[MC] Rewrite tablegen for printInstrAlias to comiple faster, NFC

Before this change, the *InstPrinter.cpp files of each target where some
of the slowest objects to compile in all of LLVM. See this snippet produced by
ClangBuildAnalyzer:
https://reviews.llvm.org/P8171$96
Search for "InstPrinter", and see that it shows up in a few places.

Tablegen was emitting a large switch containing a sequence of operand checks,
each of which created many conditions and many BBs. Register allocation and
jump threading both did not scale well with such a large repetitive sequence of
basic blocks.

So, this change essentially turns those control flow structures into
data. The previous structure looked like:

  switch (Opc) {
  case TGT::ADD:
    // check alias 1
    if (MI->getOperandCount() == N && // check num opnds
        MI->getOperand(0).isReg() && // check opnd 0
        ...
        MI->getOperand(1).isImm() && // check opnd 1
     AsmString = "foo";
     break;
   }
   // check alias 2
   if (...)
     ...
   return false;

The new structure looks like:

  OpToPatterns: Sorted table of opcodes mapping to pattern indices.
   \->
     Patterns: List of patterns. Previous table points to subrange of
               patterns to match.
      \->
        Conds: The if conditions above encoded as a kind and 32-bit value.

See MCInstPrinter.cpp for the details of how the new data structures are
interpreted.

Here are some before and after metrics.
Time to compile AArch64InstPrinter.cpp:
  0m29.062s vs. 0m2.203s
size of the obj:
  3.9M vs. 676K
size of clang.exe:
  97M vs. 96M

I have not benchmarked disassembly performance, but typically
disassemblers are bottlenecked on IO and string processing, not alias
matching, so I'm not sure it's interesting enough to be worth doing.

Reviewers: RKSimon, andreadb, xbolva00, craig.topper

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D70650

Author: rnk

llvm/include/llvm/MC/MCInstPrinter.h

@@ -16,6 +16,7 @@ namespace llvm {
 
 class MCAsmInfo;
 class MCInst;
+class MCOperand;
 class MCInstrInfo;
 class MCRegisterInfo;
 class MCSubtargetInfo;
@@ -34,6 +35,8 @@ enum Style {
 
 } // end namespace HexStyle
 
+struct AliasMatchingData;
+
 /// This is an instance of a target assembly language printer that
 /// converts an MCInst to valid target assembly syntax.
 class MCInstPrinter {
@@ -58,6 +61,10 @@ class MCInstPrinter {
   /// Utility function for printing annotations.
   void printAnnotation(raw_ostream &OS, StringRef Annot);
 
+  /// Helper for matching MCInsts to alias patterns when printing instructions.
+  const char *matchAliasPatterns(const MCInst *MI, const MCSubtargetInfo *STI,
+                                 const AliasMatchingData &M);
+
 public:
   MCInstPrinter(const MCAsmInfo &mai, const MCInstrInfo &mii,
                 const MCRegisterInfo &mri) : MAI(mai), MII(mii), MRI(mri) {}
@@ -104,6 +111,48 @@ class MCInstPrinter {
   format_object<uint64_t> formatHex(uint64_t Value) const;
 };
 
+/// Map from opcode to pattern list by binary search.
+struct PatternsForOpcode {
+  uint32_t Opcode;
+  uint16_t PatternStart;
+  uint16_t NumPatterns;
+};
+
+/// Data for each alias pattern. Includes feature bits, string, number of
+/// operands, and a variadic list of conditions to check.
+struct AliasPattern {
+  uint32_t AsmStrOffset;
+  uint32_t AliasCondStart;
+  uint8_t NumOperands;
+  uint8_t NumConds;
+};
+
+struct AliasPatternCond {
+  enum CondKind : uint8_t {
+    K_Feature,    // Match only if a feature is enabled.
+    K_NegFeature, // Match only if a feature is disabled.
+    K_Ignore,     // Match any operand.
+    K_Reg,        // Match a specific register.
+    K_TiedReg,    // Match another already matched register.
+    K_Imm,        // Match a specific immediate.
+    K_RegClass,   // Match registers in a class.
+    K_Custom,     // Call custom matcher by index.
+  };
+
+  CondKind Kind;
+  uint32_t Value;
+};
+
+/// Tablegenerated data structures needed to match alias patterns.
+struct AliasMatchingData {
+  ArrayRef<PatternsForOpcode> OpToPatterns;
+  ArrayRef<AliasPattern> Patterns;
+  ArrayRef<AliasPatternCond> PatternConds;
+  StringRef AsmStrings;
+  bool (*ValidateMCOperand)(const MCOperand &MCOp, const MCSubtargetInfo &STI,
+                            unsigned PredicateIndex);
+};
+
 } // end namespace llvm
 
 #endif // LLVM_MC_MCINSTPRINTER_H

llvm/lib/MC/MCInstPrinter.cpp

@@ -10,7 +10,9 @@
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/MC/MCAsmInfo.h"
+#include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
+#include "llvm/MC/MCSubtargetInfo.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Support/raw_ostream.h"
@@ -57,6 +59,94 @@ void MCInstPrinter::printAnnotation(raw_ostream &OS, StringRef Annot) {
   }
 }
 
+static bool matchAliasCondition(const MCInst &MI, const MCSubtargetInfo *STI,
+                                const MCRegisterInfo &MRI, unsigned &OpIdx,
+                                const AliasMatchingData &M,
+                                const AliasPatternCond &C) {
+  // Feature tests are special, they don't consume operands.
+  if (C.Kind == AliasPatternCond::K_Feature)
+    return STI->getFeatureBits().test(C.Value);
+  if (C.Kind == AliasPatternCond::K_NegFeature)
+    return !STI->getFeatureBits().test(C.Value);
+
+  // Get and consume an operand.
+  const MCOperand &Opnd = MI.getOperand(OpIdx);
+  ++OpIdx;
+
+  // Check the specific condition for the operand.
+  switch (C.Kind) {
+  case AliasPatternCond::K_Imm:
+    // Operand must be a specific immediate.
+    return Opnd.isImm() && Opnd.getImm() == int32_t(C.Value);
+  case AliasPatternCond::K_Reg:
+    // Operand must be a specific register.
+    return Opnd.isReg() && Opnd.getReg() == C.Value;
+  case AliasPatternCond::K_TiedReg:
+    // Operand must match the register of another operand.
+    return Opnd.isReg() && Opnd.getReg() == MI.getOperand(C.Value).getReg();
+  case AliasPatternCond::K_RegClass:
+    // Operand must be a register in this class. Value is a register class id.
+    return Opnd.isReg() && MRI.getRegClass(C.Value).contains(Opnd.getReg());
+  case AliasPatternCond::K_Custom:
+    // Operand must match some custom criteria.
+    return M.ValidateMCOperand(Opnd, *STI, C.Value);
+  case AliasPatternCond::K_Ignore:
+    // Operand can be anything.
+    return true;
+  case AliasPatternCond::K_Feature:
+  case AliasPatternCond::K_NegFeature:
+    llvm_unreachable("handled earlier");
+  }
+  llvm_unreachable("invalid kind");
+}
+
+const char *MCInstPrinter::matchAliasPatterns(const MCInst *MI,
+                                              const MCSubtargetInfo *STI,
+                                              const AliasMatchingData &M) {
+  // Binary search by opcode. Return false if there are no aliases for this
+  // opcode.
+  auto It = lower_bound(M.OpToPatterns, MI->getOpcode(),
+                        [](const PatternsForOpcode &L, unsigned Opcode) {
+                          return L.Opcode < Opcode;
+                        });
+  if (It == M.OpToPatterns.end() || It->Opcode != MI->getOpcode())
+    return nullptr;
+
+  // Try all patterns for this opcode.
+  uint32_t AsmStrOffset = ~0U;
+  ArrayRef<AliasPattern> Patterns =
+      M.Patterns.slice(It->PatternStart, It->NumPatterns);
+  for (const AliasPattern &P : Patterns) {
+    // Check operand count first.
+    if (MI->getNumOperands() != P.NumOperands)
+      return nullptr;
+
+    // Test all conditions for this pattern.
+    ArrayRef<AliasPatternCond> Conds =
+        M.PatternConds.slice(P.AliasCondStart, P.NumConds);
+    unsigned OpIdx = 0;
+    if (llvm::all_of(Conds, [&](const AliasPatternCond &C) {
+          return matchAliasCondition(*MI, STI, MRI, OpIdx, M, C);
+        })) {
+      // If all conditions matched, use this asm string.
+      AsmStrOffset = P.AsmStrOffset;
+      break;
+    }
+  }
+
+  // If no alias matched, don't print an alias.
+  if (AsmStrOffset == ~0U)
+    return nullptr;
+
+  // Go to offset AsmStrOffset and use the null terminated string there. The
+  // offset should point to the beginning of an alias string, so it should
+  // either be zero or be preceded by a null byte.
+  assert(AsmStrOffset < M.AsmStrings.size() &&
+         (AsmStrOffset == 0 || M.AsmStrings[AsmStrOffset - 1] == '\0') &&
+         "bad asm string offset");
+  return M.AsmStrings.data() + AsmStrOffset;
+}
+
 /// Utility functions to make adding mark ups simpler.
 StringRef MCInstPrinter::markup(StringRef s) const {
   if (getUseMarkup())