1 | //===---------------- DecoderEmitter.cpp - Decoder Generator --------------===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | // |
9 | // It contains the tablegen backend that emits the decoder functions for |
10 | // targets with fixed/variable length instruction set. |
11 | // |
12 | //===----------------------------------------------------------------------===// |
13 | |
14 | #include "Common/CodeGenHwModes.h" |
15 | #include "Common/CodeGenInstruction.h" |
16 | #include "Common/CodeGenTarget.h" |
17 | #include "Common/InfoByHwMode.h" |
18 | #include "Common/VarLenCodeEmitterGen.h" |
19 | #include "TableGenBackends.h" |
20 | #include "llvm/ADT/APInt.h" |
21 | #include "llvm/ADT/ArrayRef.h" |
22 | #include "llvm/ADT/CachedHashString.h" |
23 | #include "llvm/ADT/STLExtras.h" |
24 | #include "llvm/ADT/SetVector.h" |
25 | #include "llvm/ADT/SmallBitVector.h" |
26 | #include "llvm/ADT/SmallString.h" |
27 | #include "llvm/ADT/Statistic.h" |
28 | #include "llvm/ADT/StringExtras.h" |
29 | #include "llvm/ADT/StringRef.h" |
30 | #include "llvm/MC/MCDecoderOps.h" |
31 | #include "llvm/Support/Casting.h" |
32 | #include "llvm/Support/CommandLine.h" |
33 | #include "llvm/Support/Debug.h" |
34 | #include "llvm/Support/ErrorHandling.h" |
35 | #include "llvm/Support/FormattedStream.h" |
36 | #include "llvm/Support/LEB128.h" |
37 | #include "llvm/Support/raw_ostream.h" |
38 | #include "llvm/TableGen/Error.h" |
39 | #include "llvm/TableGen/Record.h" |
40 | #include <algorithm> |
41 | #include <cassert> |
42 | #include <cstddef> |
43 | #include <cstdint> |
44 | #include <map> |
45 | #include <memory> |
46 | #include <set> |
47 | #include <string> |
48 | #include <utility> |
49 | #include <vector> |
50 | |
51 | using namespace llvm; |
52 | |
53 | #define DEBUG_TYPE "decoder-emitter" |
54 | |
55 | extern cl::OptionCategory DisassemblerEmitterCat; |
56 | |
57 | enum SuppressLevel { |
58 | SUPPRESSION_DISABLE, |
59 | SUPPRESSION_LEVEL1, |
60 | SUPPRESSION_LEVEL2 |
61 | }; |
62 | |
63 | cl::opt<SuppressLevel> DecoderEmitterSuppressDuplicates( |
64 | "suppress-per-hwmode-duplicates" , |
65 | cl::desc("Suppress duplication of instrs into per-HwMode decoder tables" ), |
66 | cl::values( |
67 | clEnumValN( |
68 | SUPPRESSION_DISABLE, "O0" , |
69 | "Do not prevent DecoderTable duplications caused by HwModes" ), |
70 | clEnumValN( |
71 | SUPPRESSION_LEVEL1, "O1" , |
72 | "Remove duplicate DecoderTable entries generated due to HwModes" ), |
73 | clEnumValN( |
74 | SUPPRESSION_LEVEL2, "O2" , |
75 | "Extract HwModes-specific instructions into new DecoderTables, " |
76 | "significantly reducing Table Duplications" )), |
77 | cl::init(Val: SUPPRESSION_DISABLE), cl::cat(DisassemblerEmitterCat)); |
78 | |
79 | namespace { |
80 | |
81 | STATISTIC(NumEncodings, "Number of encodings considered" ); |
82 | STATISTIC(NumEncodingsLackingDisasm, |
83 | "Number of encodings without disassembler info" ); |
84 | STATISTIC(NumInstructions, "Number of instructions considered" ); |
85 | STATISTIC(NumEncodingsSupported, "Number of encodings supported" ); |
86 | STATISTIC(NumEncodingsOmitted, "Number of encodings omitted" ); |
87 | |
88 | struct EncodingField { |
89 | unsigned Base, Width, Offset; |
90 | EncodingField(unsigned B, unsigned W, unsigned O) |
91 | : Base(B), Width(W), Offset(O) {} |
92 | }; |
93 | |
94 | struct OperandInfo { |
95 | std::vector<EncodingField> Fields; |
96 | std::string Decoder; |
97 | bool HasCompleteDecoder; |
98 | uint64_t InitValue; |
99 | |
100 | OperandInfo(std::string D, bool HCD) |
101 | : Decoder(std::move(D)), HasCompleteDecoder(HCD), InitValue(0) {} |
102 | |
103 | void addField(unsigned Base, unsigned Width, unsigned Offset) { |
104 | Fields.push_back(x: EncodingField(Base, Width, Offset)); |
105 | } |
106 | |
107 | unsigned numFields() const { return Fields.size(); } |
108 | |
109 | typedef std::vector<EncodingField>::const_iterator const_iterator; |
110 | |
111 | const_iterator begin() const { return Fields.begin(); } |
112 | const_iterator end() const { return Fields.end(); } |
113 | }; |
114 | |
115 | typedef std::vector<uint8_t> DecoderTable; |
116 | typedef uint32_t DecoderFixup; |
117 | typedef std::vector<DecoderFixup> FixupList; |
118 | typedef std::vector<FixupList> FixupScopeList; |
119 | typedef SmallSetVector<CachedHashString, 16> PredicateSet; |
120 | typedef SmallSetVector<CachedHashString, 16> DecoderSet; |
121 | struct DecoderTableInfo { |
122 | DecoderTable Table; |
123 | FixupScopeList FixupStack; |
124 | PredicateSet Predicates; |
125 | DecoderSet Decoders; |
126 | }; |
127 | |
128 | struct EncodingAndInst { |
129 | const Record *EncodingDef; |
130 | const CodeGenInstruction *Inst; |
131 | StringRef HwModeName; |
132 | |
133 | EncodingAndInst(const Record *EncodingDef, const CodeGenInstruction *Inst, |
134 | StringRef HwModeName = "" ) |
135 | : EncodingDef(EncodingDef), Inst(Inst), HwModeName(HwModeName) {} |
136 | }; |
137 | |
138 | struct EncodingIDAndOpcode { |
139 | unsigned EncodingID; |
140 | unsigned Opcode; |
141 | |
142 | EncodingIDAndOpcode() : EncodingID(0), Opcode(0) {} |
143 | EncodingIDAndOpcode(unsigned EncodingID, unsigned Opcode) |
144 | : EncodingID(EncodingID), Opcode(Opcode) {} |
145 | }; |
146 | |
147 | using EncodingIDsVec = std::vector<EncodingIDAndOpcode>; |
148 | using NamespacesHwModesMap = std::map<std::string, std::set<StringRef>>; |
149 | |
150 | raw_ostream &operator<<(raw_ostream &OS, const EncodingAndInst &Value) { |
151 | if (Value.EncodingDef != Value.Inst->TheDef) |
152 | OS << Value.EncodingDef->getName() << ":" ; |
153 | OS << Value.Inst->TheDef->getName(); |
154 | return OS; |
155 | } |
156 | |
157 | class DecoderEmitter { |
158 | RecordKeeper &RK; |
159 | std::vector<EncodingAndInst> NumberedEncodings; |
160 | |
161 | public: |
162 | DecoderEmitter(RecordKeeper &R, std::string PredicateNamespace) |
163 | : RK(R), Target(R), PredicateNamespace(std::move(PredicateNamespace)) {} |
164 | |
165 | // Emit the decoder state machine table. |
166 | void emitTable(formatted_raw_ostream &o, DecoderTable &Table, |
167 | unsigned Indentation, unsigned BitWidth, StringRef Namespace, |
168 | const EncodingIDsVec &EncodingIDs) const; |
169 | void emitInstrLenTable(formatted_raw_ostream &OS, |
170 | std::vector<unsigned> &InstrLen) const; |
171 | void emitPredicateFunction(formatted_raw_ostream &OS, |
172 | PredicateSet &Predicates, |
173 | unsigned Indentation) const; |
174 | void emitDecoderFunction(formatted_raw_ostream &OS, DecoderSet &Decoders, |
175 | unsigned Indentation) const; |
176 | |
177 | // run - Output the code emitter |
178 | void run(raw_ostream &o); |
179 | |
180 | private: |
181 | CodeGenTarget Target; |
182 | |
183 | public: |
184 | std::string PredicateNamespace; |
185 | }; |
186 | |
187 | } // end anonymous namespace |
188 | |
189 | // The set (BIT_TRUE, BIT_FALSE, BIT_UNSET) represents a ternary logic system |
190 | // for a bit value. |
191 | // |
192 | // BIT_UNFILTERED is used as the init value for a filter position. It is used |
193 | // only for filter processings. |
194 | typedef enum { |
195 | BIT_TRUE, // '1' |
196 | BIT_FALSE, // '0' |
197 | BIT_UNSET, // '?' |
198 | BIT_UNFILTERED // unfiltered |
199 | } bit_value_t; |
200 | |
201 | static bool ValueSet(bit_value_t V) { |
202 | return (V == BIT_TRUE || V == BIT_FALSE); |
203 | } |
204 | |
205 | static bool ValueNotSet(bit_value_t V) { return (V == BIT_UNSET); } |
206 | |
207 | static int Value(bit_value_t V) { |
208 | return ValueNotSet(V) ? -1 : (V == BIT_FALSE ? 0 : 1); |
209 | } |
210 | |
211 | static bit_value_t bitFromBits(const BitsInit &bits, unsigned index) { |
212 | if (BitInit *bit = dyn_cast<BitInit>(Val: bits.getBit(Bit: index))) |
213 | return bit->getValue() ? BIT_TRUE : BIT_FALSE; |
214 | |
215 | // The bit is uninitialized. |
216 | return BIT_UNSET; |
217 | } |
218 | |
219 | // Prints the bit value for each position. |
220 | static void dumpBits(raw_ostream &o, const BitsInit &bits) { |
221 | for (unsigned index = bits.getNumBits(); index > 0; --index) { |
222 | switch (bitFromBits(bits, index: index - 1)) { |
223 | case BIT_TRUE: |
224 | o << "1" ; |
225 | break; |
226 | case BIT_FALSE: |
227 | o << "0" ; |
228 | break; |
229 | case BIT_UNSET: |
230 | o << "_" ; |
231 | break; |
232 | default: |
233 | llvm_unreachable("unexpected return value from bitFromBits" ); |
234 | } |
235 | } |
236 | } |
237 | |
238 | static BitsInit &getBitsField(const Record &def, StringRef str) { |
239 | const RecordVal *RV = def.getValue(Name: str); |
240 | if (BitsInit *Bits = dyn_cast<BitsInit>(Val: RV->getValue())) |
241 | return *Bits; |
242 | |
243 | // variable length instruction |
244 | VarLenInst VLI = VarLenInst(cast<DagInit>(Val: RV->getValue()), RV); |
245 | SmallVector<Init *, 16> Bits; |
246 | |
247 | for (const auto &SI : VLI) { |
248 | if (const BitsInit *BI = dyn_cast<BitsInit>(Val: SI.Value)) { |
249 | for (unsigned Idx = 0U; Idx < BI->getNumBits(); ++Idx) { |
250 | Bits.push_back(Elt: BI->getBit(Bit: Idx)); |
251 | } |
252 | } else if (const BitInit *BI = dyn_cast<BitInit>(Val: SI.Value)) { |
253 | Bits.push_back(Elt: const_cast<BitInit *>(BI)); |
254 | } else { |
255 | for (unsigned Idx = 0U; Idx < SI.BitWidth; ++Idx) |
256 | Bits.push_back(Elt: UnsetInit::get(RK&: def.getRecords())); |
257 | } |
258 | } |
259 | |
260 | return *BitsInit::get(RK&: def.getRecords(), Range: Bits); |
261 | } |
262 | |
263 | // Representation of the instruction to work on. |
264 | typedef std::vector<bit_value_t> insn_t; |
265 | |
266 | namespace { |
267 | |
268 | static const uint64_t NO_FIXED_SEGMENTS_SENTINEL = -1ULL; |
269 | |
270 | class FilterChooser; |
271 | |
272 | /// Filter - Filter works with FilterChooser to produce the decoding tree for |
273 | /// the ISA. |
274 | /// |
275 | /// It is useful to think of a Filter as governing the switch stmts of the |
276 | /// decoding tree in a certain level. Each case stmt delegates to an inferior |
277 | /// FilterChooser to decide what further decoding logic to employ, or in another |
278 | /// words, what other remaining bits to look at. The FilterChooser eventually |
279 | /// chooses a best Filter to do its job. |
280 | /// |
281 | /// This recursive scheme ends when the number of Opcodes assigned to the |
282 | /// FilterChooser becomes 1 or if there is a conflict. A conflict happens when |
283 | /// the Filter/FilterChooser combo does not know how to distinguish among the |
284 | /// Opcodes assigned. |
285 | /// |
286 | /// An example of a conflict is |
287 | /// |
288 | /// Conflict: |
289 | /// 111101000.00........00010000.... |
290 | /// 111101000.00........0001........ |
291 | /// 1111010...00........0001........ |
292 | /// 1111010...00.................... |
293 | /// 1111010......................... |
294 | /// 1111............................ |
295 | /// ................................ |
296 | /// VST4q8a 111101000_00________00010000____ |
297 | /// VST4q8b 111101000_00________00010000____ |
298 | /// |
299 | /// The Debug output shows the path that the decoding tree follows to reach the |
300 | /// the conclusion that there is a conflict. VST4q8a is a vst4 to double-spaced |
301 | /// even registers, while VST4q8b is a vst4 to double-spaced odd registers. |
302 | /// |
303 | /// The encoding info in the .td files does not specify this meta information, |
304 | /// which could have been used by the decoder to resolve the conflict. The |
305 | /// decoder could try to decode the even/odd register numbering and assign to |
306 | /// VST4q8a or VST4q8b, but for the time being, the decoder chooses the "a" |
307 | /// version and return the Opcode since the two have the same Asm format string. |
308 | class Filter { |
309 | protected: |
310 | const FilterChooser |
311 | *Owner; // points to the FilterChooser who owns this filter |
312 | unsigned StartBit; // the starting bit position |
313 | unsigned NumBits; // number of bits to filter |
314 | bool Mixed; // a mixed region contains both set and unset bits |
315 | |
316 | // Map of well-known segment value to the set of uid's with that value. |
317 | std::map<uint64_t, std::vector<EncodingIDAndOpcode>> FilteredInstructions; |
318 | |
319 | // Set of uid's with non-constant segment values. |
320 | std::vector<EncodingIDAndOpcode> VariableInstructions; |
321 | |
322 | // Map of well-known segment value to its delegate. |
323 | std::map<uint64_t, std::unique_ptr<const FilterChooser>> FilterChooserMap; |
324 | |
325 | // Number of instructions which fall under FilteredInstructions category. |
326 | unsigned NumFiltered; |
327 | |
328 | // Keeps track of the last opcode in the filtered bucket. |
329 | EncodingIDAndOpcode LastOpcFiltered; |
330 | |
331 | public: |
332 | Filter(Filter &&f); |
333 | Filter(FilterChooser &owner, unsigned startBit, unsigned numBits, bool mixed); |
334 | |
335 | ~Filter() = default; |
336 | |
337 | unsigned getNumFiltered() const { return NumFiltered; } |
338 | |
339 | EncodingIDAndOpcode getSingletonOpc() const { |
340 | assert(NumFiltered == 1); |
341 | return LastOpcFiltered; |
342 | } |
343 | |
344 | // Return the filter chooser for the group of instructions without constant |
345 | // segment values. |
346 | const FilterChooser &getVariableFC() const { |
347 | assert(NumFiltered == 1); |
348 | assert(FilterChooserMap.size() == 1); |
349 | return *(FilterChooserMap.find(x: NO_FIXED_SEGMENTS_SENTINEL)->second); |
350 | } |
351 | |
352 | // Divides the decoding task into sub tasks and delegates them to the |
353 | // inferior FilterChooser's. |
354 | // |
355 | // A special case arises when there's only one entry in the filtered |
356 | // instructions. In order to unambiguously decode the singleton, we need to |
357 | // match the remaining undecoded encoding bits against the singleton. |
358 | void recurse(); |
359 | |
360 | // Emit table entries to decode instructions given a segment or segments of |
361 | // bits. |
362 | void emitTableEntry(DecoderTableInfo &TableInfo) const; |
363 | |
364 | // Returns the number of fanout produced by the filter. More fanout implies |
365 | // the filter distinguishes more categories of instructions. |
366 | unsigned usefulness() const; |
367 | }; // end class Filter |
368 | |
369 | } // end anonymous namespace |
370 | |
371 | // These are states of our finite state machines used in FilterChooser's |
372 | // filterProcessor() which produces the filter candidates to use. |
373 | typedef enum { |
374 | ATTR_NONE, |
375 | ATTR_FILTERED, |
376 | ATTR_ALL_SET, |
377 | ATTR_ALL_UNSET, |
378 | ATTR_MIXED |
379 | } bitAttr_t; |
380 | |
381 | /// FilterChooser - FilterChooser chooses the best filter among a set of Filters |
382 | /// in order to perform the decoding of instructions at the current level. |
383 | /// |
384 | /// Decoding proceeds from the top down. Based on the well-known encoding bits |
385 | /// of instructions available, FilterChooser builds up the possible Filters that |
386 | /// can further the task of decoding by distinguishing among the remaining |
387 | /// candidate instructions. |
388 | /// |
389 | /// Once a filter has been chosen, it is called upon to divide the decoding task |
390 | /// into sub-tasks and delegates them to its inferior FilterChoosers for further |
391 | /// processings. |
392 | /// |
393 | /// It is useful to think of a Filter as governing the switch stmts of the |
394 | /// decoding tree. And each case is delegated to an inferior FilterChooser to |
395 | /// decide what further remaining bits to look at. |
396 | namespace { |
397 | |
398 | class FilterChooser { |
399 | protected: |
400 | friend class Filter; |
401 | |
402 | // Vector of codegen instructions to choose our filter. |
403 | ArrayRef<EncodingAndInst> AllInstructions; |
404 | |
405 | // Vector of uid's for this filter chooser to work on. |
406 | // The first member of the pair is the opcode id being decoded, the second is |
407 | // the opcode id that should be emitted. |
408 | const std::vector<EncodingIDAndOpcode> &Opcodes; |
409 | |
410 | // Lookup table for the operand decoding of instructions. |
411 | const std::map<unsigned, std::vector<OperandInfo>> &Operands; |
412 | |
413 | // Vector of candidate filters. |
414 | std::vector<Filter> Filters; |
415 | |
416 | // Array of bit values passed down from our parent. |
417 | // Set to all BIT_UNFILTERED's for Parent == NULL. |
418 | std::vector<bit_value_t> FilterBitValues; |
419 | |
420 | // Links to the FilterChooser above us in the decoding tree. |
421 | const FilterChooser *Parent; |
422 | |
423 | // Index of the best filter from Filters. |
424 | int BestIndex; |
425 | |
426 | // Width of instructions |
427 | unsigned BitWidth; |
428 | |
429 | // Parent emitter |
430 | const DecoderEmitter *Emitter; |
431 | |
432 | public: |
433 | FilterChooser(ArrayRef<EncodingAndInst> Insts, |
434 | const std::vector<EncodingIDAndOpcode> &IDs, |
435 | const std::map<unsigned, std::vector<OperandInfo>> &Ops, |
436 | unsigned BW, const DecoderEmitter *E) |
437 | : AllInstructions(Insts), Opcodes(IDs), Operands(Ops), |
438 | FilterBitValues(BW, BIT_UNFILTERED), Parent(nullptr), BestIndex(-1), |
439 | BitWidth(BW), Emitter(E) { |
440 | doFilter(); |
441 | } |
442 | |
443 | FilterChooser(ArrayRef<EncodingAndInst> Insts, |
444 | const std::vector<EncodingIDAndOpcode> &IDs, |
445 | const std::map<unsigned, std::vector<OperandInfo>> &Ops, |
446 | const std::vector<bit_value_t> &ParentFilterBitValues, |
447 | const FilterChooser &parent) |
448 | : AllInstructions(Insts), Opcodes(IDs), Operands(Ops), |
449 | FilterBitValues(ParentFilterBitValues), Parent(&parent), BestIndex(-1), |
450 | BitWidth(parent.BitWidth), Emitter(parent.Emitter) { |
451 | doFilter(); |
452 | } |
453 | |
454 | FilterChooser(const FilterChooser &) = delete; |
455 | void operator=(const FilterChooser &) = delete; |
456 | |
457 | unsigned getBitWidth() const { return BitWidth; } |
458 | |
459 | protected: |
460 | // Populates the insn given the uid. |
461 | void insnWithID(insn_t &Insn, unsigned Opcode) const { |
462 | const Record *EncodingDef = AllInstructions[Opcode].EncodingDef; |
463 | BitsInit &Bits = getBitsField(def: *EncodingDef, str: "Inst" ); |
464 | Insn.resize(new_size: std::max(a: BitWidth, b: Bits.getNumBits()), x: BIT_UNSET); |
465 | // We may have a SoftFail bitmask, which specifies a mask where an encoding |
466 | // may differ from the value in "Inst" and yet still be valid, but the |
467 | // disassembler should return SoftFail instead of Success. |
468 | // |
469 | // This is used for marking UNPREDICTABLE instructions in the ARM world. |
470 | const RecordVal *RV = EncodingDef->getValue(Name: "SoftFail" ); |
471 | const BitsInit *SFBits = RV ? dyn_cast<BitsInit>(Val: RV->getValue()) : nullptr; |
472 | for (unsigned i = 0; i < Bits.getNumBits(); ++i) { |
473 | if (SFBits && bitFromBits(bits: *SFBits, index: i) == BIT_TRUE) |
474 | Insn[i] = BIT_UNSET; |
475 | else |
476 | Insn[i] = bitFromBits(bits: Bits, index: i); |
477 | } |
478 | } |
479 | |
480 | // Emit the name of the encoding/instruction pair. |
481 | void emitNameWithID(raw_ostream &OS, unsigned Opcode) const { |
482 | const Record *EncodingDef = AllInstructions[Opcode].EncodingDef; |
483 | const Record *InstDef = AllInstructions[Opcode].Inst->TheDef; |
484 | if (EncodingDef != InstDef) |
485 | OS << EncodingDef->getName() << ":" ; |
486 | OS << InstDef->getName(); |
487 | } |
488 | |
489 | // Populates the field of the insn given the start position and the number of |
490 | // consecutive bits to scan for. |
491 | // |
492 | // Returns a pair of values (indicator, field), where the indicator is false |
493 | // if there exists any uninitialized bit value in the range and true if all |
494 | // bits are well-known. The second value is the potentially populated field. |
495 | std::pair<bool, uint64_t> fieldFromInsn(const insn_t &Insn, unsigned StartBit, |
496 | unsigned NumBits) const; |
497 | |
498 | /// dumpFilterArray - dumpFilterArray prints out debugging info for the given |
499 | /// filter array as a series of chars. |
500 | void dumpFilterArray(raw_ostream &o, |
501 | const std::vector<bit_value_t> &filter) const; |
502 | |
503 | /// dumpStack - dumpStack traverses the filter chooser chain and calls |
504 | /// dumpFilterArray on each filter chooser up to the top level one. |
505 | void dumpStack(raw_ostream &o, const char *prefix) const; |
506 | |
507 | Filter &bestFilter() { |
508 | assert(BestIndex != -1 && "BestIndex not set" ); |
509 | return Filters[BestIndex]; |
510 | } |
511 | |
512 | bool PositionFiltered(unsigned i) const { |
513 | return ValueSet(V: FilterBitValues[i]); |
514 | } |
515 | |
516 | // Calculates the island(s) needed to decode the instruction. |
517 | // This returns a lit of undecoded bits of an instructions, for example, |
518 | // Inst{20} = 1 && Inst{3-0} == 0b1111 represents two islands of yet-to-be |
519 | // decoded bits in order to verify that the instruction matches the Opcode. |
520 | unsigned getIslands(std::vector<unsigned> &StartBits, |
521 | std::vector<unsigned> &EndBits, |
522 | std::vector<uint64_t> &FieldVals, |
523 | const insn_t &Insn) const; |
524 | |
525 | // Emits code to check the Predicates member of an instruction are true. |
526 | // Returns true if predicate matches were emitted, false otherwise. |
527 | bool emitPredicateMatch(raw_ostream &o, unsigned &Indentation, |
528 | unsigned Opc) const; |
529 | bool emitPredicateMatchAux(const Init &Val, bool ParenIfBinOp, |
530 | raw_ostream &OS) const; |
531 | |
532 | bool doesOpcodeNeedPredicate(unsigned Opc) const; |
533 | unsigned getPredicateIndex(DecoderTableInfo &TableInfo, StringRef P) const; |
534 | void emitPredicateTableEntry(DecoderTableInfo &TableInfo, unsigned Opc) const; |
535 | |
536 | void emitSoftFailTableEntry(DecoderTableInfo &TableInfo, unsigned Opc) const; |
537 | |
538 | // Emits table entries to decode the singleton. |
539 | void emitSingletonTableEntry(DecoderTableInfo &TableInfo, |
540 | EncodingIDAndOpcode Opc) const; |
541 | |
542 | // Emits code to decode the singleton, and then to decode the rest. |
543 | void emitSingletonTableEntry(DecoderTableInfo &TableInfo, |
544 | const Filter &Best) const; |
545 | |
546 | void emitBinaryParser(raw_ostream &o, unsigned &Indentation, |
547 | const OperandInfo &OpInfo, |
548 | bool &OpHasCompleteDecoder) const; |
549 | |
550 | void emitDecoder(raw_ostream &OS, unsigned Indentation, unsigned Opc, |
551 | bool &HasCompleteDecoder) const; |
552 | unsigned getDecoderIndex(DecoderSet &Decoders, unsigned Opc, |
553 | bool &HasCompleteDecoder) const; |
554 | |
555 | // Assign a single filter and run with it. |
556 | void runSingleFilter(unsigned startBit, unsigned numBit, bool mixed); |
557 | |
558 | // reportRegion is a helper function for filterProcessor to mark a region as |
559 | // eligible for use as a filter region. |
560 | void reportRegion(bitAttr_t RA, unsigned StartBit, unsigned BitIndex, |
561 | bool AllowMixed); |
562 | |
563 | // FilterProcessor scans the well-known encoding bits of the instructions and |
564 | // builds up a list of candidate filters. It chooses the best filter and |
565 | // recursively descends down the decoding tree. |
566 | bool filterProcessor(bool AllowMixed, bool Greedy = true); |
567 | |
568 | // Decides on the best configuration of filter(s) to use in order to decode |
569 | // the instructions. A conflict of instructions may occur, in which case we |
570 | // dump the conflict set to the standard error. |
571 | void doFilter(); |
572 | |
573 | public: |
574 | // emitTableEntries - Emit state machine entries to decode our share of |
575 | // instructions. |
576 | void emitTableEntries(DecoderTableInfo &TableInfo) const; |
577 | }; |
578 | |
579 | } // end anonymous namespace |
580 | |
581 | /////////////////////////// |
582 | // // |
583 | // Filter Implementation // |
584 | // // |
585 | /////////////////////////// |
586 | |
587 | Filter::Filter(Filter &&f) |
588 | : Owner(f.Owner), StartBit(f.StartBit), NumBits(f.NumBits), Mixed(f.Mixed), |
589 | FilteredInstructions(std::move(f.FilteredInstructions)), |
590 | VariableInstructions(std::move(f.VariableInstructions)), |
591 | FilterChooserMap(std::move(f.FilterChooserMap)), |
592 | NumFiltered(f.NumFiltered), LastOpcFiltered(f.LastOpcFiltered) {} |
593 | |
594 | Filter::Filter(FilterChooser &owner, unsigned startBit, unsigned numBits, |
595 | bool mixed) |
596 | : Owner(&owner), StartBit(startBit), NumBits(numBits), Mixed(mixed) { |
597 | assert(StartBit + NumBits - 1 < Owner->BitWidth); |
598 | |
599 | NumFiltered = 0; |
600 | LastOpcFiltered = {0, 0}; |
601 | |
602 | for (const auto &OpcPair : Owner->Opcodes) { |
603 | insn_t Insn; |
604 | |
605 | // Populates the insn given the uid. |
606 | Owner->insnWithID(Insn, Opcode: OpcPair.EncodingID); |
607 | |
608 | // Scans the segment for possibly well-specified encoding bits. |
609 | auto [Ok, Field] = Owner->fieldFromInsn(Insn, StartBit, NumBits); |
610 | |
611 | if (Ok) { |
612 | // The encoding bits are well-known. Lets add the uid of the |
613 | // instruction into the bucket keyed off the constant field value. |
614 | LastOpcFiltered = OpcPair; |
615 | FilteredInstructions[Field].push_back(x: LastOpcFiltered); |
616 | ++NumFiltered; |
617 | } else { |
618 | // Some of the encoding bit(s) are unspecified. This contributes to |
619 | // one additional member of "Variable" instructions. |
620 | VariableInstructions.push_back(x: OpcPair); |
621 | } |
622 | } |
623 | |
624 | assert((FilteredInstructions.size() + VariableInstructions.size() > 0) && |
625 | "Filter returns no instruction categories" ); |
626 | } |
627 | |
628 | // Divides the decoding task into sub tasks and delegates them to the |
629 | // inferior FilterChooser's. |
630 | // |
631 | // A special case arises when there's only one entry in the filtered |
632 | // instructions. In order to unambiguously decode the singleton, we need to |
633 | // match the remaining undecoded encoding bits against the singleton. |
634 | void Filter::recurse() { |
635 | // Starts by inheriting our parent filter chooser's filter bit values. |
636 | std::vector<bit_value_t> BitValueArray(Owner->FilterBitValues); |
637 | |
638 | if (!VariableInstructions.empty()) { |
639 | // Conservatively marks each segment position as BIT_UNSET. |
640 | for (unsigned bitIndex = 0; bitIndex < NumBits; ++bitIndex) |
641 | BitValueArray[StartBit + bitIndex] = BIT_UNSET; |
642 | |
643 | // Delegates to an inferior filter chooser for further processing on this |
644 | // group of instructions whose segment values are variable. |
645 | FilterChooserMap.insert(x: std::pair( |
646 | NO_FIXED_SEGMENTS_SENTINEL, |
647 | std::make_unique<FilterChooser>(args: Owner->AllInstructions, |
648 | args&: VariableInstructions, args: Owner->Operands, |
649 | args&: BitValueArray, args: *Owner))); |
650 | } |
651 | |
652 | // No need to recurse for a singleton filtered instruction. |
653 | // See also Filter::emit*(). |
654 | if (getNumFiltered() == 1) { |
655 | assert(FilterChooserMap.size() == 1); |
656 | return; |
657 | } |
658 | |
659 | // Otherwise, create sub choosers. |
660 | for (const auto &Inst : FilteredInstructions) { |
661 | |
662 | // Marks all the segment positions with either BIT_TRUE or BIT_FALSE. |
663 | for (unsigned bitIndex = 0; bitIndex < NumBits; ++bitIndex) { |
664 | if (Inst.first & (1ULL << bitIndex)) |
665 | BitValueArray[StartBit + bitIndex] = BIT_TRUE; |
666 | else |
667 | BitValueArray[StartBit + bitIndex] = BIT_FALSE; |
668 | } |
669 | |
670 | // Delegates to an inferior filter chooser for further processing on this |
671 | // category of instructions. |
672 | FilterChooserMap.insert( |
673 | x: std::pair(Inst.first, std::make_unique<FilterChooser>( |
674 | args: Owner->AllInstructions, args: Inst.second, |
675 | args: Owner->Operands, args&: BitValueArray, args: *Owner))); |
676 | } |
677 | } |
678 | |
679 | static void resolveTableFixups(DecoderTable &Table, const FixupList &Fixups, |
680 | uint32_t DestIdx) { |
681 | // Any NumToSkip fixups in the current scope can resolve to the |
682 | // current location. |
683 | for (FixupList::const_reverse_iterator I = Fixups.rbegin(), E = Fixups.rend(); |
684 | I != E; ++I) { |
685 | // Calculate the distance from the byte following the fixup entry byte |
686 | // to the destination. The Target is calculated from after the 16-bit |
687 | // NumToSkip entry itself, so subtract two from the displacement here |
688 | // to account for that. |
689 | uint32_t FixupIdx = *I; |
690 | uint32_t Delta = DestIdx - FixupIdx - 3; |
691 | // Our NumToSkip entries are 24-bits. Make sure our table isn't too |
692 | // big. |
693 | assert(Delta < (1u << 24)); |
694 | Table[FixupIdx] = (uint8_t)Delta; |
695 | Table[FixupIdx + 1] = (uint8_t)(Delta >> 8); |
696 | Table[FixupIdx + 2] = (uint8_t)(Delta >> 16); |
697 | } |
698 | } |
699 | |
700 | // Emit table entries to decode instructions given a segment or segments |
701 | // of bits. |
702 | void Filter::emitTableEntry(DecoderTableInfo &TableInfo) const { |
703 | assert((NumBits < (1u << 8)) && "NumBits overflowed uint8 table entry!" ); |
704 | TableInfo.Table.push_back(x: MCD::OPC_ExtractField); |
705 | |
706 | SmallString<16> SBytes; |
707 | raw_svector_ostream S(SBytes); |
708 | encodeULEB128(Value: StartBit, OS&: S); |
709 | TableInfo.Table.insert(position: TableInfo.Table.end(), first: SBytes.begin(), last: SBytes.end()); |
710 | TableInfo.Table.push_back(x: NumBits); |
711 | |
712 | // A new filter entry begins a new scope for fixup resolution. |
713 | TableInfo.FixupStack.emplace_back(); |
714 | |
715 | DecoderTable &Table = TableInfo.Table; |
716 | |
717 | size_t PrevFilter = 0; |
718 | bool HasFallthrough = false; |
719 | for (const auto &Filter : FilterChooserMap) { |
720 | // Field value -1 implies a non-empty set of variable instructions. |
721 | // See also recurse(). |
722 | if (Filter.first == NO_FIXED_SEGMENTS_SENTINEL) { |
723 | HasFallthrough = true; |
724 | |
725 | // Each scope should always have at least one filter value to check |
726 | // for. |
727 | assert(PrevFilter != 0 && "empty filter set!" ); |
728 | FixupList &CurScope = TableInfo.FixupStack.back(); |
729 | // Resolve any NumToSkip fixups in the current scope. |
730 | resolveTableFixups(Table, Fixups: CurScope, DestIdx: Table.size()); |
731 | CurScope.clear(); |
732 | PrevFilter = 0; // Don't re-process the filter's fallthrough. |
733 | } else { |
734 | Table.push_back(x: MCD::OPC_FilterValue); |
735 | // Encode and emit the value to filter against. |
736 | uint8_t Buffer[16]; |
737 | unsigned Len = encodeULEB128(Value: Filter.first, p: Buffer); |
738 | Table.insert(position: Table.end(), first: Buffer, last: Buffer + Len); |
739 | // Reserve space for the NumToSkip entry. We'll backpatch the value |
740 | // later. |
741 | PrevFilter = Table.size(); |
742 | Table.push_back(x: 0); |
743 | Table.push_back(x: 0); |
744 | Table.push_back(x: 0); |
745 | } |
746 | |
747 | // We arrive at a category of instructions with the same segment value. |
748 | // Now delegate to the sub filter chooser for further decodings. |
749 | // The case may fallthrough, which happens if the remaining well-known |
750 | // encoding bits do not match exactly. |
751 | Filter.second->emitTableEntries(TableInfo); |
752 | |
753 | // Now that we've emitted the body of the handler, update the NumToSkip |
754 | // of the filter itself to be able to skip forward when false. Subtract |
755 | // two as to account for the width of the NumToSkip field itself. |
756 | if (PrevFilter) { |
757 | uint32_t NumToSkip = Table.size() - PrevFilter - 3; |
758 | assert(NumToSkip < (1u << 24) && |
759 | "disassembler decoding table too large!" ); |
760 | Table[PrevFilter] = (uint8_t)NumToSkip; |
761 | Table[PrevFilter + 1] = (uint8_t)(NumToSkip >> 8); |
762 | Table[PrevFilter + 2] = (uint8_t)(NumToSkip >> 16); |
763 | } |
764 | } |
765 | |
766 | // Any remaining unresolved fixups bubble up to the parent fixup scope. |
767 | assert(TableInfo.FixupStack.size() > 1 && "fixup stack underflow!" ); |
768 | FixupScopeList::iterator Source = TableInfo.FixupStack.end() - 1; |
769 | FixupScopeList::iterator Dest = Source - 1; |
770 | llvm::append_range(C&: *Dest, R&: *Source); |
771 | TableInfo.FixupStack.pop_back(); |
772 | |
773 | // If there is no fallthrough, then the final filter should get fixed |
774 | // up according to the enclosing scope rather than the current position. |
775 | if (!HasFallthrough) |
776 | TableInfo.FixupStack.back().push_back(x: PrevFilter); |
777 | } |
778 | |
779 | // Returns the number of fanout produced by the filter. More fanout implies |
780 | // the filter distinguishes more categories of instructions. |
781 | unsigned Filter::usefulness() const { |
782 | if (!VariableInstructions.empty()) |
783 | return FilteredInstructions.size(); |
784 | else |
785 | return FilteredInstructions.size() + 1; |
786 | } |
787 | |
788 | ////////////////////////////////// |
789 | // // |
790 | // Filterchooser Implementation // |
791 | // // |
792 | ////////////////////////////////// |
793 | |
794 | // Emit the decoder state machine table. |
795 | void DecoderEmitter::emitTable(formatted_raw_ostream &OS, DecoderTable &Table, |
796 | unsigned Indentation, unsigned BitWidth, |
797 | StringRef Namespace, |
798 | const EncodingIDsVec &EncodingIDs) const { |
799 | // We'll need to be able to map from a decoded opcode into the corresponding |
800 | // EncodingID for this specific combination of BitWidth and Namespace. This |
801 | // is used below to index into NumberedEncodings. |
802 | DenseMap<unsigned, unsigned> OpcodeToEncodingID; |
803 | OpcodeToEncodingID.reserve(NumEntries: EncodingIDs.size()); |
804 | for (const auto &EI : EncodingIDs) |
805 | OpcodeToEncodingID[EI.Opcode] = EI.EncodingID; |
806 | |
807 | OS.indent(NumSpaces: Indentation) << "static const uint8_t DecoderTable" << Namespace |
808 | << BitWidth << "[] = {\n" ; |
809 | |
810 | Indentation += 2; |
811 | |
812 | // Emit ULEB128 encoded value to OS, returning the number of bytes emitted. |
813 | auto emitULEB128 = [](DecoderTable::const_iterator I, |
814 | formatted_raw_ostream &OS) { |
815 | unsigned Len = 0; |
816 | while (*I >= 128) { |
817 | OS << (unsigned)*I++ << ", " ; |
818 | Len++; |
819 | } |
820 | OS << (unsigned)*I++ << ", " ; |
821 | return Len + 1; |
822 | }; |
823 | |
824 | // Emit 24-bit numtoskip value to OS, returning the NumToSkip value. |
825 | auto emitNumToSkip = [](DecoderTable::const_iterator I, |
826 | formatted_raw_ostream &OS) { |
827 | uint8_t Byte = *I++; |
828 | uint32_t NumToSkip = Byte; |
829 | OS << (unsigned)Byte << ", " ; |
830 | Byte = *I++; |
831 | OS << (unsigned)Byte << ", " ; |
832 | NumToSkip |= Byte << 8; |
833 | Byte = *I++; |
834 | OS << utostr(X: Byte) << ", " ; |
835 | NumToSkip |= Byte << 16; |
836 | return NumToSkip; |
837 | }; |
838 | |
839 | // FIXME: We may be able to use the NumToSkip values to recover |
840 | // appropriate indentation levels. |
841 | DecoderTable::const_iterator I = Table.begin(); |
842 | DecoderTable::const_iterator E = Table.end(); |
843 | while (I != E) { |
844 | assert(I < E && "incomplete decode table entry!" ); |
845 | |
846 | uint64_t Pos = I - Table.begin(); |
847 | OS << "/* " << Pos << " */" ; |
848 | OS.PadToColumn(NewCol: 12); |
849 | |
850 | switch (*I) { |
851 | default: |
852 | PrintFatalError(Msg: "invalid decode table opcode" ); |
853 | case MCD::OPC_ExtractField: { |
854 | ++I; |
855 | OS.indent(NumSpaces: Indentation) << "MCD::OPC_ExtractField, " ; |
856 | |
857 | // ULEB128 encoded start value. |
858 | const char *ErrMsg = nullptr; |
859 | unsigned Start = decodeULEB128(p: Table.data() + Pos + 1, n: nullptr, |
860 | end: Table.data() + Table.size(), error: &ErrMsg); |
861 | assert(ErrMsg == nullptr && "ULEB128 value too large!" ); |
862 | I += emitULEB128(I, OS); |
863 | |
864 | unsigned Len = *I++; |
865 | OS << Len << ", // Inst{" ; |
866 | if (Len > 1) |
867 | OS << (Start + Len - 1) << "-" ; |
868 | OS << Start << "} ...\n" ; |
869 | break; |
870 | } |
871 | case MCD::OPC_FilterValue: { |
872 | ++I; |
873 | OS.indent(NumSpaces: Indentation) << "MCD::OPC_FilterValue, " ; |
874 | // The filter value is ULEB128 encoded. |
875 | I += emitULEB128(I, OS); |
876 | |
877 | // 24-bit numtoskip value. |
878 | uint32_t NumToSkip = emitNumToSkip(I, OS); |
879 | I += 3; |
880 | OS << "// Skip to: " << ((I - Table.begin()) + NumToSkip) << "\n" ; |
881 | break; |
882 | } |
883 | case MCD::OPC_CheckField: { |
884 | ++I; |
885 | OS.indent(NumSpaces: Indentation) << "MCD::OPC_CheckField, " ; |
886 | // ULEB128 encoded start value. |
887 | I += emitULEB128(I, OS); |
888 | // 8-bit length. |
889 | unsigned Len = *I++; |
890 | OS << Len << ", " ; |
891 | // ULEB128 encoded field value. |
892 | I += emitULEB128(I, OS); |
893 | |
894 | // 24-bit numtoskip value. |
895 | uint32_t NumToSkip = emitNumToSkip(I, OS); |
896 | I += 3; |
897 | OS << "// Skip to: " << ((I - Table.begin()) + NumToSkip) << "\n" ; |
898 | break; |
899 | } |
900 | case MCD::OPC_CheckPredicate: { |
901 | ++I; |
902 | OS.indent(NumSpaces: Indentation) << "MCD::OPC_CheckPredicate, " ; |
903 | I += emitULEB128(I, OS); |
904 | |
905 | // 24-bit numtoskip value. |
906 | uint32_t NumToSkip = emitNumToSkip(I, OS); |
907 | I += 3; |
908 | OS << "// Skip to: " << ((I - Table.begin()) + NumToSkip) << "\n" ; |
909 | break; |
910 | } |
911 | case MCD::OPC_Decode: |
912 | case MCD::OPC_TryDecode: { |
913 | bool IsTry = *I == MCD::OPC_TryDecode; |
914 | ++I; |
915 | // Decode the Opcode value. |
916 | const char *ErrMsg = nullptr; |
917 | unsigned Opc = decodeULEB128(p: Table.data() + Pos + 1, n: nullptr, |
918 | end: Table.data() + Table.size(), error: &ErrMsg); |
919 | assert(ErrMsg == nullptr && "ULEB128 value too large!" ); |
920 | |
921 | OS.indent(NumSpaces: Indentation) |
922 | << "MCD::OPC_" << (IsTry ? "Try" : "" ) << "Decode, " ; |
923 | I += emitULEB128(I, OS); |
924 | |
925 | // Decoder index. |
926 | I += emitULEB128(I, OS); |
927 | |
928 | auto EncI = OpcodeToEncodingID.find(Val: Opc); |
929 | assert(EncI != OpcodeToEncodingID.end() && "no encoding entry" ); |
930 | auto EncodingID = EncI->second; |
931 | |
932 | if (!IsTry) { |
933 | OS << "// Opcode: " << NumberedEncodings[EncodingID] << "\n" ; |
934 | break; |
935 | } |
936 | |
937 | // Fallthrough for OPC_TryDecode. |
938 | |
939 | // 24-bit numtoskip value. |
940 | uint32_t NumToSkip = emitNumToSkip(I, OS); |
941 | I += 3; |
942 | |
943 | OS << "// Opcode: " << NumberedEncodings[EncodingID] |
944 | << ", skip to: " << ((I - Table.begin()) + NumToSkip) << "\n" ; |
945 | break; |
946 | } |
947 | case MCD::OPC_SoftFail: { |
948 | ++I; |
949 | OS.indent(NumSpaces: Indentation) << "MCD::OPC_SoftFail" ; |
950 | // Positive mask |
951 | uint64_t Value = 0; |
952 | unsigned Shift = 0; |
953 | do { |
954 | OS << ", " << (unsigned)*I; |
955 | Value += ((uint64_t)(*I & 0x7f)) << Shift; |
956 | Shift += 7; |
957 | } while (*I++ >= 128); |
958 | if (Value > 127) { |
959 | OS << " /* 0x" ; |
960 | OS.write_hex(N: Value); |
961 | OS << " */" ; |
962 | } |
963 | // Negative mask |
964 | Value = 0; |
965 | Shift = 0; |
966 | do { |
967 | OS << ", " << (unsigned)*I; |
968 | Value += ((uint64_t)(*I & 0x7f)) << Shift; |
969 | Shift += 7; |
970 | } while (*I++ >= 128); |
971 | if (Value > 127) { |
972 | OS << " /* 0x" ; |
973 | OS.write_hex(N: Value); |
974 | OS << " */" ; |
975 | } |
976 | OS << ",\n" ; |
977 | break; |
978 | } |
979 | case MCD::OPC_Fail: { |
980 | ++I; |
981 | OS.indent(NumSpaces: Indentation) << "MCD::OPC_Fail,\n" ; |
982 | break; |
983 | } |
984 | } |
985 | } |
986 | OS.indent(NumSpaces: Indentation) << "0\n" ; |
987 | |
988 | Indentation -= 2; |
989 | |
990 | OS.indent(NumSpaces: Indentation) << "};\n\n" ; |
991 | } |
992 | |
993 | void DecoderEmitter::emitInstrLenTable(formatted_raw_ostream &OS, |
994 | std::vector<unsigned> &InstrLen) const { |
995 | OS << "static const uint8_t InstrLenTable[] = {\n" ; |
996 | for (unsigned &Len : InstrLen) { |
997 | OS << Len << ",\n" ; |
998 | } |
999 | OS << "};\n\n" ; |
1000 | } |
1001 | |
1002 | void DecoderEmitter::emitPredicateFunction(formatted_raw_ostream &OS, |
1003 | PredicateSet &Predicates, |
1004 | unsigned Indentation) const { |
1005 | // The predicate function is just a big switch statement based on the |
1006 | // input predicate index. |
1007 | OS.indent(NumSpaces: Indentation) << "static bool checkDecoderPredicate(unsigned Idx, " |
1008 | << "const FeatureBitset &Bits) {\n" ; |
1009 | Indentation += 2; |
1010 | if (!Predicates.empty()) { |
1011 | OS.indent(NumSpaces: Indentation) << "switch (Idx) {\n" ; |
1012 | OS.indent(NumSpaces: Indentation) |
1013 | << "default: llvm_unreachable(\"Invalid index!\");\n" ; |
1014 | unsigned Index = 0; |
1015 | for (const auto &Predicate : Predicates) { |
1016 | OS.indent(NumSpaces: Indentation) << "case " << Index++ << ":\n" ; |
1017 | OS.indent(NumSpaces: Indentation + 2) << "return (" << Predicate << ");\n" ; |
1018 | } |
1019 | OS.indent(NumSpaces: Indentation) << "}\n" ; |
1020 | } else { |
1021 | // No case statement to emit |
1022 | OS.indent(NumSpaces: Indentation) << "llvm_unreachable(\"Invalid index!\");\n" ; |
1023 | } |
1024 | Indentation -= 2; |
1025 | OS.indent(NumSpaces: Indentation) << "}\n\n" ; |
1026 | } |
1027 | |
1028 | void DecoderEmitter::emitDecoderFunction(formatted_raw_ostream &OS, |
1029 | DecoderSet &Decoders, |
1030 | unsigned Indentation) const { |
1031 | // The decoder function is just a big switch statement based on the |
1032 | // input decoder index. |
1033 | OS.indent(NumSpaces: Indentation) << "template <typename InsnType>\n" ; |
1034 | OS.indent(NumSpaces: Indentation) << "static DecodeStatus decodeToMCInst(DecodeStatus S," |
1035 | << " unsigned Idx, InsnType insn, MCInst &MI,\n" ; |
1036 | OS.indent(NumSpaces: Indentation) |
1037 | << " uint64_t " |
1038 | << "Address, const MCDisassembler *Decoder, bool &DecodeComplete) {\n" ; |
1039 | Indentation += 2; |
1040 | OS.indent(NumSpaces: Indentation) << "DecodeComplete = true;\n" ; |
1041 | // TODO: When InsnType is large, using uint64_t limits all fields to 64 bits |
1042 | // It would be better for emitBinaryParser to use a 64-bit tmp whenever |
1043 | // possible but fall back to an InsnType-sized tmp for truly large fields. |
1044 | OS.indent(NumSpaces: Indentation) << "using TmpType = " |
1045 | "std::conditional_t<std::is_integral<InsnType>::" |
1046 | "value, InsnType, uint64_t>;\n" ; |
1047 | OS.indent(NumSpaces: Indentation) << "TmpType tmp;\n" ; |
1048 | OS.indent(NumSpaces: Indentation) << "switch (Idx) {\n" ; |
1049 | OS.indent(NumSpaces: Indentation) << "default: llvm_unreachable(\"Invalid index!\");\n" ; |
1050 | unsigned Index = 0; |
1051 | for (const auto &Decoder : Decoders) { |
1052 | OS.indent(NumSpaces: Indentation) << "case " << Index++ << ":\n" ; |
1053 | OS << Decoder; |
1054 | OS.indent(NumSpaces: Indentation + 2) << "return S;\n" ; |
1055 | } |
1056 | OS.indent(NumSpaces: Indentation) << "}\n" ; |
1057 | Indentation -= 2; |
1058 | OS.indent(NumSpaces: Indentation) << "}\n" ; |
1059 | } |
1060 | |
1061 | // Populates the field of the insn given the start position and the number of |
1062 | // consecutive bits to scan for. |
1063 | // |
1064 | // Returns a pair of values (indicator, field), where the indicator is false |
1065 | // if there exists any uninitialized bit value in the range and true if all |
1066 | // bits are well-known. The second value is the potentially populated field. |
1067 | std::pair<bool, uint64_t> FilterChooser::fieldFromInsn(const insn_t &Insn, |
1068 | unsigned StartBit, |
1069 | unsigned NumBits) const { |
1070 | uint64_t Field = 0; |
1071 | |
1072 | for (unsigned i = 0; i < NumBits; ++i) { |
1073 | if (Insn[StartBit + i] == BIT_UNSET) |
1074 | return {false, Field}; |
1075 | |
1076 | if (Insn[StartBit + i] == BIT_TRUE) |
1077 | Field = Field | (1ULL << i); |
1078 | } |
1079 | |
1080 | return {true, Field}; |
1081 | } |
1082 | |
1083 | /// dumpFilterArray - dumpFilterArray prints out debugging info for the given |
1084 | /// filter array as a series of chars. |
1085 | void FilterChooser::dumpFilterArray( |
1086 | raw_ostream &o, const std::vector<bit_value_t> &filter) const { |
1087 | for (unsigned bitIndex = BitWidth; bitIndex > 0; bitIndex--) { |
1088 | switch (filter[bitIndex - 1]) { |
1089 | case BIT_UNFILTERED: |
1090 | o << "." ; |
1091 | break; |
1092 | case BIT_UNSET: |
1093 | o << "_" ; |
1094 | break; |
1095 | case BIT_TRUE: |
1096 | o << "1" ; |
1097 | break; |
1098 | case BIT_FALSE: |
1099 | o << "0" ; |
1100 | break; |
1101 | } |
1102 | } |
1103 | } |
1104 | |
1105 | /// dumpStack - dumpStack traverses the filter chooser chain and calls |
1106 | /// dumpFilterArray on each filter chooser up to the top level one. |
1107 | void FilterChooser::dumpStack(raw_ostream &o, const char *prefix) const { |
1108 | const FilterChooser *current = this; |
1109 | |
1110 | while (current) { |
1111 | o << prefix; |
1112 | dumpFilterArray(o, filter: current->FilterBitValues); |
1113 | o << '\n'; |
1114 | current = current->Parent; |
1115 | } |
1116 | } |
1117 | |
1118 | // Calculates the island(s) needed to decode the instruction. |
1119 | // This returns a list of undecoded bits of an instructions, for example, |
1120 | // Inst{20} = 1 && Inst{3-0} == 0b1111 represents two islands of yet-to-be |
1121 | // decoded bits in order to verify that the instruction matches the Opcode. |
1122 | unsigned FilterChooser::getIslands(std::vector<unsigned> &StartBits, |
1123 | std::vector<unsigned> &EndBits, |
1124 | std::vector<uint64_t> &FieldVals, |
1125 | const insn_t &Insn) const { |
1126 | unsigned Num, BitNo; |
1127 | Num = BitNo = 0; |
1128 | |
1129 | uint64_t FieldVal = 0; |
1130 | |
1131 | // 0: Init |
1132 | // 1: Water (the bit value does not affect decoding) |
1133 | // 2: Island (well-known bit value needed for decoding) |
1134 | int State = 0; |
1135 | |
1136 | for (unsigned i = 0; i < BitWidth; ++i) { |
1137 | int64_t Val = Value(V: Insn[i]); |
1138 | bool Filtered = PositionFiltered(i); |
1139 | switch (State) { |
1140 | default: |
1141 | llvm_unreachable("Unreachable code!" ); |
1142 | case 0: |
1143 | case 1: |
1144 | if (Filtered || Val == -1) |
1145 | State = 1; // Still in Water |
1146 | else { |
1147 | State = 2; // Into the Island |
1148 | BitNo = 0; |
1149 | StartBits.push_back(x: i); |
1150 | FieldVal = Val; |
1151 | } |
1152 | break; |
1153 | case 2: |
1154 | if (Filtered || Val == -1) { |
1155 | State = 1; // Into the Water |
1156 | EndBits.push_back(x: i - 1); |
1157 | FieldVals.push_back(x: FieldVal); |
1158 | ++Num; |
1159 | } else { |
1160 | State = 2; // Still in Island |
1161 | ++BitNo; |
1162 | FieldVal = FieldVal | Val << BitNo; |
1163 | } |
1164 | break; |
1165 | } |
1166 | } |
1167 | // If we are still in Island after the loop, do some housekeeping. |
1168 | if (State == 2) { |
1169 | EndBits.push_back(x: BitWidth - 1); |
1170 | FieldVals.push_back(x: FieldVal); |
1171 | ++Num; |
1172 | } |
1173 | |
1174 | assert(StartBits.size() == Num && EndBits.size() == Num && |
1175 | FieldVals.size() == Num); |
1176 | return Num; |
1177 | } |
1178 | |
1179 | void FilterChooser::emitBinaryParser(raw_ostream &o, unsigned &Indentation, |
1180 | const OperandInfo &OpInfo, |
1181 | bool &OpHasCompleteDecoder) const { |
1182 | const std::string &Decoder = OpInfo.Decoder; |
1183 | |
1184 | bool UseInsertBits = OpInfo.numFields() != 1 || OpInfo.InitValue != 0; |
1185 | |
1186 | if (UseInsertBits) { |
1187 | o.indent(NumSpaces: Indentation) << "tmp = 0x" ; |
1188 | o.write_hex(N: OpInfo.InitValue); |
1189 | o << ";\n" ; |
1190 | } |
1191 | |
1192 | for (const EncodingField &EF : OpInfo) { |
1193 | o.indent(NumSpaces: Indentation); |
1194 | if (UseInsertBits) |
1195 | o << "insertBits(tmp, " ; |
1196 | else |
1197 | o << "tmp = " ; |
1198 | o << "fieldFromInstruction(insn, " << EF.Base << ", " << EF.Width << ')'; |
1199 | if (UseInsertBits) |
1200 | o << ", " << EF.Offset << ", " << EF.Width << ')'; |
1201 | else if (EF.Offset != 0) |
1202 | o << " << " << EF.Offset; |
1203 | o << ";\n" ; |
1204 | } |
1205 | |
1206 | if (Decoder != "" ) { |
1207 | OpHasCompleteDecoder = OpInfo.HasCompleteDecoder; |
1208 | o.indent(NumSpaces: Indentation) << "if (!Check(S, " << Decoder |
1209 | << "(MI, tmp, Address, Decoder))) { " |
1210 | << (OpHasCompleteDecoder ? "" |
1211 | : "DecodeComplete = false; " ) |
1212 | << "return MCDisassembler::Fail; }\n" ; |
1213 | } else { |
1214 | OpHasCompleteDecoder = true; |
1215 | o.indent(NumSpaces: Indentation) << "MI.addOperand(MCOperand::createImm(tmp));\n" ; |
1216 | } |
1217 | } |
1218 | |
1219 | void FilterChooser::emitDecoder(raw_ostream &OS, unsigned Indentation, |
1220 | unsigned Opc, bool &HasCompleteDecoder) const { |
1221 | HasCompleteDecoder = true; |
1222 | |
1223 | for (const auto &Op : Operands.find(x: Opc)->second) { |
1224 | // If a custom instruction decoder was specified, use that. |
1225 | if (Op.numFields() == 0 && !Op.Decoder.empty()) { |
1226 | HasCompleteDecoder = Op.HasCompleteDecoder; |
1227 | OS.indent(NumSpaces: Indentation) |
1228 | << "if (!Check(S, " << Op.Decoder |
1229 | << "(MI, insn, Address, Decoder))) { " |
1230 | << (HasCompleteDecoder ? "" : "DecodeComplete = false; " ) |
1231 | << "return MCDisassembler::Fail; }\n" ; |
1232 | break; |
1233 | } |
1234 | |
1235 | bool OpHasCompleteDecoder; |
1236 | emitBinaryParser(o&: OS, Indentation, OpInfo: Op, OpHasCompleteDecoder); |
1237 | if (!OpHasCompleteDecoder) |
1238 | HasCompleteDecoder = false; |
1239 | } |
1240 | } |
1241 | |
1242 | unsigned FilterChooser::getDecoderIndex(DecoderSet &Decoders, unsigned Opc, |
1243 | bool &HasCompleteDecoder) const { |
1244 | // Build up the predicate string. |
1245 | SmallString<256> Decoder; |
1246 | // FIXME: emitDecoder() function can take a buffer directly rather than |
1247 | // a stream. |
1248 | raw_svector_ostream S(Decoder); |
1249 | unsigned I = 4; |
1250 | emitDecoder(OS&: S, Indentation: I, Opc, HasCompleteDecoder); |
1251 | |
1252 | // Using the full decoder string as the key value here is a bit |
1253 | // heavyweight, but is effective. If the string comparisons become a |
1254 | // performance concern, we can implement a mangling of the predicate |
1255 | // data easily enough with a map back to the actual string. That's |
1256 | // overkill for now, though. |
1257 | |
1258 | // Make sure the predicate is in the table. |
1259 | Decoders.insert(X: CachedHashString(Decoder)); |
1260 | // Now figure out the index for when we write out the table. |
1261 | DecoderSet::const_iterator P = find(Range&: Decoders, Val: Decoder.str()); |
1262 | return (unsigned)(P - Decoders.begin()); |
1263 | } |
1264 | |
1265 | // If ParenIfBinOp is true, print a surrounding () if Val uses && or ||. |
1266 | bool FilterChooser::emitPredicateMatchAux(const Init &Val, bool ParenIfBinOp, |
1267 | raw_ostream &OS) const { |
1268 | if (const auto *D = dyn_cast<DefInit>(Val: &Val)) { |
1269 | if (!D->getDef()->isSubClassOf(Name: "SubtargetFeature" )) |
1270 | return true; |
1271 | OS << "Bits[" << Emitter->PredicateNamespace << "::" << D->getAsString() |
1272 | << "]" ; |
1273 | return false; |
1274 | } |
1275 | if (const auto *D = dyn_cast<DagInit>(Val: &Val)) { |
1276 | std::string Op = D->getOperator()->getAsString(); |
1277 | if (Op == "not" && D->getNumArgs() == 1) { |
1278 | OS << '!'; |
1279 | return emitPredicateMatchAux(Val: *D->getArg(Num: 0), ParenIfBinOp: true, OS); |
1280 | } |
1281 | if ((Op == "any_of" || Op == "all_of" ) && D->getNumArgs() > 0) { |
1282 | bool Paren = D->getNumArgs() > 1 && std::exchange(obj&: ParenIfBinOp, new_val: true); |
1283 | if (Paren) |
1284 | OS << '('; |
1285 | ListSeparator LS(Op == "any_of" ? " || " : " && " ); |
1286 | for (auto *Arg : D->getArgs()) { |
1287 | OS << LS; |
1288 | if (emitPredicateMatchAux(Val: *Arg, ParenIfBinOp, OS)) |
1289 | return true; |
1290 | } |
1291 | if (Paren) |
1292 | OS << ')'; |
1293 | return false; |
1294 | } |
1295 | } |
1296 | return true; |
1297 | } |
1298 | |
1299 | bool FilterChooser::emitPredicateMatch(raw_ostream &o, unsigned &Indentation, |
1300 | unsigned Opc) const { |
1301 | ListInit *Predicates = |
1302 | AllInstructions[Opc].EncodingDef->getValueAsListInit(FieldName: "Predicates" ); |
1303 | bool IsFirstEmission = true; |
1304 | for (unsigned i = 0; i < Predicates->size(); ++i) { |
1305 | Record *Pred = Predicates->getElementAsRecord(i); |
1306 | if (!Pred->getValue(Name: "AssemblerMatcherPredicate" )) |
1307 | continue; |
1308 | |
1309 | if (!isa<DagInit>(Val: Pred->getValue(Name: "AssemblerCondDag" )->getValue())) |
1310 | continue; |
1311 | |
1312 | if (!IsFirstEmission) |
1313 | o << " && " ; |
1314 | if (emitPredicateMatchAux(Val: *Pred->getValueAsDag(FieldName: "AssemblerCondDag" ), |
1315 | ParenIfBinOp: Predicates->size() > 1, OS&: o)) |
1316 | PrintFatalError(ErrorLoc: Pred->getLoc(), Msg: "Invalid AssemblerCondDag!" ); |
1317 | IsFirstEmission = false; |
1318 | } |
1319 | return !Predicates->empty(); |
1320 | } |
1321 | |
1322 | bool FilterChooser::doesOpcodeNeedPredicate(unsigned Opc) const { |
1323 | ListInit *Predicates = |
1324 | AllInstructions[Opc].EncodingDef->getValueAsListInit(FieldName: "Predicates" ); |
1325 | for (unsigned i = 0; i < Predicates->size(); ++i) { |
1326 | Record *Pred = Predicates->getElementAsRecord(i); |
1327 | if (!Pred->getValue(Name: "AssemblerMatcherPredicate" )) |
1328 | continue; |
1329 | |
1330 | if (isa<DagInit>(Val: Pred->getValue(Name: "AssemblerCondDag" )->getValue())) |
1331 | return true; |
1332 | } |
1333 | return false; |
1334 | } |
1335 | |
1336 | unsigned FilterChooser::getPredicateIndex(DecoderTableInfo &TableInfo, |
1337 | StringRef Predicate) const { |
1338 | // Using the full predicate string as the key value here is a bit |
1339 | // heavyweight, but is effective. If the string comparisons become a |
1340 | // performance concern, we can implement a mangling of the predicate |
1341 | // data easily enough with a map back to the actual string. That's |
1342 | // overkill for now, though. |
1343 | |
1344 | // Make sure the predicate is in the table. |
1345 | TableInfo.Predicates.insert(X: CachedHashString(Predicate)); |
1346 | // Now figure out the index for when we write out the table. |
1347 | PredicateSet::const_iterator P = find(Range&: TableInfo.Predicates, Val: Predicate); |
1348 | return (unsigned)(P - TableInfo.Predicates.begin()); |
1349 | } |
1350 | |
1351 | void FilterChooser::emitPredicateTableEntry(DecoderTableInfo &TableInfo, |
1352 | unsigned Opc) const { |
1353 | if (!doesOpcodeNeedPredicate(Opc)) |
1354 | return; |
1355 | |
1356 | // Build up the predicate string. |
1357 | SmallString<256> Predicate; |
1358 | // FIXME: emitPredicateMatch() functions can take a buffer directly rather |
1359 | // than a stream. |
1360 | raw_svector_ostream PS(Predicate); |
1361 | unsigned I = 0; |
1362 | emitPredicateMatch(o&: PS, Indentation&: I, Opc); |
1363 | |
1364 | // Figure out the index into the predicate table for the predicate just |
1365 | // computed. |
1366 | unsigned PIdx = getPredicateIndex(TableInfo, Predicate: PS.str()); |
1367 | SmallString<16> PBytes; |
1368 | raw_svector_ostream S(PBytes); |
1369 | encodeULEB128(Value: PIdx, OS&: S); |
1370 | |
1371 | TableInfo.Table.push_back(x: MCD::OPC_CheckPredicate); |
1372 | // Predicate index. |
1373 | for (const auto PB : PBytes) |
1374 | TableInfo.Table.push_back(x: PB); |
1375 | // Push location for NumToSkip backpatching. |
1376 | TableInfo.FixupStack.back().push_back(x: TableInfo.Table.size()); |
1377 | TableInfo.Table.push_back(x: 0); |
1378 | TableInfo.Table.push_back(x: 0); |
1379 | TableInfo.Table.push_back(x: 0); |
1380 | } |
1381 | |
1382 | void FilterChooser::emitSoftFailTableEntry(DecoderTableInfo &TableInfo, |
1383 | unsigned Opc) const { |
1384 | const Record *EncodingDef = AllInstructions[Opc].EncodingDef; |
1385 | const RecordVal *RV = EncodingDef->getValue(Name: "SoftFail" ); |
1386 | BitsInit *SFBits = RV ? dyn_cast<BitsInit>(Val: RV->getValue()) : nullptr; |
1387 | |
1388 | if (!SFBits) |
1389 | return; |
1390 | BitsInit *InstBits = EncodingDef->getValueAsBitsInit(FieldName: "Inst" ); |
1391 | |
1392 | APInt PositiveMask(BitWidth, 0ULL); |
1393 | APInt NegativeMask(BitWidth, 0ULL); |
1394 | for (unsigned i = 0; i < BitWidth; ++i) { |
1395 | bit_value_t B = bitFromBits(bits: *SFBits, index: i); |
1396 | bit_value_t IB = bitFromBits(bits: *InstBits, index: i); |
1397 | |
1398 | if (B != BIT_TRUE) |
1399 | continue; |
1400 | |
1401 | switch (IB) { |
1402 | case BIT_FALSE: |
1403 | // The bit is meant to be false, so emit a check to see if it is true. |
1404 | PositiveMask.setBit(i); |
1405 | break; |
1406 | case BIT_TRUE: |
1407 | // The bit is meant to be true, so emit a check to see if it is false. |
1408 | NegativeMask.setBit(i); |
1409 | break; |
1410 | default: |
1411 | // The bit is not set; this must be an error! |
1412 | errs() << "SoftFail Conflict: bit SoftFail{" << i << "} in " |
1413 | << AllInstructions[Opc] << " is set but Inst{" << i |
1414 | << "} is unset!\n" |
1415 | << " - You can only mark a bit as SoftFail if it is fully defined" |
1416 | << " (1/0 - not '?') in Inst\n" ; |
1417 | return; |
1418 | } |
1419 | } |
1420 | |
1421 | bool NeedPositiveMask = PositiveMask.getBoolValue(); |
1422 | bool NeedNegativeMask = NegativeMask.getBoolValue(); |
1423 | |
1424 | if (!NeedPositiveMask && !NeedNegativeMask) |
1425 | return; |
1426 | |
1427 | TableInfo.Table.push_back(x: MCD::OPC_SoftFail); |
1428 | |
1429 | SmallString<16> MaskBytes; |
1430 | raw_svector_ostream S(MaskBytes); |
1431 | if (NeedPositiveMask) { |
1432 | encodeULEB128(Value: PositiveMask.getZExtValue(), OS&: S); |
1433 | for (unsigned i = 0, e = MaskBytes.size(); i != e; ++i) |
1434 | TableInfo.Table.push_back(x: MaskBytes[i]); |
1435 | } else |
1436 | TableInfo.Table.push_back(x: 0); |
1437 | if (NeedNegativeMask) { |
1438 | MaskBytes.clear(); |
1439 | encodeULEB128(Value: NegativeMask.getZExtValue(), OS&: S); |
1440 | for (unsigned i = 0, e = MaskBytes.size(); i != e; ++i) |
1441 | TableInfo.Table.push_back(x: MaskBytes[i]); |
1442 | } else |
1443 | TableInfo.Table.push_back(x: 0); |
1444 | } |
1445 | |
1446 | // Emits table entries to decode the singleton. |
1447 | void FilterChooser::emitSingletonTableEntry(DecoderTableInfo &TableInfo, |
1448 | EncodingIDAndOpcode Opc) const { |
1449 | std::vector<unsigned> StartBits; |
1450 | std::vector<unsigned> EndBits; |
1451 | std::vector<uint64_t> FieldVals; |
1452 | insn_t Insn; |
1453 | insnWithID(Insn, Opcode: Opc.EncodingID); |
1454 | |
1455 | // Look for islands of undecoded bits of the singleton. |
1456 | getIslands(StartBits, EndBits, FieldVals, Insn); |
1457 | |
1458 | unsigned Size = StartBits.size(); |
1459 | |
1460 | // Emit the predicate table entry if one is needed. |
1461 | emitPredicateTableEntry(TableInfo, Opc: Opc.EncodingID); |
1462 | |
1463 | // Check any additional encoding fields needed. |
1464 | for (unsigned I = Size; I != 0; --I) { |
1465 | unsigned NumBits = EndBits[I - 1] - StartBits[I - 1] + 1; |
1466 | assert((NumBits < (1u << 8)) && "NumBits overflowed uint8 table entry!" ); |
1467 | TableInfo.Table.push_back(x: MCD::OPC_CheckField); |
1468 | uint8_t Buffer[16], *P; |
1469 | encodeULEB128(Value: StartBits[I - 1], p: Buffer); |
1470 | for (P = Buffer; *P >= 128; ++P) |
1471 | TableInfo.Table.push_back(x: *P); |
1472 | TableInfo.Table.push_back(x: *P); |
1473 | TableInfo.Table.push_back(x: NumBits); |
1474 | encodeULEB128(Value: FieldVals[I - 1], p: Buffer); |
1475 | for (P = Buffer; *P >= 128; ++P) |
1476 | TableInfo.Table.push_back(x: *P); |
1477 | TableInfo.Table.push_back(x: *P); |
1478 | // Push location for NumToSkip backpatching. |
1479 | TableInfo.FixupStack.back().push_back(x: TableInfo.Table.size()); |
1480 | // The fixup is always 24-bits, so go ahead and allocate the space |
1481 | // in the table so all our relative position calculations work OK even |
1482 | // before we fully resolve the real value here. |
1483 | TableInfo.Table.push_back(x: 0); |
1484 | TableInfo.Table.push_back(x: 0); |
1485 | TableInfo.Table.push_back(x: 0); |
1486 | } |
1487 | |
1488 | // Check for soft failure of the match. |
1489 | emitSoftFailTableEntry(TableInfo, Opc: Opc.EncodingID); |
1490 | |
1491 | bool HasCompleteDecoder; |
1492 | unsigned DIdx = |
1493 | getDecoderIndex(Decoders&: TableInfo.Decoders, Opc: Opc.EncodingID, HasCompleteDecoder); |
1494 | |
1495 | // Produce OPC_Decode or OPC_TryDecode opcode based on the information |
1496 | // whether the instruction decoder is complete or not. If it is complete |
1497 | // then it handles all possible values of remaining variable/unfiltered bits |
1498 | // and for any value can determine if the bitpattern is a valid instruction |
1499 | // or not. This means OPC_Decode will be the final step in the decoding |
1500 | // process. If it is not complete, then the Fail return code from the |
1501 | // decoder method indicates that additional processing should be done to see |
1502 | // if there is any other instruction that also matches the bitpattern and |
1503 | // can decode it. |
1504 | TableInfo.Table.push_back(x: HasCompleteDecoder ? MCD::OPC_Decode |
1505 | : MCD::OPC_TryDecode); |
1506 | NumEncodingsSupported++; |
1507 | uint8_t Buffer[16], *p; |
1508 | encodeULEB128(Value: Opc.Opcode, p: Buffer); |
1509 | for (p = Buffer; *p >= 128; ++p) |
1510 | TableInfo.Table.push_back(x: *p); |
1511 | TableInfo.Table.push_back(x: *p); |
1512 | |
1513 | SmallString<16> Bytes; |
1514 | raw_svector_ostream S(Bytes); |
1515 | encodeULEB128(Value: DIdx, OS&: S); |
1516 | |
1517 | // Decoder index. |
1518 | for (const auto B : Bytes) |
1519 | TableInfo.Table.push_back(x: B); |
1520 | |
1521 | if (!HasCompleteDecoder) { |
1522 | // Push location for NumToSkip backpatching. |
1523 | TableInfo.FixupStack.back().push_back(x: TableInfo.Table.size()); |
1524 | // Allocate the space for the fixup. |
1525 | TableInfo.Table.push_back(x: 0); |
1526 | TableInfo.Table.push_back(x: 0); |
1527 | TableInfo.Table.push_back(x: 0); |
1528 | } |
1529 | } |
1530 | |
1531 | // Emits table entries to decode the singleton, and then to decode the rest. |
1532 | void FilterChooser::emitSingletonTableEntry(DecoderTableInfo &TableInfo, |
1533 | const Filter &Best) const { |
1534 | EncodingIDAndOpcode Opc = Best.getSingletonOpc(); |
1535 | |
1536 | // complex singletons need predicate checks from the first singleton |
1537 | // to refer forward to the variable filterchooser that follows. |
1538 | TableInfo.FixupStack.emplace_back(); |
1539 | |
1540 | emitSingletonTableEntry(TableInfo, Opc); |
1541 | |
1542 | resolveTableFixups(Table&: TableInfo.Table, Fixups: TableInfo.FixupStack.back(), |
1543 | DestIdx: TableInfo.Table.size()); |
1544 | TableInfo.FixupStack.pop_back(); |
1545 | |
1546 | Best.getVariableFC().emitTableEntries(TableInfo); |
1547 | } |
1548 | |
1549 | // Assign a single filter and run with it. Top level API client can initialize |
1550 | // with a single filter to start the filtering process. |
1551 | void FilterChooser::runSingleFilter(unsigned startBit, unsigned numBit, |
1552 | bool mixed) { |
1553 | Filters.clear(); |
1554 | Filters.emplace_back(args&: *this, args&: startBit, args&: numBit, args: true); |
1555 | BestIndex = 0; // Sole Filter instance to choose from. |
1556 | bestFilter().recurse(); |
1557 | } |
1558 | |
1559 | // reportRegion is a helper function for filterProcessor to mark a region as |
1560 | // eligible for use as a filter region. |
1561 | void FilterChooser::reportRegion(bitAttr_t RA, unsigned StartBit, |
1562 | unsigned BitIndex, bool AllowMixed) { |
1563 | if (RA == ATTR_MIXED && AllowMixed) |
1564 | Filters.emplace_back(args&: *this, args&: StartBit, args: BitIndex - StartBit, args: true); |
1565 | else if (RA == ATTR_ALL_SET && !AllowMixed) |
1566 | Filters.emplace_back(args&: *this, args&: StartBit, args: BitIndex - StartBit, args: false); |
1567 | } |
1568 | |
1569 | // FilterProcessor scans the well-known encoding bits of the instructions and |
1570 | // builds up a list of candidate filters. It chooses the best filter and |
1571 | // recursively descends down the decoding tree. |
1572 | bool FilterChooser::filterProcessor(bool AllowMixed, bool Greedy) { |
1573 | Filters.clear(); |
1574 | BestIndex = -1; |
1575 | unsigned numInstructions = Opcodes.size(); |
1576 | |
1577 | assert(numInstructions && "Filter created with no instructions" ); |
1578 | |
1579 | // No further filtering is necessary. |
1580 | if (numInstructions == 1) |
1581 | return true; |
1582 | |
1583 | // Heuristics. See also doFilter()'s "Heuristics" comment when num of |
1584 | // instructions is 3. |
1585 | if (AllowMixed && !Greedy) { |
1586 | assert(numInstructions == 3); |
1587 | |
1588 | for (const auto &Opcode : Opcodes) { |
1589 | std::vector<unsigned> StartBits; |
1590 | std::vector<unsigned> EndBits; |
1591 | std::vector<uint64_t> FieldVals; |
1592 | insn_t Insn; |
1593 | |
1594 | insnWithID(Insn, Opcode: Opcode.EncodingID); |
1595 | |
1596 | // Look for islands of undecoded bits of any instruction. |
1597 | if (getIslands(StartBits, EndBits, FieldVals, Insn) > 0) { |
1598 | // Found an instruction with island(s). Now just assign a filter. |
1599 | runSingleFilter(startBit: StartBits[0], numBit: EndBits[0] - StartBits[0] + 1, mixed: true); |
1600 | return true; |
1601 | } |
1602 | } |
1603 | } |
1604 | |
1605 | unsigned BitIndex; |
1606 | |
1607 | // We maintain BIT_WIDTH copies of the bitAttrs automaton. |
1608 | // The automaton consumes the corresponding bit from each |
1609 | // instruction. |
1610 | // |
1611 | // Input symbols: 0, 1, and _ (unset). |
1612 | // States: NONE, FILTERED, ALL_SET, ALL_UNSET, and MIXED. |
1613 | // Initial state: NONE. |
1614 | // |
1615 | // (NONE) ------- [01] -> (ALL_SET) |
1616 | // (NONE) ------- _ ----> (ALL_UNSET) |
1617 | // (ALL_SET) ---- [01] -> (ALL_SET) |
1618 | // (ALL_SET) ---- _ ----> (MIXED) |
1619 | // (ALL_UNSET) -- [01] -> (MIXED) |
1620 | // (ALL_UNSET) -- _ ----> (ALL_UNSET) |
1621 | // (MIXED) ------ . ----> (MIXED) |
1622 | // (FILTERED)---- . ----> (FILTERED) |
1623 | |
1624 | std::vector<bitAttr_t> bitAttrs; |
1625 | |
1626 | // FILTERED bit positions provide no entropy and are not worthy of pursuing. |
1627 | // Filter::recurse() set either BIT_TRUE or BIT_FALSE for each position. |
1628 | for (BitIndex = 0; BitIndex < BitWidth; ++BitIndex) |
1629 | if (FilterBitValues[BitIndex] == BIT_TRUE || |
1630 | FilterBitValues[BitIndex] == BIT_FALSE) |
1631 | bitAttrs.push_back(x: ATTR_FILTERED); |
1632 | else |
1633 | bitAttrs.push_back(x: ATTR_NONE); |
1634 | |
1635 | for (const auto &OpcPair : Opcodes) { |
1636 | insn_t insn; |
1637 | |
1638 | insnWithID(Insn&: insn, Opcode: OpcPair.EncodingID); |
1639 | |
1640 | for (BitIndex = 0; BitIndex < BitWidth; ++BitIndex) { |
1641 | switch (bitAttrs[BitIndex]) { |
1642 | case ATTR_NONE: |
1643 | if (insn[BitIndex] == BIT_UNSET) |
1644 | bitAttrs[BitIndex] = ATTR_ALL_UNSET; |
1645 | else |
1646 | bitAttrs[BitIndex] = ATTR_ALL_SET; |
1647 | break; |
1648 | case ATTR_ALL_SET: |
1649 | if (insn[BitIndex] == BIT_UNSET) |
1650 | bitAttrs[BitIndex] = ATTR_MIXED; |
1651 | break; |
1652 | case ATTR_ALL_UNSET: |
1653 | if (insn[BitIndex] != BIT_UNSET) |
1654 | bitAttrs[BitIndex] = ATTR_MIXED; |
1655 | break; |
1656 | case ATTR_MIXED: |
1657 | case ATTR_FILTERED: |
1658 | break; |
1659 | } |
1660 | } |
1661 | } |
1662 | |
1663 | // The regionAttr automaton consumes the bitAttrs automatons' state, |
1664 | // lowest-to-highest. |
1665 | // |
1666 | // Input symbols: F(iltered), (all_)S(et), (all_)U(nset), M(ixed) |
1667 | // States: NONE, ALL_SET, MIXED |
1668 | // Initial state: NONE |
1669 | // |
1670 | // (NONE) ----- F --> (NONE) |
1671 | // (NONE) ----- S --> (ALL_SET) ; and set region start |
1672 | // (NONE) ----- U --> (NONE) |
1673 | // (NONE) ----- M --> (MIXED) ; and set region start |
1674 | // (ALL_SET) -- F --> (NONE) ; and report an ALL_SET region |
1675 | // (ALL_SET) -- S --> (ALL_SET) |
1676 | // (ALL_SET) -- U --> (NONE) ; and report an ALL_SET region |
1677 | // (ALL_SET) -- M --> (MIXED) ; and report an ALL_SET region |
1678 | // (MIXED) ---- F --> (NONE) ; and report a MIXED region |
1679 | // (MIXED) ---- S --> (ALL_SET) ; and report a MIXED region |
1680 | // (MIXED) ---- U --> (NONE) ; and report a MIXED region |
1681 | // (MIXED) ---- M --> (MIXED) |
1682 | |
1683 | bitAttr_t RA = ATTR_NONE; |
1684 | unsigned StartBit = 0; |
1685 | |
1686 | for (BitIndex = 0; BitIndex < BitWidth; ++BitIndex) { |
1687 | bitAttr_t bitAttr = bitAttrs[BitIndex]; |
1688 | |
1689 | assert(bitAttr != ATTR_NONE && "Bit without attributes" ); |
1690 | |
1691 | switch (RA) { |
1692 | case ATTR_NONE: |
1693 | switch (bitAttr) { |
1694 | case ATTR_FILTERED: |
1695 | break; |
1696 | case ATTR_ALL_SET: |
1697 | StartBit = BitIndex; |
1698 | RA = ATTR_ALL_SET; |
1699 | break; |
1700 | case ATTR_ALL_UNSET: |
1701 | break; |
1702 | case ATTR_MIXED: |
1703 | StartBit = BitIndex; |
1704 | RA = ATTR_MIXED; |
1705 | break; |
1706 | default: |
1707 | llvm_unreachable("Unexpected bitAttr!" ); |
1708 | } |
1709 | break; |
1710 | case ATTR_ALL_SET: |
1711 | switch (bitAttr) { |
1712 | case ATTR_FILTERED: |
1713 | reportRegion(RA, StartBit, BitIndex, AllowMixed); |
1714 | RA = ATTR_NONE; |
1715 | break; |
1716 | case ATTR_ALL_SET: |
1717 | break; |
1718 | case ATTR_ALL_UNSET: |
1719 | reportRegion(RA, StartBit, BitIndex, AllowMixed); |
1720 | RA = ATTR_NONE; |
1721 | break; |
1722 | case ATTR_MIXED: |
1723 | reportRegion(RA, StartBit, BitIndex, AllowMixed); |
1724 | StartBit = BitIndex; |
1725 | RA = ATTR_MIXED; |
1726 | break; |
1727 | default: |
1728 | llvm_unreachable("Unexpected bitAttr!" ); |
1729 | } |
1730 | break; |
1731 | case ATTR_MIXED: |
1732 | switch (bitAttr) { |
1733 | case ATTR_FILTERED: |
1734 | reportRegion(RA, StartBit, BitIndex, AllowMixed); |
1735 | StartBit = BitIndex; |
1736 | RA = ATTR_NONE; |
1737 | break; |
1738 | case ATTR_ALL_SET: |
1739 | reportRegion(RA, StartBit, BitIndex, AllowMixed); |
1740 | StartBit = BitIndex; |
1741 | RA = ATTR_ALL_SET; |
1742 | break; |
1743 | case ATTR_ALL_UNSET: |
1744 | reportRegion(RA, StartBit, BitIndex, AllowMixed); |
1745 | RA = ATTR_NONE; |
1746 | break; |
1747 | case ATTR_MIXED: |
1748 | break; |
1749 | default: |
1750 | llvm_unreachable("Unexpected bitAttr!" ); |
1751 | } |
1752 | break; |
1753 | case ATTR_ALL_UNSET: |
1754 | llvm_unreachable("regionAttr state machine has no ATTR_UNSET state" ); |
1755 | case ATTR_FILTERED: |
1756 | llvm_unreachable("regionAttr state machine has no ATTR_FILTERED state" ); |
1757 | } |
1758 | } |
1759 | |
1760 | // At the end, if we're still in ALL_SET or MIXED states, report a region |
1761 | switch (RA) { |
1762 | case ATTR_NONE: |
1763 | break; |
1764 | case ATTR_FILTERED: |
1765 | break; |
1766 | case ATTR_ALL_SET: |
1767 | reportRegion(RA, StartBit, BitIndex, AllowMixed); |
1768 | break; |
1769 | case ATTR_ALL_UNSET: |
1770 | break; |
1771 | case ATTR_MIXED: |
1772 | reportRegion(RA, StartBit, BitIndex, AllowMixed); |
1773 | break; |
1774 | } |
1775 | |
1776 | // We have finished with the filter processings. Now it's time to choose |
1777 | // the best performing filter. |
1778 | BestIndex = 0; |
1779 | bool AllUseless = true; |
1780 | unsigned BestScore = 0; |
1781 | |
1782 | for (const auto &[Idx, Filter] : enumerate(First&: Filters)) { |
1783 | unsigned Usefulness = Filter.usefulness(); |
1784 | |
1785 | if (Usefulness) |
1786 | AllUseless = false; |
1787 | |
1788 | if (Usefulness > BestScore) { |
1789 | BestIndex = Idx; |
1790 | BestScore = Usefulness; |
1791 | } |
1792 | } |
1793 | |
1794 | if (!AllUseless) |
1795 | bestFilter().recurse(); |
1796 | |
1797 | return !AllUseless; |
1798 | } // end of FilterChooser::filterProcessor(bool) |
1799 | |
1800 | // Decides on the best configuration of filter(s) to use in order to decode |
1801 | // the instructions. A conflict of instructions may occur, in which case we |
1802 | // dump the conflict set to the standard error. |
1803 | void FilterChooser::doFilter() { |
1804 | unsigned Num = Opcodes.size(); |
1805 | assert(Num && "FilterChooser created with no instructions" ); |
1806 | |
1807 | // Try regions of consecutive known bit values first. |
1808 | if (filterProcessor(AllowMixed: false)) |
1809 | return; |
1810 | |
1811 | // Then regions of mixed bits (both known and unitialized bit values allowed). |
1812 | if (filterProcessor(AllowMixed: true)) |
1813 | return; |
1814 | |
1815 | // Heuristics to cope with conflict set {t2CMPrs, t2SUBSrr, t2SUBSrs} where |
1816 | // no single instruction for the maximum ATTR_MIXED region Inst{14-4} has a |
1817 | // well-known encoding pattern. In such case, we backtrack and scan for the |
1818 | // the very first consecutive ATTR_ALL_SET region and assign a filter to it. |
1819 | if (Num == 3 && filterProcessor(AllowMixed: true, Greedy: false)) |
1820 | return; |
1821 | |
1822 | // If we come to here, the instruction decoding has failed. |
1823 | // Set the BestIndex to -1 to indicate so. |
1824 | BestIndex = -1; |
1825 | } |
1826 | |
1827 | // emitTableEntries - Emit state machine entries to decode our share of |
1828 | // instructions. |
1829 | void FilterChooser::emitTableEntries(DecoderTableInfo &TableInfo) const { |
1830 | if (Opcodes.size() == 1) { |
1831 | // There is only one instruction in the set, which is great! |
1832 | // Call emitSingletonDecoder() to see whether there are any remaining |
1833 | // encodings bits. |
1834 | emitSingletonTableEntry(TableInfo, Opc: Opcodes[0]); |
1835 | return; |
1836 | } |
1837 | |
1838 | // Choose the best filter to do the decodings! |
1839 | if (BestIndex != -1) { |
1840 | const Filter &Best = Filters[BestIndex]; |
1841 | if (Best.getNumFiltered() == 1) |
1842 | emitSingletonTableEntry(TableInfo, Best); |
1843 | else |
1844 | Best.emitTableEntry(TableInfo); |
1845 | return; |
1846 | } |
1847 | |
1848 | // We don't know how to decode these instructions! Dump the |
1849 | // conflict set and bail. |
1850 | |
1851 | // Print out useful conflict information for postmortem analysis. |
1852 | errs() << "Decoding Conflict:\n" ; |
1853 | |
1854 | dumpStack(o&: errs(), prefix: "\t\t" ); |
1855 | |
1856 | for (auto Opcode : Opcodes) { |
1857 | errs() << '\t'; |
1858 | emitNameWithID(OS&: errs(), Opcode: Opcode.EncodingID); |
1859 | errs() << " " ; |
1860 | dumpBits( |
1861 | o&: errs(), |
1862 | bits: getBitsField(def: *AllInstructions[Opcode.EncodingID].EncodingDef, str: "Inst" )); |
1863 | errs() << '\n'; |
1864 | } |
1865 | } |
1866 | |
1867 | static std::string findOperandDecoderMethod(Record *Record) { |
1868 | std::string Decoder; |
1869 | |
1870 | RecordVal *DecoderString = Record->getValue(Name: "DecoderMethod" ); |
1871 | StringInit *String = |
1872 | DecoderString ? dyn_cast<StringInit>(Val: DecoderString->getValue()) : nullptr; |
1873 | if (String) { |
1874 | Decoder = std::string(String->getValue()); |
1875 | if (!Decoder.empty()) |
1876 | return Decoder; |
1877 | } |
1878 | |
1879 | if (Record->isSubClassOf(Name: "RegisterOperand" )) |
1880 | // Allows use of a DecoderMethod in referenced RegisterClass if set. |
1881 | return findOperandDecoderMethod(Record: Record->getValueAsDef(FieldName: "RegClass" )); |
1882 | |
1883 | if (Record->isSubClassOf(Name: "RegisterClass" )) { |
1884 | Decoder = "Decode" + Record->getName().str() + "RegisterClass" ; |
1885 | } else if (Record->isSubClassOf(Name: "PointerLikeRegClass" )) { |
1886 | Decoder = "DecodePointerLikeRegClass" + |
1887 | utostr(X: Record->getValueAsInt(FieldName: "RegClassKind" )); |
1888 | } |
1889 | |
1890 | return Decoder; |
1891 | } |
1892 | |
1893 | OperandInfo getOpInfo(Record *TypeRecord) { |
1894 | std::string Decoder = findOperandDecoderMethod(Record: TypeRecord); |
1895 | |
1896 | RecordVal *HasCompleteDecoderVal = TypeRecord->getValue(Name: "hasCompleteDecoder" ); |
1897 | BitInit *HasCompleteDecoderBit = |
1898 | HasCompleteDecoderVal |
1899 | ? dyn_cast<BitInit>(Val: HasCompleteDecoderVal->getValue()) |
1900 | : nullptr; |
1901 | bool HasCompleteDecoder = |
1902 | HasCompleteDecoderBit ? HasCompleteDecoderBit->getValue() : true; |
1903 | |
1904 | return OperandInfo(Decoder, HasCompleteDecoder); |
1905 | } |
1906 | |
1907 | void parseVarLenInstOperand(const Record &Def, |
1908 | std::vector<OperandInfo> &Operands, |
1909 | const CodeGenInstruction &CGI) { |
1910 | |
1911 | const RecordVal *RV = Def.getValue(Name: "Inst" ); |
1912 | VarLenInst VLI(cast<DagInit>(Val: RV->getValue()), RV); |
1913 | SmallVector<int> TiedTo; |
1914 | |
1915 | for (const auto &[Idx, Op] : enumerate(First: CGI.Operands)) { |
1916 | if (Op.MIOperandInfo && Op.MIOperandInfo->getNumArgs() > 0) |
1917 | for (auto *Arg : Op.MIOperandInfo->getArgs()) |
1918 | Operands.push_back(x: getOpInfo(TypeRecord: cast<DefInit>(Val: Arg)->getDef())); |
1919 | else |
1920 | Operands.push_back(x: getOpInfo(TypeRecord: Op.Rec)); |
1921 | |
1922 | int TiedReg = Op.getTiedRegister(); |
1923 | TiedTo.push_back(Elt: -1); |
1924 | if (TiedReg != -1) { |
1925 | TiedTo[Idx] = TiedReg; |
1926 | TiedTo[TiedReg] = Idx; |
1927 | } |
1928 | } |
1929 | |
1930 | unsigned CurrBitPos = 0; |
1931 | for (const auto &EncodingSegment : VLI) { |
1932 | unsigned Offset = 0; |
1933 | StringRef OpName; |
1934 | |
1935 | if (const StringInit *SI = dyn_cast<StringInit>(Val: EncodingSegment.Value)) { |
1936 | OpName = SI->getValue(); |
1937 | } else if (const DagInit *DI = dyn_cast<DagInit>(Val: EncodingSegment.Value)) { |
1938 | OpName = cast<StringInit>(Val: DI->getArg(Num: 0))->getValue(); |
1939 | Offset = cast<IntInit>(Val: DI->getArg(Num: 2))->getValue(); |
1940 | } |
1941 | |
1942 | if (!OpName.empty()) { |
1943 | auto OpSubOpPair = |
1944 | const_cast<CodeGenInstruction &>(CGI).Operands.ParseOperandName( |
1945 | Op: OpName); |
1946 | unsigned OpIdx = CGI.Operands.getFlattenedOperandNumber(Op: OpSubOpPair); |
1947 | Operands[OpIdx].addField(Base: CurrBitPos, Width: EncodingSegment.BitWidth, Offset); |
1948 | if (!EncodingSegment.CustomDecoder.empty()) |
1949 | Operands[OpIdx].Decoder = EncodingSegment.CustomDecoder.str(); |
1950 | |
1951 | int TiedReg = TiedTo[OpSubOpPair.first]; |
1952 | if (TiedReg != -1) { |
1953 | unsigned OpIdx = CGI.Operands.getFlattenedOperandNumber( |
1954 | Op: std::pair(TiedReg, OpSubOpPair.second)); |
1955 | Operands[OpIdx].addField(Base: CurrBitPos, Width: EncodingSegment.BitWidth, Offset); |
1956 | } |
1957 | } |
1958 | |
1959 | CurrBitPos += EncodingSegment.BitWidth; |
1960 | } |
1961 | } |
1962 | |
1963 | static void debugDumpRecord(const Record &Rec) { |
1964 | // Dump the record, so we can see what's going on... |
1965 | std::string E; |
1966 | raw_string_ostream S(E); |
1967 | S << "Dumping record for previous error:\n" ; |
1968 | S << Rec; |
1969 | PrintNote(Msg: E); |
1970 | } |
1971 | |
1972 | /// For an operand field named OpName: populate OpInfo.InitValue with the |
1973 | /// constant-valued bit values, and OpInfo.Fields with the ranges of bits to |
1974 | /// insert from the decoded instruction. |
1975 | static void addOneOperandFields(const Record &EncodingDef, const BitsInit &Bits, |
1976 | std::map<std::string, std::string> &TiedNames, |
1977 | StringRef OpName, OperandInfo &OpInfo) { |
1978 | // Some bits of the operand may be required to be 1 depending on the |
1979 | // instruction's encoding. Collect those bits. |
1980 | if (const RecordVal *EncodedValue = EncodingDef.getValue(Name: OpName)) |
1981 | if (const BitsInit *OpBits = dyn_cast<BitsInit>(Val: EncodedValue->getValue())) |
1982 | for (unsigned I = 0; I < OpBits->getNumBits(); ++I) |
1983 | if (const BitInit *OpBit = dyn_cast<BitInit>(Val: OpBits->getBit(Bit: I))) |
1984 | if (OpBit->getValue()) |
1985 | OpInfo.InitValue |= 1ULL << I; |
1986 | |
1987 | for (unsigned I = 0, J = 0; I != Bits.getNumBits(); I = J) { |
1988 | VarInit *Var; |
1989 | unsigned Offset = 0; |
1990 | for (; J != Bits.getNumBits(); ++J) { |
1991 | VarBitInit *BJ = dyn_cast<VarBitInit>(Val: Bits.getBit(Bit: J)); |
1992 | if (BJ) { |
1993 | Var = dyn_cast<VarInit>(Val: BJ->getBitVar()); |
1994 | if (I == J) |
1995 | Offset = BJ->getBitNum(); |
1996 | else if (BJ->getBitNum() != Offset + J - I) |
1997 | break; |
1998 | } else { |
1999 | Var = dyn_cast<VarInit>(Val: Bits.getBit(Bit: J)); |
2000 | } |
2001 | if (!Var || (Var->getName() != OpName && |
2002 | Var->getName() != TiedNames[std::string(OpName)])) |
2003 | break; |
2004 | } |
2005 | if (I == J) |
2006 | ++J; |
2007 | else |
2008 | OpInfo.addField(Base: I, Width: J - I, Offset); |
2009 | } |
2010 | } |
2011 | |
2012 | static unsigned |
2013 | populateInstruction(CodeGenTarget &Target, const Record &EncodingDef, |
2014 | const CodeGenInstruction &CGI, unsigned Opc, |
2015 | std::map<unsigned, std::vector<OperandInfo>> &Operands, |
2016 | bool IsVarLenInst) { |
2017 | const Record &Def = *CGI.TheDef; |
2018 | // If all the bit positions are not specified; do not decode this instruction. |
2019 | // We are bound to fail! For proper disassembly, the well-known encoding bits |
2020 | // of the instruction must be fully specified. |
2021 | |
2022 | BitsInit &Bits = getBitsField(def: EncodingDef, str: "Inst" ); |
2023 | if (Bits.allInComplete()) |
2024 | return 0; |
2025 | |
2026 | std::vector<OperandInfo> InsnOperands; |
2027 | |
2028 | // If the instruction has specified a custom decoding hook, use that instead |
2029 | // of trying to auto-generate the decoder. |
2030 | StringRef InstDecoder = EncodingDef.getValueAsString(FieldName: "DecoderMethod" ); |
2031 | if (InstDecoder != "" ) { |
2032 | bool HasCompleteInstDecoder = |
2033 | EncodingDef.getValueAsBit(FieldName: "hasCompleteDecoder" ); |
2034 | InsnOperands.push_back( |
2035 | x: OperandInfo(std::string(InstDecoder), HasCompleteInstDecoder)); |
2036 | Operands[Opc] = InsnOperands; |
2037 | return Bits.getNumBits(); |
2038 | } |
2039 | |
2040 | // Generate a description of the operand of the instruction that we know |
2041 | // how to decode automatically. |
2042 | // FIXME: We'll need to have a way to manually override this as needed. |
2043 | |
2044 | // Gather the outputs/inputs of the instruction, so we can find their |
2045 | // positions in the encoding. This assumes for now that they appear in the |
2046 | // MCInst in the order that they're listed. |
2047 | std::vector<std::pair<Init *, StringRef>> InOutOperands; |
2048 | DagInit *Out = Def.getValueAsDag(FieldName: "OutOperandList" ); |
2049 | DagInit *In = Def.getValueAsDag(FieldName: "InOperandList" ); |
2050 | for (const auto &[Idx, Arg] : enumerate(First: Out->getArgs())) |
2051 | InOutOperands.push_back(x: std::pair(Arg, Out->getArgNameStr(Num: Idx))); |
2052 | for (const auto &[Idx, Arg] : enumerate(First: In->getArgs())) |
2053 | InOutOperands.push_back(x: std::pair(Arg, In->getArgNameStr(Num: Idx))); |
2054 | |
2055 | // Search for tied operands, so that we can correctly instantiate |
2056 | // operands that are not explicitly represented in the encoding. |
2057 | std::map<std::string, std::string> TiedNames; |
2058 | for (const auto &[I, Op] : enumerate(First: CGI.Operands)) { |
2059 | for (const auto &[J, CI] : enumerate(First: Op.Constraints)) { |
2060 | if (CI.isTied()) { |
2061 | std::pair<unsigned, unsigned> SO = |
2062 | CGI.Operands.getSubOperandNumber(Op: CI.getTiedOperand()); |
2063 | std::string TiedName = CGI.Operands[SO.first].SubOpNames[SO.second]; |
2064 | if (TiedName.empty()) |
2065 | TiedName = CGI.Operands[SO.first].Name; |
2066 | std::string MyName = Op.SubOpNames[J]; |
2067 | if (MyName.empty()) |
2068 | MyName = Op.Name; |
2069 | |
2070 | TiedNames[MyName] = TiedName; |
2071 | TiedNames[TiedName] = MyName; |
2072 | } |
2073 | } |
2074 | } |
2075 | |
2076 | if (IsVarLenInst) { |
2077 | parseVarLenInstOperand(Def: EncodingDef, Operands&: InsnOperands, CGI); |
2078 | } else { |
2079 | // For each operand, see if we can figure out where it is encoded. |
2080 | for (const auto &Op : InOutOperands) { |
2081 | Init *OpInit = Op.first; |
2082 | StringRef OpName = Op.second; |
2083 | |
2084 | // We're ready to find the instruction encoding locations for this |
2085 | // operand. |
2086 | |
2087 | // First, find the operand type ("OpInit"), and sub-op names |
2088 | // ("SubArgDag") if present. |
2089 | DagInit *SubArgDag = dyn_cast<DagInit>(Val: OpInit); |
2090 | if (SubArgDag) |
2091 | OpInit = SubArgDag->getOperator(); |
2092 | Record *OpTypeRec = cast<DefInit>(Val: OpInit)->getDef(); |
2093 | // Lookup the sub-operands from the operand type record (note that only |
2094 | // Operand subclasses have MIOperandInfo, see CodeGenInstruction.cpp). |
2095 | DagInit *SubOps = OpTypeRec->isSubClassOf(Name: "Operand" ) |
2096 | ? OpTypeRec->getValueAsDag(FieldName: "MIOperandInfo" ) |
2097 | : nullptr; |
2098 | |
2099 | // Lookup the decoder method and construct a new OperandInfo to hold our |
2100 | // result. |
2101 | OperandInfo OpInfo = getOpInfo(TypeRecord: OpTypeRec); |
2102 | |
2103 | // If we have named sub-operands... |
2104 | if (SubArgDag) { |
2105 | // Then there should not be a custom decoder specified on the top-level |
2106 | // type. |
2107 | if (!OpInfo.Decoder.empty()) { |
2108 | PrintError(ErrorLoc: EncodingDef.getLoc(), |
2109 | Msg: "DecoderEmitter: operand \"" + OpName + "\" has type \"" + |
2110 | OpInit->getAsString() + |
2111 | "\" with a custom DecoderMethod, but also named " |
2112 | "sub-operands." ); |
2113 | continue; |
2114 | } |
2115 | |
2116 | // Decode each of the sub-ops separately. |
2117 | assert(SubOps && SubArgDag->getNumArgs() == SubOps->getNumArgs()); |
2118 | for (const auto &[I, Arg] : enumerate(First: SubOps->getArgs())) { |
2119 | StringRef SubOpName = SubArgDag->getArgNameStr(Num: I); |
2120 | OperandInfo SubOpInfo = getOpInfo(TypeRecord: cast<DefInit>(Val: Arg)->getDef()); |
2121 | |
2122 | addOneOperandFields(EncodingDef, Bits, TiedNames, OpName: SubOpName, |
2123 | OpInfo&: SubOpInfo); |
2124 | InsnOperands.push_back(x: SubOpInfo); |
2125 | } |
2126 | continue; |
2127 | } |
2128 | |
2129 | // Otherwise, if we have an operand with sub-operands, but they aren't |
2130 | // named... |
2131 | if (SubOps && OpInfo.Decoder.empty()) { |
2132 | // If it's a single sub-operand, and no custom decoder, use the decoder |
2133 | // from the one sub-operand. |
2134 | if (SubOps->getNumArgs() == 1) |
2135 | OpInfo = getOpInfo(TypeRecord: cast<DefInit>(Val: SubOps->getArg(Num: 0))->getDef()); |
2136 | |
2137 | // If we have multiple sub-ops, there'd better have a custom |
2138 | // decoder. (Otherwise we don't know how to populate them properly...) |
2139 | if (SubOps->getNumArgs() > 1) { |
2140 | PrintError(ErrorLoc: EncodingDef.getLoc(), |
2141 | Msg: "DecoderEmitter: operand \"" + OpName + |
2142 | "\" uses MIOperandInfo with multiple ops, but doesn't " |
2143 | "have a custom decoder!" ); |
2144 | debugDumpRecord(Rec: EncodingDef); |
2145 | continue; |
2146 | } |
2147 | } |
2148 | |
2149 | addOneOperandFields(EncodingDef, Bits, TiedNames, OpName, OpInfo); |
2150 | // FIXME: it should be an error not to find a definition for a given |
2151 | // operand, rather than just failing to add it to the resulting |
2152 | // instruction! (This is a longstanding bug, which will be addressed in an |
2153 | // upcoming change.) |
2154 | if (OpInfo.numFields() > 0) |
2155 | InsnOperands.push_back(x: OpInfo); |
2156 | } |
2157 | } |
2158 | Operands[Opc] = InsnOperands; |
2159 | |
2160 | #if 0 |
2161 | LLVM_DEBUG({ |
2162 | // Dumps the instruction encoding bits. |
2163 | dumpBits(errs(), Bits); |
2164 | |
2165 | errs() << '\n'; |
2166 | |
2167 | // Dumps the list of operand info. |
2168 | for (unsigned i = 0, e = CGI.Operands.size(); i != e; ++i) { |
2169 | const CGIOperandList::OperandInfo &Info = CGI.Operands[i]; |
2170 | const std::string &OperandName = Info.Name; |
2171 | const Record &OperandDef = *Info.Rec; |
2172 | |
2173 | errs() << "\t" << OperandName << " (" << OperandDef.getName() << ")\n" ; |
2174 | } |
2175 | }); |
2176 | #endif |
2177 | |
2178 | return Bits.getNumBits(); |
2179 | } |
2180 | |
2181 | // emitFieldFromInstruction - Emit the templated helper function |
2182 | // fieldFromInstruction(). |
2183 | // On Windows we make sure that this function is not inlined when |
2184 | // using the VS compiler. It has a bug which causes the function |
2185 | // to be optimized out in some circumstances. See llvm.org/pr38292 |
2186 | static void emitFieldFromInstruction(formatted_raw_ostream &OS) { |
2187 | OS << R"( |
2188 | // Helper functions for extracting fields from encoded instructions. |
2189 | // InsnType must either be integral or an APInt-like object that must: |
2190 | // * be default-constructible and copy-constructible |
2191 | // * be constructible from an APInt (this can be private) |
2192 | // * Support insertBits(bits, startBit, numBits) |
2193 | // * Support extractBitsAsZExtValue(numBits, startBit) |
2194 | // * Support the ~, &, ==, and != operators with other objects of the same type |
2195 | // * Support the != and bitwise & with uint64_t |
2196 | // * Support put (<<) to raw_ostream& |
2197 | template <typename InsnType> |
2198 | #if defined(_MSC_VER) && !defined(__clang__) |
2199 | __declspec(noinline) |
2200 | #endif |
2201 | static std::enable_if_t<std::is_integral<InsnType>::value, InsnType> |
2202 | fieldFromInstruction(const InsnType &insn, unsigned startBit, |
2203 | unsigned numBits) { |
2204 | assert(startBit + numBits <= 64 && "Cannot support >64-bit extractions!"); |
2205 | assert(startBit + numBits <= (sizeof(InsnType) * 8) && |
2206 | "Instruction field out of bounds!"); |
2207 | InsnType fieldMask; |
2208 | if (numBits == sizeof(InsnType) * 8) |
2209 | fieldMask = (InsnType)(-1LL); |
2210 | else |
2211 | fieldMask = (((InsnType)1 << numBits) - 1) << startBit; |
2212 | return (insn & fieldMask) >> startBit; |
2213 | } |
2214 | |
2215 | template <typename InsnType> |
2216 | static std::enable_if_t<!std::is_integral<InsnType>::value, uint64_t> |
2217 | fieldFromInstruction(const InsnType &insn, unsigned startBit, |
2218 | unsigned numBits) { |
2219 | return insn.extractBitsAsZExtValue(numBits, startBit); |
2220 | } |
2221 | )" ; |
2222 | } |
2223 | |
2224 | // emitInsertBits - Emit the templated helper function insertBits(). |
2225 | static void emitInsertBits(formatted_raw_ostream &OS) { |
2226 | OS << R"( |
2227 | // Helper function for inserting bits extracted from an encoded instruction into |
2228 | // a field. |
2229 | template <typename InsnType> |
2230 | static std::enable_if_t<std::is_integral<InsnType>::value> |
2231 | insertBits(InsnType &field, InsnType bits, unsigned startBit, unsigned numBits) { |
2232 | assert(startBit + numBits <= sizeof field * 8); |
2233 | field |= (InsnType)bits << startBit; |
2234 | } |
2235 | |
2236 | template <typename InsnType> |
2237 | static std::enable_if_t<!std::is_integral<InsnType>::value> |
2238 | insertBits(InsnType &field, uint64_t bits, unsigned startBit, unsigned numBits) { |
2239 | field.insertBits(bits, startBit, numBits); |
2240 | } |
2241 | )" ; |
2242 | } |
2243 | |
2244 | // emitDecodeInstruction - Emit the templated helper function |
2245 | // decodeInstruction(). |
2246 | static void emitDecodeInstruction(formatted_raw_ostream &OS, |
2247 | bool IsVarLenInst) { |
2248 | OS << R"( |
2249 | template <typename InsnType> |
2250 | static DecodeStatus decodeInstruction(const uint8_t DecodeTable[], MCInst &MI, |
2251 | InsnType insn, uint64_t Address, |
2252 | const MCDisassembler *DisAsm, |
2253 | const MCSubtargetInfo &STI)" ; |
2254 | if (IsVarLenInst) { |
2255 | OS << ",\n " |
2256 | "llvm::function_ref<void(APInt &, uint64_t)> makeUp" ; |
2257 | } |
2258 | OS << R"() { |
2259 | const FeatureBitset &Bits = STI.getFeatureBits(); |
2260 | |
2261 | const uint8_t *Ptr = DecodeTable; |
2262 | uint64_t CurFieldValue = 0; |
2263 | DecodeStatus S = MCDisassembler::Success; |
2264 | while (true) { |
2265 | ptrdiff_t Loc = Ptr - DecodeTable; |
2266 | switch (*Ptr) { |
2267 | default: |
2268 | errs() << Loc << ": Unexpected decode table opcode!\n"; |
2269 | return MCDisassembler::Fail; |
2270 | case MCD::OPC_ExtractField: { |
2271 | // Decode the start value. |
2272 | unsigned Start = decodeULEB128AndIncUnsafe(++Ptr); |
2273 | unsigned Len = *Ptr++;)" ; |
2274 | if (IsVarLenInst) |
2275 | OS << "\n makeUp(insn, Start + Len);" ; |
2276 | OS << R"( |
2277 | CurFieldValue = fieldFromInstruction(insn, Start, Len); |
2278 | LLVM_DEBUG(dbgs() << Loc << ": OPC_ExtractField(" << Start << ", " |
2279 | << Len << "): " << CurFieldValue << "\n"); |
2280 | break; |
2281 | } |
2282 | case MCD::OPC_FilterValue: { |
2283 | // Decode the field value. |
2284 | uint64_t Val = decodeULEB128AndIncUnsafe(++Ptr); |
2285 | // NumToSkip is a plain 24-bit integer. |
2286 | unsigned NumToSkip = *Ptr++; |
2287 | NumToSkip |= (*Ptr++) << 8; |
2288 | NumToSkip |= (*Ptr++) << 16; |
2289 | |
2290 | // Perform the filter operation. |
2291 | if (Val != CurFieldValue) |
2292 | Ptr += NumToSkip; |
2293 | LLVM_DEBUG(dbgs() << Loc << ": OPC_FilterValue(" << Val << ", " << NumToSkip |
2294 | << "): " << ((Val != CurFieldValue) ? "FAIL:" : "PASS:") |
2295 | << " continuing at " << (Ptr - DecodeTable) << "\n"); |
2296 | |
2297 | break; |
2298 | } |
2299 | case MCD::OPC_CheckField: { |
2300 | // Decode the start value. |
2301 | unsigned Start = decodeULEB128AndIncUnsafe(++Ptr); |
2302 | unsigned Len = *Ptr;)" ; |
2303 | if (IsVarLenInst) |
2304 | OS << "\n makeUp(insn, Start + Len);" ; |
2305 | OS << R"( |
2306 | uint64_t FieldValue = fieldFromInstruction(insn, Start, Len); |
2307 | // Decode the field value. |
2308 | unsigned PtrLen = 0; |
2309 | uint64_t ExpectedValue = decodeULEB128(++Ptr, &PtrLen); |
2310 | Ptr += PtrLen; |
2311 | // NumToSkip is a plain 24-bit integer. |
2312 | unsigned NumToSkip = *Ptr++; |
2313 | NumToSkip |= (*Ptr++) << 8; |
2314 | NumToSkip |= (*Ptr++) << 16; |
2315 | |
2316 | // If the actual and expected values don't match, skip. |
2317 | if (ExpectedValue != FieldValue) |
2318 | Ptr += NumToSkip; |
2319 | LLVM_DEBUG(dbgs() << Loc << ": OPC_CheckField(" << Start << ", " |
2320 | << Len << ", " << ExpectedValue << ", " << NumToSkip |
2321 | << "): FieldValue = " << FieldValue << ", ExpectedValue = " |
2322 | << ExpectedValue << ": " |
2323 | << ((ExpectedValue == FieldValue) ? "PASS\n" : "FAIL\n")); |
2324 | break; |
2325 | } |
2326 | case MCD::OPC_CheckPredicate: { |
2327 | // Decode the Predicate Index value. |
2328 | unsigned PIdx = decodeULEB128AndIncUnsafe(++Ptr); |
2329 | // NumToSkip is a plain 24-bit integer. |
2330 | unsigned NumToSkip = *Ptr++; |
2331 | NumToSkip |= (*Ptr++) << 8; |
2332 | NumToSkip |= (*Ptr++) << 16; |
2333 | // Check the predicate. |
2334 | bool Pred; |
2335 | if (!(Pred = checkDecoderPredicate(PIdx, Bits))) |
2336 | Ptr += NumToSkip; |
2337 | (void)Pred; |
2338 | LLVM_DEBUG(dbgs() << Loc << ": OPC_CheckPredicate(" << PIdx << "): " |
2339 | << (Pred ? "PASS\n" : "FAIL\n")); |
2340 | |
2341 | break; |
2342 | } |
2343 | case MCD::OPC_Decode: { |
2344 | // Decode the Opcode value. |
2345 | unsigned Opc = decodeULEB128AndIncUnsafe(++Ptr); |
2346 | unsigned DecodeIdx = decodeULEB128AndIncUnsafe(Ptr); |
2347 | |
2348 | MI.clear(); |
2349 | MI.setOpcode(Opc); |
2350 | bool DecodeComplete;)" ; |
2351 | if (IsVarLenInst) { |
2352 | OS << "\n unsigned Len = InstrLenTable[Opc];\n" |
2353 | << " makeUp(insn, Len);" ; |
2354 | } |
2355 | OS << R"( |
2356 | S = decodeToMCInst(S, DecodeIdx, insn, MI, Address, DisAsm, DecodeComplete); |
2357 | assert(DecodeComplete); |
2358 | |
2359 | LLVM_DEBUG(dbgs() << Loc << ": OPC_Decode: opcode " << Opc |
2360 | << ", using decoder " << DecodeIdx << ": " |
2361 | << (S != MCDisassembler::Fail ? "PASS" : "FAIL") << "\n"); |
2362 | return S; |
2363 | } |
2364 | case MCD::OPC_TryDecode: { |
2365 | // Decode the Opcode value. |
2366 | unsigned Opc = decodeULEB128AndIncUnsafe(++Ptr); |
2367 | unsigned DecodeIdx = decodeULEB128AndIncUnsafe(Ptr); |
2368 | // NumToSkip is a plain 24-bit integer. |
2369 | unsigned NumToSkip = *Ptr++; |
2370 | NumToSkip |= (*Ptr++) << 8; |
2371 | NumToSkip |= (*Ptr++) << 16; |
2372 | |
2373 | // Perform the decode operation. |
2374 | MCInst TmpMI; |
2375 | TmpMI.setOpcode(Opc); |
2376 | bool DecodeComplete; |
2377 | S = decodeToMCInst(S, DecodeIdx, insn, TmpMI, Address, DisAsm, DecodeComplete); |
2378 | LLVM_DEBUG(dbgs() << Loc << ": OPC_TryDecode: opcode " << Opc |
2379 | << ", using decoder " << DecodeIdx << ": "); |
2380 | |
2381 | if (DecodeComplete) { |
2382 | // Decoding complete. |
2383 | LLVM_DEBUG(dbgs() << (S != MCDisassembler::Fail ? "PASS" : "FAIL") << "\n"); |
2384 | MI = TmpMI; |
2385 | return S; |
2386 | } else { |
2387 | assert(S == MCDisassembler::Fail); |
2388 | // If the decoding was incomplete, skip. |
2389 | Ptr += NumToSkip; |
2390 | LLVM_DEBUG(dbgs() << "FAIL: continuing at " << (Ptr - DecodeTable) << "\n"); |
2391 | // Reset decode status. This also drops a SoftFail status that could be |
2392 | // set before the decode attempt. |
2393 | S = MCDisassembler::Success; |
2394 | } |
2395 | break; |
2396 | } |
2397 | case MCD::OPC_SoftFail: { |
2398 | // Decode the mask values. |
2399 | uint64_t PositiveMask = decodeULEB128AndIncUnsafe(++Ptr); |
2400 | uint64_t NegativeMask = decodeULEB128AndIncUnsafe(Ptr); |
2401 | bool Fail = (insn & PositiveMask) != 0 || (~insn & NegativeMask) != 0; |
2402 | if (Fail) |
2403 | S = MCDisassembler::SoftFail; |
2404 | LLVM_DEBUG(dbgs() << Loc << ": OPC_SoftFail: " << (Fail ? "FAIL\n" : "PASS\n")); |
2405 | break; |
2406 | } |
2407 | case MCD::OPC_Fail: { |
2408 | LLVM_DEBUG(dbgs() << Loc << ": OPC_Fail\n"); |
2409 | return MCDisassembler::Fail; |
2410 | } |
2411 | } |
2412 | } |
2413 | llvm_unreachable("bogosity detected in disassembler state machine!"); |
2414 | } |
2415 | |
2416 | )" ; |
2417 | } |
2418 | |
2419 | // Helper to propagate SoftFail status. Returns false if the status is Fail; |
2420 | // callers are expected to early-exit in that condition. (Note, the '&' operator |
2421 | // is correct to propagate the values of this enum; see comment on 'enum |
2422 | // DecodeStatus'.) |
2423 | static void emitCheck(formatted_raw_ostream &OS) { |
2424 | OS << R"( |
2425 | static bool Check(DecodeStatus &Out, DecodeStatus In) { |
2426 | Out = static_cast<DecodeStatus>(Out & In); |
2427 | return Out != MCDisassembler::Fail; |
2428 | } |
2429 | |
2430 | )" ; |
2431 | } |
2432 | |
2433 | // Collect all HwModes referenced by the target for encoding purposes, |
2434 | // returning a vector of corresponding names. |
2435 | static void collectHwModesReferencedForEncodings( |
2436 | const CodeGenHwModes &HWM, std::vector<StringRef> &Names, |
2437 | NamespacesHwModesMap &NamespacesWithHwModes) { |
2438 | SmallBitVector BV(HWM.getNumModeIds()); |
2439 | for (const auto &MS : HWM.getHwModeSelects()) { |
2440 | for (const HwModeSelect::PairType &P : MS.second.Items) { |
2441 | if (P.second->isSubClassOf(Name: "InstructionEncoding" )) { |
2442 | std::string DecoderNamespace = |
2443 | std::string(P.second->getValueAsString(FieldName: "DecoderNamespace" )); |
2444 | if (P.first == DefaultMode) { |
2445 | NamespacesWithHwModes[DecoderNamespace].insert(x: "" ); |
2446 | } else { |
2447 | NamespacesWithHwModes[DecoderNamespace].insert( |
2448 | x: HWM.getMode(Id: P.first).Name); |
2449 | } |
2450 | BV.set(P.first); |
2451 | } |
2452 | } |
2453 | } |
2454 | transform(Range: BV.set_bits(), d_first: std::back_inserter(x&: Names), F: [&HWM](const int &M) { |
2455 | if (M == DefaultMode) |
2456 | return StringRef("" ); |
2457 | return HWM.getModeName(Id: M, /*IncludeDefault=*/true); |
2458 | }); |
2459 | } |
2460 | |
2461 | static void |
2462 | handleHwModesUnrelatedEncodings(const CodeGenInstruction *Instr, |
2463 | const std::vector<StringRef> &HwModeNames, |
2464 | NamespacesHwModesMap &NamespacesWithHwModes, |
2465 | std::vector<EncodingAndInst> &GlobalEncodings) { |
2466 | const Record *InstDef = Instr->TheDef; |
2467 | |
2468 | switch (DecoderEmitterSuppressDuplicates) { |
2469 | case SUPPRESSION_DISABLE: { |
2470 | for (StringRef HwModeName : HwModeNames) |
2471 | GlobalEncodings.emplace_back(args&: InstDef, args&: Instr, args&: HwModeName); |
2472 | break; |
2473 | } |
2474 | case SUPPRESSION_LEVEL1: { |
2475 | std::string DecoderNamespace = |
2476 | std::string(InstDef->getValueAsString(FieldName: "DecoderNamespace" )); |
2477 | auto It = NamespacesWithHwModes.find(x: DecoderNamespace); |
2478 | if (It != NamespacesWithHwModes.end()) { |
2479 | for (StringRef HwModeName : It->second) |
2480 | GlobalEncodings.emplace_back(args&: InstDef, args&: Instr, args&: HwModeName); |
2481 | } else { |
2482 | // Only emit the encoding once, as it's DecoderNamespace doesn't |
2483 | // contain any HwModes. |
2484 | GlobalEncodings.emplace_back(args&: InstDef, args&: Instr, args: "" ); |
2485 | } |
2486 | break; |
2487 | } |
2488 | case SUPPRESSION_LEVEL2: |
2489 | GlobalEncodings.emplace_back(args&: InstDef, args&: Instr, args: "" ); |
2490 | break; |
2491 | } |
2492 | } |
2493 | |
2494 | // Emits disassembler code for instruction decoding. |
2495 | void DecoderEmitter::run(raw_ostream &o) { |
2496 | formatted_raw_ostream OS(o); |
2497 | OS << R"( |
2498 | #include "llvm/MC/MCInst.h" |
2499 | #include "llvm/MC/MCSubtargetInfo.h" |
2500 | #include "llvm/Support/DataTypes.h" |
2501 | #include "llvm/Support/Debug.h" |
2502 | #include "llvm/Support/LEB128.h" |
2503 | #include "llvm/Support/raw_ostream.h" |
2504 | #include "llvm/TargetParser/SubtargetFeature.h" |
2505 | #include <assert.h> |
2506 | |
2507 | namespace llvm { |
2508 | )" ; |
2509 | |
2510 | emitFieldFromInstruction(OS); |
2511 | emitInsertBits(OS); |
2512 | emitCheck(OS); |
2513 | |
2514 | Target.reverseBitsForLittleEndianEncoding(); |
2515 | |
2516 | // Parameterize the decoders based on namespace and instruction width. |
2517 | |
2518 | // First, collect all encoding-related HwModes referenced by the target. |
2519 | // And establish a mapping table between DecoderNamespace and HwMode. |
2520 | // If HwModeNames is empty, add the empty string so we always have one HwMode. |
2521 | const CodeGenHwModes &HWM = Target.getHwModes(); |
2522 | std::vector<StringRef> HwModeNames; |
2523 | NamespacesHwModesMap NamespacesWithHwModes; |
2524 | collectHwModesReferencedForEncodings(HWM, Names&: HwModeNames, NamespacesWithHwModes); |
2525 | if (HwModeNames.empty()) |
2526 | HwModeNames.push_back(x: "" ); |
2527 | |
2528 | const auto &NumberedInstructions = Target.getInstructionsByEnumValue(); |
2529 | NumberedEncodings.reserve(n: NumberedInstructions.size()); |
2530 | for (const auto &NumberedInstruction : NumberedInstructions) { |
2531 | const Record *InstDef = NumberedInstruction->TheDef; |
2532 | if (const RecordVal *RV = InstDef->getValue(Name: "EncodingInfos" )) { |
2533 | if (DefInit *DI = dyn_cast_or_null<DefInit>(Val: RV->getValue())) { |
2534 | EncodingInfoByHwMode EBM(DI->getDef(), HWM); |
2535 | for (auto &[ModeId, Encoding] : EBM) { |
2536 | // DecoderTables with DefaultMode should not have any suffix. |
2537 | if (ModeId == DefaultMode) { |
2538 | NumberedEncodings.emplace_back(args&: Encoding, args: NumberedInstruction, args: "" ); |
2539 | } else { |
2540 | NumberedEncodings.emplace_back(args&: Encoding, args: NumberedInstruction, |
2541 | args: HWM.getMode(Id: ModeId).Name); |
2542 | } |
2543 | } |
2544 | continue; |
2545 | } |
2546 | } |
2547 | // This instruction is encoded the same on all HwModes. |
2548 | // According to user needs, provide varying degrees of suppression. |
2549 | handleHwModesUnrelatedEncodings(Instr: NumberedInstruction, HwModeNames, |
2550 | NamespacesWithHwModes, GlobalEncodings&: NumberedEncodings); |
2551 | } |
2552 | for (const auto &NumberedAlias : |
2553 | RK.getAllDerivedDefinitions(ClassName: "AdditionalEncoding" )) |
2554 | NumberedEncodings.emplace_back( |
2555 | args: NumberedAlias, |
2556 | args: &Target.getInstruction(InstRec: NumberedAlias->getValueAsDef(FieldName: "AliasOf" ))); |
2557 | |
2558 | std::map<std::pair<std::string, unsigned>, std::vector<EncodingIDAndOpcode>> |
2559 | OpcMap; |
2560 | std::map<unsigned, std::vector<OperandInfo>> Operands; |
2561 | std::vector<unsigned> InstrLen; |
2562 | bool IsVarLenInst = Target.hasVariableLengthEncodings(); |
2563 | unsigned MaxInstLen = 0; |
2564 | |
2565 | for (const auto &[NEI, NumberedEncoding] : enumerate(First&: NumberedEncodings)) { |
2566 | const Record *EncodingDef = NumberedEncoding.EncodingDef; |
2567 | const CodeGenInstruction *Inst = NumberedEncoding.Inst; |
2568 | const Record *Def = Inst->TheDef; |
2569 | unsigned Size = EncodingDef->getValueAsInt(FieldName: "Size" ); |
2570 | if (Def->getValueAsString(FieldName: "Namespace" ) == "TargetOpcode" || |
2571 | Def->getValueAsBit(FieldName: "isPseudo" ) || |
2572 | Def->getValueAsBit(FieldName: "isAsmParserOnly" ) || |
2573 | Def->getValueAsBit(FieldName: "isCodeGenOnly" )) { |
2574 | NumEncodingsLackingDisasm++; |
2575 | continue; |
2576 | } |
2577 | |
2578 | if (NEI < NumberedInstructions.size()) |
2579 | NumInstructions++; |
2580 | NumEncodings++; |
2581 | |
2582 | if (!Size && !IsVarLenInst) |
2583 | continue; |
2584 | |
2585 | if (IsVarLenInst) |
2586 | InstrLen.resize(new_size: NumberedInstructions.size(), x: 0); |
2587 | |
2588 | if (unsigned Len = populateInstruction(Target, EncodingDef: *EncodingDef, CGI: *Inst, Opc: NEI, |
2589 | Operands, IsVarLenInst)) { |
2590 | if (IsVarLenInst) { |
2591 | MaxInstLen = std::max(a: MaxInstLen, b: Len); |
2592 | InstrLen[NEI] = Len; |
2593 | } |
2594 | std::string DecoderNamespace = |
2595 | std::string(EncodingDef->getValueAsString(FieldName: "DecoderNamespace" )); |
2596 | if (!NumberedEncoding.HwModeName.empty()) |
2597 | DecoderNamespace += |
2598 | std::string("_" ) + NumberedEncoding.HwModeName.str(); |
2599 | OpcMap[std::pair(DecoderNamespace, Size)].emplace_back( |
2600 | args&: NEI, args: Target.getInstrIntValue(R: Def)); |
2601 | } else { |
2602 | NumEncodingsOmitted++; |
2603 | } |
2604 | } |
2605 | |
2606 | DecoderTableInfo TableInfo; |
2607 | for (const auto &Opc : OpcMap) { |
2608 | // Emit the decoder for this namespace+width combination. |
2609 | ArrayRef<EncodingAndInst> NumberedEncodingsRef(NumberedEncodings.data(), |
2610 | NumberedEncodings.size()); |
2611 | FilterChooser FC(NumberedEncodingsRef, Opc.second, Operands, |
2612 | IsVarLenInst ? MaxInstLen : 8 * Opc.first.second, this); |
2613 | |
2614 | // The decode table is cleared for each top level decoder function. The |
2615 | // predicates and decoders themselves, however, are shared across all |
2616 | // decoders to give more opportunities for uniqueing. |
2617 | TableInfo.Table.clear(); |
2618 | TableInfo.FixupStack.clear(); |
2619 | TableInfo.Table.reserve(n: 16384); |
2620 | TableInfo.FixupStack.emplace_back(); |
2621 | FC.emitTableEntries(TableInfo); |
2622 | // Any NumToSkip fixups in the top level scope can resolve to the |
2623 | // OPC_Fail at the end of the table. |
2624 | assert(TableInfo.FixupStack.size() == 1 && "fixup stack phasing error!" ); |
2625 | // Resolve any NumToSkip fixups in the current scope. |
2626 | resolveTableFixups(Table&: TableInfo.Table, Fixups: TableInfo.FixupStack.back(), |
2627 | DestIdx: TableInfo.Table.size()); |
2628 | TableInfo.FixupStack.clear(); |
2629 | |
2630 | TableInfo.Table.push_back(x: MCD::OPC_Fail); |
2631 | |
2632 | // Print the table to the output stream. |
2633 | emitTable(OS, Table&: TableInfo.Table, Indentation: 0, BitWidth: FC.getBitWidth(), Namespace: Opc.first.first, |
2634 | EncodingIDs: Opc.second); |
2635 | } |
2636 | |
2637 | // For variable instruction, we emit a instruction length table |
2638 | // to let the decoder know how long the instructions are. |
2639 | // You can see example usage in M68k's disassembler. |
2640 | if (IsVarLenInst) |
2641 | emitInstrLenTable(OS, InstrLen); |
2642 | // Emit the predicate function. |
2643 | emitPredicateFunction(OS, Predicates&: TableInfo.Predicates, Indentation: 0); |
2644 | |
2645 | // Emit the decoder function. |
2646 | emitDecoderFunction(OS, Decoders&: TableInfo.Decoders, Indentation: 0); |
2647 | |
2648 | // Emit the main entry point for the decoder, decodeInstruction(). |
2649 | emitDecodeInstruction(OS, IsVarLenInst); |
2650 | |
2651 | OS << "\n} // end namespace llvm\n" ; |
2652 | } |
2653 | |
2654 | namespace llvm { |
2655 | |
2656 | void EmitDecoder(RecordKeeper &RK, raw_ostream &OS, |
2657 | const std::string &PredicateNamespace) { |
2658 | DecoderEmitter(RK, PredicateNamespace).run(o&: OS); |
2659 | } |
2660 | |
2661 | } // end namespace llvm |
2662 | |