| 1 | /* |
| 2 | * VIS op helpers |
| 3 | * |
| 4 | * Copyright (c) 2003-2005 Fabrice Bellard |
| 5 | * |
| 6 | * This library is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Lesser General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2.1 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * This library is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with this library; if not, see <http://www.gnu.org/licenses/>. |
| 18 | */ |
| 19 | |
| 20 | #include "qemu/osdep.h" |
| 21 | #include "cpu.h" |
| 22 | #include "exec/helper-proto.h" |
| 23 | #include "crypto/clmul.h" |
| 24 | |
| 25 | target_ulong helper_array8(target_ulong rs1, target_ulong rs2) |
| 26 | { |
| 27 | /* |
| 28 | * From Oracle SPARC Architecture 2015: |
| 29 | * Architecturally, an illegal R[rs2] value (>5) causes the array |
| 30 | * instructions to produce undefined results. For historic reference, |
| 31 | * past implementations of these instructions have ignored R[rs2]{63:3} |
| 32 | * and have treated R[rs2] values of 6 and 7 as if they were 5. |
| 33 | */ |
| 34 | target_ulong n = MIN(rs2 & 7, 5); |
| 35 | |
| 36 | target_ulong x_int = (rs1 >> 11) & 0x7ff; |
| 37 | target_ulong y_int = (rs1 >> 33) & 0x7ff; |
| 38 | target_ulong z_int = rs1 >> 55; |
| 39 | |
| 40 | target_ulong lower_x = x_int & 3; |
| 41 | target_ulong lower_y = y_int & 3; |
| 42 | target_ulong lower_z = z_int & 1; |
| 43 | |
| 44 | target_ulong middle_x = (x_int >> 2) & 15; |
| 45 | target_ulong middle_y = (y_int >> 2) & 15; |
| 46 | target_ulong middle_z = (z_int >> 1) & 15; |
| 47 | |
| 48 | target_ulong upper_x = (x_int >> 6) & ((1 << n) - 1); |
| 49 | target_ulong upper_y = (y_int >> 6) & ((1 << n) - 1); |
| 50 | target_ulong upper_z = z_int >> 5; |
| 51 | |
| 52 | return (upper_z << (17 + 2 * n)) |
| 53 | | (upper_y << (17 + n)) |
| 54 | | (upper_x << 17) |
| 55 | | (middle_z << 13) |
| 56 | | (middle_y << 9) |
| 57 | | (middle_x << 5) |
| 58 | | (lower_z << 4) |
| 59 | | (lower_y << 2) |
| 60 | | lower_x; |
| 61 | } |
| 62 | |
| 63 | #if HOST_BIG_ENDIAN |
| 64 | #define VIS_B64(n) b[7 - (n)] |
| 65 | #define VIS_SB64(n) sb[7 - (n)] |
| 66 | #define VIS_W64(n) w[3 - (n)] |
| 67 | #define VIS_SW64(n) sw[3 - (n)] |
| 68 | #define VIS_L64(n) l[1 - (n)] |
| 69 | #define VIS_SL64(n) sl[1 - (n)] |
| 70 | #define VIS_B32(n) b[3 - (n)] |
| 71 | #define VIS_W32(n) w[1 - (n)] |
| 72 | #else |
| 73 | #define VIS_B64(n) b[n] |
| 74 | #define VIS_SB64(n) sb[n] |
| 75 | #define VIS_W64(n) w[n] |
| 76 | #define VIS_SW64(n) sw[n] |
| 77 | #define VIS_L64(n) l[n] |
| 78 | #define VIS_SL64(n) sl[n] |
| 79 | #define VIS_B32(n) b[n] |
| 80 | #define VIS_W32(n) w[n] |
| 81 | #endif |
| 82 | |
| 83 | typedef union { |
| 84 | uint8_t b[8]; |
| 85 | int8_t sb[8]; |
| 86 | uint16_t w[4]; |
| 87 | int16_t sw[4]; |
| 88 | uint32_t l[2]; |
| 89 | int32_t sl[2]; |
| 90 | uint64_t ll; |
| 91 | float64 d; |
| 92 | } VIS64; |
| 93 | |
| 94 | typedef union { |
| 95 | uint8_t b[4]; |
| 96 | uint16_t w[2]; |
| 97 | uint32_t l; |
| 98 | float32 f; |
| 99 | } VIS32; |
| 100 | |
| 101 | uint64_t helper_fpmerge(uint32_t src1, uint32_t src2) |
| 102 | { |
| 103 | VIS32 s1, s2; |
| 104 | VIS64 d; |
| 105 | |
| 106 | s1.l = src1; |
| 107 | s2.l = src2; |
| 108 | d.ll = 0; |
| 109 | |
| 110 | d.VIS_B64(7) = s1.VIS_B32(3); |
| 111 | d.VIS_B64(6) = s2.VIS_B32(3); |
| 112 | d.VIS_B64(5) = s1.VIS_B32(2); |
| 113 | d.VIS_B64(4) = s2.VIS_B32(2); |
| 114 | d.VIS_B64(3) = s1.VIS_B32(1); |
| 115 | d.VIS_B64(2) = s2.VIS_B32(1); |
| 116 | d.VIS_B64(1) = s1.VIS_B32(0); |
| 117 | d.VIS_B64(0) = s2.VIS_B32(0); |
| 118 | |
| 119 | return d.ll; |
| 120 | } |
| 121 | |
| 122 | static inline int do_ms16b(int x, int y) |
| 123 | { |
| 124 | return ((x * y) + 0x80) >> 8; |
| 125 | } |
| 126 | |
| 127 | uint64_t helper_fmul8x16(uint32_t src1, uint64_t src2) |
| 128 | { |
| 129 | VIS64 d; |
| 130 | VIS32 s; |
| 131 | |
| 132 | s.l = src1; |
| 133 | d.ll = src2; |
| 134 | |
| 135 | d.VIS_W64(0) = do_ms16b(s.VIS_B32(0), d.VIS_SW64(0)); |
| 136 | d.VIS_W64(1) = do_ms16b(s.VIS_B32(1), d.VIS_SW64(1)); |
| 137 | d.VIS_W64(2) = do_ms16b(s.VIS_B32(2), d.VIS_SW64(2)); |
| 138 | d.VIS_W64(3) = do_ms16b(s.VIS_B32(3), d.VIS_SW64(3)); |
| 139 | |
| 140 | return d.ll; |
| 141 | } |
| 142 | |
| 143 | uint64_t helper_fmul8x16a(uint32_t src1, int32_t src2) |
| 144 | { |
| 145 | VIS32 s; |
| 146 | VIS64 d; |
| 147 | |
| 148 | s.l = src1; |
| 149 | d.ll = 0; |
| 150 | |
| 151 | d.VIS_W64(0) = do_ms16b(s.VIS_B32(0), src2); |
| 152 | d.VIS_W64(1) = do_ms16b(s.VIS_B32(1), src2); |
| 153 | d.VIS_W64(2) = do_ms16b(s.VIS_B32(2), src2); |
| 154 | d.VIS_W64(3) = do_ms16b(s.VIS_B32(3), src2); |
| 155 | |
| 156 | return d.ll; |
| 157 | } |
| 158 | |
| 159 | uint64_t helper_fmul8sux16(uint64_t src1, uint64_t src2) |
| 160 | { |
| 161 | VIS64 s, d; |
| 162 | |
| 163 | s.ll = src1; |
| 164 | d.ll = src2; |
| 165 | |
| 166 | d.VIS_W64(0) = do_ms16b(s.VIS_SB64(1), d.VIS_SW64(0)); |
| 167 | d.VIS_W64(1) = do_ms16b(s.VIS_SB64(3), d.VIS_SW64(1)); |
| 168 | d.VIS_W64(2) = do_ms16b(s.VIS_SB64(5), d.VIS_SW64(2)); |
| 169 | d.VIS_W64(3) = do_ms16b(s.VIS_SB64(7), d.VIS_SW64(3)); |
| 170 | |
| 171 | return d.ll; |
| 172 | } |
| 173 | |
| 174 | uint64_t helper_fmul8ulx16(uint64_t src1, uint64_t src2) |
| 175 | { |
| 176 | VIS64 s, d; |
| 177 | |
| 178 | s.ll = src1; |
| 179 | d.ll = src2; |
| 180 | |
| 181 | d.VIS_W64(0) = (s.VIS_B64(0) * d.VIS_SW64(0) + 0x8000) >> 16; |
| 182 | d.VIS_W64(1) = (s.VIS_B64(2) * d.VIS_SW64(1) + 0x8000) >> 16; |
| 183 | d.VIS_W64(2) = (s.VIS_B64(4) * d.VIS_SW64(2) + 0x8000) >> 16; |
| 184 | d.VIS_W64(3) = (s.VIS_B64(6) * d.VIS_SW64(3) + 0x8000) >> 16; |
| 185 | |
| 186 | return d.ll; |
| 187 | } |
| 188 | |
| 189 | uint64_t helper_fexpand(uint32_t src2) |
| 190 | { |
| 191 | VIS32 s; |
| 192 | VIS64 d; |
| 193 | |
| 194 | s.l = src2; |
| 195 | d.ll = 0; |
| 196 | d.VIS_W64(0) = s.VIS_B32(0) << 4; |
| 197 | d.VIS_W64(1) = s.VIS_B32(1) << 4; |
| 198 | d.VIS_W64(2) = s.VIS_B32(2) << 4; |
| 199 | d.VIS_W64(3) = s.VIS_B32(3) << 4; |
| 200 | |
| 201 | return d.ll; |
| 202 | } |
| 203 | |
| 204 | uint64_t helper_fcmpeq8(uint64_t src1, uint64_t src2) |
| 205 | { |
| 206 | uint64_t a = src1 ^ src2; |
| 207 | uint64_t m = 0x7f7f7f7f7f7f7f7fULL; |
| 208 | uint64_t c = ~(((a & m) + m) | a | m); |
| 209 | |
| 210 | /* a.......b.......c.......d.......e.......f.......g.......h....... */ |
| 211 | c |= c << 7; |
| 212 | /* ab......bc......cd......de......ef......fg......gh......h....... */ |
| 213 | c |= c << 14; |
| 214 | /* abcd....bcde....cdef....defg....efgh....fgh.....gh......h....... */ |
| 215 | c |= c << 28; |
| 216 | /* abcdefghbcdefgh.cdefgh..defgh...efgh....fgh.....gh......h....... */ |
| 217 | return c >> 56; |
| 218 | } |
| 219 | |
| 220 | uint64_t helper_fcmpne8(uint64_t src1, uint64_t src2) |
| 221 | { |
| 222 | return helper_fcmpeq8(src1, src2) ^ 0xff; |
| 223 | } |
| 224 | |
| 225 | uint64_t helper_fcmple8(uint64_t src1, uint64_t src2) |
| 226 | { |
| 227 | VIS64 s1, s2; |
| 228 | uint64_t r = 0; |
| 229 | |
| 230 | s1.ll = src1; |
| 231 | s2.ll = src2; |
| 232 | |
| 233 | for (int i = 0; i < 8; ++i) { |
| 234 | r |= (s1.VIS_SB64(i) <= s2.VIS_SB64(i)) << i; |
| 235 | } |
| 236 | return r; |
| 237 | } |
| 238 | |
| 239 | uint64_t helper_fcmpgt8(uint64_t src1, uint64_t src2) |
| 240 | { |
| 241 | return helper_fcmple8(src1, src2) ^ 0xff; |
| 242 | } |
| 243 | |
| 244 | uint64_t helper_fcmpule8(uint64_t src1, uint64_t src2) |
| 245 | { |
| 246 | VIS64 s1, s2; |
| 247 | uint64_t r = 0; |
| 248 | |
| 249 | s1.ll = src1; |
| 250 | s2.ll = src2; |
| 251 | |
| 252 | for (int i = 0; i < 8; ++i) { |
| 253 | r |= (s1.VIS_B64(i) <= s2.VIS_B64(i)) << i; |
| 254 | } |
| 255 | return r; |
| 256 | } |
| 257 | |
| 258 | uint64_t helper_fcmpugt8(uint64_t src1, uint64_t src2) |
| 259 | { |
| 260 | return helper_fcmpule8(src1, src2) ^ 0xff; |
| 261 | } |
| 262 | |
| 263 | uint64_t helper_fcmpeq16(uint64_t src1, uint64_t src2) |
| 264 | { |
| 265 | uint64_t a = src1 ^ src2; |
| 266 | uint64_t m = 0x7fff7fff7fff7fffULL; |
| 267 | uint64_t c = ~(((a & m) + m) | a | m); |
| 268 | |
| 269 | /* a...............b...............c...............d............... */ |
| 270 | c |= c << 15; |
| 271 | /* ab..............bc..............cd..............d............... */ |
| 272 | c |= c << 30; |
| 273 | /* abcd............bcd.............cd..............d............... */ |
| 274 | return c >> 60; |
| 275 | } |
| 276 | |
| 277 | uint64_t helper_fcmpne16(uint64_t src1, uint64_t src2) |
| 278 | { |
| 279 | return helper_fcmpeq16(src1, src2) ^ 0xf; |
| 280 | } |
| 281 | |
| 282 | uint64_t helper_fcmple16(uint64_t src1, uint64_t src2) |
| 283 | { |
| 284 | VIS64 s1, s2; |
| 285 | uint64_t r = 0; |
| 286 | |
| 287 | s1.ll = src1; |
| 288 | s2.ll = src2; |
| 289 | |
| 290 | for (int i = 0; i < 4; ++i) { |
| 291 | r |= (s1.VIS_SW64(i) <= s2.VIS_SW64(i)) << i; |
| 292 | } |
| 293 | return r; |
| 294 | } |
| 295 | |
| 296 | uint64_t helper_fcmpgt16(uint64_t src1, uint64_t src2) |
| 297 | { |
| 298 | return helper_fcmple16(src1, src2) ^ 0xf; |
| 299 | } |
| 300 | |
| 301 | uint64_t helper_fcmpule16(uint64_t src1, uint64_t src2) |
| 302 | { |
| 303 | VIS64 s1, s2; |
| 304 | uint64_t r = 0; |
| 305 | |
| 306 | s1.ll = src1; |
| 307 | s2.ll = src2; |
| 308 | |
| 309 | for (int i = 0; i < 4; ++i) { |
| 310 | r |= (s1.VIS_W64(i) <= s2.VIS_W64(i)) << i; |
| 311 | } |
| 312 | return r; |
| 313 | } |
| 314 | |
| 315 | uint64_t helper_fcmpugt16(uint64_t src1, uint64_t src2) |
| 316 | { |
| 317 | return helper_fcmpule16(src1, src2) ^ 0xf; |
| 318 | } |
| 319 | |
| 320 | uint64_t helper_fcmpeq32(uint64_t src1, uint64_t src2) |
| 321 | { |
| 322 | uint64_t a = src1 ^ src2; |
| 323 | return ((uint32_t)a == 0) | (a >> 32 ? 0 : 2); |
| 324 | } |
| 325 | |
| 326 | uint64_t helper_fcmpne32(uint64_t src1, uint64_t src2) |
| 327 | { |
| 328 | uint64_t a = src1 ^ src2; |
| 329 | return ((uint32_t)a != 0) | (a >> 32 ? 2 : 0); |
| 330 | } |
| 331 | |
| 332 | uint64_t helper_fcmple32(uint64_t src1, uint64_t src2) |
| 333 | { |
| 334 | VIS64 s1, s2; |
| 335 | uint64_t r = 0; |
| 336 | |
| 337 | s1.ll = src1; |
| 338 | s2.ll = src2; |
| 339 | |
| 340 | for (int i = 0; i < 2; ++i) { |
| 341 | r |= (s1.VIS_SL64(i) <= s2.VIS_SL64(i)) << i; |
| 342 | } |
| 343 | return r; |
| 344 | } |
| 345 | |
| 346 | uint64_t helper_fcmpgt32(uint64_t src1, uint64_t src2) |
| 347 | { |
| 348 | return helper_fcmple32(src1, src2) ^ 3; |
| 349 | } |
| 350 | |
| 351 | uint64_t helper_fcmpule32(uint64_t src1, uint64_t src2) |
| 352 | { |
| 353 | VIS64 s1, s2; |
| 354 | uint64_t r = 0; |
| 355 | |
| 356 | s1.ll = src1; |
| 357 | s2.ll = src2; |
| 358 | |
| 359 | for (int i = 0; i < 2; ++i) { |
| 360 | r |= (s1.VIS_L64(i) <= s2.VIS_L64(i)) << i; |
| 361 | } |
| 362 | return r; |
| 363 | } |
| 364 | |
| 365 | uint64_t helper_fcmpugt32(uint64_t src1, uint64_t src2) |
| 366 | { |
| 367 | return helper_fcmpule32(src1, src2) ^ 3; |
| 368 | } |
| 369 | |
| 370 | uint64_t helper_pdist(uint64_t sum, uint64_t src1, uint64_t src2) |
| 371 | { |
| 372 | int i; |
| 373 | for (i = 0; i < 8; i++) { |
| 374 | int s1, s2; |
| 375 | |
| 376 | s1 = (src1 >> (56 - (i * 8))) & 0xff; |
| 377 | s2 = (src2 >> (56 - (i * 8))) & 0xff; |
| 378 | |
| 379 | /* Absolute value of difference. */ |
| 380 | s1 -= s2; |
| 381 | if (s1 < 0) { |
| 382 | s1 = -s1; |
| 383 | } |
| 384 | |
| 385 | sum += s1; |
| 386 | } |
| 387 | |
| 388 | return sum; |
| 389 | } |
| 390 | |
| 391 | uint32_t helper_fpack16(uint64_t gsr, uint64_t rs2) |
| 392 | { |
| 393 | int scale = (gsr >> 3) & 0xf; |
| 394 | uint32_t ret = 0; |
| 395 | int byte; |
| 396 | |
| 397 | for (byte = 0; byte < 4; byte++) { |
| 398 | uint32_t val; |
| 399 | int16_t src = rs2 >> (byte * 16); |
| 400 | int32_t scaled = src << scale; |
| 401 | int32_t from_fixed = scaled >> 7; |
| 402 | |
| 403 | val = (from_fixed < 0 ? 0 : |
| 404 | from_fixed > 255 ? 255 : from_fixed); |
| 405 | |
| 406 | ret |= val << (8 * byte); |
| 407 | } |
| 408 | |
| 409 | return ret; |
| 410 | } |
| 411 | |
| 412 | uint64_t helper_fpack32(uint64_t gsr, uint64_t rs1, uint64_t rs2) |
| 413 | { |
| 414 | int scale = (gsr >> 3) & 0x1f; |
| 415 | uint64_t ret = 0; |
| 416 | int word; |
| 417 | |
| 418 | ret = (rs1 << 8) & ~(0x000000ff000000ffULL); |
| 419 | for (word = 0; word < 2; word++) { |
| 420 | uint64_t val; |
| 421 | int32_t src = rs2 >> (word * 32); |
| 422 | int64_t scaled = (int64_t)src << scale; |
| 423 | int64_t from_fixed = scaled >> 23; |
| 424 | |
| 425 | val = (from_fixed < 0 ? 0 : |
| 426 | (from_fixed > 255) ? 255 : from_fixed); |
| 427 | |
| 428 | ret |= val << (32 * word); |
| 429 | } |
| 430 | |
| 431 | return ret; |
| 432 | } |
| 433 | |
| 434 | uint32_t helper_fpackfix(uint64_t gsr, uint64_t rs2) |
| 435 | { |
| 436 | int scale = (gsr >> 3) & 0x1f; |
| 437 | uint32_t ret = 0; |
| 438 | int word; |
| 439 | |
| 440 | for (word = 0; word < 2; word++) { |
| 441 | uint32_t val; |
| 442 | int32_t src = rs2 >> (word * 32); |
| 443 | int64_t scaled = (int64_t)src << scale; |
| 444 | int64_t from_fixed = scaled >> 16; |
| 445 | |
| 446 | val = (from_fixed < -32768 ? -32768 : |
| 447 | from_fixed > 32767 ? 32767 : from_fixed); |
| 448 | |
| 449 | ret |= (val & 0xffff) << (word * 16); |
| 450 | } |
| 451 | |
| 452 | return ret; |
| 453 | } |
| 454 | |
| 455 | uint64_t helper_bshuffle(uint64_t gsr, uint64_t src1, uint64_t src2) |
| 456 | { |
| 457 | union { |
| 458 | uint64_t ll[2]; |
| 459 | uint8_t b[16]; |
| 460 | } s; |
| 461 | VIS64 r; |
| 462 | uint32_t i, mask, host; |
| 463 | |
| 464 | /* Set up S such that we can index across all of the bytes. */ |
| 465 | #if HOST_BIG_ENDIAN |
| 466 | s.ll[0] = src1; |
| 467 | s.ll[1] = src2; |
| 468 | host = 0; |
| 469 | #else |
| 470 | s.ll[1] = src1; |
| 471 | s.ll[0] = src2; |
| 472 | host = 15; |
| 473 | #endif |
| 474 | mask = gsr >> 32; |
| 475 | |
| 476 | for (i = 0; i < 8; ++i) { |
| 477 | unsigned e = (mask >> (28 - i*4)) & 0xf; |
| 478 | r.VIS_B64(i) = s.b[e ^ host]; |
| 479 | } |
| 480 | |
| 481 | return r.ll; |
| 482 | } |
| 483 | |
| 484 | uint64_t helper_cmask8(uint64_t gsr, uint64_t src) |
| 485 | { |
| 486 | uint32_t mask = 0; |
| 487 | |
| 488 | mask |= (src & 0x01 ? 0x00000007 : 0x0000000f); |
| 489 | mask |= (src & 0x02 ? 0x00000060 : 0x000000e0); |
| 490 | mask |= (src & 0x04 ? 0x00000500 : 0x00000d00); |
| 491 | mask |= (src & 0x08 ? 0x00004000 : 0x0000c000); |
| 492 | mask |= (src & 0x10 ? 0x00030000 : 0x000b0000); |
| 493 | mask |= (src & 0x20 ? 0x00200000 : 0x00a00000); |
| 494 | mask |= (src & 0x40 ? 0x01000000 : 0x09000000); |
| 495 | mask |= (src & 0x80 ? 0x00000000 : 0x80000000); |
| 496 | |
| 497 | return deposit64(gsr, 32, 32, mask); |
| 498 | } |
| 499 | |
| 500 | uint64_t helper_cmask16(uint64_t gsr, uint64_t src) |
| 501 | { |
| 502 | uint32_t mask = 0; |
| 503 | |
| 504 | mask |= (src & 0x1 ? 0x00000067 : 0x000000ef); |
| 505 | mask |= (src & 0x2 ? 0x00004500 : 0x0000cd00); |
| 506 | mask |= (src & 0x4 ? 0x00230000 : 0x00ab0000); |
| 507 | mask |= (src & 0x8 ? 0x01000000 : 0x89000000); |
| 508 | |
| 509 | return deposit64(gsr, 32, 32, mask); |
| 510 | } |
| 511 | |
| 512 | uint64_t helper_cmask32(uint64_t gsr, uint64_t src) |
| 513 | { |
| 514 | uint32_t mask = 0; |
| 515 | |
| 516 | mask |= (src & 0x1 ? 0x00004567 : 0x0000cdef); |
| 517 | mask |= (src & 0x2 ? 0x01230000 : 0x89ab0000); |
| 518 | |
| 519 | return deposit64(gsr, 32, 32, mask); |
| 520 | } |
| 521 | |
| 522 | static inline uint16_t do_fchksm16(uint16_t src1, uint16_t src2) |
| 523 | { |
| 524 | uint16_t a = src1 + src2; |
| 525 | uint16_t c = a < src1; |
| 526 | return a + c; |
| 527 | } |
| 528 | |
| 529 | uint64_t helper_fchksm16(uint64_t src1, uint64_t src2) |
| 530 | { |
| 531 | VIS64 r, s1, s2; |
| 532 | |
| 533 | s1.ll = src1; |
| 534 | s2.ll = src2; |
| 535 | r.ll = 0; |
| 536 | |
| 537 | r.VIS_W64(0) = do_fchksm16(s1.VIS_W64(0), s2.VIS_W64(0)); |
| 538 | r.VIS_W64(1) = do_fchksm16(s1.VIS_W64(1), s2.VIS_W64(1)); |
| 539 | r.VIS_W64(2) = do_fchksm16(s1.VIS_W64(2), s2.VIS_W64(2)); |
| 540 | r.VIS_W64(3) = do_fchksm16(s1.VIS_W64(3), s2.VIS_W64(3)); |
| 541 | |
| 542 | return r.ll; |
| 543 | } |
| 544 | |
| 545 | static inline int16_t do_fmean16(int16_t src1, int16_t src2) |
| 546 | { |
| 547 | return (src1 + src2 + 1) / 2; |
| 548 | } |
| 549 | |
| 550 | uint64_t helper_fmean16(uint64_t src1, uint64_t src2) |
| 551 | { |
| 552 | VIS64 r, s1, s2; |
| 553 | |
| 554 | s1.ll = src1; |
| 555 | s2.ll = src2; |
| 556 | r.ll = 0; |
| 557 | |
| 558 | r.VIS_SW64(0) = do_fmean16(s1.VIS_SW64(0), s2.VIS_SW64(0)); |
| 559 | r.VIS_SW64(1) = do_fmean16(s1.VIS_SW64(1), s2.VIS_SW64(1)); |
| 560 | r.VIS_SW64(2) = do_fmean16(s1.VIS_SW64(2), s2.VIS_SW64(2)); |
| 561 | r.VIS_SW64(3) = do_fmean16(s1.VIS_SW64(3), s2.VIS_SW64(3)); |
| 562 | |
| 563 | return r.ll; |
| 564 | } |
| 565 | |
| 566 | uint64_t helper_fslas16(uint64_t src1, uint64_t src2) |
| 567 | { |
| 568 | VIS64 r, s1, s2; |
| 569 | |
| 570 | s1.ll = src1; |
| 571 | s2.ll = src2; |
| 572 | r.ll = 0; |
| 573 | |
| 574 | for (int i = 0; i < 4; ++i) { |
| 575 | int t = s1.VIS_SW64(i) << (s2.VIS_W64(i) % 16); |
| 576 | t = MIN(t, INT16_MAX); |
| 577 | t = MAX(t, INT16_MIN); |
| 578 | r.VIS_SW64(i) = t; |
| 579 | } |
| 580 | |
| 581 | return r.ll; |
| 582 | } |
| 583 | |
| 584 | uint64_t helper_fslas32(uint64_t src1, uint64_t src2) |
| 585 | { |
| 586 | VIS64 r, s1, s2; |
| 587 | |
| 588 | s1.ll = src1; |
| 589 | s2.ll = src2; |
| 590 | r.ll = 0; |
| 591 | |
| 592 | for (int i = 0; i < 2; ++i) { |
| 593 | int64_t t = (int64_t)(int32_t)s1.VIS_L64(i) << (s2.VIS_L64(i) % 32); |
| 594 | t = MIN(t, INT32_MAX); |
| 595 | t = MAX(t, INT32_MIN); |
| 596 | r.VIS_L64(i) = t; |
| 597 | } |
| 598 | |
| 599 | return r.ll; |
| 600 | } |
| 601 | |
| 602 | uint64_t helper_xmulx(uint64_t src1, uint64_t src2) |
| 603 | { |
| 604 | return int128_getlo(clmul_64(src1, src2)); |
| 605 | } |
| 606 | |
| 607 | uint64_t helper_xmulxhi(uint64_t src1, uint64_t src2) |
| 608 | { |
| 609 | return int128_gethi(clmul_64(src1, src2)); |
| 610 | } |