1 /* visemul.c: Emulation of VIS instructions. 2 * 3 * Copyright (C) 2006 David S. Miller (davem@davemloft.net) 4 */ 5 #include <linux/kernel.h> 6 #include <linux/errno.h> 7 #include <linux/thread_info.h> 8 9 #include <asm/ptrace.h> 10 #include <asm/pstate.h> 11 #include <asm/system.h> 12 #include <asm/fpumacro.h> 13 #include <asm/uaccess.h> 14 15 /* OPF field of various VIS instructions. */ 16 17 /* 000111011 - four 16-bit packs */ 18 #define FPACK16_OPF 0x03b 19 20 /* 000111010 - two 32-bit packs */ 21 #define FPACK32_OPF 0x03a 22 23 /* 000111101 - four 16-bit packs */ 24 #define FPACKFIX_OPF 0x03d 25 26 /* 001001101 - four 16-bit expands */ 27 #define FEXPAND_OPF 0x04d 28 29 /* 001001011 - two 32-bit merges */ 30 #define FPMERGE_OPF 0x04b 31 32 /* 000110001 - 8-by-16-bit partitoned product */ 33 #define FMUL8x16_OPF 0x031 34 35 /* 000110011 - 8-by-16-bit upper alpha partitioned product */ 36 #define FMUL8x16AU_OPF 0x033 37 38 /* 000110101 - 8-by-16-bit lower alpha partitioned product */ 39 #define FMUL8x16AL_OPF 0x035 40 41 /* 000110110 - upper 8-by-16-bit partitioned product */ 42 #define FMUL8SUx16_OPF 0x036 43 44 /* 000110111 - lower 8-by-16-bit partitioned product */ 45 #define FMUL8ULx16_OPF 0x037 46 47 /* 000111000 - upper 8-by-16-bit partitioned product */ 48 #define FMULD8SUx16_OPF 0x038 49 50 /* 000111001 - lower unsigned 8-by-16-bit partitioned product */ 51 #define FMULD8ULx16_OPF 0x039 52 53 /* 000101000 - four 16-bit compare; set rd if src1 > src2 */ 54 #define FCMPGT16_OPF 0x028 55 56 /* 000101100 - two 32-bit compare; set rd if src1 > src2 */ 57 #define FCMPGT32_OPF 0x02c 58 59 /* 000100000 - four 16-bit compare; set rd if src1 <= src2 */ 60 #define FCMPLE16_OPF 0x020 61 62 /* 000100100 - two 32-bit compare; set rd if src1 <= src2 */ 63 #define FCMPLE32_OPF 0x024 64 65 /* 000100010 - four 16-bit compare; set rd if src1 != src2 */ 66 #define FCMPNE16_OPF 0x022 67 68 /* 000100110 - two 32-bit compare; set rd if src1 != src2 */ 69 #define FCMPNE32_OPF 0x026 70 71 /* 000101010 - four 16-bit compare; set rd if src1 == src2 */ 72 #define FCMPEQ16_OPF 0x02a 73 74 /* 000101110 - two 32-bit compare; set rd if src1 == src2 */ 75 #define FCMPEQ32_OPF 0x02e 76 77 /* 000000000 - Eight 8-bit edge boundary processing */ 78 #define EDGE8_OPF 0x000 79 80 /* 000000001 - Eight 8-bit edge boundary processing, no CC */ 81 #define EDGE8N_OPF 0x001 82 83 /* 000000010 - Eight 8-bit edge boundary processing, little-endian */ 84 #define EDGE8L_OPF 0x002 85 86 /* 000000011 - Eight 8-bit edge boundary processing, little-endian, no CC */ 87 #define EDGE8LN_OPF 0x003 88 89 /* 000000100 - Four 16-bit edge boundary processing */ 90 #define EDGE16_OPF 0x004 91 92 /* 000000101 - Four 16-bit edge boundary processing, no CC */ 93 #define EDGE16N_OPF 0x005 94 95 /* 000000110 - Four 16-bit edge boundary processing, little-endian */ 96 #define EDGE16L_OPF 0x006 97 98 /* 000000111 - Four 16-bit edge boundary processing, little-endian, no CC */ 99 #define EDGE16LN_OPF 0x007 100 101 /* 000001000 - Two 32-bit edge boundary processing */ 102 #define EDGE32_OPF 0x008 103 104 /* 000001001 - Two 32-bit edge boundary processing, no CC */ 105 #define EDGE32N_OPF 0x009 106 107 /* 000001010 - Two 32-bit edge boundary processing, little-endian */ 108 #define EDGE32L_OPF 0x00a 109 110 /* 000001011 - Two 32-bit edge boundary processing, little-endian, no CC */ 111 #define EDGE32LN_OPF 0x00b 112 113 /* 000111110 - distance between 8 8-bit components */ 114 #define PDIST_OPF 0x03e 115 116 /* 000010000 - convert 8-bit 3-D address to blocked byte address */ 117 #define ARRAY8_OPF 0x010 118 119 /* 000010010 - convert 16-bit 3-D address to blocked byte address */ 120 #define ARRAY16_OPF 0x012 121 122 /* 000010100 - convert 32-bit 3-D address to blocked byte address */ 123 #define ARRAY32_OPF 0x014 124 125 /* 000011001 - Set the GSR.MASK field in preparation for a BSHUFFLE */ 126 #define BMASK_OPF 0x019 127 128 /* 001001100 - Permute bytes as specified by GSR.MASK */ 129 #define BSHUFFLE_OPF 0x04c 130 131 #define VIS_OPF_SHIFT 5 132 #define VIS_OPF_MASK (0x1ff << VIS_OPF_SHIFT) 133 134 #define RS1(INSN) (((INSN) >> 14) & 0x1f) 135 #define RS2(INSN) (((INSN) >> 0) & 0x1f) 136 #define RD(INSN) (((INSN) >> 25) & 0x1f) 137 138 static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2, 139 unsigned int rd, int from_kernel) 140 { 141 if (rs2 >= 16 || rs1 >= 16 || rd >= 16) { 142 if (from_kernel != 0) 143 __asm__ __volatile__("flushw"); 144 else 145 flushw_user(); 146 } 147 } 148 149 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs) 150 { 151 unsigned long value; 152 153 if (reg < 16) 154 return (!reg ? 0 : regs->u_regs[reg]); 155 if (regs->tstate & TSTATE_PRIV) { 156 struct reg_window *win; 157 win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS); 158 value = win->locals[reg - 16]; 159 } else if (test_thread_flag(TIF_32BIT)) { 160 struct reg_window32 __user *win32; 161 win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP])); 162 get_user(value, &win32->locals[reg - 16]); 163 } else { 164 struct reg_window __user *win; 165 win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS); 166 get_user(value, &win->locals[reg - 16]); 167 } 168 return value; 169 } 170 171 static inline unsigned long __user *__fetch_reg_addr_user(unsigned int reg, 172 struct pt_regs *regs) 173 { 174 BUG_ON(reg < 16); 175 BUG_ON(regs->tstate & TSTATE_PRIV); 176 177 if (test_thread_flag(TIF_32BIT)) { 178 struct reg_window32 __user *win32; 179 win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP])); 180 return (unsigned long __user *)&win32->locals[reg - 16]; 181 } else { 182 struct reg_window __user *win; 183 win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS); 184 return &win->locals[reg - 16]; 185 } 186 } 187 188 static inline unsigned long *__fetch_reg_addr_kern(unsigned int reg, 189 struct pt_regs *regs) 190 { 191 BUG_ON(reg >= 16); 192 BUG_ON(regs->tstate & TSTATE_PRIV); 193 194 return ®s->u_regs[reg]; 195 } 196 197 static void store_reg(struct pt_regs *regs, unsigned long val, unsigned long rd) 198 { 199 if (rd < 16) { 200 unsigned long *rd_kern = __fetch_reg_addr_kern(rd, regs); 201 202 *rd_kern = val; 203 } else { 204 unsigned long __user *rd_user = __fetch_reg_addr_user(rd, regs); 205 206 if (test_thread_flag(TIF_32BIT)) 207 __put_user((u32)val, (u32 __user *)rd_user); 208 else 209 __put_user(val, rd_user); 210 } 211 } 212 213 static inline unsigned long fpd_regval(struct fpustate *f, 214 unsigned int insn_regnum) 215 { 216 insn_regnum = (((insn_regnum & 1) << 5) | 217 (insn_regnum & 0x1e)); 218 219 return *(unsigned long *) &f->regs[insn_regnum]; 220 } 221 222 static inline unsigned long *fpd_regaddr(struct fpustate *f, 223 unsigned int insn_regnum) 224 { 225 insn_regnum = (((insn_regnum & 1) << 5) | 226 (insn_regnum & 0x1e)); 227 228 return (unsigned long *) &f->regs[insn_regnum]; 229 } 230 231 static inline unsigned int fps_regval(struct fpustate *f, 232 unsigned int insn_regnum) 233 { 234 return f->regs[insn_regnum]; 235 } 236 237 static inline unsigned int *fps_regaddr(struct fpustate *f, 238 unsigned int insn_regnum) 239 { 240 return &f->regs[insn_regnum]; 241 } 242 243 struct edge_tab { 244 u16 left, right; 245 }; 246 static struct edge_tab edge8_tab[8] = { 247 { 0xff, 0x80 }, 248 { 0x7f, 0xc0 }, 249 { 0x3f, 0xe0 }, 250 { 0x1f, 0xf0 }, 251 { 0x0f, 0xf8 }, 252 { 0x07, 0xfc }, 253 { 0x03, 0xfe }, 254 { 0x01, 0xff }, 255 }; 256 static struct edge_tab edge8_tab_l[8] = { 257 { 0xff, 0x01 }, 258 { 0xfe, 0x03 }, 259 { 0xfc, 0x07 }, 260 { 0xf8, 0x0f }, 261 { 0xf0, 0x1f }, 262 { 0xe0, 0x3f }, 263 { 0xc0, 0x7f }, 264 { 0x80, 0xff }, 265 }; 266 static struct edge_tab edge16_tab[4] = { 267 { 0xf, 0x8 }, 268 { 0x7, 0xc }, 269 { 0x3, 0xe }, 270 { 0x1, 0xf }, 271 }; 272 static struct edge_tab edge16_tab_l[4] = { 273 { 0xf, 0x1 }, 274 { 0xe, 0x3 }, 275 { 0xc, 0x7 }, 276 { 0x8, 0xf }, 277 }; 278 static struct edge_tab edge32_tab[2] = { 279 { 0x3, 0x2 }, 280 { 0x1, 0x3 }, 281 }; 282 static struct edge_tab edge32_tab_l[2] = { 283 { 0x3, 0x1 }, 284 { 0x2, 0x3 }, 285 }; 286 287 static void edge(struct pt_regs *regs, unsigned int insn, unsigned int opf) 288 { 289 unsigned long orig_rs1, rs1, orig_rs2, rs2, rd_val; 290 u16 left, right; 291 292 maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0); 293 orig_rs1 = rs1 = fetch_reg(RS1(insn), regs); 294 orig_rs2 = rs2 = fetch_reg(RS2(insn), regs); 295 296 if (test_thread_flag(TIF_32BIT)) { 297 rs1 = rs1 & 0xffffffff; 298 rs2 = rs2 & 0xffffffff; 299 } 300 switch (opf) { 301 default: 302 case EDGE8_OPF: 303 case EDGE8N_OPF: 304 left = edge8_tab[rs1 & 0x7].left; 305 right = edge8_tab[rs2 & 0x7].right; 306 break; 307 case EDGE8L_OPF: 308 case EDGE8LN_OPF: 309 left = edge8_tab_l[rs1 & 0x7].left; 310 right = edge8_tab_l[rs2 & 0x7].right; 311 break; 312 313 case EDGE16_OPF: 314 case EDGE16N_OPF: 315 left = edge16_tab[(rs1 >> 1) & 0x3].left; 316 right = edge16_tab[(rs2 >> 1) & 0x3].right; 317 break; 318 319 case EDGE16L_OPF: 320 case EDGE16LN_OPF: 321 left = edge16_tab_l[(rs1 >> 1) & 0x3].left; 322 right = edge16_tab_l[(rs2 >> 1) & 0x3].right; 323 break; 324 325 case EDGE32_OPF: 326 case EDGE32N_OPF: 327 left = edge32_tab[(rs1 >> 2) & 0x1].left; 328 right = edge32_tab[(rs2 >> 2) & 0x1].right; 329 break; 330 331 case EDGE32L_OPF: 332 case EDGE32LN_OPF: 333 left = edge32_tab_l[(rs1 >> 2) & 0x1].left; 334 right = edge32_tab_l[(rs2 >> 2) & 0x1].right; 335 break; 336 }; 337 338 if ((rs1 & ~0x7UL) == (rs2 & ~0x7UL)) 339 rd_val = right & left; 340 else 341 rd_val = left; 342 343 store_reg(regs, rd_val, RD(insn)); 344 345 switch (opf) { 346 case EDGE8_OPF: 347 case EDGE8L_OPF: 348 case EDGE16_OPF: 349 case EDGE16L_OPF: 350 case EDGE32_OPF: 351 case EDGE32L_OPF: { 352 unsigned long ccr, tstate; 353 354 __asm__ __volatile__("subcc %1, %2, %%g0\n\t" 355 "rd %%ccr, %0" 356 : "=r" (ccr) 357 : "r" (orig_rs1), "r" (orig_rs2) 358 : "cc"); 359 tstate = regs->tstate & ~(TSTATE_XCC | TSTATE_ICC); 360 regs->tstate = tstate | (ccr << 32UL); 361 } 362 }; 363 } 364 365 static void array(struct pt_regs *regs, unsigned int insn, unsigned int opf) 366 { 367 unsigned long rs1, rs2, rd_val; 368 unsigned int bits, bits_mask; 369 370 maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0); 371 rs1 = fetch_reg(RS1(insn), regs); 372 rs2 = fetch_reg(RS2(insn), regs); 373 374 bits = (rs2 > 5 ? 5 : rs2); 375 bits_mask = (1UL << bits) - 1UL; 376 377 rd_val = ((((rs1 >> 11) & 0x3) << 0) | 378 (((rs1 >> 33) & 0x3) << 2) | 379 (((rs1 >> 55) & 0x1) << 4) | 380 (((rs1 >> 13) & 0xf) << 5) | 381 (((rs1 >> 35) & 0xf) << 9) | 382 (((rs1 >> 56) & 0xf) << 13) | 383 (((rs1 >> 17) & bits_mask) << 17) | 384 (((rs1 >> 39) & bits_mask) << (17 + bits)) | 385 (((rs1 >> 60) & 0xf) << (17 + (2*bits)))); 386 387 switch (opf) { 388 case ARRAY16_OPF: 389 rd_val <<= 1; 390 break; 391 392 case ARRAY32_OPF: 393 rd_val <<= 2; 394 }; 395 396 store_reg(regs, rd_val, RD(insn)); 397 } 398 399 static void bmask(struct pt_regs *regs, unsigned int insn) 400 { 401 unsigned long rs1, rs2, rd_val, gsr; 402 403 maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0); 404 rs1 = fetch_reg(RS1(insn), regs); 405 rs2 = fetch_reg(RS2(insn), regs); 406 rd_val = rs1 + rs2; 407 408 store_reg(regs, rd_val, RD(insn)); 409 410 gsr = current_thread_info()->gsr[0] & 0xffffffff; 411 gsr |= rd_val << 32UL; 412 current_thread_info()->gsr[0] = gsr; 413 } 414 415 static void bshuffle(struct pt_regs *regs, unsigned int insn) 416 { 417 struct fpustate *f = FPUSTATE; 418 unsigned long rs1, rs2, rd_val; 419 unsigned long bmask, i; 420 421 bmask = current_thread_info()->gsr[0] >> 32UL; 422 423 rs1 = fpd_regval(f, RS1(insn)); 424 rs2 = fpd_regval(f, RS2(insn)); 425 426 rd_val = 0UL; 427 for (i = 0; i < 8; i++) { 428 unsigned long which = (bmask >> (i * 4)) & 0xf; 429 unsigned long byte; 430 431 if (which < 8) 432 byte = (rs1 >> (which * 8)) & 0xff; 433 else 434 byte = (rs2 >> ((which-8)*8)) & 0xff; 435 rd_val |= (byte << (i * 8)); 436 } 437 438 *fpd_regaddr(f, RD(insn)) = rd_val; 439 } 440 441 static void pdist(struct pt_regs *regs, unsigned int insn) 442 { 443 struct fpustate *f = FPUSTATE; 444 unsigned long rs1, rs2, *rd, rd_val; 445 unsigned long i; 446 447 rs1 = fpd_regval(f, RS1(insn)); 448 rs2 = fpd_regval(f, RS2(insn)); 449 rd = fpd_regaddr(f, RD(insn)); 450 451 rd_val = *rd; 452 453 for (i = 0; i < 8; i++) { 454 s16 s1, s2; 455 456 s1 = (rs1 >> (56 - (i * 8))) & 0xff; 457 s2 = (rs2 >> (56 - (i * 8))) & 0xff; 458 459 /* Absolute value of difference. */ 460 s1 -= s2; 461 if (s1 < 0) 462 s1 = ~s1 + 1; 463 464 rd_val += s1; 465 } 466 467 *rd = rd_val; 468 } 469 470 static void pformat(struct pt_regs *regs, unsigned int insn, unsigned int opf) 471 { 472 struct fpustate *f = FPUSTATE; 473 unsigned long rs1, rs2, gsr, scale, rd_val; 474 475 gsr = current_thread_info()->gsr[0]; 476 scale = (gsr >> 3) & (opf == FPACK16_OPF ? 0xf : 0x1f); 477 switch (opf) { 478 case FPACK16_OPF: { 479 unsigned long byte; 480 481 rs2 = fpd_regval(f, RS2(insn)); 482 rd_val = 0; 483 for (byte = 0; byte < 4; byte++) { 484 unsigned int val; 485 s16 src = (rs2 >> (byte * 16UL)) & 0xffffUL; 486 int scaled = src << scale; 487 int from_fixed = scaled >> 7; 488 489 val = ((from_fixed < 0) ? 490 0 : 491 (from_fixed > 255) ? 492 255 : from_fixed); 493 494 rd_val |= (val << (8 * byte)); 495 } 496 *fps_regaddr(f, RD(insn)) = rd_val; 497 break; 498 } 499 500 case FPACK32_OPF: { 501 unsigned long word; 502 503 rs1 = fpd_regval(f, RS1(insn)); 504 rs2 = fpd_regval(f, RS2(insn)); 505 rd_val = (rs1 << 8) & ~(0x000000ff000000ffUL); 506 for (word = 0; word < 2; word++) { 507 unsigned long val; 508 s32 src = (rs2 >> (word * 32UL)); 509 s64 scaled = src << scale; 510 s64 from_fixed = scaled >> 23; 511 512 val = ((from_fixed < 0) ? 513 0 : 514 (from_fixed > 255) ? 515 255 : from_fixed); 516 517 rd_val |= (val << (32 * word)); 518 } 519 *fpd_regaddr(f, RD(insn)) = rd_val; 520 break; 521 } 522 523 case FPACKFIX_OPF: { 524 unsigned long word; 525 526 rs2 = fpd_regval(f, RS2(insn)); 527 528 rd_val = 0; 529 for (word = 0; word < 2; word++) { 530 long val; 531 s32 src = (rs2 >> (word * 32UL)); 532 s64 scaled = src << scale; 533 s64 from_fixed = scaled >> 16; 534 535 val = ((from_fixed < -32768) ? 536 -32768 : 537 (from_fixed > 32767) ? 538 32767 : from_fixed); 539 540 rd_val |= ((val & 0xffff) << (word * 16)); 541 } 542 *fps_regaddr(f, RD(insn)) = rd_val; 543 break; 544 } 545 546 case FEXPAND_OPF: { 547 unsigned long byte; 548 549 rs2 = fps_regval(f, RS2(insn)); 550 551 rd_val = 0; 552 for (byte = 0; byte < 4; byte++) { 553 unsigned long val; 554 u8 src = (rs2 >> (byte * 8)) & 0xff; 555 556 val = src << 4; 557 558 rd_val |= (val << (byte * 16)); 559 } 560 *fpd_regaddr(f, RD(insn)) = rd_val; 561 break; 562 } 563 564 case FPMERGE_OPF: { 565 rs1 = fps_regval(f, RS1(insn)); 566 rs2 = fps_regval(f, RS2(insn)); 567 568 rd_val = (((rs2 & 0x000000ff) << 0) | 569 ((rs1 & 0x000000ff) << 8) | 570 ((rs2 & 0x0000ff00) << 8) | 571 ((rs1 & 0x0000ff00) << 16) | 572 ((rs2 & 0x00ff0000) << 16) | 573 ((rs1 & 0x00ff0000) << 24) | 574 ((rs2 & 0xff000000) << 24) | 575 ((rs1 & 0xff000000) << 32)); 576 *fpd_regaddr(f, RD(insn)) = rd_val; 577 break; 578 } 579 }; 580 } 581 582 static void pmul(struct pt_regs *regs, unsigned int insn, unsigned int opf) 583 { 584 struct fpustate *f = FPUSTATE; 585 unsigned long rs1, rs2, rd_val; 586 587 switch (opf) { 588 case FMUL8x16_OPF: { 589 unsigned long byte; 590 591 rs1 = fps_regval(f, RS1(insn)); 592 rs2 = fpd_regval(f, RS2(insn)); 593 594 rd_val = 0; 595 for (byte = 0; byte < 4; byte++) { 596 u16 src1 = (rs1 >> (byte * 8)) & 0x00ff; 597 s16 src2 = (rs2 >> (byte * 16)) & 0xffff; 598 u32 prod = src1 * src2; 599 u16 scaled = ((prod & 0x00ffff00) >> 8); 600 601 /* Round up. */ 602 if (prod & 0x80) 603 scaled++; 604 rd_val |= ((scaled & 0xffffUL) << (byte * 16UL)); 605 } 606 607 *fpd_regaddr(f, RD(insn)) = rd_val; 608 break; 609 } 610 611 case FMUL8x16AU_OPF: 612 case FMUL8x16AL_OPF: { 613 unsigned long byte; 614 s16 src2; 615 616 rs1 = fps_regval(f, RS1(insn)); 617 rs2 = fps_regval(f, RS2(insn)); 618 619 rd_val = 0; 620 src2 = (rs2 >> (opf == FMUL8x16AU_OPF) ? 16 : 0); 621 for (byte = 0; byte < 4; byte++) { 622 u16 src1 = (rs1 >> (byte * 8)) & 0x00ff; 623 u32 prod = src1 * src2; 624 u16 scaled = ((prod & 0x00ffff00) >> 8); 625 626 /* Round up. */ 627 if (prod & 0x80) 628 scaled++; 629 rd_val |= ((scaled & 0xffffUL) << (byte * 16UL)); 630 } 631 632 *fpd_regaddr(f, RD(insn)) = rd_val; 633 break; 634 } 635 636 case FMUL8SUx16_OPF: 637 case FMUL8ULx16_OPF: { 638 unsigned long byte, ushift; 639 640 rs1 = fpd_regval(f, RS1(insn)); 641 rs2 = fpd_regval(f, RS2(insn)); 642 643 rd_val = 0; 644 ushift = (opf == FMUL8SUx16_OPF) ? 8 : 0; 645 for (byte = 0; byte < 4; byte++) { 646 u16 src1; 647 s16 src2; 648 u32 prod; 649 u16 scaled; 650 651 src1 = ((rs1 >> ((16 * byte) + ushift)) & 0x00ff); 652 src2 = ((rs2 >> (16 * byte)) & 0xffff); 653 prod = src1 * src2; 654 scaled = ((prod & 0x00ffff00) >> 8); 655 656 /* Round up. */ 657 if (prod & 0x80) 658 scaled++; 659 rd_val |= ((scaled & 0xffffUL) << (byte * 16UL)); 660 } 661 662 *fpd_regaddr(f, RD(insn)) = rd_val; 663 break; 664 } 665 666 case FMULD8SUx16_OPF: 667 case FMULD8ULx16_OPF: { 668 unsigned long byte, ushift; 669 670 rs1 = fps_regval(f, RS1(insn)); 671 rs2 = fps_regval(f, RS2(insn)); 672 673 rd_val = 0; 674 ushift = (opf == FMULD8SUx16_OPF) ? 8 : 0; 675 for (byte = 0; byte < 2; byte++) { 676 u16 src1; 677 s16 src2; 678 u32 prod; 679 u16 scaled; 680 681 src1 = ((rs1 >> ((16 * byte) + ushift)) & 0x00ff); 682 src2 = ((rs2 >> (16 * byte)) & 0xffff); 683 prod = src1 * src2; 684 scaled = ((prod & 0x00ffff00) >> 8); 685 686 /* Round up. */ 687 if (prod & 0x80) 688 scaled++; 689 rd_val |= ((scaled & 0xffffUL) << 690 ((byte * 32UL) + 7UL)); 691 } 692 *fpd_regaddr(f, RD(insn)) = rd_val; 693 break; 694 } 695 }; 696 } 697 698 static void pcmp(struct pt_regs *regs, unsigned int insn, unsigned int opf) 699 { 700 struct fpustate *f = FPUSTATE; 701 unsigned long rs1, rs2, rd_val, i; 702 703 rs1 = fpd_regval(f, RS1(insn)); 704 rs2 = fpd_regval(f, RS2(insn)); 705 706 rd_val = 0; 707 708 switch (opf) { 709 case FCMPGT16_OPF: 710 for (i = 0; i < 4; i++) { 711 s16 a = (rs1 >> (i * 16)) & 0xffff; 712 s16 b = (rs2 >> (i * 16)) & 0xffff; 713 714 if (a > b) 715 rd_val |= 1 << i; 716 } 717 break; 718 719 case FCMPGT32_OPF: 720 for (i = 0; i < 2; i++) { 721 s32 a = (rs1 >> (i * 32)) & 0xffff; 722 s32 b = (rs2 >> (i * 32)) & 0xffff; 723 724 if (a > b) 725 rd_val |= 1 << i; 726 } 727 break; 728 729 case FCMPLE16_OPF: 730 for (i = 0; i < 4; i++) { 731 s16 a = (rs1 >> (i * 16)) & 0xffff; 732 s16 b = (rs2 >> (i * 16)) & 0xffff; 733 734 if (a <= b) 735 rd_val |= 1 << i; 736 } 737 break; 738 739 case FCMPLE32_OPF: 740 for (i = 0; i < 2; i++) { 741 s32 a = (rs1 >> (i * 32)) & 0xffff; 742 s32 b = (rs2 >> (i * 32)) & 0xffff; 743 744 if (a <= b) 745 rd_val |= 1 << i; 746 } 747 break; 748 749 case FCMPNE16_OPF: 750 for (i = 0; i < 4; i++) { 751 s16 a = (rs1 >> (i * 16)) & 0xffff; 752 s16 b = (rs2 >> (i * 16)) & 0xffff; 753 754 if (a != b) 755 rd_val |= 1 << i; 756 } 757 break; 758 759 case FCMPNE32_OPF: 760 for (i = 0; i < 2; i++) { 761 s32 a = (rs1 >> (i * 32)) & 0xffff; 762 s32 b = (rs2 >> (i * 32)) & 0xffff; 763 764 if (a != b) 765 rd_val |= 1 << i; 766 } 767 break; 768 769 case FCMPEQ16_OPF: 770 for (i = 0; i < 4; i++) { 771 s16 a = (rs1 >> (i * 16)) & 0xffff; 772 s16 b = (rs2 >> (i * 16)) & 0xffff; 773 774 if (a == b) 775 rd_val |= 1 << i; 776 } 777 break; 778 779 case FCMPEQ32_OPF: 780 for (i = 0; i < 2; i++) { 781 s32 a = (rs1 >> (i * 32)) & 0xffff; 782 s32 b = (rs2 >> (i * 32)) & 0xffff; 783 784 if (a == b) 785 rd_val |= 1 << i; 786 } 787 break; 788 }; 789 790 maybe_flush_windows(0, 0, RD(insn), 0); 791 store_reg(regs, rd_val, RD(insn)); 792 } 793 794 /* Emulate the VIS instructions which are not implemented in 795 * hardware on Niagara. 796 */ 797 int vis_emul(struct pt_regs *regs, unsigned int insn) 798 { 799 unsigned long pc = regs->tpc; 800 unsigned int opf; 801 802 BUG_ON(regs->tstate & TSTATE_PRIV); 803 804 if (test_thread_flag(TIF_32BIT)) 805 pc = (u32)pc; 806 807 if (get_user(insn, (u32 __user *) pc)) 808 return -EFAULT; 809 810 save_and_clear_fpu(); 811 812 opf = (insn & VIS_OPF_MASK) >> VIS_OPF_SHIFT; 813 switch (opf) { 814 default: 815 return -EINVAL; 816 817 /* Pixel Formatting Instructions. */ 818 case FPACK16_OPF: 819 case FPACK32_OPF: 820 case FPACKFIX_OPF: 821 case FEXPAND_OPF: 822 case FPMERGE_OPF: 823 pformat(regs, insn, opf); 824 break; 825 826 /* Partitioned Multiply Instructions */ 827 case FMUL8x16_OPF: 828 case FMUL8x16AU_OPF: 829 case FMUL8x16AL_OPF: 830 case FMUL8SUx16_OPF: 831 case FMUL8ULx16_OPF: 832 case FMULD8SUx16_OPF: 833 case FMULD8ULx16_OPF: 834 pmul(regs, insn, opf); 835 break; 836 837 /* Pixel Compare Instructions */ 838 case FCMPGT16_OPF: 839 case FCMPGT32_OPF: 840 case FCMPLE16_OPF: 841 case FCMPLE32_OPF: 842 case FCMPNE16_OPF: 843 case FCMPNE32_OPF: 844 case FCMPEQ16_OPF: 845 case FCMPEQ32_OPF: 846 pcmp(regs, insn, opf); 847 break; 848 849 /* Edge Handling Instructions */ 850 case EDGE8_OPF: 851 case EDGE8N_OPF: 852 case EDGE8L_OPF: 853 case EDGE8LN_OPF: 854 case EDGE16_OPF: 855 case EDGE16N_OPF: 856 case EDGE16L_OPF: 857 case EDGE16LN_OPF: 858 case EDGE32_OPF: 859 case EDGE32N_OPF: 860 case EDGE32L_OPF: 861 case EDGE32LN_OPF: 862 edge(regs, insn, opf); 863 break; 864 865 /* Pixel Component Distance */ 866 case PDIST_OPF: 867 pdist(regs, insn); 868 break; 869 870 /* Three-Dimensional Array Addressing Instructions */ 871 case ARRAY8_OPF: 872 case ARRAY16_OPF: 873 case ARRAY32_OPF: 874 array(regs, insn, opf); 875 break; 876 877 /* Byte Mask and Shuffle Instructions */ 878 case BMASK_OPF: 879 bmask(regs, insn); 880 break; 881 882 case BSHUFFLE_OPF: 883 bshuffle(regs, insn); 884 break; 885 }; 886 887 regs->tpc = regs->tnpc; 888 regs->tnpc += 4; 889 return 0; 890 } 891