + stxvx vs45, o0, T1 |
| 110 | + stxvx vs44, o16, T1 |
| 111 | + stxvx vs47, o32, T1 |
| 112 | + stxvx vs46, o48, T1 |
| 113 | +#endif |
| 114 | + |
| 115 | +#if defined(_AIX) |
| 116 | +') |
| 117 | +#else |
| 118 | +.endm |
| 119 | +#endif |
| 120 | + |
| 121 | +/********************************************************************************************** |
| 122 | +* Macros for N=4 and M=8 |
| 123 | +**********************************************************************************************/ |
| 124 | + |
| 125 | +#if defined(_AIX) |
| 126 | +define(`COPY_4x8', ` |
| 127 | +#else |
| 128 | +.macro COPY_4x8 |
| 129 | +#endif |
| 130 | + |
| 131 | + lxvpx vs32, o0, A0 |
| 132 | + |
| 133 | + lxvpx vs34, o0, A1 |
| 134 | + |
| 135 | + lxvpx vs36, o0, A2 |
| 136 | + |
| 137 | + lxvpx vs38, o0, A3 |
| 138 | + |
| 139 | + mr T1, BO |
| 140 | + |
| 141 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 142 | + stxvx vs32, o0, T1 |
| 143 | + stxvx vs33, o16, T1 |
| 144 | + |
| 145 | + stxvx vs34, o32, T1 |
| 146 | + stxvx vs35, o48, T1 |
| 147 | +#else |
| 148 | + stxvx vs33, o0, T1 |
| 149 | + stxvx vs32, o16, T1 |
| 150 | + |
| 151 | + stxvx vs35, o32, T1 |
| 152 | + stxvx vs34, o48, T1 |
| 153 | +#endif |
| 154 | + |
| 155 | + addi T1, T1, 64 |
| 156 | + |
| 157 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 158 | + stxvx vs36, o0, T1 |
| 159 | + stxvx vs37, o16, T1 |
| 160 | + |
| 161 | + stxvx vs38, o32, T1 |
| 162 | + stxvx vs39, o48, T1 |
| 163 | +#else |
| 164 | + stxvx vs37, o0, T1 |
| 165 | + stxvx vs36, o16, T1 |
| 166 | + |
| 167 | + stxvx vs39, o32, T1 |
| 168 | + stxvx vs38, o48, T1 |
| 169 | +#endif |
| 170 | + |
| 171 | +#if defined(_AIX) |
| 172 | +') |
| 173 | +#else |
| 174 | +.endm |
| 175 | +#endif |
| 176 | + |
| 177 | +/********************************************************************************************** |
| 178 | +* Macros for N=2 and M=16 |
| 179 | +**********************************************************************************************/ |
| 180 | + |
| 181 | +#if defined(_AIX) |
| 182 | +define(`COPY_2x16', ` |
| 183 | +#else |
| 184 | +.macro COPY_2x16 |
| 185 | +#endif |
| 186 | + |
| 187 | + lxvpx vs32, o0, A0 |
| 188 | + lxvpx vs34, o32, A0 |
| 189 | + |
| 190 | + lxvpx vs36, o0, A1 |
| 191 | + lxvpx vs38, o32, A1 |
| 192 | + |
| 193 | + mr T1, BO |
| 194 | + |
| 195 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 196 | + stxvx vs32, o0, T1 |
| 197 | + stxvx vs33, o16, T1 |
| 198 | + stxvx vs34, o32, T1 |
| 199 | + stxvx vs35, o48, T1 |
| 200 | +#else |
| 201 | + stxvx vs33, o0, T1 |
| 202 | + stxvx vs32, o16, T1 |
| 203 | + stxvx vs35, o32, T1 |
| 204 | + stxvx vs34, o48, T1 |
| 205 | +#endif |
| 206 | + |
| 207 | + addi T1, T1, 64 |
| 208 | + |
| 209 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 210 | + stxvx vs36, o0, T1 |
| 211 | + stxvx vs37, o16, T1 |
| 212 | + stxvx vs38, o32, T1 |
| 213 | + stxvx vs39, o48, T1 |
| 214 | +#else |
| 215 | + stxvx vs37, o0, T1 |
| 216 | + stxvx vs36, o16, T1 |
| 217 | + stxvx vs39, o32, T1 |
| 218 | + stxvx vs38, o48, T1 |
| 219 | +#endif |
| 220 | + |
| 221 | +#if defined(_AIX) |
| 222 | +') |
| 223 | +#else |
| 224 | +.endm |
| 225 | +#endif |
| 226 | + |
| 227 | +/********************************************************************************************** |
| 228 | +* Macros for N=2 and M=8 |
| 229 | +**********************************************************************************************/ |
| 230 | + |
| 231 | +#if defined(_AIX) |
| 232 | +define(`COPY_2x8', ` |
| 233 | +#else |
| 234 | +.macro COPY_2x8 |
| 235 | +#endif |
| 236 | + |
| 237 | + lxvpx vs32, o0, A0 |
| 238 | + |
| 239 | + lxvpx vs34, o0, A1 |
| 240 | + |
| 241 | + mr T1, BO |
| 242 | + |
| 243 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 244 | + stxvx vs32, o0, T1 |
| 245 | + stxvx vs33, o16, T1 |
| 246 | + |
| 247 | + stxvx vs34, o32, T1 |
| 248 | + stxvx vs35, o48, T1 |
| 249 | +#else |
| 250 | + stxvx vs33, o0, T1 |
| 251 | + stxvx vs32, o16, T1 |
| 252 | + |
| 253 | + stxvx vs35, o32, T1 |
| 254 | + stxvx vs34, o48, T1 |
| 255 | +#endif |
| 256 | + |
| 257 | +#if defined(_AIX) |
| 258 | +') |
| 259 | +#else |
| 260 | +.endm |
| 261 | +#endif |
| 262 | + |
| 263 | +/********************************************************************************************** |
| 264 | +* Macros for N=1 and M=16 |
| 265 | +**********************************************************************************************/ |
| 266 | + |
| 267 | +#if defined(_AIX) |
| 268 | +define(`COPY_1x16', ` |
| 269 | +#else |
| 270 | +.macro COPY_1x16 |
| 271 | +#endif |
| 272 | + |
| 273 | + lxvpx vs32, o0, A0 |
| 274 | + lxvpx vs34, o32, A0 |
| 275 | + |
| 276 | + mr T1, BO |
| 277 | + |
| 278 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 279 | + stxvx vs32, o0, T1 |
| 280 | + stxvx vs33, o16, T1 |
| 281 | + stxvx vs34, o32, T1 |
| 282 | + stxvx vs35, o48, T1 |
| 283 | +#else |
| 284 | + stxvx vs33, o0, T1 |
| 285 | + stxvx vs32, o16, T1 |
| 286 | + stxvx vs35, o32, T1 |
| 287 | + stxvx vs34, o48, T1 |
| 288 | +#endif |
| 289 | + |
| 290 | +#if defined(_AIX) |
| 291 | +') |
| 292 | +#else |
| 293 | +.endm |
| 294 | +#endif |
| 295 | + |
| 296 | +/********************************************************************************************** |
| 297 | +* Macros for N=1 and M=8 |
| 298 | +**********************************************************************************************/ |
| 299 | + |
| 300 | +#if defined(_AIX) |
| 301 | +define(`COPY_1x8', ` |
| 302 | +#else |
| 303 | +.macro COPY_1x8 |
| 304 | +#endif |
| 305 | + |
| 306 | + lxvpx vs32, o0, A0 |
| 307 | + |
| 308 | + mr T1, BO |
| 309 | + |
| 310 | +#if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 311 | + stxvx vs32, o0, T1 |
| 312 | + stxvx vs33, o16, T1 |
| 313 | +#else |
| 314 | + stxvx vs33, o0, T1 |
| 315 | + stxvx vs32, o16, T1 |
| 316 | +#endif |
| 317 | + |
| 318 | +#if defined(_AIX) |
| 319 | +') |
| 320 | +#else |
| 321 | +.endm |
| 322 | +#endif |
| 323 | + |
0 commit comments