decoder.isa revision 2068
1// -*- mode:c++ -*-
2
3// Copyright (c) 2003-2005 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({{ EA = Rb + disp; }}, {{ Ra.uq = Mem.ub; }});
38        0x0c: ldwu({{ EA = Rb + disp; }}, {{ Ra.uq = Mem.uw; }});
39        0x0b: ldq_u({{ EA = (Rb + disp) & ~7; }}, {{ Ra = Mem.uq; }});
40        0x23: ldt({{ EA = Rb + disp; }}, {{ Fa = Mem.df; }});
41        0x2a: ldl_l({{ EA = Rb + disp; }}, {{ Ra.sl = Mem.sl; }}, LOCKED);
42        0x2b: ldq_l({{ EA = Rb + disp; }}, {{ Ra.uq = Mem.uq; }}, LOCKED);
43        0x20: copy_load({{EA = Ra;}},
44                        {{fault = xc->copySrcTranslate(EA);}},
45                        IsMemRef, IsLoad, IsCopy);
46    }
47
48    format LoadOrPrefetch {
49        0x28: ldl({{ EA = Rb + disp; }}, {{ Ra.sl = Mem.sl; }});
50        0x29: ldq({{ EA = Rb + disp; }}, {{ Ra.uq = Mem.uq; }}, 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({{ EA = Rb + disp; }}, {{ Fa.uq = s_to_t(Mem.ul); }},
54                  PF_EXCLUSIVE, IsFloating);
55    }
56
57    format Store {
58        0x0e: stb({{ EA = Rb + disp; }}, {{ Mem.ub = Ra<7:0>; }});
59        0x0d: stw({{ EA = Rb + disp; }}, {{ Mem.uw = Ra<15:0>; }});
60        0x2c: stl({{ EA = Rb + disp; }}, {{ Mem.ul = Ra<31:0>; }});
61        0x2d: stq({{ EA = Rb + disp; }}, {{ Mem.uq = Ra.uq; }});
62        0x0f: stq_u({{ EA = (Rb + disp) & ~7; }}, {{ Mem.uq = Ra.uq; }});
63        0x26: sts({{ EA = Rb + disp; }}, {{ Mem.ul = t_to_s(Fa.uq); }});
64        0x27: stt({{ EA = Rb + disp; }}, {{ Mem.df = Fa; }});
65        0x24: copy_store({{EA = Rb;}},
66                         {{fault = xc->copy(EA);}},
67                         IsMemRef, IsStore, IsCopy);
68    }
69
70    format StoreCond {
71        0x2e: stl_c({{ EA = Rb + disp; }}, {{ Mem.ul = Ra<31:0>; }},
72                    {{
73                        uint64_t tmp = Mem_write_result;
74                        // see stq_c
75                        Ra = (tmp == 0 || tmp == 1) ? tmp : Ra;
76                    }}, LOCKED);
77        0x2f: stq_c({{ EA = Rb + disp; }}, {{ Mem.uq = Ra; }},
78                    {{
79                        uint64_t tmp = Mem_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                    }}, 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 = Integer_Overflow_Fault;
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 = Integer_Overflow_Fault;
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 = Integer_Overflow_Fault;
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 = Integer_Overflow_Fault;
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 = Integer_Overflow_Fault;
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 = Integer_Overflow_Fault;
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 = Arithmetic_Fault;
431                        Fc = sqrt(Fb);
432                    }}, FloatSqrtOp);
433#else
434                    0x0b: sqrts({{
435                        if (Fb.sf < 0.0)
436                            fault = Arithmetic_Fault;
437                        Fc.sf = sqrt(Fb.sf);
438                    }}, FloatSqrtOp);
439#endif
440                    0x2b: sqrtt({{
441                        if (Fb < 0.0)
442                            fault = Arithmetic_Fault;
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 = Integer_Overflow_Fault;
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                         IsMemRef, IsDataPrefetch, IsStore, MemWriteOp,
611                         NO_FAULT);
612        }
613
614        format BasicOperate {
615            0xc000: rpcc({{
616#if FULL_SYSTEM
617        /* Rb is a fake dependency so here is a fun way to get
618         * the parser to understand that.
619         */
620                Ra = xc->readIpr(AlphaISA::IPR_CC, fault) + (Rb & 0);
621
622#else
623                Ra = curTick;
624#endif
625            }});
626
627            // All of the barrier instructions below do nothing in
628            // their execute() methods (hence the empty code blocks).
629            // All of their functionality is hard-coded in the
630            // pipeline based on the flags IsSerializing,
631            // IsMemBarrier, and IsWriteBarrier.  In the current
632            // detailed CPU model, the execute() function only gets
633            // called at fetch, so there's no way to generate pipeline
634            // behavior at any other stage.  Once we go to an
635            // exec-in-exec CPU model we should be able to get rid of
636            // these flags and implement this behavior via the
637            // execute() methods.
638
639            // trapb is just a barrier on integer traps, where excb is
640            // a barrier on integer and FP traps.  "EXCB is thus a
641            // superset of TRAPB." (Alpha ARM, Sec 4.11.4) We treat
642            // them the same though.
643            0x0000: trapb({{ }}, IsSerializing, No_OpClass);
644            0x0400: excb({{ }}, IsSerializing, No_OpClass);
645            0x4000: mb({{ }}, IsMemBarrier, MemReadOp);
646            0x4400: wmb({{ }}, IsWriteBarrier, MemWriteOp);
647        }
648
649#if FULL_SYSTEM
650        format BasicOperate {
651            0xe000: rc({{
652                Ra = xc->readIntrFlag();
653                xc->setIntrFlag(0);
654            }}, IsNonSpeculative);
655            0xf000: rs({{
656                Ra = xc->readIntrFlag();
657                xc->setIntrFlag(1);
658            }}, IsNonSpeculative);
659        }
660#else
661        format FailUnimpl {
662            0xe000: rc();
663            0xf000: rs();
664        }
665#endif
666    }
667
668#if FULL_SYSTEM
669    0x00: CallPal::call_pal({{
670        if (!palValid ||
671            (palPriv
672             && xc->readIpr(AlphaISA::IPR_ICM, fault) != AlphaISA::mode_kernel)) {
673            // invalid pal function code, or attempt to do privileged
674            // PAL call in non-kernel mode
675            fault = Unimplemented_Opcode_Fault;
676        }
677        else {
678            // check to see if simulator wants to do something special
679            // on this PAL call (including maybe suppress it)
680            bool dopal = xc->simPalCheck(palFunc);
681
682            if (dopal) {
683                AlphaISA::swap_palshadow(&xc->xcBase()->regs, true);
684                xc->setIpr(AlphaISA::IPR_EXC_ADDR, NPC);
685                NPC = xc->readIpr(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();
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    format HwLoadStore {
708        0x1b: decode HW_LDST_QUAD {
709            0: hw_ld({{ EA = (Rb + disp) & ~3; }}, {{ Ra = Mem.ul; }}, L);
710            1: hw_ld({{ EA = (Rb + disp) & ~7; }}, {{ Ra = Mem.uq; }}, Q);
711        }
712
713        0x1f: decode HW_LDST_COND {
714            0: decode HW_LDST_QUAD {
715                0: hw_st({{ EA = (Rb + disp) & ~3; }},
716                         {{ Mem.ul = Ra<31:0>; }}, L);
717                1: hw_st({{ EA = (Rb + disp) & ~7; }},
718                         {{ Mem.uq = Ra.uq; }}, Q);
719            }
720
721            1: FailUnimpl::hw_st_cond();
722        }
723    }
724
725    format HwMoveIPR {
726        0x19: hw_mfpr({{
727            // this instruction is only valid in PAL mode
728            if (!xc->inPalMode()) {
729                fault = Unimplemented_Opcode_Fault;
730            }
731            else {
732                Ra = xc->readIpr(ipr_index, fault);
733            }
734        }});
735        0x1d: hw_mtpr({{
736            // this instruction is only valid in PAL mode
737            if (!xc->inPalMode()) {
738                fault = Unimplemented_Opcode_Fault;
739            }
740            else {
741                xc->setIpr(ipr_index, Ra);
742                if (traceData) { traceData->setData(Ra); }
743            }
744        }});
745    }
746
747    format BasicOperate {
748        0x1e: hw_rei({{ xc->hwrei(); }}, IsSerializing);
749
750        // M5 special opcodes use the reserved 0x01 opcode space
751        0x01: decode M5FUNC {
752            0x00: arm({{
753                AlphaPseudo::arm(xc->xcBase());
754            }}, IsNonSpeculative);
755            0x01: quiesce({{
756                AlphaPseudo::quiesce(xc->xcBase());
757            }}, IsNonSpeculative);
758            0x10: ivlb({{
759                AlphaPseudo::ivlb(xc->xcBase());
760            }}, No_OpClass, IsNonSpeculative);
761            0x11: ivle({{
762                AlphaPseudo::ivle(xc->xcBase());
763            }}, No_OpClass, IsNonSpeculative);
764            0x20: m5exit_old({{
765                AlphaPseudo::m5exit_old(xc->xcBase());
766            }}, No_OpClass, IsNonSpeculative);
767            0x21: m5exit({{
768                AlphaPseudo::m5exit(xc->xcBase());
769            }}, No_OpClass, IsNonSpeculative);
770            0x30: initparam({{ Ra = xc->xcBase()->cpu->system->init_param; }});
771            0x40: resetstats({{
772                AlphaPseudo::resetstats(xc->xcBase());
773            }}, IsNonSpeculative);
774            0x41: dumpstats({{
775                AlphaPseudo::dumpstats(xc->xcBase());
776            }}, IsNonSpeculative);
777            0x42: dumpresetstats({{
778                AlphaPseudo::dumpresetstats(xc->xcBase());
779            }}, IsNonSpeculative);
780            0x43: m5checkpoint({{
781                AlphaPseudo::m5checkpoint(xc->xcBase());
782            }}, IsNonSpeculative);
783            0x50: m5readfile({{
784                AlphaPseudo::readfile(xc->xcBase());
785            }}, IsNonSpeculative);
786            0x51: m5break({{
787                AlphaPseudo::debugbreak(xc->xcBase());
788            }}, IsNonSpeculative);
789            0x52: m5switchcpu({{
790                AlphaPseudo::switchcpu(xc->xcBase());
791            }}, IsNonSpeculative);
792            0x53: m5addsymbol({{
793                AlphaPseudo::addsymbol(xc->xcBase());
794            }}, IsNonSpeculative);
795
796        }
797    }
798#endif
799}
800