decoder.isa revision 8564
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//
29// Authors: Steve Reinhardt
30
31////////////////////////////////////////////////////////////////////
32//
33// The actual decoder specification
34//
35
36decode OPCODE default Unknown::unknown() {
37
38    format LoadAddress {
39        0x08: lda({{ Ra = Rb + disp; }});
40        0x09: ldah({{ Ra = Rb + (disp << 16); }});
41    }
42
43    format LoadOrNop {
44        0x0a: ldbu({{ Ra.uq = Mem.ub; }});
45        0x0c: ldwu({{ Ra.uq = Mem.uw; }});
46        0x0b: ldq_u({{ Ra = Mem.uq; }}, ea_code = {{ EA = (Rb + disp) & ~7; }});
47        0x23: ldt({{ Fa = Mem.df; }});
48        0x2a: ldl_l({{ Ra.sl = Mem.sl; }}, mem_flags = LLSC);
49        0x2b: ldq_l({{ Ra.uq = Mem.uq; }}, mem_flags = LLSC);
50    }
51
52    format LoadOrPrefetch {
53        0x28: ldl({{ Ra.sl = Mem.sl; }});
54        0x29: ldq({{ Ra.uq = Mem.uq; }}, pf_flags = EVICT_NEXT);
55        // IsFloating flag on lds gets the prefetch to disassemble
56        // using f31 instead of r31... funcitonally it's unnecessary
57        0x22: lds({{ Fa.uq = s_to_t(Mem.ul); }},
58                  pf_flags = PF_EXCLUSIVE, inst_flags = IsFloating);
59    }
60
61    format Store {
62        0x0e: stb({{ Mem.ub = Ra<7:0>; }});
63        0x0d: stw({{ Mem.uw = Ra<15:0>; }});
64        0x2c: stl({{ Mem.ul = Ra<31:0>; }});
65        0x2d: stq({{ Mem.uq = Ra.uq; }});
66        0x0f: stq_u({{ Mem.uq = Ra.uq; }}, {{ EA = (Rb + disp) & ~7; }});
67        0x26: sts({{ Mem.ul = t_to_s(Fa.uq); }});
68        0x27: stt({{ Mem.df = Fa; }});
69    }
70
71    format StoreCond {
72        0x2e: stl_c({{ Mem.ul = Ra<31:0>; }},
73                    {{
74                        uint64_t tmp = write_result;
75                        // see stq_c
76                        Ra = (tmp == 0 || tmp == 1) ? tmp : Ra;
77                        if (tmp == 1) {
78                            xc->setStCondFailures(0);
79                        }
80                    }}, mem_flags = LLSC, inst_flags = IsStoreConditional);
81        0x2f: stq_c({{ Mem.uq = Ra; }},
82                    {{
83                        uint64_t tmp = write_result;
84                        // If the write operation returns 0 or 1, then
85                        // this was a conventional store conditional,
86                        // and the value indicates the success/failure
87                        // of the operation.  If another value is
88                        // returned, then this was a Turbolaser
89                        // mailbox access, and we don't update the
90                        // result register at all.
91                        Ra = (tmp == 0 || tmp == 1) ? tmp : Ra;
92                        if (tmp == 1) {
93                            // clear failure counter... this is
94                            // non-architectural and for debugging
95                            // only.
96                            xc->setStCondFailures(0);
97                        }
98                    }}, mem_flags = LLSC, inst_flags = IsStoreConditional);
99    }
100
101    format IntegerOperate {
102
103        0x10: decode INTFUNC {  // integer arithmetic operations
104
105            0x00: addl({{ Rc.sl = Ra.sl + Rb_or_imm.sl; }});
106            0x40: addlv({{
107                int32_t tmp  = Ra.sl + Rb_or_imm.sl;
108                // signed overflow occurs when operands have same sign
109                // and sign of result does not match.
110                if (Ra.sl<31:> == Rb_or_imm.sl<31:> && tmp<31:> != Ra.sl<31:>)
111                    fault = new IntegerOverflowFault;
112                Rc.sl = tmp;
113            }});
114            0x02: s4addl({{ Rc.sl = (Ra.sl << 2) + Rb_or_imm.sl; }});
115            0x12: s8addl({{ Rc.sl = (Ra.sl << 3) + Rb_or_imm.sl; }});
116
117            0x20: addq({{ Rc = Ra + Rb_or_imm; }});
118            0x60: addqv({{
119                uint64_t tmp = Ra + Rb_or_imm;
120                // signed overflow occurs when operands have same sign
121                // and sign of result does not match.
122                if (Ra<63:> == Rb_or_imm<63:> && tmp<63:> != Ra<63:>)
123                    fault = new IntegerOverflowFault;
124                Rc = tmp;
125            }});
126            0x22: s4addq({{ Rc = (Ra << 2) + Rb_or_imm; }});
127            0x32: s8addq({{ Rc = (Ra << 3) + Rb_or_imm; }});
128
129            0x09: subl({{ Rc.sl = Ra.sl - Rb_or_imm.sl; }});
130            0x49: sublv({{
131                int32_t tmp  = Ra.sl - Rb_or_imm.sl;
132                // signed overflow detection is same as for add,
133                // except we need to look at the *complemented*
134                // sign bit of the subtrahend (Rb), i.e., if the initial
135                // signs are the *same* then no overflow can occur
136                if (Ra.sl<31:> != Rb_or_imm.sl<31:> && tmp<31:> != Ra.sl<31:>)
137                    fault = new IntegerOverflowFault;
138                Rc.sl = tmp;
139            }});
140            0x0b: s4subl({{ Rc.sl = (Ra.sl << 2) - Rb_or_imm.sl; }});
141            0x1b: s8subl({{ Rc.sl = (Ra.sl << 3) - Rb_or_imm.sl; }});
142
143            0x29: subq({{ Rc = Ra - Rb_or_imm; }});
144            0x69: subqv({{
145                uint64_t tmp  = Ra - Rb_or_imm;
146                // signed overflow detection is same as for add,
147                // except we need to look at the *complemented*
148                // sign bit of the subtrahend (Rb), i.e., if the initial
149                // signs are the *same* then no overflow can occur
150                if (Ra<63:> != Rb_or_imm<63:> && tmp<63:> != Ra<63:>)
151                    fault = new IntegerOverflowFault;
152                Rc = tmp;
153            }});
154            0x2b: s4subq({{ Rc = (Ra << 2) - Rb_or_imm; }});
155            0x3b: s8subq({{ Rc = (Ra << 3) - Rb_or_imm; }});
156
157            0x2d: cmpeq({{ Rc = (Ra == Rb_or_imm); }});
158            0x6d: cmple({{ Rc = (Ra.sq <= Rb_or_imm.sq); }});
159            0x4d: cmplt({{ Rc = (Ra.sq <  Rb_or_imm.sq); }});
160            0x3d: cmpule({{ Rc = (Ra.uq <= Rb_or_imm.uq); }});
161            0x1d: cmpult({{ Rc = (Ra.uq <  Rb_or_imm.uq); }});
162
163            0x0f: cmpbge({{
164                int hi = 7;
165                int lo = 0;
166                uint64_t tmp = 0;
167                for (int i = 0; i < 8; ++i) {
168                    tmp |= (Ra.uq<hi:lo> >= Rb_or_imm.uq<hi:lo>) << i;
169                    hi += 8;
170                    lo += 8;
171                }
172                Rc = tmp;
173            }});
174        }
175
176        0x11: decode INTFUNC {  // integer logical operations
177
178            0x00: and({{ Rc = Ra & Rb_or_imm; }});
179            0x08: bic({{ Rc = Ra & ~Rb_or_imm; }});
180            0x20: bis({{ Rc = Ra | Rb_or_imm; }});
181            0x28: ornot({{ Rc = Ra | ~Rb_or_imm; }});
182            0x40: xor({{ Rc = Ra ^ Rb_or_imm; }});
183            0x48: eqv({{ Rc = Ra ^ ~Rb_or_imm; }});
184
185            // conditional moves
186            0x14: cmovlbs({{ Rc = ((Ra & 1) == 1) ? Rb_or_imm : Rc; }});
187            0x16: cmovlbc({{ Rc = ((Ra & 1) == 0) ? Rb_or_imm : Rc; }});
188            0x24: cmoveq({{ Rc = (Ra == 0) ? Rb_or_imm : Rc; }});
189            0x26: cmovne({{ Rc = (Ra != 0) ? Rb_or_imm : Rc; }});
190            0x44: cmovlt({{ Rc = (Ra.sq <  0) ? Rb_or_imm : Rc; }});
191            0x46: cmovge({{ Rc = (Ra.sq >= 0) ? Rb_or_imm : Rc; }});
192            0x64: cmovle({{ Rc = (Ra.sq <= 0) ? Rb_or_imm : Rc; }});
193            0x66: cmovgt({{ Rc = (Ra.sq >  0) ? Rb_or_imm : Rc; }});
194
195            // For AMASK, RA must be R31.
196            0x61: decode RA {
197                31: amask({{ Rc = Rb_or_imm & ~ULL(0x17); }});
198            }
199
200            // For IMPLVER, RA must be R31 and the B operand
201            // must be the immediate value 1.
202            0x6c: decode RA {
203                31: decode IMM {
204                    1: decode INTIMM {
205                        // return EV5 for FULL_SYSTEM and EV6 otherwise
206                        1: implver({{ Rc = FULL_SYSTEM ? 1 : 2 }});
207                    }
208                }
209            }
210
211            // The mysterious 11.25...
212            0x25: WarnUnimpl::eleven25();
213        }
214
215        0x12: decode INTFUNC {
216            0x39: sll({{ Rc = Ra << Rb_or_imm<5:0>; }});
217            0x34: srl({{ Rc = Ra.uq >> Rb_or_imm<5:0>; }});
218            0x3c: sra({{ Rc = Ra.sq >> Rb_or_imm<5:0>; }});
219
220            0x02: mskbl({{ Rc = Ra & ~(mask( 8) << (Rb_or_imm<2:0> * 8)); }});
221            0x12: mskwl({{ Rc = Ra & ~(mask(16) << (Rb_or_imm<2:0> * 8)); }});
222            0x22: mskll({{ Rc = Ra & ~(mask(32) << (Rb_or_imm<2:0> * 8)); }});
223            0x32: mskql({{ Rc = Ra & ~(mask(64) << (Rb_or_imm<2:0> * 8)); }});
224
225            0x52: mskwh({{
226                int bv = Rb_or_imm<2:0>;
227                Rc =  bv ? (Ra & ~(mask(16) >> (64 - 8 * bv))) : Ra;
228            }});
229            0x62: msklh({{
230                int bv = Rb_or_imm<2:0>;
231                Rc =  bv ? (Ra & ~(mask(32) >> (64 - 8 * bv))) : Ra;
232            }});
233            0x72: mskqh({{
234                int bv = Rb_or_imm<2:0>;
235                Rc =  bv ? (Ra & ~(mask(64) >> (64 - 8 * bv))) : Ra;
236            }});
237
238            0x06: extbl({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))< 7:0>; }});
239            0x16: extwl({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))<15:0>; }});
240            0x26: extll({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8))<31:0>; }});
241            0x36: extql({{ Rc = (Ra.uq >> (Rb_or_imm<2:0> * 8)); }});
242
243            0x5a: extwh({{
244                Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<15:0>; }});
245            0x6a: extlh({{
246                Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>)<31:0>; }});
247            0x7a: extqh({{
248                Rc = (Ra << (64 - (Rb_or_imm<2:0> * 8))<5:0>); }});
249
250            0x0b: insbl({{ Rc = Ra< 7:0> << (Rb_or_imm<2:0> * 8); }});
251            0x1b: inswl({{ Rc = Ra<15:0> << (Rb_or_imm<2:0> * 8); }});
252            0x2b: insll({{ Rc = Ra<31:0> << (Rb_or_imm<2:0> * 8); }});
253            0x3b: insql({{ Rc = Ra       << (Rb_or_imm<2:0> * 8); }});
254
255            0x57: inswh({{
256                int bv = Rb_or_imm<2:0>;
257                Rc = bv ? (Ra.uq<15:0> >> (64 - 8 * bv)) : 0;
258            }});
259            0x67: inslh({{
260                int bv = Rb_or_imm<2:0>;
261                Rc = bv ? (Ra.uq<31:0> >> (64 - 8 * bv)) : 0;
262            }});
263            0x77: insqh({{
264                int bv = Rb_or_imm<2:0>;
265                Rc = bv ? (Ra.uq       >> (64 - 8 * bv)) : 0;
266            }});
267
268            0x30: zap({{
269                uint64_t zapmask = 0;
270                for (int i = 0; i < 8; ++i) {
271                    if (Rb_or_imm<i:>)
272                        zapmask |= (mask(8) << (i * 8));
273                }
274                Rc = Ra & ~zapmask;
275            }});
276            0x31: zapnot({{
277                uint64_t zapmask = 0;
278                for (int i = 0; i < 8; ++i) {
279                    if (!Rb_or_imm<i:>)
280                        zapmask |= (mask(8) << (i * 8));
281                }
282                Rc = Ra & ~zapmask;
283            }});
284        }
285
286        0x13: decode INTFUNC {  // integer multiplies
287            0x00: mull({{ Rc.sl = Ra.sl * Rb_or_imm.sl; }}, IntMultOp);
288            0x20: mulq({{ Rc    = Ra    * Rb_or_imm;    }}, IntMultOp);
289            0x30: umulh({{
290                uint64_t hi, lo;
291                mul128(Ra, Rb_or_imm, hi, lo);
292                Rc = hi;
293            }}, IntMultOp);
294            0x40: mullv({{
295                // 32-bit multiply with trap on overflow
296                int64_t Rax = Ra.sl;    // sign extended version of Ra.sl
297                int64_t Rbx = Rb_or_imm.sl;
298                int64_t tmp = Rax * Rbx;
299                // To avoid overflow, all the upper 32 bits must match
300                // the sign bit of the lower 32.  We code this as
301                // checking the upper 33 bits for all 0s or all 1s.
302                uint64_t sign_bits = tmp<63:31>;
303                if (sign_bits != 0 && sign_bits != mask(33))
304                    fault = new IntegerOverflowFault;
305                Rc.sl = tmp<31:0>;
306            }}, IntMultOp);
307            0x60: mulqv({{
308                // 64-bit multiply with trap on overflow
309                uint64_t hi, lo;
310                mul128(Ra, Rb_or_imm, hi, lo);
311                // all the upper 64 bits must match the sign bit of
312                // the lower 64
313                if (!((hi == 0 && lo<63:> == 0) ||
314                      (hi == mask(64) && lo<63:> == 1)))
315                    fault = new IntegerOverflowFault;
316                Rc = lo;
317            }}, IntMultOp);
318        }
319
320        0x1c: decode INTFUNC {
321            0x00: decode RA { 31: sextb({{ Rc.sb = Rb_or_imm< 7:0>; }}); }
322            0x01: decode RA { 31: sextw({{ Rc.sw = Rb_or_imm<15:0>; }}); }
323
324            0x30: ctpop({{
325                             uint64_t count = 0;
326                             for (int i = 0; Rb<63:i>; ++i) {
327                                 if (Rb<i:i> == 0x1)
328                                     ++count;
329                             }
330                             Rc = count;
331                           }}, IntAluOp);
332
333            0x31: perr({{
334                             uint64_t temp = 0;
335                             int hi = 7;
336                             int lo = 0;
337                             for (int i = 0; i < 8; ++i) {
338                                 uint8_t ra_ub = Ra.uq<hi:lo>;
339                                 uint8_t rb_ub = Rb.uq<hi:lo>;
340                                 temp += (ra_ub >= rb_ub) ? 
341                                         (ra_ub - rb_ub) : (rb_ub - ra_ub);
342                                 hi += 8;
343                                 lo += 8;
344                             }
345                             Rc = temp;
346                           }});
347
348            0x32: ctlz({{
349                             uint64_t count = 0;
350                             uint64_t temp = Rb;
351                             if (temp<63:32>) temp >>= 32; else count += 32;
352                             if (temp<31:16>) temp >>= 16; else count += 16;
353                             if (temp<15:8>) temp >>= 8; else count += 8;
354                             if (temp<7:4>) temp >>= 4; else count += 4;
355                             if (temp<3:2>) temp >>= 2; else count += 2;
356                             if (temp<1:1>) temp >>= 1; else count += 1;
357                             if ((temp<0:0>) != 0x1) count += 1;
358                             Rc = count;
359                           }}, IntAluOp);
360
361            0x33: cttz({{
362                             uint64_t count = 0;
363                             uint64_t temp = Rb;
364                             if (!(temp<31:0>)) { temp >>= 32; count += 32; }
365                             if (!(temp<15:0>)) { temp >>= 16; count += 16; }
366                             if (!(temp<7:0>)) { temp >>= 8; count += 8; }
367                             if (!(temp<3:0>)) { temp >>= 4; count += 4; }
368                             if (!(temp<1:0>)) { temp >>= 2; count += 2; }
369                             if (!(temp<0:0> & ULL(0x1))) { 
370                                 temp >>= 1; count += 1; 
371                             }
372                             if (!(temp<0:0> & ULL(0x1))) count += 1;
373                             Rc = count;
374                           }}, IntAluOp);
375
376
377            0x34: unpkbw({{ 
378                             Rc = (Rb.uq<7:0> 
379                                   | (Rb.uq<15:8> << 16) 
380                                   | (Rb.uq<23:16> << 32) 
381                                   | (Rb.uq<31:24> << 48)); 
382                           }}, IntAluOp);
383
384            0x35: unpkbl({{
385                             Rc = (Rb.uq<7:0> | (Rb.uq<15:8> << 32)); 
386                           }}, IntAluOp);
387
388            0x36: pkwb({{
389                             Rc = (Rb.uq<7:0> 
390                                   | (Rb.uq<23:16> << 8) 
391                                   | (Rb.uq<39:32> << 16) 
392                                   | (Rb.uq<55:48> << 24)); 
393                           }}, IntAluOp);
394
395            0x37: pklb({{
396                             Rc = (Rb.uq<7:0> | (Rb.uq<39:32> << 8)); 
397                           }}, IntAluOp);
398
399            0x38: minsb8({{
400                             uint64_t temp = 0;
401                             int hi = 63;
402                             int lo = 56;
403                             for (int i = 7; i >= 0; --i) {
404                                 int8_t ra_sb = Ra.uq<hi:lo>;
405                                 int8_t rb_sb = Rb.uq<hi:lo>;
406                                 temp = ((temp << 8) 
407                                         | ((ra_sb < rb_sb) ? Ra.uq<hi:lo>
408                                                          : Rb.uq<hi:lo>));
409                                 hi -= 8;
410                                 lo -= 8;
411                             }
412                             Rc = temp;
413                          }});
414
415            0x39: minsw4({{
416                             uint64_t temp = 0;
417                             int hi = 63;
418                             int lo = 48;
419                             for (int i = 3; i >= 0; --i) {
420                                 int16_t ra_sw = Ra.uq<hi:lo>;
421                                 int16_t rb_sw = Rb.uq<hi:lo>;
422                                 temp = ((temp << 16) 
423                                         | ((ra_sw < rb_sw) ? Ra.uq<hi:lo>
424                                                          : Rb.uq<hi:lo>));
425                                 hi -= 16;
426                                 lo -= 16;
427                             }
428                             Rc = temp;
429                          }});
430
431            0x3a: minub8({{
432                             uint64_t temp = 0;
433                             int hi = 63;
434                             int lo = 56;
435                             for (int i = 7; i >= 0; --i) {
436                                 uint8_t ra_ub = Ra.uq<hi:lo>;
437                                 uint8_t rb_ub = Rb.uq<hi:lo>;
438                                 temp = ((temp << 8) 
439                                         | ((ra_ub < rb_ub) ? Ra.uq<hi:lo>
440                                                          : Rb.uq<hi:lo>));
441                                 hi -= 8;
442                                 lo -= 8;
443                             }
444                             Rc = temp;
445                          }});
446
447            0x3b: minuw4({{
448                             uint64_t temp = 0;
449                             int hi = 63;
450                             int lo = 48;
451                             for (int i = 3; i >= 0; --i) {
452                                 uint16_t ra_sw = Ra.uq<hi:lo>;
453                                 uint16_t rb_sw = Rb.uq<hi:lo>;
454                                 temp = ((temp << 16) 
455                                         | ((ra_sw < rb_sw) ? Ra.uq<hi:lo>
456                                                          : Rb.uq<hi:lo>));
457                                 hi -= 16;
458                                 lo -= 16;
459                             }
460                             Rc = temp;
461                          }});
462
463            0x3c: maxub8({{
464                             uint64_t temp = 0;
465                             int hi = 63;
466                             int lo = 56;
467                             for (int i = 7; i >= 0; --i) {
468                                 uint8_t ra_ub = Ra.uq<hi:lo>;
469                                 uint8_t rb_ub = Rb.uq<hi:lo>;
470                                 temp = ((temp << 8) 
471                                         | ((ra_ub > rb_ub) ? Ra.uq<hi:lo>
472                                                          : Rb.uq<hi:lo>));
473                                 hi -= 8;
474                                 lo -= 8;
475                             }
476                             Rc = temp;
477                          }});
478
479            0x3d: maxuw4({{
480                             uint64_t temp = 0;
481                             int hi = 63;
482                             int lo = 48;
483                             for (int i = 3; i >= 0; --i) {
484                                 uint16_t ra_uw = Ra.uq<hi:lo>;
485                                 uint16_t rb_uw = Rb.uq<hi:lo>;
486                                 temp = ((temp << 16) 
487                                         | ((ra_uw > rb_uw) ? Ra.uq<hi:lo>
488                                                          : Rb.uq<hi:lo>));
489                                 hi -= 16;
490                                 lo -= 16;
491                             }
492                             Rc = temp;
493                          }});
494
495            0x3e: maxsb8({{
496                             uint64_t temp = 0;
497                             int hi = 63;
498                             int lo = 56;
499                             for (int i = 7; i >= 0; --i) {
500                                 int8_t ra_sb = Ra.uq<hi:lo>;
501                                 int8_t rb_sb = Rb.uq<hi:lo>;
502                                 temp = ((temp << 8) 
503                                         | ((ra_sb > rb_sb) ? Ra.uq<hi:lo>
504                                                          : Rb.uq<hi:lo>));
505                                 hi -= 8;
506                                 lo -= 8;
507                             }
508                             Rc = temp;
509                          }});
510
511            0x3f: maxsw4({{
512                             uint64_t temp = 0;
513                             int hi = 63;
514                             int lo = 48;
515                             for (int i = 3; i >= 0; --i) {
516                                 int16_t ra_sw = Ra.uq<hi:lo>;
517                                 int16_t rb_sw = Rb.uq<hi:lo>;
518                                 temp = ((temp << 16) 
519                                         | ((ra_sw > rb_sw) ? Ra.uq<hi:lo>
520                                                          : Rb.uq<hi:lo>));
521                                 hi -= 16;
522                                 lo -= 16;
523                             }
524                             Rc = temp;
525                          }});
526
527            format BasicOperateWithNopCheck {
528                0x70: decode RB {
529                    31: ftoit({{ Rc = Fa.uq; }}, FloatCvtOp);
530                }
531                0x78: decode RB {
532                    31: ftois({{ Rc.sl = t_to_s(Fa.uq); }},
533                              FloatCvtOp);
534                }
535            }
536        }
537    }
538
539    // Conditional branches.
540    format CondBranch {
541        0x39: beq({{ cond = (Ra == 0); }});
542        0x3d: bne({{ cond = (Ra != 0); }});
543        0x3e: bge({{ cond = (Ra.sq >= 0); }});
544        0x3f: bgt({{ cond = (Ra.sq >  0); }});
545        0x3b: ble({{ cond = (Ra.sq <= 0); }});
546        0x3a: blt({{ cond = (Ra.sq < 0); }});
547        0x38: blbc({{ cond = ((Ra & 1) == 0); }});
548        0x3c: blbs({{ cond = ((Ra & 1) == 1); }});
549
550        0x31: fbeq({{ cond = (Fa == 0); }});
551        0x35: fbne({{ cond = (Fa != 0); }});
552        0x36: fbge({{ cond = (Fa >= 0); }});
553        0x37: fbgt({{ cond = (Fa >  0); }});
554        0x33: fble({{ cond = (Fa <= 0); }});
555        0x32: fblt({{ cond = (Fa < 0); }});
556    }
557
558    // unconditional branches
559    format UncondBranch {
560        0x30: br();
561        0x34: bsr(IsCall);
562    }
563
564    // indirect branches
565    0x1a: decode JMPFUNC {
566        format Jump {
567            0: jmp();
568            1: jsr(IsCall);
569            2: ret(IsReturn);
570            3: jsr_coroutine(IsCall, IsReturn);
571        }
572    }
573
574    // Square root and integer-to-FP moves
575    0x14: decode FP_SHORTFUNC {
576        // Integer to FP register moves must have RB == 31
577        0x4: decode RB {
578            31: decode FP_FULLFUNC {
579                format BasicOperateWithNopCheck {
580                    0x004: itofs({{ Fc.uq = s_to_t(Ra.ul); }}, FloatCvtOp);
581                    0x024: itoft({{ Fc.uq = Ra.uq; }}, FloatCvtOp);
582                    0x014: FailUnimpl::itoff(); // VAX-format conversion
583                }
584            }
585        }
586
587        // Square root instructions must have FA == 31
588        0xb: decode FA {
589            31: decode FP_TYPEFUNC {
590                format FloatingPointOperate {
591#if SS_COMPATIBLE_FP
592                    0x0b: sqrts({{
593                        if (Fb < 0.0)
594                            fault = new ArithmeticFault;
595                        Fc = sqrt(Fb);
596                    }}, FloatSqrtOp);
597#else
598                    0x0b: sqrts({{
599                        if (Fb.sf < 0.0)
600                            fault = new ArithmeticFault;
601                        Fc.sf = sqrt(Fb.sf);
602                    }}, FloatSqrtOp);
603#endif
604                    0x2b: sqrtt({{
605                        if (Fb < 0.0)
606                            fault = new ArithmeticFault;
607                        Fc = sqrt(Fb);
608                    }}, FloatSqrtOp);
609                }
610            }
611        }
612
613        // VAX-format sqrtf and sqrtg are not implemented
614        0xa: FailUnimpl::sqrtfg();
615    }
616
617    // IEEE floating point
618    0x16: decode FP_SHORTFUNC_TOP2 {
619        // The top two bits of the short function code break this
620        // space into four groups: binary ops, compares, reserved, and
621        // conversions.  See Table 4-12 of AHB.  There are different
622        // special cases in these different groups, so we decode on
623        // these top two bits first just to select a decode strategy.
624        // Most of these instructions may have various trapping and
625        // rounding mode flags set; these are decoded in the
626        // FloatingPointDecode template used by the
627        // FloatingPointOperate format.
628
629        // add/sub/mul/div: just decode on the short function code
630        // and source type.  All valid trapping and rounding modes apply.
631        0: decode FP_TRAPMODE {
632            // check for valid trapping modes here
633            0,1,5,7: decode FP_TYPEFUNC {
634                   format FloatingPointOperate {
635#if SS_COMPATIBLE_FP
636                       0x00: adds({{ Fc = Fa + Fb; }});
637                       0x01: subs({{ Fc = Fa - Fb; }});
638                       0x02: muls({{ Fc = Fa * Fb; }}, FloatMultOp);
639                       0x03: divs({{ Fc = Fa / Fb; }}, FloatDivOp);
640#else
641                       0x00: adds({{ Fc.sf = Fa.sf + Fb.sf; }});
642                       0x01: subs({{ Fc.sf = Fa.sf - Fb.sf; }});
643                       0x02: muls({{ Fc.sf = Fa.sf * Fb.sf; }}, FloatMultOp);
644                       0x03: divs({{ Fc.sf = Fa.sf / Fb.sf; }}, FloatDivOp);
645#endif
646
647                       0x20: addt({{ Fc = Fa + Fb; }});
648                       0x21: subt({{ Fc = Fa - Fb; }});
649                       0x22: mult({{ Fc = Fa * Fb; }}, FloatMultOp);
650                       0x23: divt({{ Fc = Fa / Fb; }}, FloatDivOp);
651                   }
652             }
653        }
654
655        // Floating-point compare instructions must have the default
656        // rounding mode, and may use the default trapping mode or
657        // /SU.  Both trapping modes are treated the same by M5; the
658        // only difference on the real hardware (as far a I can tell)
659        // is that without /SU you'd get an imprecise trap if you
660        // tried to compare a NaN with something else (instead of an
661        // "unordered" result).
662        1: decode FP_FULLFUNC {
663            format BasicOperateWithNopCheck {
664                0x0a5, 0x5a5: cmpteq({{ Fc = (Fa == Fb) ? 2.0 : 0.0; }},
665                                     FloatCmpOp);
666                0x0a7, 0x5a7: cmptle({{ Fc = (Fa <= Fb) ? 2.0 : 0.0; }},
667                                     FloatCmpOp);
668                0x0a6, 0x5a6: cmptlt({{ Fc = (Fa <  Fb) ? 2.0 : 0.0; }},
669                                     FloatCmpOp);
670                0x0a4, 0x5a4: cmptun({{ // unordered
671                    Fc = (!(Fa < Fb) && !(Fa == Fb) && !(Fa > Fb)) ? 2.0 : 0.0;
672                }}, FloatCmpOp);
673            }
674        }
675
676        // The FP-to-integer and integer-to-FP conversion insts
677        // require that FA be 31.
678        3: decode FA {
679            31: decode FP_TYPEFUNC {
680                format FloatingPointOperate {
681                    0x2f: decode FP_ROUNDMODE {
682                        format FPFixedRounding {
683                            // "chopped" i.e. round toward zero
684                            0: cvttq({{ Fc.sq = (int64_t)trunc(Fb); }},
685                                     Chopped);
686                            // round to minus infinity
687                            1: cvttq({{ Fc.sq = (int64_t)floor(Fb); }},
688                                     MinusInfinity);
689                        }
690                      default: cvttq({{ Fc.sq = (int64_t)nearbyint(Fb); }});
691                    }
692
693                    // The cvtts opcode is overloaded to be cvtst if the trap
694                    // mode is 2 or 6 (which are not valid otherwise)
695                    0x2c: decode FP_FULLFUNC {
696                        format BasicOperateWithNopCheck {
697                            // trap on denorm version "cvtst/s" is
698                            // simulated same as cvtst
699                            0x2ac, 0x6ac: cvtst({{ Fc = Fb.sf; }});
700                        }
701                      default: cvtts({{ Fc.sf = Fb; }});
702                    }
703
704                    // The trapping mode for integer-to-FP conversions
705                    // must be /SUI or nothing; /U and /SU are not
706                    // allowed.  The full set of rounding modes are
707                    // supported though.
708                    0x3c: decode FP_TRAPMODE {
709                        0,7: cvtqs({{ Fc.sf = Fb.sq; }});
710                    }
711                    0x3e: decode FP_TRAPMODE {
712                        0,7: cvtqt({{ Fc    = Fb.sq; }});
713                    }
714                }
715            }
716        }
717    }
718
719    // misc FP operate
720    0x17: decode FP_FULLFUNC {
721        format BasicOperateWithNopCheck {
722            0x010: cvtlq({{
723                Fc.sl = (Fb.uq<63:62> << 30) | Fb.uq<58:29>;
724            }});
725            0x030: cvtql({{
726                Fc.uq = (Fb.uq<31:30> << 62) | (Fb.uq<29:0> << 29);
727            }});
728
729            // We treat the precise & imprecise trapping versions of
730            // cvtql identically.
731            0x130, 0x530: cvtqlv({{
732                // To avoid overflow, all the upper 32 bits must match
733                // the sign bit of the lower 32.  We code this as
734                // checking the upper 33 bits for all 0s or all 1s.
735                uint64_t sign_bits = Fb.uq<63:31>;
736                if (sign_bits != 0 && sign_bits != mask(33))
737                    fault = new IntegerOverflowFault;
738                Fc.uq = (Fb.uq<31:30> << 62) | (Fb.uq<29:0> << 29);
739            }});
740
741            0x020: cpys({{  // copy sign
742                Fc.uq = (Fa.uq<63:> << 63) | Fb.uq<62:0>;
743            }});
744            0x021: cpysn({{ // copy sign negated
745                Fc.uq = (~Fa.uq<63:> << 63) | Fb.uq<62:0>;
746            }});
747            0x022: cpyse({{ // copy sign and exponent
748                Fc.uq = (Fa.uq<63:52> << 52) | Fb.uq<51:0>;
749            }});
750
751            0x02a: fcmoveq({{ Fc = (Fa == 0) ? Fb : Fc; }});
752            0x02b: fcmovne({{ Fc = (Fa != 0) ? Fb : Fc; }});
753            0x02c: fcmovlt({{ Fc = (Fa <  0) ? Fb : Fc; }});
754            0x02d: fcmovge({{ Fc = (Fa >= 0) ? Fb : Fc; }});
755            0x02e: fcmovle({{ Fc = (Fa <= 0) ? Fb : Fc; }});
756            0x02f: fcmovgt({{ Fc = (Fa >  0) ? Fb : Fc; }});
757
758            0x024: mt_fpcr({{ FPCR = Fa.uq; }}, IsIprAccess);
759            0x025: mf_fpcr({{ Fa.uq = FPCR; }}, IsIprAccess);
760        }
761    }
762
763    // miscellaneous mem-format ops
764    0x18: decode MEMFUNC {
765        format WarnUnimpl {
766            0x8000: fetch();
767            0xa000: fetch_m();
768            0xe800: ecb();
769        }
770
771        format MiscPrefetch {
772            0xf800: wh64({{ EA = Rb & ~ULL(63); }},
773                         {{ ; }},
774                         mem_flags = PREFETCH);
775        }
776
777        format BasicOperate {
778            0xc000: rpcc({{
779                /* Rb is a fake dependency so here is a fun way to get
780                 * the parser to understand that.
781                 */
782                uint64_t unused_var M5_VAR_USED = Rb;
783                Ra = FULL_SYSTEM ? xc->readMiscReg(IPR_CC) : curTick();
784            }}, IsUnverifiable);
785
786            // All of the barrier instructions below do nothing in
787            // their execute() methods (hence the empty code blocks).
788            // All of their functionality is hard-coded in the
789            // pipeline based on the flags IsSerializing,
790            // IsMemBarrier, and IsWriteBarrier.  In the current
791            // detailed CPU model, the execute() function only gets
792            // called at fetch, so there's no way to generate pipeline
793            // behavior at any other stage.  Once we go to an
794            // exec-in-exec CPU model we should be able to get rid of
795            // these flags and implement this behavior via the
796            // execute() methods.
797
798            // trapb is just a barrier on integer traps, where excb is
799            // a barrier on integer and FP traps.  "EXCB is thus a
800            // superset of TRAPB." (Alpha ARM, Sec 4.11.4) We treat
801            // them the same though.
802            0x0000: trapb({{ }}, IsSerializing, IsSerializeBefore, No_OpClass);
803            0x0400: excb({{ }}, IsSerializing, IsSerializeBefore, No_OpClass);
804            0x4000: mb({{ }}, IsMemBarrier, MemReadOp);
805            0x4400: wmb({{ }}, IsWriteBarrier, MemWriteOp);
806        }
807
808        0xe000: decode FULL_SYSTEM {
809            0: FailUnimpl::rc_se();
810            default: BasicOperate::rc({{
811                Ra = IntrFlag;
812                IntrFlag = 0;
813            }}, IsNonSpeculative, IsUnverifiable);
814        }
815        0xf000: decode FULL_SYSTEM {
816            0: FailUnimpl::rs_se();
817            default: BasicOperate::rs({{
818                Ra = IntrFlag;
819                IntrFlag = 1;
820            }}, IsNonSpeculative, IsUnverifiable);
821        }
822    }
823
824#if FULL_SYSTEM
825    0x00: CallPal::call_pal({{
826        if (!palValid ||
827            (palPriv
828             && xc->readMiscReg(IPR_ICM) != mode_kernel)) {
829            // invalid pal function code, or attempt to do privileged
830            // PAL call in non-kernel mode
831            fault = new UnimplementedOpcodeFault;
832        } else {
833            // check to see if simulator wants to do something special
834            // on this PAL call (including maybe suppress it)
835            bool dopal = xc->simPalCheck(palFunc);
836
837            if (dopal) {
838                xc->setMiscReg(IPR_EXC_ADDR, NPC);
839                NPC = xc->readMiscReg(IPR_PAL_BASE) + palOffset;
840            }
841        }
842    }}, IsNonSpeculative);
843#else
844    0x00: decode PALFUNC {
845        format EmulatedCallPal {
846            0x00: halt ({{
847                exitSimLoop("halt instruction encountered");
848            }}, IsNonSpeculative);
849            0x83: callsys({{
850                xc->syscall(R0);
851            }}, IsSerializeAfter, IsNonSpeculative, IsSyscall);
852            // Read uniq reg into ABI return value register (r0)
853            0x9e: rduniq({{ R0 = Runiq; }}, IsIprAccess);
854            // Write uniq reg with value from ABI arg register (r16)
855            0x9f: wruniq({{ Runiq = R16; }}, IsIprAccess);
856        }
857    }
858#endif
859
860#if FULL_SYSTEM
861    0x1b: decode PALMODE {
862        0: OpcdecFault::hw_st_quad();
863        1: decode HW_LDST_QUAD {
864            format HwLoad {
865                0: hw_ld({{ EA = (Rb + disp) & ~3; }}, {{ Ra = Mem.ul; }},
866                         L, IsSerializing, IsSerializeBefore);
867                1: hw_ld({{ EA = (Rb + disp) & ~7; }}, {{ Ra = Mem.uq; }},
868                         Q, IsSerializing, IsSerializeBefore);
869            }
870        }
871    }
872
873    0x1f: decode PALMODE {
874        0: OpcdecFault::hw_st_cond();
875        format HwStore {
876            1: decode HW_LDST_COND {
877                0: decode HW_LDST_QUAD {
878                    0: hw_st({{ EA = (Rb + disp) & ~3; }},
879                {{ Mem.ul = Ra<31:0>; }}, L, IsSerializing, IsSerializeBefore);
880                    1: hw_st({{ EA = (Rb + disp) & ~7; }},
881                {{ Mem.uq = Ra.uq; }}, Q, IsSerializing, IsSerializeBefore);
882                }
883
884                1: FailUnimpl::hw_st_cond();
885            }
886        }
887    }
888
889    0x19: decode PALMODE {
890        0: OpcdecFault::hw_mfpr();
891        format HwMoveIPR {
892            1: hw_mfpr({{
893                int miscRegIndex = (ipr_index < MaxInternalProcRegs) ?
894                        IprToMiscRegIndex[ipr_index] : -1;
895                if(miscRegIndex < 0 || !IprIsReadable(miscRegIndex) ||
896                    miscRegIndex >= NumInternalProcRegs)
897                        fault = new UnimplementedOpcodeFault;
898                else
899                    Ra = xc->readMiscReg(miscRegIndex);
900            }}, IsIprAccess);
901        }
902    }
903
904    0x1d: decode PALMODE {
905        0: OpcdecFault::hw_mtpr();
906        format HwMoveIPR {
907            1: hw_mtpr({{
908                int miscRegIndex = (ipr_index < MaxInternalProcRegs) ?
909                        IprToMiscRegIndex[ipr_index] : -1;
910                if(miscRegIndex < 0 || !IprIsWritable(miscRegIndex) ||
911                    miscRegIndex >= NumInternalProcRegs)
912                        fault = new UnimplementedOpcodeFault;
913                else
914                    xc->setMiscReg(miscRegIndex, Ra);
915                if (traceData) { traceData->setData(Ra); }
916            }}, IsIprAccess);
917        }
918    }
919
920  0x1e: decode PALMODE {
921      0: OpcdecFault::hw_rei();
922        format BasicOperate {
923          1: hw_rei({{ xc->hwrei(); }}, IsSerializing, IsSerializeBefore);
924        }
925    }
926
927#endif
928
929    format BasicOperate {
930        // M5 special opcodes use the reserved 0x01 opcode space
931        0x01: decode M5FUNC {
932            0x00: arm({{
933                PseudoInst::arm(xc->tcBase());
934            }}, IsNonSpeculative);
935            0x01: quiesce({{
936                PseudoInst::quiesce(xc->tcBase());
937            }}, IsNonSpeculative, IsQuiesce);
938            0x02: quiesceNs({{
939                PseudoInst::quiesceNs(xc->tcBase(), R16);
940            }}, IsNonSpeculative, IsQuiesce);
941            0x03: quiesceCycles({{
942                PseudoInst::quiesceCycles(xc->tcBase(), R16);
943            }}, IsNonSpeculative, IsQuiesce, IsUnverifiable);
944            0x04: quiesceTime({{
945                R0 = PseudoInst::quiesceTime(xc->tcBase());
946            }}, IsNonSpeculative, IsUnverifiable);
947            0x07: rpns({{
948                R0 = PseudoInst::rpns(xc->tcBase());
949            }}, IsNonSpeculative, IsUnverifiable);
950            0x09: wakeCPU({{
951                PseudoInst::wakeCPU(xc->tcBase(), R16);
952            }}, IsNonSpeculative, IsUnverifiable);
953            0x10: deprecated_ivlb({{
954                warn_once("Obsolete M5 ivlb instruction encountered.\n");
955            }});
956            0x11: deprecated_ivle({{
957                warn_once("Obsolete M5 ivlb instruction encountered.\n");
958            }});
959            0x20: deprecated_exit ({{
960                warn_once("deprecated M5 exit instruction encountered.\n");
961                PseudoInst::m5exit(xc->tcBase(), 0);
962            }}, No_OpClass, IsNonSpeculative);
963            0x21: m5exit({{
964                PseudoInst::m5exit(xc->tcBase(), R16);
965            }}, No_OpClass, IsNonSpeculative);
966            0x31: loadsymbol({{
967                PseudoInst::loadsymbol(xc->tcBase());
968            }}, No_OpClass, IsNonSpeculative);
969            0x30: initparam({{
970                Ra = PseudoInst::initParam(xc->tcBase());
971            }});
972            0x40: resetstats({{
973                PseudoInst::resetstats(xc->tcBase(), R16, R17);
974            }}, IsNonSpeculative);
975            0x41: dumpstats({{
976                PseudoInst::dumpstats(xc->tcBase(), R16, R17);
977            }}, IsNonSpeculative);
978            0x42: dumpresetstats({{
979                PseudoInst::dumpresetstats(xc->tcBase(), R16, R17);
980            }}, IsNonSpeculative);
981            0x43: m5checkpoint({{
982                PseudoInst::m5checkpoint(xc->tcBase(), R16, R17);
983            }}, IsNonSpeculative);
984            0x50: m5readfile({{
985                R0 = PseudoInst::readfile(xc->tcBase(), R16, R17, R18);
986            }}, IsNonSpeculative);
987            0x51: m5break({{
988                PseudoInst::debugbreak(xc->tcBase());
989            }}, IsNonSpeculative);
990            0x52: m5switchcpu({{
991                PseudoInst::switchcpu(xc->tcBase());
992            }}, IsNonSpeculative);
993            0x53: m5addsymbol({{
994                PseudoInst::addsymbol(xc->tcBase(), R16, R17);
995            }}, IsNonSpeculative);
996            0x54: m5panic({{
997                panic("M5 panic instruction called at pc = %#x.", PC);
998            }}, IsNonSpeculative);
999#define  CPANN(lbl) CPA::cpa()->lbl(xc->tcBase())
1000            0x55: decode RA {
1001                0x00: m5a_old({{
1002                    panic("Deprecated M5 annotate instruction executed "
1003                          "at pc = %#x\n", PC);
1004                }}, IsNonSpeculative);
1005                0x01: m5a_bsm({{
1006                    CPANN(swSmBegin);
1007                }}, IsNonSpeculative);
1008                0x02: m5a_esm({{
1009                    CPANN(swSmEnd);
1010                }}, IsNonSpeculative);
1011                0x03: m5a_begin({{
1012                    CPANN(swExplictBegin);
1013                }}, IsNonSpeculative);
1014                0x04: m5a_end({{
1015                    CPANN(swEnd);
1016                }}, IsNonSpeculative);
1017                0x06: m5a_q({{
1018                    CPANN(swQ);
1019                }}, IsNonSpeculative);
1020                0x07: m5a_dq({{
1021                    CPANN(swDq);
1022                }}, IsNonSpeculative);
1023                0x08: m5a_wf({{
1024                    CPANN(swWf);
1025                }}, IsNonSpeculative);
1026                0x09: m5a_we({{
1027                    CPANN(swWe);
1028                }}, IsNonSpeculative);
1029                0x0C: m5a_sq({{
1030                    CPANN(swSq);
1031                }}, IsNonSpeculative);
1032                0x0D: m5a_aq({{
1033                    CPANN(swAq);
1034                }}, IsNonSpeculative);
1035                0x0E: m5a_pq({{
1036                    CPANN(swPq);
1037                }}, IsNonSpeculative);
1038                0x0F: m5a_l({{
1039                    CPANN(swLink);
1040                }}, IsNonSpeculative);
1041                0x10: m5a_identify({{
1042                    CPANN(swIdentify);
1043                }}, IsNonSpeculative);
1044                0x11: m5a_getid({{
1045                    R0 = CPANN(swGetId);
1046                }}, IsNonSpeculative);
1047                0x13: m5a_scl({{
1048                    CPANN(swSyscallLink);
1049                }}, IsNonSpeculative);
1050                0x14: m5a_rq({{
1051                    CPANN(swRq);
1052                }}, IsNonSpeculative);
1053            } // M5 Annotate Operations
1054#undef CPANN
1055            0x56: m5reserved2({{
1056                warn("M5 reserved opcode ignored");
1057            }}, IsNonSpeculative);
1058            0x57: m5reserved3({{
1059                warn("M5 reserved opcode ignored");
1060            }}, IsNonSpeculative);
1061            0x58: m5reserved4({{
1062                warn("M5 reserved opcode ignored");
1063            }}, IsNonSpeculative);
1064            0x59: m5reserved5({{
1065                warn("M5 reserved opcode ignored");
1066            }}, IsNonSpeculative);
1067        }
1068    }
1069}
1070