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