decoder.isa revision 2312
1// -*- mode:c++ -*- 2 3// Copyright (c) 2003-2006 The Regents of The University of Michigan 4// All rights reserved. 5// 6// Redistribution and use in source and binary forms, with or without 7// modification, are permitted provided that the following conditions are 8// met: redistributions of source code must retain the above copyright 9// notice, this list of conditions and the following disclaimer; 10// redistributions in binary form must reproduce the above copyright 11// notice, this list of conditions and the following disclaimer in the 12// documentation and/or other materials provided with the distribution; 13// neither the name of the copyright holders nor the names of its 14// contributors may be used to endorse or promote products derived from 15// this software without specific prior written permission. 16// 17// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 18// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 19// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 20// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 21// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 22// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 23// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 27// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 29decode OPCODE default Unknown::unknown() { 30 31 format LoadAddress { 32 0x08: lda({{ Ra = Rb + disp; }}); 33 0x09: ldah({{ Ra = Rb + (disp << 16); }}); 34 } 35 36 format LoadOrNop { 37 0x0a: ldbu({{ Ra.uq = Mem.ub; }}); 38 0x0c: ldwu({{ Ra.uq = Mem.uw; }}); 39 0x0b: ldq_u({{ Ra = Mem.uq; }}, ea_code = {{ EA = (Rb + disp) & ~7; }}); 40 0x23: ldt({{ Fa = Mem.df; }}); 41 0x2a: ldl_l({{ Ra.sl = Mem.sl; }}, mem_flags = LOCKED); 42 0x2b: ldq_l({{ Ra.uq = Mem.uq; }}, mem_flags = LOCKED); 43 0x20: MiscPrefetch::copy_load({{ EA = Ra; }}, 44 {{ fault = xc->copySrcTranslate(EA); }}, 45 inst_flags = [IsMemRef, IsLoad, IsCopy]); 46 } 47 48 format LoadOrPrefetch { 49 0x28: ldl({{ Ra.sl = Mem.sl; }}); 50 0x29: ldq({{ Ra.uq = Mem.uq; }}, pf_flags = EVICT_NEXT); 51 // IsFloating flag on lds gets the prefetch to disassemble 52 // using f31 instead of r31... funcitonally it's unnecessary 53 0x22: lds({{ Fa.uq = s_to_t(Mem.ul); }}, 54 pf_flags = PF_EXCLUSIVE, inst_flags = IsFloating); 55 } 56 57 format Store { 58 0x0e: stb({{ Mem.ub = Ra<7:0>; }}); 59 0x0d: stw({{ Mem.uw = Ra<15:0>; }}); 60 0x2c: stl({{ Mem.ul = Ra<31:0>; }}); 61 0x2d: stq({{ Mem.uq = Ra.uq; }}); 62 0x0f: stq_u({{ Mem.uq = Ra.uq; }}, {{ EA = (Rb + disp) & ~7; }}); 63 0x26: sts({{ Mem.ul = t_to_s(Fa.uq); }}); 64 0x27: stt({{ Mem.df = Fa; }}); 65 0x24: MiscPrefetch::copy_store({{ EA = Rb; }}, 66 {{ fault = xc->copy(EA); }}, 67 inst_flags = [IsMemRef, IsStore, IsCopy]); 68 } 69 70 format StoreCond { 71 0x2e: stl_c({{ Mem.ul = Ra<31:0>; }}, 72 {{ 73 uint64_t tmp = write_result; 74 // see stq_c 75 Ra = (tmp == 0 || tmp == 1) ? tmp : Ra; 76 }}, mem_flags = LOCKED, inst_flags = [IsNonSpeculative, 77 IsSerializing, 78 IsSerializeAfter]); 79 0x2f: stq_c({{ Mem.uq = Ra; }}, 80 {{ 81 uint64_t tmp = write_result; 82 // If the write operation returns 0 or 1, then 83 // this was a conventional store conditional, 84 // and the value indicates the success/failure 85 // of the operation. If another value is 86 // returned, then this was a Turbolaser 87 // mailbox access, and we don't update the 88 // result register at all. 89 Ra = (tmp == 0 || tmp == 1) ? tmp : Ra; 90 }}, mem_flags = LOCKED, inst_flags = [IsNonSpeculative, 91 IsSerializing, 92 IsSerializeAfter]); 93 } 94 95 format IntegerOperate { 96 97 0x10: decode INTFUNC { // integer arithmetic operations 98 99 0x00: addl({{ Rc.sl = Ra.sl + Rb_or_imm.sl; }}); 100 0x40: addlv({{ 101 uint32_t tmp = Ra.sl + Rb_or_imm.sl; 102 // signed overflow occurs when operands have same sign 103 // and sign of result does not match. 104 if (Ra.sl<31:> == Rb_or_imm.sl<31:> && tmp<31:> != Ra.sl<31:>) 105 fault = new IntegerOverflowFault; 106 Rc.sl = tmp; 107 }}); 108 0x02: s4addl({{ Rc.sl = (Ra.sl << 2) + Rb_or_imm.sl; }}); 109 0x12: s8addl({{ Rc.sl = (Ra.sl << 3) + Rb_or_imm.sl; }}); 110 111 0x20: addq({{ Rc = Ra + Rb_or_imm; }}); 112 0x60: addqv({{ 113 uint64_t tmp = Ra + Rb_or_imm; 114 // signed overflow occurs when operands have same sign 115 // and sign of result does not match. 116 if (Ra<63:> == Rb_or_imm<63:> && tmp<63:> != Ra<63:>) 117 fault = new IntegerOverflowFault; 118 Rc = tmp; 119 }}); 120 0x22: s4addq({{ Rc = (Ra << 2) + Rb_or_imm; }}); 121 0x32: s8addq({{ Rc = (Ra << 3) + Rb_or_imm; }}); 122 123 0x09: subl({{ Rc.sl = Ra.sl - Rb_or_imm.sl; }}); 124 0x49: sublv({{ 125 uint32_t tmp = Ra.sl - Rb_or_imm.sl; 126 // signed overflow detection is same as for add, 127 // except we need to look at the *complemented* 128 // sign bit of the subtrahend (Rb), i.e., if the initial 129 // signs are the *same* then no overflow can occur 130 if (Ra.sl<31:> != Rb_or_imm.sl<31:> && tmp<31:> != Ra.sl<31:>) 131 fault = new IntegerOverflowFault; 132 Rc.sl = tmp; 133 }}); 134 0x0b: s4subl({{ Rc.sl = (Ra.sl << 2) - Rb_or_imm.sl; }}); 135 0x1b: s8subl({{ Rc.sl = (Ra.sl << 3) - Rb_or_imm.sl; }}); 136 137 0x29: subq({{ Rc = Ra - Rb_or_imm; }}); 138 0x69: subqv({{ 139 uint64_t tmp = Ra - Rb_or_imm; 140 // signed overflow detection is same as for add, 141 // except we need to look at the *complemented* 142 // sign bit of the subtrahend (Rb), i.e., if the initial 143 // signs are the *same* then no overflow can occur 144 if (Ra<63:> != Rb_or_imm<63:> && tmp<63:> != Ra<63:>) 145 fault = new IntegerOverflowFault; 146 Rc = tmp; 147 }}); 148 0x2b: s4subq({{ Rc = (Ra << 2) - Rb_or_imm; }}); 149 0x3b: s8subq({{ Rc = (Ra << 3) - Rb_or_imm; }}); 150 151 0x2d: cmpeq({{ Rc = (Ra == Rb_or_imm); }}); 152 0x6d: cmple({{ Rc = (Ra.sq <= Rb_or_imm.sq); }}); 153 0x4d: cmplt({{ Rc = (Ra.sq < Rb_or_imm.sq); }}); 154 0x3d: cmpule({{ Rc = (Ra.uq <= Rb_or_imm.uq); }}); 155 0x1d: cmpult({{ Rc = (Ra.uq < Rb_or_imm.uq); }}); 156 157 0x0f: cmpbge({{ 158 int hi = 7; 159 int lo = 0; 160 uint64_t tmp = 0; 161 for (int i = 0; i < 8; ++i) { 162 tmp |= (Ra.uq<hi:lo> >= Rb_or_imm.uq<hi:lo>) << i; 163 hi += 8; 164 lo += 8; 165 } 166 Rc = tmp; 167 }}); 168 } 169 170 0x11: decode INTFUNC { // integer logical operations 171 172 0x00: and({{ Rc = Ra & Rb_or_imm; }}); 173 0x08: bic({{ Rc = Ra & ~Rb_or_imm; }}); 174 0x20: bis({{ Rc = Ra | Rb_or_imm; }}); 175 0x28: ornot({{ Rc = Ra | ~Rb_or_imm; }}); 176 0x40: xor({{ Rc = Ra ^ Rb_or_imm; }}); 177 0x48: eqv({{ Rc = Ra ^ ~Rb_or_imm; }}); 178 179 // conditional moves 180 0x14: cmovlbs({{ Rc = ((Ra & 1) == 1) ? Rb_or_imm : Rc; }}); 181 0x16: cmovlbc({{ Rc = ((Ra & 1) == 0) ? Rb_or_imm : Rc; }}); 182 0x24: cmoveq({{ Rc = (Ra == 0) ? Rb_or_imm : Rc; }}); 183 0x26: cmovne({{ Rc = (Ra != 0) ? Rb_or_imm : Rc; }}); 184 0x44: cmovlt({{ Rc = (Ra.sq < 0) ? Rb_or_imm : Rc; }}); 185 0x46: cmovge({{ Rc = (Ra.sq >= 0) ? Rb_or_imm : Rc; }}); 186 0x64: cmovle({{ Rc = (Ra.sq <= 0) ? Rb_or_imm : Rc; }}); 187 0x66: cmovgt({{ Rc = (Ra.sq > 0) ? Rb_or_imm : Rc; }}); 188 189 // For AMASK, RA must be R31. 190 0x61: decode RA { 191 31: amask({{ Rc = Rb_or_imm & ~ULL(0x17); }}); 192 } 193 194 // For IMPLVER, RA must be R31 and the B operand 195 // must be the immediate value 1. 196 0x6c: decode RA { 197 31: decode IMM { 198 1: decode INTIMM { 199 // return EV5 for FULL_SYSTEM and EV6 otherwise 200 1: implver({{ 201#if FULL_SYSTEM 202 Rc = 1; 203#else 204 Rc = 2; 205#endif 206 }}); 207 } 208 } 209 } 210 211#if FULL_SYSTEM 212 // The mysterious 11.25... 213 0x25: WarnUnimpl::eleven25(); 214#endif 215 } 216 217 0x12: decode INTFUNC { 218 0x39: sll({{ Rc = Ra << Rb_or_imm<5:0>; }}); 219 0x34: srl({{ Rc = Ra.uq >> Rb_or_imm<5:0>; }}); 220 0x3c: sra({{ Rc = Ra.sq >> Rb_or_imm<5:0>; }}); 221 222 0x02: mskbl({{ Rc = Ra & ~(mask( 8) << (Rb_or_imm<2:0> * 8)); }}); 223 0x12: mskwl({{ Rc = Ra & ~(mask(16) << (Rb_or_imm<2:0> * 8)); }}); 224 0x22: mskll({{ Rc = Ra & ~(mask(32) << (Rb_or_imm<2:0> * 8)); }}); 225 0x32: mskql({{ Rc = Ra & ~(mask(64) << (Rb_or_imm<2:0> * 8)); }}); 226 227 0x52: mskwh({{ 228 int bv = Rb_or_imm<2:0>; 229 Rc = bv ? (Ra & ~(mask(16) >> (64 - 8 * bv))) : Ra; 230 }}); 231 0x62: msklh({{ 232 int bv = Rb_or_imm<2:0>; 233 Rc = bv ? (Ra & ~(mask(32) >> (64 - 8 * bv))) : Ra; 234 }}); 235 0x72: mskqh({{ 236 int bv = Rb_or_imm<2:0>; 237 Rc = bv ? (Ra & ~(mask(64) >> (64 - 8 * bv))) : Ra; 238 }}); 239 240 0x06: extbl({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))< 7:0>; }}); 241 0x16: extwl({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))<15:0>; }}); 242 0x26: extll({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))<31:0>; }}); 243 0x36: extql({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8)); }}); 244 245 0x5a: extwh({{ 246 Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<15:0>; }}); 247 0x6a: extlh({{ 248 Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<31:0>; }}); 249 0x7a: extqh({{ 250 Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>); }}); 251 252 0x0b: insbl({{ Rc = Ra< 7:0> << (Rb_or_imm<2:0> * 8); }}); 253 0x1b: inswl({{ Rc = Ra<15:0> << (Rb_or_imm<2:0> * 8); }}); 254 0x2b: insll({{ Rc = Ra<31:0> << (Rb_or_imm<2:0> * 8); }}); 255 0x3b: insql({{ Rc = Ra << (Rb_or_imm<2:0> * 8); }}); 256 257 0x57: inswh({{ 258 int bv = Rb_or_imm<2:0>; 259 Rc = bv ? (Ra.uq<15:0> >> (64 - 8 * bv)) : 0; 260 }}); 261 0x67: inslh({{ 262 int bv = Rb_or_imm<2:0>; 263 Rc = bv ? (Ra.uq<31:0> >> (64 - 8 * bv)) : 0; 264 }}); 265 0x77: insqh({{ 266 int bv = Rb_or_imm<2:0>; 267 Rc = bv ? (Ra.uq >> (64 - 8 * bv)) : 0; 268 }}); 269 270 0x30: zap({{ 271 uint64_t zapmask = 0; 272 for (int i = 0; i < 8; ++i) { 273 if (Rb_or_imm<i:>) 274 zapmask |= (mask(8) << (i * 8)); 275 } 276 Rc = Ra & ~zapmask; 277 }}); 278 0x31: zapnot({{ 279 uint64_t zapmask = 0; 280 for (int i = 0; i < 8; ++i) { 281 if (!Rb_or_imm<i:>) 282 zapmask |= (mask(8) << (i * 8)); 283 } 284 Rc = Ra & ~zapmask; 285 }}); 286 } 287 288 0x13: decode INTFUNC { // integer multiplies 289 0x00: mull({{ Rc.sl = Ra.sl * Rb_or_imm.sl; }}, IntMultOp); 290 0x20: mulq({{ Rc = Ra * Rb_or_imm; }}, IntMultOp); 291 0x30: umulh({{ 292 uint64_t hi, lo; 293 mul128(Ra, Rb_or_imm, hi, lo); 294 Rc = hi; 295 }}, IntMultOp); 296 0x40: mullv({{ 297 // 32-bit multiply with trap on overflow 298 int64_t Rax = Ra.sl; // sign extended version of Ra.sl 299 int64_t Rbx = Rb_or_imm.sl; 300 int64_t tmp = Rax * Rbx; 301 // To avoid overflow, all the upper 32 bits must match 302 // the sign bit of the lower 32. We code this as 303 // checking the upper 33 bits for all 0s or all 1s. 304 uint64_t sign_bits = tmp<63:31>; 305 if (sign_bits != 0 && sign_bits != mask(33)) 306 fault = new IntegerOverflowFault; 307 Rc.sl = tmp<31:0>; 308 }}, IntMultOp); 309 0x60: mulqv({{ 310 // 64-bit multiply with trap on overflow 311 uint64_t hi, lo; 312 mul128(Ra, Rb_or_imm, hi, lo); 313 // all the upper 64 bits must match the sign bit of 314 // the lower 64 315 if (!((hi == 0 && lo<63:> == 0) || 316 (hi == mask(64) && lo<63:> == 1))) 317 fault = new IntegerOverflowFault; 318 Rc = lo; 319 }}, IntMultOp); 320 } 321 322 0x1c: decode INTFUNC { 323 0x00: decode RA { 31: sextb({{ Rc.sb = Rb_or_imm< 7:0>; }}); } 324 0x01: decode RA { 31: sextw({{ Rc.sw = Rb_or_imm<15:0>; }}); } 325 0x32: ctlz({{ 326 uint64_t count = 0; 327 uint64_t temp = Rb; 328 if (temp<63:32>) temp >>= 32; else count += 32; 329 if (temp<31:16>) temp >>= 16; else count += 16; 330 if (temp<15:8>) temp >>= 8; else count += 8; 331 if (temp<7:4>) temp >>= 4; else count += 4; 332 if (temp<3:2>) temp >>= 2; else count += 2; 333 if (temp<1:1>) temp >>= 1; else count += 1; 334 if ((temp<0:0>) != 0x1) count += 1; 335 Rc = count; 336 }}, IntAluOp); 337 338 0x33: cttz({{ 339 uint64_t count = 0; 340 uint64_t temp = Rb; 341 if (!(temp<31:0>)) { temp >>= 32; count += 32; } 342 if (!(temp<15:0>)) { temp >>= 16; count += 16; } 343 if (!(temp<7:0>)) { temp >>= 8; count += 8; } 344 if (!(temp<3:0>)) { temp >>= 4; count += 4; } 345 if (!(temp<1:0>)) { temp >>= 2; count += 2; } 346 if (!(temp<0:0> & ULL(0x1))) count += 1; 347 Rc = count; 348 }}, IntAluOp); 349 350 format FailUnimpl { 351 0x30: ctpop(); 352 0x31: perr(); 353 0x34: unpkbw(); 354 0x35: unpkbl(); 355 0x36: pkwb(); 356 0x37: pklb(); 357 0x38: minsb8(); 358 0x39: minsw4(); 359 0x3a: minub8(); 360 0x3b: minuw4(); 361 0x3c: maxub8(); 362 0x3d: maxuw4(); 363 0x3e: maxsb8(); 364 0x3f: maxsw4(); 365 } 366 367 format BasicOperateWithNopCheck { 368 0x70: decode RB { 369 31: ftoit({{ Rc = Fa.uq; }}, FloatCvtOp); 370 } 371 0x78: decode RB { 372 31: ftois({{ Rc.sl = t_to_s(Fa.uq); }}, 373 FloatCvtOp); 374 } 375 } 376 } 377 } 378 379 // Conditional branches. 380 format CondBranch { 381 0x39: beq({{ cond = (Ra == 0); }}); 382 0x3d: bne({{ cond = (Ra != 0); }}); 383 0x3e: bge({{ cond = (Ra.sq >= 0); }}); 384 0x3f: bgt({{ cond = (Ra.sq > 0); }}); 385 0x3b: ble({{ cond = (Ra.sq <= 0); }}); 386 0x3a: blt({{ cond = (Ra.sq < 0); }}); 387 0x38: blbc({{ cond = ((Ra & 1) == 0); }}); 388 0x3c: blbs({{ cond = ((Ra & 1) == 1); }}); 389 390 0x31: fbeq({{ cond = (Fa == 0); }}); 391 0x35: fbne({{ cond = (Fa != 0); }}); 392 0x36: fbge({{ cond = (Fa >= 0); }}); 393 0x37: fbgt({{ cond = (Fa > 0); }}); 394 0x33: fble({{ cond = (Fa <= 0); }}); 395 0x32: fblt({{ cond = (Fa < 0); }}); 396 } 397 398 // unconditional branches 399 format UncondBranch { 400 0x30: br(); 401 0x34: bsr(IsCall); 402 } 403 404 // indirect branches 405 0x1a: decode JMPFUNC { 406 format Jump { 407 0: jmp(); 408 1: jsr(IsCall); 409 2: ret(IsReturn); 410 3: jsr_coroutine(IsCall, IsReturn); 411 } 412 } 413 414 // Square root and integer-to-FP moves 415 0x14: decode FP_SHORTFUNC { 416 // Integer to FP register moves must have RB == 31 417 0x4: decode RB { 418 31: decode FP_FULLFUNC { 419 format BasicOperateWithNopCheck { 420 0x004: itofs({{ Fc.uq = s_to_t(Ra.ul); }}, FloatCvtOp); 421 0x024: itoft({{ Fc.uq = Ra.uq; }}, FloatCvtOp); 422 0x014: FailUnimpl::itoff(); // VAX-format conversion 423 } 424 } 425 } 426 427 // Square root instructions must have FA == 31 428 0xb: decode FA { 429 31: decode FP_TYPEFUNC { 430 format FloatingPointOperate { 431#if SS_COMPATIBLE_FP 432 0x0b: sqrts({{ 433 if (Fb < 0.0) 434 fault = new ArithmeticFault; 435 Fc = sqrt(Fb); 436 }}, FloatSqrtOp); 437#else 438 0x0b: sqrts({{ 439 if (Fb.sf < 0.0) 440 fault = new ArithmeticFault; 441 Fc.sf = sqrt(Fb.sf); 442 }}, FloatSqrtOp); 443#endif 444 0x2b: sqrtt({{ 445 if (Fb < 0.0) 446 fault = new ArithmeticFault; 447 Fc = sqrt(Fb); 448 }}, FloatSqrtOp); 449 } 450 } 451 } 452 453 // VAX-format sqrtf and sqrtg are not implemented 454 0xa: FailUnimpl::sqrtfg(); 455 } 456 457 // IEEE floating point 458 0x16: decode FP_SHORTFUNC_TOP2 { 459 // The top two bits of the short function code break this 460 // space into four groups: binary ops, compares, reserved, and 461 // conversions. See Table 4-12 of AHB. There are different 462 // special cases in these different groups, so we decode on 463 // these top two bits first just to select a decode strategy. 464 // Most of these instructions may have various trapping and 465 // rounding mode flags set; these are decoded in the 466 // FloatingPointDecode template used by the 467 // FloatingPointOperate format. 468 469 // add/sub/mul/div: just decode on the short function code 470 // and source type. All valid trapping and rounding modes apply. 471 0: decode FP_TRAPMODE { 472 // check for valid trapping modes here 473 0,1,5,7: decode FP_TYPEFUNC { 474 format FloatingPointOperate { 475#if SS_COMPATIBLE_FP 476 0x00: adds({{ Fc = Fa + Fb; }}); 477 0x01: subs({{ Fc = Fa - Fb; }}); 478 0x02: muls({{ Fc = Fa * Fb; }}, FloatMultOp); 479 0x03: divs({{ Fc = Fa / Fb; }}, FloatDivOp); 480#else 481 0x00: adds({{ Fc.sf = Fa.sf + Fb.sf; }}); 482 0x01: subs({{ Fc.sf = Fa.sf - Fb.sf; }}); 483 0x02: muls({{ Fc.sf = Fa.sf * Fb.sf; }}, FloatMultOp); 484 0x03: divs({{ Fc.sf = Fa.sf / Fb.sf; }}, FloatDivOp); 485#endif 486 487 0x20: addt({{ Fc = Fa + Fb; }}); 488 0x21: subt({{ Fc = Fa - Fb; }}); 489 0x22: mult({{ Fc = Fa * Fb; }}, FloatMultOp); 490 0x23: divt({{ Fc = Fa / Fb; }}, FloatDivOp); 491 } 492 } 493 } 494 495 // Floating-point compare instructions must have the default 496 // rounding mode, and may use the default trapping mode or 497 // /SU. Both trapping modes are treated the same by M5; the 498 // only difference on the real hardware (as far a I can tell) 499 // is that without /SU you'd get an imprecise trap if you 500 // tried to compare a NaN with something else (instead of an 501 // "unordered" result). 502 1: decode FP_FULLFUNC { 503 format BasicOperateWithNopCheck { 504 0x0a5, 0x5a5: cmpteq({{ Fc = (Fa == Fb) ? 2.0 : 0.0; }}, 505 FloatCmpOp); 506 0x0a7, 0x5a7: cmptle({{ Fc = (Fa <= Fb) ? 2.0 : 0.0; }}, 507 FloatCmpOp); 508 0x0a6, 0x5a6: cmptlt({{ Fc = (Fa < Fb) ? 2.0 : 0.0; }}, 509 FloatCmpOp); 510 0x0a4, 0x5a4: cmptun({{ // unordered 511 Fc = (!(Fa < Fb) && !(Fa == Fb) && !(Fa > Fb)) ? 2.0 : 0.0; 512 }}, FloatCmpOp); 513 } 514 } 515 516 // The FP-to-integer and integer-to-FP conversion insts 517 // require that FA be 31. 518 3: decode FA { 519 31: decode FP_TYPEFUNC { 520 format FloatingPointOperate { 521 0x2f: decode FP_ROUNDMODE { 522 format FPFixedRounding { 523 // "chopped" i.e. round toward zero 524 0: cvttq({{ Fc.sq = (int64_t)trunc(Fb); }}, 525 Chopped); 526 // round to minus infinity 527 1: cvttq({{ Fc.sq = (int64_t)floor(Fb); }}, 528 MinusInfinity); 529 } 530 default: cvttq({{ Fc.sq = (int64_t)nearbyint(Fb); }}); 531 } 532 533 // The cvtts opcode is overloaded to be cvtst if the trap 534 // mode is 2 or 6 (which are not valid otherwise) 535 0x2c: decode FP_FULLFUNC { 536 format BasicOperateWithNopCheck { 537 // trap on denorm version "cvtst/s" is 538 // simulated same as cvtst 539 0x2ac, 0x6ac: cvtst({{ Fc = Fb.sf; }}); 540 } 541 default: cvtts({{ Fc.sf = Fb; }}); 542 } 543 544 // The trapping mode for integer-to-FP conversions 545 // must be /SUI or nothing; /U and /SU are not 546 // allowed. The full set of rounding modes are 547 // supported though. 548 0x3c: decode FP_TRAPMODE { 549 0,7: cvtqs({{ Fc.sf = Fb.sq; }}); 550 } 551 0x3e: decode FP_TRAPMODE { 552 0,7: cvtqt({{ Fc = Fb.sq; }}); 553 } 554 } 555 } 556 } 557 } 558 559 // misc FP operate 560 0x17: decode FP_FULLFUNC { 561 format BasicOperateWithNopCheck { 562 0x010: cvtlq({{ 563 Fc.sl = (Fb.uq<63:62> << 30) | Fb.uq<58:29>; 564 }}); 565 0x030: cvtql({{ 566 Fc.uq = (Fb.uq<31:30> << 62) | (Fb.uq<29:0> << 29); 567 }}); 568 569 // We treat the precise & imprecise trapping versions of 570 // cvtql identically. 571 0x130, 0x530: cvtqlv({{ 572 // To avoid overflow, all the upper 32 bits must match 573 // the sign bit of the lower 32. We code this as 574 // checking the upper 33 bits for all 0s or all 1s. 575 uint64_t sign_bits = Fb.uq<63:31>; 576 if (sign_bits != 0 && sign_bits != mask(33)) 577 fault = new IntegerOverflowFault; 578 Fc.uq = (Fb.uq<31:30> << 62) | (Fb.uq<29:0> << 29); 579 }}); 580 581 0x020: cpys({{ // copy sign 582 Fc.uq = (Fa.uq<63:> << 63) | Fb.uq<62:0>; 583 }}); 584 0x021: cpysn({{ // copy sign negated 585 Fc.uq = (~Fa.uq<63:> << 63) | Fb.uq<62:0>; 586 }}); 587 0x022: cpyse({{ // copy sign and exponent 588 Fc.uq = (Fa.uq<63:52> << 52) | Fb.uq<51:0>; 589 }}); 590 591 0x02a: fcmoveq({{ Fc = (Fa == 0) ? Fb : Fc; }}); 592 0x02b: fcmovne({{ Fc = (Fa != 0) ? Fb : Fc; }}); 593 0x02c: fcmovlt({{ Fc = (Fa < 0) ? Fb : Fc; }}); 594 0x02d: fcmovge({{ Fc = (Fa >= 0) ? Fb : Fc; }}); 595 0x02e: fcmovle({{ Fc = (Fa <= 0) ? Fb : Fc; }}); 596 0x02f: fcmovgt({{ Fc = (Fa > 0) ? Fb : Fc; }}); 597 598 0x024: mt_fpcr({{ FPCR = Fa.uq; }}, IsSerializing, IsSerializeBefore); 599 0x025: mf_fpcr({{ Fa.uq = FPCR; }}, IsSerializing, IsSerializeBefore); 600 } 601 } 602 603 // miscellaneous mem-format ops 604 0x18: decode MEMFUNC { 605 format WarnUnimpl { 606 0x8000: fetch(); 607 0xa000: fetch_m(); 608 0xe800: ecb(); 609 } 610 611 format MiscPrefetch { 612 0xf800: wh64({{ EA = Rb & ~ULL(63); }}, 613 {{ xc->writeHint(EA, 64, memAccessFlags); }}, 614 mem_flags = NO_FAULT, 615 inst_flags = [IsMemRef, IsDataPrefetch, 616 IsStore, MemWriteOp]); 617 } 618 619 format BasicOperate { 620 0xc000: rpcc({{ 621#if FULL_SYSTEM 622 /* Rb is a fake dependency so here is a fun way to get 623 * the parser to understand that. 624 */ 625 Ra = xc->readMiscRegWithEffect(AlphaISA::IPR_CC, fault) + (Rb & 0); 626 627#else 628 Ra = curTick; 629#endif 630 }}, IsUnverifiable); 631 632 // All of the barrier instructions below do nothing in 633 // their execute() methods (hence the empty code blocks). 634 // All of their functionality is hard-coded in the 635 // pipeline based on the flags IsSerializing, 636 // IsMemBarrier, and IsWriteBarrier. In the current 637 // detailed CPU model, the execute() function only gets 638 // called at fetch, so there's no way to generate pipeline 639 // behavior at any other stage. Once we go to an 640 // exec-in-exec CPU model we should be able to get rid of 641 // these flags and implement this behavior via the 642 // execute() methods. 643 644 // trapb is just a barrier on integer traps, where excb is 645 // a barrier on integer and FP traps. "EXCB is thus a 646 // superset of TRAPB." (Alpha ARM, Sec 4.11.4) We treat 647 // them the same though. 648 0x0000: trapb({{ }}, IsSerializing, IsSerializeBefore, No_OpClass); 649 0x0400: excb({{ }}, IsSerializing, IsSerializeBefore, No_OpClass); 650 0x4000: mb({{ }}, IsMemBarrier, MemReadOp); 651 0x4400: wmb({{ }}, IsWriteBarrier, MemWriteOp); 652 } 653 654#if FULL_SYSTEM 655 format BasicOperate { 656 0xe000: rc({{ 657 Ra = xc->readIntrFlag(); 658 xc->setIntrFlag(0); 659 }}, IsNonSpeculative); 660 0xf000: rs({{ 661 Ra = xc->readIntrFlag(); 662 xc->setIntrFlag(1); 663 }}, IsNonSpeculative); 664 } 665#else 666 format FailUnimpl { 667 0xe000: rc(); 668 0xf000: rs(); 669 } 670#endif 671 } 672 673#if FULL_SYSTEM 674 0x00: CallPal::call_pal({{ 675 if (!palValid || 676 (palPriv 677 && xc->readMiscRegWithEffect(AlphaISA::IPR_ICM, fault) != AlphaISA::mode_kernel)) { 678 // invalid pal function code, or attempt to do privileged 679 // PAL call in non-kernel mode 680 fault = new UnimplementedOpcodeFault; 681 } 682 else { 683 // check to see if simulator wants to do something special 684 // on this PAL call (including maybe suppress it) 685 bool dopal = xc->simPalCheck(palFunc); 686 687 if (dopal) { 688 xc->setMiscRegWithEffect(AlphaISA::IPR_EXC_ADDR, NPC); 689 NPC = xc->readMiscRegWithEffect(AlphaISA::IPR_PAL_BASE, fault) + palOffset; 690 } 691 } 692 }}, IsNonSpeculative); 693#else 694 0x00: decode PALFUNC { 695 format EmulatedCallPal { 696 0x00: halt ({{ 697 SimExit(curTick, "halt instruction encountered"); 698 }}, IsNonSpeculative); 699 0x83: callsys({{ 700 xc->syscall(); 701 }}, IsNonSpeculative, IsSerializeAfter); 702 // Read uniq reg into ABI return value register (r0) 703 0x9e: rduniq({{ R0 = Runiq; }}, IsSerializing, IsSerializeBefore); 704 // Write uniq reg with value from ABI arg register (r16) 705 0x9f: wruniq({{ Runiq = R16; }}, IsSerializing, IsSerializeBefore); 706 } 707 } 708#endif 709 710#if FULL_SYSTEM 711 0x1b: decode PALMODE { 712 0: OpcdecFault::hw_st_quad(); 713 1: decode HW_LDST_QUAD { 714 format HwLoad { 715 0: hw_ld({{ EA = (Rb + disp) & ~3; }}, {{ Ra = Mem.ul; }}, L); 716 1: hw_ld({{ EA = (Rb + disp) & ~7; }}, {{ Ra = Mem.uq; }}, Q); 717 } 718 } 719 } 720 721 0x1f: decode PALMODE { 722 0: OpcdecFault::hw_st_cond(); 723 format HwStore { 724 1: decode HW_LDST_COND { 725 0: decode HW_LDST_QUAD { 726 0: hw_st({{ EA = (Rb + disp) & ~3; }}, 727 {{ Mem.ul = Ra<31:0>; }}, L); 728 1: hw_st({{ EA = (Rb + disp) & ~7; }}, 729 {{ Mem.uq = Ra.uq; }}, Q); 730 } 731 732 1: FailUnimpl::hw_st_cond(); 733 } 734 } 735 } 736 737 0x19: decode PALMODE { 738 0: OpcdecFault::hw_mfpr(); 739 format HwMoveIPR { 740 1: hw_mfpr({{ 741 Ra = xc->readMiscRegWithEffect(ipr_index, fault); 742 }}, IsSerializing, IsSerializeBefore); 743 } 744 } 745 746 0x1d: decode PALMODE { 747 0: OpcdecFault::hw_mtpr(); 748 format HwMoveIPR { 749 1: hw_mtpr({{ 750 xc->setMiscRegWithEffect(ipr_index, Ra); 751 if (traceData) { traceData->setData(Ra); } 752 }}, IsSerializing, IsSerializeBefore); 753 } 754 } 755 756 format BasicOperate { 757 0x1e: decode PALMODE { 758 0: OpcdecFault::hw_rei(); 759 1:hw_rei({{ xc->hwrei(); }}, IsSerializing, IsSerializeBefore); 760 } 761 762 // M5 special opcodes use the reserved 0x01 opcode space 763 0x01: decode M5FUNC { 764 0x00: arm({{ 765 AlphaPseudo::arm(xc->xcBase()); 766 }}, IsNonSpeculative); 767 0x01: quiesce({{ 768 AlphaPseudo::quiesce(xc->xcBase()); 769 }}, IsNonSpeculative, IsQuiesce); 770 0x02: quiesceNs({{ 771 AlphaPseudo::quiesceNs(xc->xcBase(), R16); 772 }}, IsNonSpeculative, IsQuiesce); 773 0x03: quiesceCycles({{ 774 AlphaPseudo::quiesceCycles(xc->xcBase(), R16); 775 }}, IsNonSpeculative, IsQuiesce); 776 0x04: quiesceTime({{ 777 R0 = AlphaPseudo::quiesceTime(xc->xcBase()); 778 }}, IsNonSpeculative); 779 0x10: ivlb({{ 780 AlphaPseudo::ivlb(xc->xcBase()); 781 }}, No_OpClass, IsNonSpeculative); 782 0x11: ivle({{ 783 AlphaPseudo::ivle(xc->xcBase()); 784 }}, No_OpClass, IsNonSpeculative); 785 0x20: m5exit_old({{ 786 AlphaPseudo::m5exit_old(xc->xcBase()); 787 }}, No_OpClass, IsNonSpeculative); 788 0x21: m5exit({{ 789 AlphaPseudo::m5exit(xc->xcBase(), R16); 790 }}, No_OpClass, IsNonSpeculative); 791 0x30: initparam({{ Ra = xc->xcBase()->getCpuPtr()->system->init_param; }}); 792 0x40: resetstats({{ 793 AlphaPseudo::resetstats(xc->xcBase(), R16, R17); 794 }}, IsNonSpeculative); 795 0x41: dumpstats({{ 796 AlphaPseudo::dumpstats(xc->xcBase(), R16, R17); 797 }}, IsNonSpeculative); 798 0x42: dumpresetstats({{ 799 AlphaPseudo::dumpresetstats(xc->xcBase(), R16, R17); 800 }}, IsNonSpeculative); 801 0x43: m5checkpoint({{ 802 AlphaPseudo::m5checkpoint(xc->xcBase(), R16, R17); 803 }}, IsNonSpeculative); 804 0x50: m5readfile({{ 805 R0 = AlphaPseudo::readfile(xc->xcBase(), R16, R17, R18); 806 }}, IsNonSpeculative); 807 0x51: m5break({{ 808 AlphaPseudo::debugbreak(xc->xcBase()); 809 }}, IsNonSpeculative); 810 0x52: m5switchcpu({{ 811 AlphaPseudo::switchcpu(xc->xcBase()); 812 }}, IsNonSpeculative); 813 0x53: m5addsymbol({{ 814 AlphaPseudo::addsymbol(xc->xcBase(), R16, R17); 815 }}, IsNonSpeculative); 816 0x54: m5panic({{ 817 panic("M5 panic instruction called."); 818 }}, IsNonSpeculative); 819 820 } 821 } 822#endif 823} 824