// Some example inputs, for X86:
// 'addl' (immediate ...) (register ...)
// 'add' (immediate ...) (memory ...)
-// 'call' '*' %epc
+// 'call' '*' %epc
//
// The assembly matcher is responsible for converting this input into a precise
// machine instruction (i.e., an instruction with a well defined encoding). This
#include "AsmMatcherEmitter.h"
#include "CodeGenTarget.h"
#include "Record.h"
+#include "StringMatcher.h"
#include "llvm/ADT/OwningPtr.h"
+#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/StringExtras.h"
#include <set>
using namespace llvm;
-namespace {
static cl::opt<std::string>
MatchPrefix("match-prefix", cl::init(""),
cl::desc("Only match instructions with the given prefix"));
-}
-
-/// FlattenVariants - Flatten an .td file assembly string by selecting the
-/// variant at index \arg N.
-static std::string FlattenVariants(const std::string &AsmString,
- unsigned N) {
- StringRef Cur = AsmString;
- std::string Res = "";
-
- for (;;) {
- // Find the start of the next variant string.
- size_t VariantsStart = 0;
- for (size_t e = Cur.size(); VariantsStart != e; ++VariantsStart)
- if (Cur[VariantsStart] == '{' &&
- (VariantsStart == 0 || (Cur[VariantsStart-1] != '$' &&
- Cur[VariantsStart-1] != '\\')))
- break;
-
- // Add the prefix to the result.
- Res += Cur.slice(0, VariantsStart);
- if (VariantsStart == Cur.size())
- break;
-
- ++VariantsStart; // Skip the '{'.
-
- // Scan to the end of the variants string.
- size_t VariantsEnd = VariantsStart;
- unsigned NestedBraces = 1;
- for (size_t e = Cur.size(); VariantsEnd != e; ++VariantsEnd) {
- if (Cur[VariantsEnd] == '}' && Cur[VariantsEnd-1] != '\\') {
- if (--NestedBraces == 0)
- break;
- } else if (Cur[VariantsEnd] == '{')
- ++NestedBraces;
- }
-
- // Select the Nth variant (or empty).
- StringRef Selection = Cur.slice(VariantsStart, VariantsEnd);
- for (unsigned i = 0; i != N; ++i)
- Selection = Selection.split('|').second;
- Res += Selection.split('|').first;
-
- assert(VariantsEnd != Cur.size() &&
- "Unterminated variants in assembly string!");
- Cur = Cur.substr(VariantsEnd + 1);
- }
-
- return Res;
-}
/// TokenizeAsmString - Tokenize a simplified assembly string.
-static void TokenizeAsmString(const StringRef &AsmString,
+static void TokenizeAsmString(StringRef AsmString,
SmallVectorImpl<StringRef> &Tokens) {
unsigned Prev = 0;
bool InTok = true;
Tokens.push_back(AsmString.substr(i, 1));
Prev = i + 1;
break;
-
+
case '\\':
if (InTok) {
Tokens.push_back(AsmString.slice(Prev, i));
break;
}
+ case '.':
+ if (InTok) {
+ Tokens.push_back(AsmString.slice(Prev, i));
+ }
+ Prev = i;
+ InTok = true;
+ break;
+
default:
InTok = true;
}
Tokens.push_back(AsmString.substr(Prev));
}
-static bool IsAssemblerInstruction(const StringRef &Name,
- const CodeGenInstruction &CGI,
- const SmallVectorImpl<StringRef> &Tokens) {
- // Ignore psuedo ops.
- //
- // FIXME: This is a hack.
- if (const RecordVal *Form = CGI.TheDef->getValue("Form"))
- if (Form->getValue()->getAsString() == "Pseudo")
- return false;
-
- // Ignore "PHI" node.
- //
- // FIXME: This is also a hack.
- if (Name == "PHI")
- return false;
-
- // Ignore instructions with no .s string.
- //
- // FIXME: What are these?
- if (CGI.AsmString.empty())
- return false;
-
- // FIXME: Hack; ignore any instructions with a newline in them.
- if (std::find(CGI.AsmString.begin(),
- CGI.AsmString.end(), '\n') != CGI.AsmString.end())
- return false;
-
- // Ignore instructions with attributes, these are always fake instructions for
- // simplifying codegen.
- //
- // FIXME: Is this true?
- //
- // Also, we ignore instructions which reference the operand multiple times;
- // this implies a constraint we would not currently honor. These are
- // currently always fake instructions for simplifying codegen.
- //
- // FIXME: Encode this assumption in the .td, so we can error out here.
- std::set<std::string> OperandNames;
- for (unsigned i = 1, e = Tokens.size(); i < e; ++i) {
- if (Tokens[i][0] == '$' &&
- std::find(Tokens[i].begin(),
- Tokens[i].end(), ':') != Tokens[i].end()) {
- DEBUG({
- errs() << "warning: '" << Name << "': "
- << "ignoring instruction; operand with attribute '"
- << Tokens[i] << "', \n";
- });
- return false;
- }
-
- if (Tokens[i][0] == '$' && !OperandNames.insert(Tokens[i]).second) {
- DEBUG({
- errs() << "warning: '" << Name << "': "
- << "ignoring instruction; tied operand '"
- << Tokens[i] << "', \n";
- });
- return false;
- }
- }
-
- return true;
-}
namespace {
+ class AsmMatcherInfo;
+struct SubtargetFeatureInfo;
/// ClassInfo - Helper class for storing the information about a particular
/// class of operands which can be matched.
struct ClassInfo {
enum ClassInfoKind {
- Invalid = 0, ///< Invalid kind, for use as a sentinel value.
- Token, ///< The class for a particular token.
- Register, ///< A register class.
- UserClass0 ///< The (first) user defined class, subsequent user defined
- /// classes are UserClass0+1, and so on.
+ /// Invalid kind, for use as a sentinel value.
+ Invalid = 0,
+
+ /// The class for a particular token.
+ Token,
+
+ /// The (first) register class, subsequent register classes are
+ /// RegisterClass0+1, and so on.
+ RegisterClass0,
+
+ /// The (first) user defined class, subsequent user defined classes are
+ /// UserClass0+1, and so on.
+ UserClass0 = 1<<16
};
/// Kind - The class kind, which is either a predefined kind, or (UserClass0 +
/// N) for the Nth user defined class.
unsigned Kind;
+ /// SuperClasses - The super classes of this class. Note that for simplicities
+ /// sake user operands only record their immediate super class, while register
+ /// operands include all superclasses.
+ std::vector<ClassInfo*> SuperClasses;
+
/// Name - The full class name, suitable for use in an enum.
std::string Name;
std::string ValueName;
/// PredicateMethod - The name of the operand method to test whether the
- /// operand matches this class; this is not valid for Token kinds.
+ /// operand matches this class; this is not valid for Token or register kinds.
std::string PredicateMethod;
/// RenderMethod - The name of the operand method to add this operand to an
- /// MCInst; this is not valid for Token kinds.
+ /// MCInst; this is not valid for Token or register kinds.
std::string RenderMethod;
+ /// For register classes, the records for all the registers in this class.
+ std::set<Record*> Registers;
+
+public:
+ /// isRegisterClass() - Check if this is a register class.
+ bool isRegisterClass() const {
+ return Kind >= RegisterClass0 && Kind < UserClass0;
+ }
+
+ /// isUserClass() - Check if this is a user defined class.
+ bool isUserClass() const {
+ return Kind >= UserClass0;
+ }
+
+ /// isRelatedTo - Check whether this class is "related" to \arg RHS. Classes
+ /// are related if they are in the same class hierarchy.
+ bool isRelatedTo(const ClassInfo &RHS) const {
+ // Tokens are only related to tokens.
+ if (Kind == Token || RHS.Kind == Token)
+ return Kind == Token && RHS.Kind == Token;
+
+ // Registers classes are only related to registers classes, and only if
+ // their intersection is non-empty.
+ if (isRegisterClass() || RHS.isRegisterClass()) {
+ if (!isRegisterClass() || !RHS.isRegisterClass())
+ return false;
+
+ std::set<Record*> Tmp;
+ std::insert_iterator< std::set<Record*> > II(Tmp, Tmp.begin());
+ std::set_intersection(Registers.begin(), Registers.end(),
+ RHS.Registers.begin(), RHS.Registers.end(),
+ II);
+
+ return !Tmp.empty();
+ }
+
+ // Otherwise we have two users operands; they are related if they are in the
+ // same class hierarchy.
+ //
+ // FIXME: This is an oversimplification, they should only be related if they
+ // intersect, however we don't have that information.
+ assert(isUserClass() && RHS.isUserClass() && "Unexpected class!");
+ const ClassInfo *Root = this;
+ while (!Root->SuperClasses.empty())
+ Root = Root->SuperClasses.front();
+
+ const ClassInfo *RHSRoot = &RHS;
+ while (!RHSRoot->SuperClasses.empty())
+ RHSRoot = RHSRoot->SuperClasses.front();
+
+ return Root == RHSRoot;
+ }
+
+ /// isSubsetOf - Test whether this class is a subset of \arg RHS;
+ bool isSubsetOf(const ClassInfo &RHS) const {
+ // This is a subset of RHS if it is the same class...
+ if (this == &RHS)
+ return true;
+
+ // ... or if any of its super classes are a subset of RHS.
+ for (std::vector<ClassInfo*>::const_iterator it = SuperClasses.begin(),
+ ie = SuperClasses.end(); it != ie; ++it)
+ if ((*it)->isSubsetOf(RHS))
+ return true;
+
+ return false;
+ }
+
/// operator< - Compare two classes.
bool operator<(const ClassInfo &RHS) const {
- // Incompatible kinds are comparable.
- if (Kind != RHS.Kind)
+ if (this == &RHS)
+ return false;
+
+ // Unrelated classes can be ordered by kind.
+ if (!isRelatedTo(RHS))
return Kind < RHS.Kind;
switch (Kind) {
case Invalid:
assert(0 && "Invalid kind!");
case Token:
- // Tokens are always comparable.
+ // Tokens are comparable by value.
//
// FIXME: Compare by enum value.
return ValueName < RHS.ValueName;
- case Register:
- // FIXME: Compare by subset relation.
- return false;
-
default:
- // FIXME: Allow user defined relation.
- return false;
+ // This class preceeds the RHS if it is a proper subset of the RHS.
+ if (isSubsetOf(RHS))
+ return true;
+ if (RHS.isSubsetOf(*this))
+ return false;
+
+ // Otherwise, order by name to ensure we have a total ordering.
+ return ValueName < RHS.ValueName;
}
}
};
-/// InstructionInfo - Helper class for storing the necessary information for an
-/// instruction which is capable of being matched.
-struct InstructionInfo {
+/// MatchableInfo - Helper class for storing the necessary information for an
+/// instruction or alias which is capable of being matched.
+struct MatchableInfo {
struct Operand {
/// The unique class instance this operand should match.
ClassInfo *Class;
/// The original operand this corresponds to, if any.
- const CodeGenInstruction::OperandInfo *OperandInfo;
+ const CGIOperandList::OperandInfo *OperandInfo;
+
+ Operand(ClassInfo *C, const CGIOperandList::OperandInfo *OpInfo)
+ : Class(C), OperandInfo(OpInfo) {}
};
/// InstrName - The target name for this instruction.
std::string InstrName;
- /// Instr - The instruction this matches.
- const CodeGenInstruction *Instr;
+ Record *const TheDef;
+ const CGIOperandList &OperandList;
/// AsmString - The assembly string for this instruction (with variants
/// removed).
/// Operands - The operands that this instruction matches.
SmallVector<Operand, 4> Operands;
+ /// Predicates - The required subtarget features to match this instruction.
+ SmallVector<SubtargetFeatureInfo*, 4> RequiredFeatures;
+
/// ConversionFnKind - The enum value which is passed to the generated
/// ConvertToMCInst to convert parsed operands into an MCInst for this
/// function.
std::string ConversionFnKind;
+
+ MatchableInfo(const CodeGenInstruction &CGI)
+ : TheDef(CGI.TheDef), OperandList(CGI.Operands), AsmString(CGI.AsmString) {
+ InstrName = TheDef->getName();
+ }
+
+ MatchableInfo(const CodeGenInstAlias *Alias)
+ : TheDef(Alias->TheDef), OperandList(Alias->Operands),
+ AsmString(Alias->AsmString) {
+
+ // FIXME: Huge hack.
+ DefInit *DI = dynamic_cast<DefInit*>(Alias->Result->getOperator());
+ assert(DI);
+
+ InstrName = DI->getDef()->getName();
+ }
+
+ void Initialize(const AsmMatcherInfo &Info,
+ SmallPtrSet<Record*, 16> &SingletonRegisters);
+
+ /// Validate - Return true if this matchable is a valid thing to match against
+ /// and perform a bunch of validity checking.
+ bool Validate(StringRef CommentDelimiter, bool Hack) const;
+
+ /// getSingletonRegisterForToken - If the specified token is a singleton
+ /// register, return the Record for it, otherwise return null.
+ Record *getSingletonRegisterForToken(unsigned i,
+ const AsmMatcherInfo &Info) const;
+
+ /// operator< - Compare two matchables.
+ bool operator<(const MatchableInfo &RHS) const {
+ // The primary comparator is the instruction mnemonic.
+ if (Tokens[0] != RHS.Tokens[0])
+ return Tokens[0] < RHS.Tokens[0];
- /// operator< - Compare two instructions.
- bool operator<(const InstructionInfo &RHS) const {
- // Order first by the number of operands (which is unambiguous).
if (Operands.size() != RHS.Operands.size())
return Operands.size() < RHS.Operands.size();
-
- // Otherwise, order by lexicographic comparison of tokens and operand kinds
- // (these can never be ambiguous).
- for (unsigned i = 0, e = Operands.size(); i != e; ++i)
- if (Operands[i].Class->Kind != RHS.Operands[i].Class->Kind ||
- Operands[i].Class->Kind == ClassInfo::Token)
- if (*Operands[i].Class < *RHS.Operands[i].Class)
- return true;
-
- // Finally, order by the component wise comparison of operand classes. We
- // don't want to rely on the lexigraphic ordering of elements, so we define
- // only define the ordering when it is unambiguous. That is, when some pair
- // compares less than and no pair compares greater than.
- // Check that no pair compares greater than.
- for (unsigned i = 0, e = Operands.size(); i != e; ++i)
+ // Compare lexicographically by operand. The matcher validates that other
+ // orderings wouldn't be ambiguous using \see CouldMatchAmiguouslyWith().
+ for (unsigned i = 0, e = Operands.size(); i != e; ++i) {
+ if (*Operands[i].Class < *RHS.Operands[i].Class)
+ return true;
if (*RHS.Operands[i].Class < *Operands[i].Class)
return false;
+ }
- // Otherwise, return true if some pair compares less than.
+ return false;
+ }
+
+ /// CouldMatchAmiguouslyWith - Check whether this matchable could
+ /// ambiguously match the same set of operands as \arg RHS (without being a
+ /// strictly superior match).
+ bool CouldMatchAmiguouslyWith(const MatchableInfo &RHS) {
+ // The primary comparator is the instruction mnemonic.
+ if (Tokens[0] != RHS.Tokens[0])
+ return false;
+
+ // The number of operands is unambiguous.
+ if (Operands.size() != RHS.Operands.size())
+ return false;
+
+ // Otherwise, make sure the ordering of the two instructions is unambiguous
+ // by checking that either (a) a token or operand kind discriminates them,
+ // or (b) the ordering among equivalent kinds is consistent.
+
+ // Tokens and operand kinds are unambiguous (assuming a correct target
+ // specific parser).
for (unsigned i = 0, e = Operands.size(); i != e; ++i)
+ if (Operands[i].Class->Kind != RHS.Operands[i].Class->Kind ||
+ Operands[i].Class->Kind == ClassInfo::Token)
+ if (*Operands[i].Class < *RHS.Operands[i].Class ||
+ *RHS.Operands[i].Class < *Operands[i].Class)
+ return false;
+
+ // Otherwise, this operand could commute if all operands are equivalent, or
+ // there is a pair of operands that compare less than and a pair that
+ // compare greater than.
+ bool HasLT = false, HasGT = false;
+ for (unsigned i = 0, e = Operands.size(); i != e; ++i) {
if (*Operands[i].Class < *RHS.Operands[i].Class)
- return true;
+ HasLT = true;
+ if (*RHS.Operands[i].Class < *Operands[i].Class)
+ HasGT = true;
+ }
- return false;
+ return !(HasLT ^ HasGT);
}
-public:
void dump();
};
+/// SubtargetFeatureInfo - Helper class for storing information on a subtarget
+/// feature which participates in instruction matching.
+struct SubtargetFeatureInfo {
+ /// \brief The predicate record for this feature.
+ Record *TheDef;
+
+ /// \brief An unique index assigned to represent this feature.
+ unsigned Index;
+
+ SubtargetFeatureInfo(Record *D, unsigned Idx) : TheDef(D), Index(Idx) {}
+
+ /// \brief The name of the enumerated constant identifying this feature.
+ std::string getEnumName() const {
+ return "Feature_" + TheDef->getName();
+ }
+};
+
class AsmMatcherInfo {
public:
+ /// The tablegen AsmParser record.
+ Record *AsmParser;
+
+ /// Target - The target information.
+ CodeGenTarget &Target;
+
+ /// The AsmParser "RegisterPrefix" value.
+ std::string RegisterPrefix;
+
/// The classes which are needed for matching.
std::vector<ClassInfo*> Classes;
-
- /// The information on the instruction to match.
- std::vector<InstructionInfo*> Instructions;
+ /// The information on the matchables to match.
+ std::vector<MatchableInfo*> Matchables;
+
+ /// Map of Register records to their class information.
+ std::map<Record*, ClassInfo*> RegisterClasses;
+
+ /// Map of Predicate records to their subtarget information.
+ std::map<Record*, SubtargetFeatureInfo*> SubtargetFeatures;
+
private:
/// Map of token to class information which has already been constructed.
std::map<std::string, ClassInfo*> TokenClasses;
- /// Map of operand name to class information which has already been
- /// constructed.
- std::map<std::string, ClassInfo*> OperandClasses;
+ /// Map of RegisterClass records to their class information.
+ std::map<Record*, ClassInfo*> RegisterClassClasses;
- /// Map of user class names to kind value.
- std::map<std::string, unsigned> UserClasses;
+ /// Map of AsmOperandClass records to their class information.
+ std::map<Record*, ClassInfo*> AsmOperandClasses;
private:
/// getTokenClass - Lookup or create the class for the given token.
- ClassInfo *getTokenClass(const StringRef &Token);
-
- /// getUserClassKind - Lookup or create the kind value for the given class
- /// name.
- unsigned getUserClassKind(const StringRef &Name);
+ ClassInfo *getTokenClass(StringRef Token);
/// getOperandClass - Lookup or create the class for the given operand.
- ClassInfo *getOperandClass(const StringRef &Token,
- const CodeGenInstruction::OperandInfo &OI);
+ ClassInfo *getOperandClass(StringRef Token,
+ const CGIOperandList::OperandInfo &OI);
+
+ /// BuildRegisterClasses - Build the ClassInfo* instances for register
+ /// classes.
+ void BuildRegisterClasses(SmallPtrSet<Record*, 16> &SingletonRegisters);
+
+ /// BuildOperandClasses - Build the ClassInfo* instances for user defined
+ /// operand classes.
+ void BuildOperandClasses();
public:
+ AsmMatcherInfo(Record *AsmParser, CodeGenTarget &Target);
+
/// BuildInfo - Construct the various tables used during matching.
- void BuildInfo(CodeGenTarget &Target);
+ void BuildInfo();
+
+ /// getSubtargetFeature - Lookup or create the subtarget feature info for the
+ /// given operand.
+ SubtargetFeatureInfo *getSubtargetFeature(Record *Def) const {
+ assert(Def->isSubClassOf("Predicate") && "Invalid predicate type!");
+ std::map<Record*, SubtargetFeatureInfo*>::const_iterator I =
+ SubtargetFeatures.find(Def);
+ return I == SubtargetFeatures.end() ? 0 : I->second;
+ }
};
}
-void InstructionInfo::dump() {
+void MatchableInfo::dump() {
errs() << InstrName << " -- " << "flattened:\"" << AsmString << '\"'
<< ", tokens:[";
for (unsigned i = 0, e = Tokens.size(); i != e; ++i) {
continue;
}
- const CodeGenInstruction::OperandInfo &OI = *Op.OperandInfo;
+ if (!Op.OperandInfo) {
+ errs() << "(singleton register)\n";
+ continue;
+ }
+
+ const CGIOperandList::OperandInfo &OI = *Op.OperandInfo;
errs() << OI.Name << " " << OI.Rec->getName()
<< " (" << OI.MIOperandNo << ", " << OI.MINumOperands << ")\n";
}
}
-static std::string getEnumNameForToken(const StringRef &Str) {
- std::string Res;
+void MatchableInfo::Initialize(const AsmMatcherInfo &Info,
+ SmallPtrSet<Record*, 16> &SingletonRegisters) {
+ // TODO: Eventually support asmparser for Variant != 0.
+ AsmString = CodeGenInstruction::FlattenAsmStringVariants(AsmString, 0);
+
+ TokenizeAsmString(AsmString, Tokens);
+
+ // Compute the require features.
+ std::vector<Record*> Predicates =TheDef->getValueAsListOfDefs("Predicates");
+ for (unsigned i = 0, e = Predicates.size(); i != e; ++i)
+ if (SubtargetFeatureInfo *Feature =
+ Info.getSubtargetFeature(Predicates[i]))
+ RequiredFeatures.push_back(Feature);
+
+ // Collect singleton registers, if used.
+ for (unsigned i = 0, e = Tokens.size(); i != e; ++i) {
+ if (Record *Reg = getSingletonRegisterForToken(i, Info))
+ SingletonRegisters.insert(Reg);
+ }
+}
+
+
+/// getRegisterRecord - Get the register record for \arg name, or 0.
+static Record *getRegisterRecord(CodeGenTarget &Target, StringRef Name) {
+ for (unsigned i = 0, e = Target.getRegisters().size(); i != e; ++i) {
+ const CodeGenRegister &Reg = Target.getRegisters()[i];
+ if (Name == Reg.TheDef->getValueAsString("AsmName"))
+ return Reg.TheDef;
+ }
+
+ return 0;
+}
+
+bool MatchableInfo::Validate(StringRef CommentDelimiter, bool Hack) const {
+ // Reject matchables with no .s string.
+ if (AsmString.empty())
+ throw TGError(TheDef->getLoc(), "instruction with empty asm string");
+
+ // Reject any matchables with a newline in them, they should be marked
+ // isCodeGenOnly if they are pseudo instructions.
+ if (AsmString.find('\n') != std::string::npos)
+ throw TGError(TheDef->getLoc(),
+ "multiline instruction is not valid for the asmparser, "
+ "mark it isCodeGenOnly");
+
+ // Remove comments from the asm string. We know that the asmstring only
+ // has one line.
+ if (!CommentDelimiter.empty() &&
+ StringRef(AsmString).find(CommentDelimiter) != StringRef::npos)
+ throw TGError(TheDef->getLoc(),
+ "asmstring for instruction has comment character in it, "
+ "mark it isCodeGenOnly");
+
+ // Reject matchables with operand modifiers, these aren't something we can
+ /// handle, the target should be refactored to use operands instead of
+ /// modifiers.
+ //
+ // Also, check for instructions which reference the operand multiple times;
+ // this implies a constraint we would not honor.
+ std::set<std::string> OperandNames;
+ for (unsigned i = 1, e = Tokens.size(); i < e; ++i) {
+ if (Tokens[i][0] == '$' && Tokens[i].find(':') != StringRef::npos)
+ throw TGError(TheDef->getLoc(),
+ "matchable with operand modifier '" + Tokens[i].str() +
+ "' not supported by asm matcher. Mark isCodeGenOnly!");
+
+ // Verify that any operand is only mentioned once.
+ if (Tokens[i][0] == '$' && !OperandNames.insert(Tokens[i]).second) {
+ if (!Hack)
+ throw TGError(TheDef->getLoc(),
+ "ERROR: matchable with tied operand '" + Tokens[i].str() +
+ "' can never be matched!");
+ // FIXME: Should reject these. The ARM backend hits this with $lane in a
+ // bunch of instructions. It is unclear what the right answer is.
+ DEBUG({
+ errs() << "warning: '" << InstrName << "': "
+ << "ignoring instruction with tied operand '"
+ << Tokens[i].str() << "'\n";
+ });
+ return false;
+ }
+ }
+
+ return true;
+}
+
+
+/// getSingletonRegisterForToken - If the specified token is a singleton
+/// register, return the register name, otherwise return a null StringRef.
+Record *MatchableInfo::
+getSingletonRegisterForToken(unsigned i, const AsmMatcherInfo &Info) const {
+ StringRef Tok = Tokens[i];
+ if (!Tok.startswith(Info.RegisterPrefix))
+ return 0;
+
+ StringRef RegName = Tok.substr(Info.RegisterPrefix.size());
+ if (Record *Rec = getRegisterRecord(Info.Target, RegName))
+ return Rec;
+ // If there is no register prefix (i.e. "%" in "%eax"), then this may
+ // be some random non-register token, just ignore it.
+ if (Info.RegisterPrefix.empty())
+ return 0;
+
+ std::string Err = "unable to find register for '" + RegName.str() +
+ "' (which matches register prefix)";
+ throw TGError(TheDef->getLoc(), Err);
+}
+
+
+static std::string getEnumNameForToken(StringRef Str) {
+ std::string Res;
+
for (StringRef::iterator it = Str.begin(), ie = Str.end(); it != ie; ++it) {
switch (*it) {
case '*': Res += "_STAR_"; break;
case '%': Res += "_PCT_"; break;
case ':': Res += "_COLON_"; break;
-
default:
- if (isalnum(*it)) {
+ if (isalnum(*it))
Res += *it;
- } else {
+ else
Res += "_" + utostr((unsigned) *it) + "_";
- }
}
}
return Res;
}
-ClassInfo *AsmMatcherInfo::getTokenClass(const StringRef &Token) {
+ClassInfo *AsmMatcherInfo::getTokenClass(StringRef Token) {
ClassInfo *&Entry = TokenClasses[Token];
-
+
if (!Entry) {
Entry = new ClassInfo();
Entry->Kind = ClassInfo::Token;
return Entry;
}
-unsigned AsmMatcherInfo::getUserClassKind(const StringRef &Name) {
- unsigned &Entry = UserClasses[Name];
-
- if (!Entry)
- Entry = ClassInfo::UserClass0 + UserClasses.size() - 1;
+ClassInfo *
+AsmMatcherInfo::getOperandClass(StringRef Token,
+ const CGIOperandList::OperandInfo &OI) {
+ if (OI.Rec->isSubClassOf("RegisterClass")) {
+ ClassInfo *CI = RegisterClassClasses[OI.Rec];
- return Entry;
+ if (!CI)
+ throw TGError(OI.Rec->getLoc(), "register class has no class info!");
+
+ return CI;
+ }
+
+ assert(OI.Rec->isSubClassOf("Operand") && "Unexpected operand!");
+ Record *MatchClass = OI.Rec->getValueAsDef("ParserMatchClass");
+ ClassInfo *CI = AsmOperandClasses[MatchClass];
+
+ if (!CI)
+ throw TGError(OI.Rec->getLoc(), "operand has no match class!");
+
+ return CI;
}
-ClassInfo *
-AsmMatcherInfo::getOperandClass(const StringRef &Token,
- const CodeGenInstruction::OperandInfo &OI) {
- std::string ClassName;
- if (OI.Rec->isSubClassOf("RegisterClass")) {
- ClassName = "Reg";
- } else {
- try {
- ClassName = OI.Rec->getValueAsString("ParserMatchClass");
- assert(ClassName != "Reg" && "'Reg' class name is reserved!");
- } catch(...) {
- PrintError(OI.Rec->getLoc(), "operand has no match class!");
- ClassName = "Invalid";
+void AsmMatcherInfo::
+BuildRegisterClasses(SmallPtrSet<Record*, 16> &SingletonRegisters) {
+ std::vector<CodeGenRegisterClass> RegisterClasses;
+ std::vector<CodeGenRegister> Registers;
+
+ RegisterClasses = Target.getRegisterClasses();
+ Registers = Target.getRegisters();
+
+ // The register sets used for matching.
+ std::set< std::set<Record*> > RegisterSets;
+
+ // Gather the defined sets.
+ for (std::vector<CodeGenRegisterClass>::iterator it = RegisterClasses.begin(),
+ ie = RegisterClasses.end(); it != ie; ++it)
+ RegisterSets.insert(std::set<Record*>(it->Elements.begin(),
+ it->Elements.end()));
+
+ // Add any required singleton sets.
+ for (SmallPtrSet<Record*, 16>::iterator it = SingletonRegisters.begin(),
+ ie = SingletonRegisters.end(); it != ie; ++it) {
+ Record *Rec = *it;
+ RegisterSets.insert(std::set<Record*>(&Rec, &Rec + 1));
+ }
+
+ // Introduce derived sets where necessary (when a register does not determine
+ // a unique register set class), and build the mapping of registers to the set
+ // they should classify to.
+ std::map<Record*, std::set<Record*> > RegisterMap;
+ for (std::vector<CodeGenRegister>::iterator it = Registers.begin(),
+ ie = Registers.end(); it != ie; ++it) {
+ CodeGenRegister &CGR = *it;
+ // Compute the intersection of all sets containing this register.
+ std::set<Record*> ContainingSet;
+
+ for (std::set< std::set<Record*> >::iterator it = RegisterSets.begin(),
+ ie = RegisterSets.end(); it != ie; ++it) {
+ if (!it->count(CGR.TheDef))
+ continue;
+
+ if (ContainingSet.empty()) {
+ ContainingSet = *it;
+ } else {
+ std::set<Record*> Tmp;
+ std::swap(Tmp, ContainingSet);
+ std::insert_iterator< std::set<Record*> > II(ContainingSet,
+ ContainingSet.begin());
+ std::set_intersection(Tmp.begin(), Tmp.end(), it->begin(), it->end(),
+ II);
+ }
+ }
+
+ if (!ContainingSet.empty()) {
+ RegisterSets.insert(ContainingSet);
+ RegisterMap.insert(std::make_pair(CGR.TheDef, ContainingSet));
}
}
- ClassInfo *&Entry = OperandClasses[ClassName];
-
- if (!Entry) {
- Entry = new ClassInfo();
- // FIXME: Hack.
- if (ClassName == "Reg") {
- Entry->Kind = ClassInfo::Register;
+ // Construct the register classes.
+ std::map<std::set<Record*>, ClassInfo*> RegisterSetClasses;
+ unsigned Index = 0;
+ for (std::set< std::set<Record*> >::iterator it = RegisterSets.begin(),
+ ie = RegisterSets.end(); it != ie; ++it, ++Index) {
+ ClassInfo *CI = new ClassInfo();
+ CI->Kind = ClassInfo::RegisterClass0 + Index;
+ CI->ClassName = "Reg" + utostr(Index);
+ CI->Name = "MCK_Reg" + utostr(Index);
+ CI->ValueName = "";
+ CI->PredicateMethod = ""; // unused
+ CI->RenderMethod = "addRegOperands";
+ CI->Registers = *it;
+ Classes.push_back(CI);
+ RegisterSetClasses.insert(std::make_pair(*it, CI));
+ }
+
+ // Find the superclasses; we could compute only the subgroup lattice edges,
+ // but there isn't really a point.
+ for (std::set< std::set<Record*> >::iterator it = RegisterSets.begin(),
+ ie = RegisterSets.end(); it != ie; ++it) {
+ ClassInfo *CI = RegisterSetClasses[*it];
+ for (std::set< std::set<Record*> >::iterator it2 = RegisterSets.begin(),
+ ie2 = RegisterSets.end(); it2 != ie2; ++it2)
+ if (*it != *it2 &&
+ std::includes(it2->begin(), it2->end(), it->begin(), it->end()))
+ CI->SuperClasses.push_back(RegisterSetClasses[*it2]);
+ }
+
+ // Name the register classes which correspond to a user defined RegisterClass.
+ for (std::vector<CodeGenRegisterClass>::iterator it = RegisterClasses.begin(),
+ ie = RegisterClasses.end(); it != ie; ++it) {
+ ClassInfo *CI = RegisterSetClasses[std::set<Record*>(it->Elements.begin(),
+ it->Elements.end())];
+ if (CI->ValueName.empty()) {
+ CI->ClassName = it->getName();
+ CI->Name = "MCK_" + it->getName();
+ CI->ValueName = it->getName();
+ } else
+ CI->ValueName = CI->ValueName + "," + it->getName();
+
+ RegisterClassClasses.insert(std::make_pair(it->TheDef, CI));
+ }
+
+ // Populate the map for individual registers.
+ for (std::map<Record*, std::set<Record*> >::iterator it = RegisterMap.begin(),
+ ie = RegisterMap.end(); it != ie; ++it)
+ this->RegisterClasses[it->first] = RegisterSetClasses[it->second];
+
+ // Name the register classes which correspond to singleton registers.
+ for (SmallPtrSet<Record*, 16>::iterator it = SingletonRegisters.begin(),
+ ie = SingletonRegisters.end(); it != ie; ++it) {
+ Record *Rec = *it;
+ ClassInfo *CI = this->RegisterClasses[Rec];
+ assert(CI && "Missing singleton register class info!");
+
+ if (CI->ValueName.empty()) {
+ CI->ClassName = Rec->getName();
+ CI->Name = "MCK_" + Rec->getName();
+ CI->ValueName = Rec->getName();
+ } else
+ CI->ValueName = CI->ValueName + "," + Rec->getName();
+ }
+}
+
+void AsmMatcherInfo::BuildOperandClasses() {
+ std::vector<Record*> AsmOperands =
+ Records.getAllDerivedDefinitions("AsmOperandClass");
+
+ // Pre-populate AsmOperandClasses map.
+ for (std::vector<Record*>::iterator it = AsmOperands.begin(),
+ ie = AsmOperands.end(); it != ie; ++it)
+ AsmOperandClasses[*it] = new ClassInfo();
+
+ unsigned Index = 0;
+ for (std::vector<Record*>::iterator it = AsmOperands.begin(),
+ ie = AsmOperands.end(); it != ie; ++it, ++Index) {
+ ClassInfo *CI = AsmOperandClasses[*it];
+ CI->Kind = ClassInfo::UserClass0 + Index;
+
+ ListInit *Supers = (*it)->getValueAsListInit("SuperClasses");
+ for (unsigned i = 0, e = Supers->getSize(); i != e; ++i) {
+ DefInit *DI = dynamic_cast<DefInit*>(Supers->getElement(i));
+ if (!DI) {
+ PrintError((*it)->getLoc(), "Invalid super class reference!");
+ continue;
+ }
+
+ ClassInfo *SC = AsmOperandClasses[DI->getDef()];
+ if (!SC)
+ PrintError((*it)->getLoc(), "Invalid super class reference!");
+ else
+ CI->SuperClasses.push_back(SC);
+ }
+ CI->ClassName = (*it)->getValueAsString("Name");
+ CI->Name = "MCK_" + CI->ClassName;
+ CI->ValueName = (*it)->getName();
+
+ // Get or construct the predicate method name.
+ Init *PMName = (*it)->getValueInit("PredicateMethod");
+ if (StringInit *SI = dynamic_cast<StringInit*>(PMName)) {
+ CI->PredicateMethod = SI->getValue();
} else {
- Entry->Kind = getUserClassKind(ClassName);
+ assert(dynamic_cast<UnsetInit*>(PMName) &&
+ "Unexpected PredicateMethod field!");
+ CI->PredicateMethod = "is" + CI->ClassName;
}
- Entry->ClassName = ClassName;
- Entry->Name = "MCK_" + ClassName;
- Entry->ValueName = OI.Rec->getName();
- Entry->PredicateMethod = "is" + ClassName;
- Entry->RenderMethod = "add" + ClassName + "Operands";
- Classes.push_back(Entry);
+
+ // Get or construct the render method name.
+ Init *RMName = (*it)->getValueInit("RenderMethod");
+ if (StringInit *SI = dynamic_cast<StringInit*>(RMName)) {
+ CI->RenderMethod = SI->getValue();
+ } else {
+ assert(dynamic_cast<UnsetInit*>(RMName) &&
+ "Unexpected RenderMethod field!");
+ CI->RenderMethod = "add" + CI->ClassName + "Operands";
+ }
+
+ AsmOperandClasses[*it] = CI;
+ Classes.push_back(CI);
}
-
- return Entry;
}
-void AsmMatcherInfo::BuildInfo(CodeGenTarget &Target) {
- for (std::map<std::string, CodeGenInstruction>::const_iterator
- it = Target.getInstructions().begin(),
- ie = Target.getInstructions().end();
- it != ie; ++it) {
- const CodeGenInstruction &CGI = it->second;
+AsmMatcherInfo::AsmMatcherInfo(Record *asmParser, CodeGenTarget &target)
+ : AsmParser(asmParser), Target(target),
+ RegisterPrefix(AsmParser->getValueAsString("RegisterPrefix")) {
+}
- if (!StringRef(it->first).startswith(MatchPrefix))
- continue;
- OwningPtr<InstructionInfo> II(new InstructionInfo);
+void AsmMatcherInfo::BuildInfo() {
+ // Build information about all of the AssemblerPredicates.
+ std::vector<Record*> AllPredicates =
+ Records.getAllDerivedDefinitions("Predicate");
+ for (unsigned i = 0, e = AllPredicates.size(); i != e; ++i) {
+ Record *Pred = AllPredicates[i];
+ // Ignore predicates that are not intended for the assembler.
+ if (!Pred->getValueAsBit("AssemblerMatcherPredicate"))
+ continue;
- II->InstrName = it->first;
- II->Instr = &it->second;
- II->AsmString = FlattenVariants(CGI.AsmString, 0);
+ if (Pred->getName().empty())
+ throw TGError(Pred->getLoc(), "Predicate has no name!");
+
+ unsigned FeatureNo = SubtargetFeatures.size();
+ SubtargetFeatures[Pred] = new SubtargetFeatureInfo(Pred, FeatureNo);
+ assert(FeatureNo < 32 && "Too many subtarget features!");
+ }
- TokenizeAsmString(II->AsmString, II->Tokens);
+ StringRef CommentDelimiter = AsmParser->getValueAsString("CommentDelimiter");
+
+ // Parse the instructions; we need to do this first so that we can gather the
+ // singleton register classes.
+ SmallPtrSet<Record*, 16> SingletonRegisters;
+ for (CodeGenTarget::inst_iterator I = Target.inst_begin(),
+ E = Target.inst_end(); I != E; ++I) {
+ const CodeGenInstruction &CGI = **I;
+
+ // If the tblgen -match-prefix option is specified (for tblgen hackers),
+ // filter the set of instructions we consider.
+ if (!StringRef(CGI.TheDef->getName()).startswith(MatchPrefix))
+ continue;
- // Ignore instructions which shouldn't be matched.
- if (!IsAssemblerInstruction(it->first, CGI, II->Tokens))
+ // Ignore "codegen only" instructions.
+ if (CGI.TheDef->getValueAsBit("isCodeGenOnly"))
continue;
+
+ OwningPtr<MatchableInfo> II(new MatchableInfo(CGI));
- for (unsigned i = 0, e = II->Tokens.size(); i != e; ++i) {
+ II->Initialize(*this, SingletonRegisters);
+
+ // Ignore instructions which shouldn't be matched and diagnose invalid
+ // instruction definitions with an error.
+ if (!II->Validate(CommentDelimiter, true))
+ continue;
+
+ // Ignore "Int_*" and "*_Int" instructions, which are internal aliases.
+ //
+ // FIXME: This is a total hack.
+ if (StringRef(II->InstrName).startswith("Int_") ||
+ StringRef(II->InstrName).endswith("_Int"))
+ continue;
+
+ Matchables.push_back(II.take());
+ }
+
+ // Parse all of the InstAlias definitions and stick them in the list of
+ // matchables.
+ std::vector<Record*> AllInstAliases =
+ Records.getAllDerivedDefinitions("InstAlias");
+ for (unsigned i = 0, e = AllInstAliases.size(); i != e; ++i) {
+ CodeGenInstAlias *Alias = new CodeGenInstAlias(AllInstAliases[i]);
+
+ OwningPtr<MatchableInfo> II(new MatchableInfo(Alias));
+
+ II->Initialize(*this, SingletonRegisters);
+
+ // Validate the alias definitions.
+ II->Validate(CommentDelimiter, false);
+
+ Matchables.push_back(II.take());
+ }
+
+ // Build info for the register classes.
+ BuildRegisterClasses(SingletonRegisters);
+
+ // Build info for the user defined assembly operand classes.
+ BuildOperandClasses();
+
+ // Build the information about matchables.
+ for (std::vector<MatchableInfo*>::iterator it = Matchables.begin(),
+ ie = Matchables.end(); it != ie; ++it) {
+ MatchableInfo *II = *it;
+
+ // The first token of the instruction is the mnemonic, which must be a
+ // simple string, not a $foo variable or a singleton register.
+ assert(!II->Tokens.empty() && "Instruction has no tokens?");
+ StringRef Mnemonic = II->Tokens[0];
+ if (Mnemonic[0] == '$' || II->getSingletonRegisterForToken(0, *this))
+ throw TGError(II->TheDef->getLoc(),
+ "Invalid instruction mnemonic '" + Mnemonic.str() + "'!");
+
+ // Parse the tokens after the mnemonic.
+ for (unsigned i = 1, e = II->Tokens.size(); i != e; ++i) {
StringRef Token = II->Tokens[i];
+ // Check for singleton registers.
+ if (Record *RegRecord = II->getSingletonRegisterForToken(i, *this)) {
+ MatchableInfo::Operand Op(RegisterClasses[RegRecord], 0);
+ assert(Op.Class && Op.Class->Registers.size() == 1 &&
+ "Unexpected class for singleton register");
+ II->Operands.push_back(Op);
+ continue;
+ }
+
// Check for simple tokens.
if (Token[0] != '$') {
- InstructionInfo::Operand Op;
- Op.Class = getTokenClass(Token);
- Op.OperandInfo = 0;
- II->Operands.push_back(Op);
+ II->Operands.push_back(MatchableInfo::Operand(getTokenClass(Token), 0));
continue;
}
// Map this token to an operand. FIXME: Move elsewhere.
unsigned Idx;
- try {
- Idx = CGI.getOperandNamed(OperandName);
- } catch(...) {
- errs() << "error: unable to find operand: '" << OperandName << "'!\n";
- break;
+ if (!II->OperandList.hasOperandNamed(OperandName, Idx))
+ throw TGError(II->TheDef->getLoc(), "error: unable to find operand: '" +
+ OperandName.str() + "'");
+
+ // FIXME: This is annoying, the named operand may be tied (e.g.,
+ // XCHG8rm). What we want is the untied operand, which we now have to
+ // grovel for. Only worry about this for single entry operands, we have to
+ // clean this up anyway.
+ const CGIOperandList::OperandInfo *OI = &II->OperandList[Idx];
+ if (OI->Constraints[0].isTied()) {
+ unsigned TiedOp = OI->Constraints[0].getTiedOperand();
+
+ // The tied operand index is an MIOperand index, find the operand that
+ // contains it.
+ for (unsigned i = 0, e = II->OperandList.size(); i != e; ++i) {
+ if (II->OperandList[i].MIOperandNo == TiedOp) {
+ OI = &II->OperandList[i];
+ break;
+ }
+ }
+
+ assert(OI && "Unable to find tied operand target!");
}
- const CodeGenInstruction::OperandInfo &OI = CGI.OperandList[Idx];
- InstructionInfo::Operand Op;
- Op.Class = getOperandClass(Token, OI);
- Op.OperandInfo = &OI;
- II->Operands.push_back(Op);
+ II->Operands.push_back(MatchableInfo::Operand(getOperandClass(Token,
+ *OI), OI));
}
+ }
- // If we broke out, ignore the instruction.
- if (II->Operands.size() != II->Tokens.size())
- continue;
+ // Reorder classes so that classes preceed super classes.
+ std::sort(Classes.begin(), Classes.end(), less_ptr<ClassInfo>());
+}
- Instructions.push_back(II.take());
- }
+static std::pair<unsigned, unsigned> *
+GetTiedOperandAtIndex(SmallVectorImpl<std::pair<unsigned, unsigned> > &List,
+ unsigned Index) {
+ for (unsigned i = 0, e = List.size(); i != e; ++i)
+ if (Index == List[i].first)
+ return &List[i];
+
+ return 0;
}
static void EmitConvertToMCInst(CodeGenTarget &Target,
- std::vector<InstructionInfo*> &Infos,
+ std::vector<MatchableInfo*> &Infos,
raw_ostream &OS) {
// Write the convert function to a separate stream, so we can drop it after
// the enum.
// Start the unified conversion function.
- CvtOS << "static bool ConvertToMCInst(ConversionKind Kind, MCInst &Inst, "
+ CvtOS << "static void ConvertToMCInst(ConversionKind Kind, MCInst &Inst, "
<< "unsigned Opcode,\n"
- << " SmallVectorImpl<"
- << Target.getName() << "Operand> &Operands) {\n";
+ << " const SmallVectorImpl<MCParsedAsmOperand*"
+ << "> &Operands) {\n";
CvtOS << " Inst.setOpcode(Opcode);\n";
CvtOS << " switch (Kind) {\n";
CvtOS << " default:\n";
OS << "// Unified function for converting operants to MCInst instances.\n\n";
OS << "enum ConversionKind {\n";
-
- for (std::vector<InstructionInfo*>::const_iterator it = Infos.begin(),
+
+ // TargetOperandClass - This is the target's operand class, like X86Operand.
+ std::string TargetOperandClass = Target.getName() + "Operand";
+
+ for (std::vector<MatchableInfo*>::const_iterator it = Infos.begin(),
ie = Infos.end(); it != ie; ++it) {
- InstructionInfo &II = **it;
+ MatchableInfo &II = **it;
// Order the (class) operands by the order to convert them into an MCInst.
SmallVector<std::pair<unsigned, unsigned>, 4> MIOperandList;
for (unsigned i = 0, e = II.Operands.size(); i != e; ++i) {
- InstructionInfo::Operand &Op = II.Operands[i];
+ MatchableInfo::Operand &Op = II.Operands[i];
if (Op.OperandInfo)
MIOperandList.push_back(std::make_pair(Op.OperandInfo->MIOperandNo, i));
}
- std::sort(MIOperandList.begin(), MIOperandList.end());
+
+ // Find any tied operands.
+ SmallVector<std::pair<unsigned, unsigned>, 4> TiedOperands;
+ for (unsigned i = 0, e = II.OperandList.size(); i != e; ++i) {
+ const CGIOperandList::OperandInfo &OpInfo = II.OperandList[i];
+ for (unsigned j = 0, e = OpInfo.Constraints.size(); j != e; ++j) {
+ const CGIOperandList::ConstraintInfo &CI = OpInfo.Constraints[j];
+ if (CI.isTied())
+ TiedOperands.push_back(std::make_pair(OpInfo.MIOperandNo + j,
+ CI.getTiedOperand()));
+ }
+ }
+
+ array_pod_sort(MIOperandList.begin(), MIOperandList.end());
// Compute the total number of operands.
unsigned NumMIOperands = 0;
- for (unsigned i = 0, e = II.Instr->OperandList.size(); i != e; ++i) {
- const CodeGenInstruction::OperandInfo &OI = II.Instr->OperandList[i];
- NumMIOperands = std::max(NumMIOperands,
+ for (unsigned i = 0, e = II.OperandList.size(); i != e; ++i) {
+ const CGIOperandList::OperandInfo &OI = II.OperandList[i];
+ NumMIOperands = std::max(NumMIOperands,
OI.MIOperandNo + OI.MINumOperands);
}
std::string Signature = "Convert";
unsigned CurIndex = 0;
for (unsigned i = 0, e = MIOperandList.size(); i != e; ++i) {
- InstructionInfo::Operand &Op = II.Operands[MIOperandList[i].second];
+ MatchableInfo::Operand &Op = II.Operands[MIOperandList[i].second];
assert(CurIndex <= Op.OperandInfo->MIOperandNo &&
"Duplicate match for instruction operand!");
-
- Signature += "_";
// Skip operands which weren't matched by anything, this occurs when the
// .td file encodes "implicit" operands as explicit ones.
//
// FIXME: This should be removed from the MCInst structure.
- for (; CurIndex != Op.OperandInfo->MIOperandNo; ++CurIndex)
- Signature += "Imp";
+ for (; CurIndex != Op.OperandInfo->MIOperandNo; ++CurIndex) {
+ std::pair<unsigned, unsigned> *Tie = GetTiedOperandAtIndex(TiedOperands,
+ CurIndex);
+ if (!Tie)
+ Signature += "__Imp";
+ else
+ Signature += "__Tie" + utostr(Tie->second);
+ }
+
+ Signature += "__";
- Signature += Op.Class->ClassName;
+ // Registers are always converted the same, don't duplicate the conversion
+ // function based on them.
+ //
+ // FIXME: We could generalize this based on the render method, if it
+ // mattered.
+ if (Op.Class->isRegisterClass())
+ Signature += "Reg";
+ else
+ Signature += Op.Class->ClassName;
Signature += utostr(Op.OperandInfo->MINumOperands);
Signature += "_" + utostr(MIOperandList[i].second);
}
// Add any trailing implicit operands.
- for (; CurIndex != NumMIOperands; ++CurIndex)
- Signature += "Imp";
+ for (; CurIndex != NumMIOperands; ++CurIndex) {
+ std::pair<unsigned, unsigned> *Tie = GetTiedOperandAtIndex(TiedOperands,
+ CurIndex);
+ if (!Tie)
+ Signature += "__Imp";
+ else
+ Signature += "__Tie" + utostr(Tie->second);
+ }
II.ConversionFnKind = Signature;
CvtOS << " case " << Signature << ":\n";
CurIndex = 0;
for (unsigned i = 0, e = MIOperandList.size(); i != e; ++i) {
- InstructionInfo::Operand &Op = II.Operands[MIOperandList[i].second];
+ MatchableInfo::Operand &Op = II.Operands[MIOperandList[i].second];
// Add the implicit operands.
- for (; CurIndex != Op.OperandInfo->MIOperandNo; ++CurIndex)
- CvtOS << " Inst.addOperand(MCOperand::CreateReg(0));\n";
+ for (; CurIndex != Op.OperandInfo->MIOperandNo; ++CurIndex) {
+ // See if this is a tied operand.
+ std::pair<unsigned, unsigned> *Tie = GetTiedOperandAtIndex(TiedOperands,
+ CurIndex);
+
+ if (!Tie) {
+ // If not, this is some implicit operand. Just assume it is a register
+ // for now.
+ CvtOS << " Inst.addOperand(MCOperand::CreateReg(0));\n";
+ } else {
+ // Copy the tied operand.
+ assert(Tie->first>Tie->second && "Tied operand preceeds its target!");
+ CvtOS << " Inst.addOperand(Inst.getOperand("
+ << Tie->second << "));\n";
+ }
+ }
- CvtOS << " Operands[" << MIOperandList[i].second
- << "]." << Op.Class->RenderMethod
+ CvtOS << " ((" << TargetOperandClass << "*)Operands["
+ << MIOperandList[i].second
+ << "+1])->" << Op.Class->RenderMethod
<< "(Inst, " << Op.OperandInfo->MINumOperands << ");\n";
CurIndex += Op.OperandInfo->MINumOperands;
}
-
+
// And add trailing implicit operands.
- for (; CurIndex != NumMIOperands; ++CurIndex)
- CvtOS << " Inst.addOperand(MCOperand::CreateReg(0));\n";
- CvtOS << " break;\n";
+ for (; CurIndex != NumMIOperands; ++CurIndex) {
+ std::pair<unsigned, unsigned> *Tie = GetTiedOperandAtIndex(TiedOperands,
+ CurIndex);
+
+ if (!Tie) {
+ // If not, this is some implicit operand. Just assume it is a register
+ // for now.
+ CvtOS << " Inst.addOperand(MCOperand::CreateReg(0));\n";
+ } else {
+ // Copy the tied operand.
+ assert(Tie->first>Tie->second && "Tied operand preceeds its target!");
+ CvtOS << " Inst.addOperand(Inst.getOperand("
+ << Tie->second << "));\n";
+ }
+ }
+
+ CvtOS << " return;\n";
}
// Finish the convert function.
CvtOS << " }\n";
- CvtOS << " return false;\n";
CvtOS << "}\n\n";
// Finish the enum, and drop the convert function after it.
OS << " NumConversionVariants\n";
OS << "};\n\n";
-
+
OS << CvtOS.str();
}
<< "/// instruction matching.\n";
OS << "enum MatchClassKind {\n";
OS << " InvalidMatchClass = 0,\n";
- for (std::vector<ClassInfo*>::iterator it = Infos.begin(),
+ for (std::vector<ClassInfo*>::iterator it = Infos.begin(),
ie = Infos.end(); it != ie; ++it) {
ClassInfo &CI = **it;
OS << " " << CI.Name << ", // ";
if (CI.Kind == ClassInfo::Token) {
OS << "'" << CI.ValueName << "'\n";
- } else if (CI.Kind == ClassInfo::Register) {
+ } else if (CI.isRegisterClass()) {
if (!CI.ValueName.empty())
OS << "register class '" << CI.ValueName << "'\n";
else
}
/// EmitClassifyOperand - Emit the function to classify an operand.
-static void EmitClassifyOperand(CodeGenTarget &Target,
- std::vector<ClassInfo*> &Infos,
+static void EmitClassifyOperand(AsmMatcherInfo &Info,
raw_ostream &OS) {
- OS << "static MatchClassKind ClassifyOperand("
- << Target.getName() << "Operand &Operand) {\n";
+ OS << "static MatchClassKind ClassifyOperand(MCParsedAsmOperand *GOp) {\n"
+ << " " << Info.Target.getName() << "Operand &Operand = *("
+ << Info.Target.getName() << "Operand*)GOp;\n";
+
+ // Classify tokens.
OS << " if (Operand.isToken())\n";
OS << " return MatchTokenString(Operand.getToken());\n\n";
- for (std::vector<ClassInfo*>::iterator it = Infos.begin(),
- ie = Infos.end(); it != ie; ++it) {
+
+ // Classify registers.
+ //
+ // FIXME: Don't hardcode isReg, getReg.
+ OS << " if (Operand.isReg()) {\n";
+ OS << " switch (Operand.getReg()) {\n";
+ OS << " default: return InvalidMatchClass;\n";
+ for (std::map<Record*, ClassInfo*>::iterator
+ it = Info.RegisterClasses.begin(), ie = Info.RegisterClasses.end();
+ it != ie; ++it)
+ OS << " case " << Info.Target.getName() << "::"
+ << it->first->getName() << ": return " << it->second->Name << ";\n";
+ OS << " }\n";
+ OS << " }\n\n";
+
+ // Classify user defined operands.
+ for (std::vector<ClassInfo*>::iterator it = Info.Classes.begin(),
+ ie = Info.Classes.end(); it != ie; ++it) {
ClassInfo &CI = **it;
- if (CI.Kind != ClassInfo::Token) {
- OS << " if (Operand." << CI.PredicateMethod << "())\n";
- OS << " return " << CI.Name << ";\n\n";
+ if (!CI.isUserClass())
+ continue;
+
+ OS << " // '" << CI.ClassName << "' class";
+ if (!CI.SuperClasses.empty()) {
+ OS << ", subclass of ";
+ for (unsigned i = 0, e = CI.SuperClasses.size(); i != e; ++i) {
+ if (i) OS << ", ";
+ OS << "'" << CI.SuperClasses[i]->ClassName << "'";
+ assert(CI < *CI.SuperClasses[i] && "Invalid class relation!");
+ }
}
+ OS << "\n";
+
+ OS << " if (Operand." << CI.PredicateMethod << "()) {\n";
+
+ // Validate subclass relationships.
+ if (!CI.SuperClasses.empty()) {
+ for (unsigned i = 0, e = CI.SuperClasses.size(); i != e; ++i)
+ OS << " assert(Operand." << CI.SuperClasses[i]->PredicateMethod
+ << "() && \"Invalid class relationship!\");\n";
+ }
+
+ OS << " return " << CI.Name << ";\n";
+ OS << " }\n\n";
}
OS << " return InvalidMatchClass;\n";
OS << "}\n\n";
}
-typedef std::pair<std::string, std::string> StringPair;
-
-/// FindFirstNonCommonLetter - Find the first character in the keys of the
-/// string pairs that is not shared across the whole set of strings. All
-/// strings are assumed to have the same length.
-static unsigned
-FindFirstNonCommonLetter(const std::vector<const StringPair*> &Matches) {
- assert(!Matches.empty());
- for (unsigned i = 0, e = Matches[0]->first.size(); i != e; ++i) {
- // Check to see if letter i is the same across the set.
- char Letter = Matches[0]->first[i];
-
- for (unsigned str = 0, e = Matches.size(); str != e; ++str)
- if (Matches[str]->first[i] != Letter)
- return i;
- }
-
- return Matches[0]->first.size();
-}
+/// EmitIsSubclass - Emit the subclass predicate function.
+static void EmitIsSubclass(CodeGenTarget &Target,
+ std::vector<ClassInfo*> &Infos,
+ raw_ostream &OS) {
+ OS << "/// IsSubclass - Compute whether \\arg A is a subclass of \\arg B.\n";
+ OS << "static bool IsSubclass(MatchClassKind A, MatchClassKind B) {\n";
+ OS << " if (A == B)\n";
+ OS << " return true;\n\n";
-/// EmitStringMatcherForChar - Given a set of strings that are known to be the
-/// same length and whose characters leading up to CharNo are the same, emit
-/// code to verify that CharNo and later are the same.
-///
-/// \return - True if control can leave the emitted code fragment.
-static bool EmitStringMatcherForChar(const std::string &StrVariableName,
- const std::vector<const StringPair*> &Matches,
- unsigned CharNo, unsigned IndentCount,
- raw_ostream &OS) {
- assert(!Matches.empty() && "Must have at least one string to match!");
- std::string Indent(IndentCount*2+4, ' ');
-
- // If we have verified that the entire string matches, we're done: output the
- // matching code.
- if (CharNo == Matches[0]->first.size()) {
- assert(Matches.size() == 1 && "Had duplicate keys to match on");
-
- // FIXME: If Matches[0].first has embeded \n, this will be bad.
- OS << Indent << Matches[0]->second << "\t // \"" << Matches[0]->first
- << "\"\n";
- return false;
- }
-
- // Bucket the matches by the character we are comparing.
- std::map<char, std::vector<const StringPair*> > MatchesByLetter;
-
- for (unsigned i = 0, e = Matches.size(); i != e; ++i)
- MatchesByLetter[Matches[i]->first[CharNo]].push_back(Matches[i]);
-
+ OS << " switch (A) {\n";
+ OS << " default:\n";
+ OS << " return false;\n";
+ for (std::vector<ClassInfo*>::iterator it = Infos.begin(),
+ ie = Infos.end(); it != ie; ++it) {
+ ClassInfo &A = **it;
- // If we have exactly one bucket to match, see how many characters are common
- // across the whole set and match all of them at once.
- if (MatchesByLetter.size() == 1) {
- unsigned FirstNonCommonLetter = FindFirstNonCommonLetter(Matches);
- unsigned NumChars = FirstNonCommonLetter-CharNo;
-
- // Emit code to break out if the prefix doesn't match.
- if (NumChars == 1) {
- // Do the comparison with if (Str[1] != 'f')
- // FIXME: Need to escape general characters.
- OS << Indent << "if (" << StrVariableName << "[" << CharNo << "] != '"
- << Matches[0]->first[CharNo] << "')\n";
- OS << Indent << " break;\n";
- } else {
- // Do the comparison with if (Str.substr(1,3) != "foo").
- // FIXME: Need to escape general strings.
- OS << Indent << "if (" << StrVariableName << ".substr(" << CharNo << ","
- << NumChars << ") != \"";
- OS << Matches[0]->first.substr(CharNo, NumChars) << "\")\n";
- OS << Indent << " break;\n";
- }
-
- return EmitStringMatcherForChar(StrVariableName, Matches,
- FirstNonCommonLetter, IndentCount, OS);
- }
-
- // Otherwise, we have multiple possible things, emit a switch on the
- // character.
- OS << Indent << "switch (" << StrVariableName << "[" << CharNo << "]) {\n";
- OS << Indent << "default: break;\n";
-
- for (std::map<char, std::vector<const StringPair*> >::iterator LI =
- MatchesByLetter.begin(), E = MatchesByLetter.end(); LI != E; ++LI) {
- // TODO: escape hard stuff (like \n) if we ever care about it.
- OS << Indent << "case '" << LI->first << "':\t // "
- << LI->second.size() << " strings to match.\n";
- if (EmitStringMatcherForChar(StrVariableName, LI->second, CharNo+1,
- IndentCount+1, OS))
- OS << Indent << " break;\n";
- }
-
- OS << Indent << "}\n";
- return true;
-}
+ if (A.Kind != ClassInfo::Token) {
+ std::vector<StringRef> SuperClasses;
+ for (std::vector<ClassInfo*>::iterator it = Infos.begin(),
+ ie = Infos.end(); it != ie; ++it) {
+ ClassInfo &B = **it;
+ if (&A != &B && A.isSubsetOf(B))
+ SuperClasses.push_back(B.Name);
+ }
-/// EmitStringMatcher - Given a list of strings and code to execute when they
-/// match, output a simple switch tree to classify the input string.
-///
-/// If a match is found, the code in Vals[i].second is executed; control must
-/// not exit this code fragment. If nothing matches, execution falls through.
-///
-/// \param StrVariableName - The name of the variable to test.
-static void EmitStringMatcher(const std::string &StrVariableName,
- const std::vector<StringPair> &Matches,
- raw_ostream &OS) {
- // First level categorization: group strings by length.
- std::map<unsigned, std::vector<const StringPair*> > MatchesByLength;
-
- for (unsigned i = 0, e = Matches.size(); i != e; ++i)
- MatchesByLength[Matches[i].first.size()].push_back(&Matches[i]);
-
- // Output a switch statement on length and categorize the elements within each
- // bin.
- OS << " switch (" << StrVariableName << ".size()) {\n";
- OS << " default: break;\n";
-
- for (std::map<unsigned, std::vector<const StringPair*> >::iterator LI =
- MatchesByLength.begin(), E = MatchesByLength.end(); LI != E; ++LI) {
- OS << " case " << LI->first << ":\t // " << LI->second.size()
- << " strings to match.\n";
- if (EmitStringMatcherForChar(StrVariableName, LI->second, 0, 0, OS))
- OS << " break;\n";
+ if (SuperClasses.empty())
+ continue;
+
+ OS << "\n case " << A.Name << ":\n";
+
+ if (SuperClasses.size() == 1) {
+ OS << " return B == " << SuperClasses.back() << ";\n";
+ continue;
+ }
+
+ OS << " switch (B) {\n";
+ OS << " default: return false;\n";
+ for (unsigned i = 0, e = SuperClasses.size(); i != e; ++i)
+ OS << " case " << SuperClasses[i] << ": return true;\n";
+ OS << " }\n";
+ }
}
-
OS << " }\n";
+ OS << "}\n\n";
}
+
/// EmitMatchTokenString - Emit the function to match a token string to the
/// appropriate match class value.
static void EmitMatchTokenString(CodeGenTarget &Target,
std::vector<ClassInfo*> &Infos,
raw_ostream &OS) {
// Construct the match list.
- std::vector<StringPair> Matches;
- for (std::vector<ClassInfo*>::iterator it = Infos.begin(),
+ std::vector<StringMatcher::StringPair> Matches;
+ for (std::vector<ClassInfo*>::iterator it = Infos.begin(),
ie = Infos.end(); it != ie; ++it) {
ClassInfo &CI = **it;
if (CI.Kind == ClassInfo::Token)
- Matches.push_back(StringPair(CI.ValueName, "return " + CI.Name + ";"));
+ Matches.push_back(StringMatcher::StringPair(CI.ValueName,
+ "return " + CI.Name + ";"));
}
- OS << "static MatchClassKind MatchTokenString(const StringRef &Name) {\n";
+ OS << "static MatchClassKind MatchTokenString(StringRef Name) {\n";
- EmitStringMatcher("Name", Matches, OS);
+ StringMatcher("Name", Matches, OS).Emit();
OS << " return InvalidMatchClass;\n";
OS << "}\n\n";
static void EmitMatchRegisterName(CodeGenTarget &Target, Record *AsmParser,
raw_ostream &OS) {
// Construct the match list.
- std::vector<StringPair> Matches;
+ std::vector<StringMatcher::StringPair> Matches;
for (unsigned i = 0, e = Target.getRegisters().size(); i != e; ++i) {
const CodeGenRegister &Reg = Target.getRegisters()[i];
if (Reg.TheDef->getValueAsString("AsmName").empty())
continue;
- Matches.push_back(StringPair(Reg.TheDef->getValueAsString("AsmName"),
- "return " + utostr(i + 1) + ";"));
+ Matches.push_back(StringMatcher::StringPair(
+ Reg.TheDef->getValueAsString("AsmName"),
+ "return " + utostr(i + 1) + ";"));
}
-
- OS << "unsigned " << Target.getName()
- << AsmParser->getValueAsString("AsmParserClassName")
- << "::MatchRegisterName(const StringRef &Name) {\n";
- EmitStringMatcher("Name", Matches, OS);
-
+ OS << "static unsigned MatchRegisterName(StringRef Name) {\n";
+
+ StringMatcher("Name", Matches, OS).Emit();
+
OS << " return 0;\n";
OS << "}\n\n";
}
+/// EmitSubtargetFeatureFlagEnumeration - Emit the subtarget feature flag
+/// definitions.
+static void EmitSubtargetFeatureFlagEnumeration(AsmMatcherInfo &Info,
+ raw_ostream &OS) {
+ OS << "// Flags for subtarget features that participate in "
+ << "instruction matching.\n";
+ OS << "enum SubtargetFeatureFlag {\n";
+ for (std::map<Record*, SubtargetFeatureInfo*>::const_iterator
+ it = Info.SubtargetFeatures.begin(),
+ ie = Info.SubtargetFeatures.end(); it != ie; ++it) {
+ SubtargetFeatureInfo &SFI = *it->second;
+ OS << " " << SFI.getEnumName() << " = (1 << " << SFI.Index << "),\n";
+ }
+ OS << " Feature_None = 0\n";
+ OS << "};\n\n";
+}
+
+/// EmitComputeAvailableFeatures - Emit the function to compute the list of
+/// available features given a subtarget.
+static void EmitComputeAvailableFeatures(AsmMatcherInfo &Info,
+ raw_ostream &OS) {
+ std::string ClassName =
+ Info.AsmParser->getValueAsString("AsmParserClassName");
+
+ OS << "unsigned " << Info.Target.getName() << ClassName << "::\n"
+ << "ComputeAvailableFeatures(const " << Info.Target.getName()
+ << "Subtarget *Subtarget) const {\n";
+ OS << " unsigned Features = 0;\n";
+ for (std::map<Record*, SubtargetFeatureInfo*>::const_iterator
+ it = Info.SubtargetFeatures.begin(),
+ ie = Info.SubtargetFeatures.end(); it != ie; ++it) {
+ SubtargetFeatureInfo &SFI = *it->second;
+ OS << " if (" << SFI.TheDef->getValueAsString("CondString")
+ << ")\n";
+ OS << " Features |= " << SFI.getEnumName() << ";\n";
+ }
+ OS << " return Features;\n";
+ OS << "}\n\n";
+}
+
+static std::string GetAliasRequiredFeatures(Record *R,
+ const AsmMatcherInfo &Info) {
+ std::vector<Record*> ReqFeatures = R->getValueAsListOfDefs("Predicates");
+ std::string Result;
+ unsigned NumFeatures = 0;
+ for (unsigned i = 0, e = ReqFeatures.size(); i != e; ++i) {
+ SubtargetFeatureInfo *F = Info.getSubtargetFeature(ReqFeatures[i]);
+
+ if (F == 0)
+ throw TGError(R->getLoc(), "Predicate '" + ReqFeatures[i]->getName() +
+ "' is not marked as an AssemblerPredicate!");
+
+ if (NumFeatures)
+ Result += '|';
+
+ Result += F->getEnumName();
+ ++NumFeatures;
+ }
+
+ if (NumFeatures > 1)
+ Result = '(' + Result + ')';
+ return Result;
+}
+
+/// EmitMnemonicAliases - If the target has any MnemonicAlias<> definitions,
+/// emit a function for them and return true, otherwise return false.
+static bool EmitMnemonicAliases(raw_ostream &OS, const AsmMatcherInfo &Info) {
+ std::vector<Record*> Aliases =
+ Records.getAllDerivedDefinitions("MnemonicAlias");
+ if (Aliases.empty()) return false;
+
+ OS << "static void ApplyMnemonicAliases(StringRef &Mnemonic, "
+ "unsigned Features) {\n";
+
+ // Keep track of all the aliases from a mnemonic. Use an std::map so that the
+ // iteration order of the map is stable.
+ std::map<std::string, std::vector<Record*> > AliasesFromMnemonic;
+
+ for (unsigned i = 0, e = Aliases.size(); i != e; ++i) {
+ Record *R = Aliases[i];
+ AliasesFromMnemonic[R->getValueAsString("FromMnemonic")].push_back(R);
+ }
+
+ // Process each alias a "from" mnemonic at a time, building the code executed
+ // by the string remapper.
+ std::vector<StringMatcher::StringPair> Cases;
+ for (std::map<std::string, std::vector<Record*> >::iterator
+ I = AliasesFromMnemonic.begin(), E = AliasesFromMnemonic.end();
+ I != E; ++I) {
+ const std::vector<Record*> &ToVec = I->second;
+
+ // Loop through each alias and emit code that handles each case. If there
+ // are two instructions without predicates, emit an error. If there is one,
+ // emit it last.
+ std::string MatchCode;
+ int AliasWithNoPredicate = -1;
+
+ for (unsigned i = 0, e = ToVec.size(); i != e; ++i) {
+ Record *R = ToVec[i];
+ std::string FeatureMask = GetAliasRequiredFeatures(R, Info);
+
+ // If this unconditionally matches, remember it for later and diagnose
+ // duplicates.
+ if (FeatureMask.empty()) {
+ if (AliasWithNoPredicate != -1) {
+ // We can't have two aliases from the same mnemonic with no predicate.
+ PrintError(ToVec[AliasWithNoPredicate]->getLoc(),
+ "two MnemonicAliases with the same 'from' mnemonic!");
+ throw TGError(R->getLoc(), "this is the other MnemonicAlias.");
+ }
+
+ AliasWithNoPredicate = i;
+ continue;
+ }
+
+ if (!MatchCode.empty())
+ MatchCode += "else ";
+ MatchCode += "if ((Features & " + FeatureMask + ") == "+FeatureMask+")\n";
+ MatchCode += " Mnemonic = \"" +R->getValueAsString("ToMnemonic")+"\";\n";
+ }
+
+ if (AliasWithNoPredicate != -1) {
+ Record *R = ToVec[AliasWithNoPredicate];
+ if (!MatchCode.empty())
+ MatchCode += "else\n ";
+ MatchCode += "Mnemonic = \"" + R->getValueAsString("ToMnemonic")+"\";\n";
+ }
+
+ MatchCode += "return;";
+
+ Cases.push_back(std::make_pair(I->first, MatchCode));
+ }
+
+
+ StringMatcher("Mnemonic", Cases, OS).Emit();
+ OS << "}\n";
+
+ return true;
+}
+
void AsmMatcherEmitter::run(raw_ostream &OS) {
CodeGenTarget Target;
Record *AsmParser = Target.getAsmParser();
std::string ClassName = AsmParser->getValueAsString("AsmParserClassName");
- EmitSourceFileHeader("Assembly Matcher Source Fragment", OS);
-
- // Emit the function to match a register name to number.
- EmitMatchRegisterName(Target, AsmParser, OS);
-
// Compute the information on the instructions to match.
- AsmMatcherInfo Info;
- Info.BuildInfo(Target);
+ AsmMatcherInfo Info(AsmParser, Target);
+ Info.BuildInfo();
+
+ // Sort the instruction table using the partial order on classes. We use
+ // stable_sort to ensure that ambiguous instructions are still
+ // deterministically ordered.
+ std::stable_sort(Info.Matchables.begin(), Info.Matchables.end(),
+ less_ptr<MatchableInfo>());
- // Sort the instruction table using the partial order on classes.
- std::sort(Info.Instructions.begin(), Info.Instructions.end(),
- less_ptr<InstructionInfo>());
-
DEBUG_WITH_TYPE("instruction_info", {
- for (std::vector<InstructionInfo*>::iterator
- it = Info.Instructions.begin(), ie = Info.Instructions.end();
+ for (std::vector<MatchableInfo*>::iterator
+ it = Info.Matchables.begin(), ie = Info.Matchables.end();
it != ie; ++it)
(*it)->dump();
});
- // Check for ambiguous instructions.
- unsigned NumAmbiguous = 0;
- for (std::vector<InstructionInfo*>::const_iterator it =
- Info.Instructions.begin(), ie = Info.Instructions.end() - 1;
- it != ie;) {
- InstructionInfo &II = **it;
- ++it;
-
- InstructionInfo &Next = **it;
-
- if (!(II < Next)){
- DEBUG_WITH_TYPE("ambiguous_instrs", {
- errs() << "warning: ambiguous instruction match:\n";
- II.dump();
+ // Check for ambiguous matchables.
+ DEBUG_WITH_TYPE("ambiguous_instrs", {
+ unsigned NumAmbiguous = 0;
+ for (unsigned i = 0, e = Info.Matchables.size(); i != e; ++i) {
+ for (unsigned j = i + 1; j != e; ++j) {
+ MatchableInfo &A = *Info.Matchables[i];
+ MatchableInfo &B = *Info.Matchables[j];
+
+ if (A.CouldMatchAmiguouslyWith(B)) {
+ errs() << "warning: ambiguous matchables:\n";
+ A.dump();
errs() << "\nis incomparable with:\n";
- Next.dump();
+ B.dump();
errs() << "\n\n";
- });
- ++NumAmbiguous;
+ ++NumAmbiguous;
+ }
+ }
}
- }
- if (NumAmbiguous)
- DEBUG_WITH_TYPE("ambiguous_instrs", {
- errs() << "warning: " << NumAmbiguous
- << " ambiguous instructions!\n";
- });
+ if (NumAmbiguous)
+ errs() << "warning: " << NumAmbiguous
+ << " ambiguous matchables!\n";
+ });
+
+ // Write the output.
+
+ EmitSourceFileHeader("Assembly Matcher Source Fragment", OS);
+
+ // Information for the class declaration.
+ OS << "\n#ifdef GET_ASSEMBLER_HEADER\n";
+ OS << "#undef GET_ASSEMBLER_HEADER\n";
+ OS << " // This should be included into the middle of the declaration of \n";
+ OS << " // your subclasses implementation of TargetAsmParser.\n";
+ OS << " unsigned ComputeAvailableFeatures(const " <<
+ Target.getName() << "Subtarget *Subtarget) const;\n";
+ OS << " enum MatchResultTy {\n";
+ OS << " Match_Success, Match_MnemonicFail, Match_InvalidOperand,\n";
+ OS << " Match_MissingFeature\n";
+ OS << " };\n";
+ OS << " MatchResultTy MatchInstructionImpl(const "
+ << "SmallVectorImpl<MCParsedAsmOperand*>"
+ << " &Operands, MCInst &Inst, unsigned &ErrorInfo);\n\n";
+ OS << "#endif // GET_ASSEMBLER_HEADER_INFO\n\n";
+
+
+
+
+ OS << "\n#ifdef GET_REGISTER_MATCHER\n";
+ OS << "#undef GET_REGISTER_MATCHER\n\n";
+
+ // Emit the subtarget feature enumeration.
+ EmitSubtargetFeatureFlagEnumeration(Info, OS);
+
+ // Emit the function to match a register name to number.
+ EmitMatchRegisterName(Target, AsmParser, OS);
+ OS << "#endif // GET_REGISTER_MATCHER\n\n";
+
+
+ OS << "\n#ifdef GET_MATCHER_IMPLEMENTATION\n";
+ OS << "#undef GET_MATCHER_IMPLEMENTATION\n\n";
+
+ // Generate the function that remaps for mnemonic aliases.
+ bool HasMnemonicAliases = EmitMnemonicAliases(OS, Info);
+
// Generate the unified function to convert operands into an MCInst.
- EmitConvertToMCInst(Target, Info.Instructions, OS);
+ EmitConvertToMCInst(Target, Info.Matchables, OS);
// Emit the enumeration for classes which participate in matching.
EmitMatchClassEnumeration(Target, Info.Classes, OS);
EmitMatchTokenString(Target, Info.Classes, OS);
// Emit the routine to classify an operand.
- EmitClassifyOperand(Target, Info.Classes, OS);
+ EmitClassifyOperand(Info, OS);
+
+ // Emit the subclass predicate routine.
+ EmitIsSubclass(Target, Info.Classes, OS);
+
+ // Emit the available features compute function.
+ EmitComputeAvailableFeatures(Info, OS);
- // Finally, build the match function.
size_t MaxNumOperands = 0;
- for (std::vector<InstructionInfo*>::const_iterator it =
- Info.Instructions.begin(), ie = Info.Instructions.end();
+ for (std::vector<MatchableInfo*>::const_iterator it =
+ Info.Matchables.begin(), ie = Info.Matchables.end();
it != ie; ++it)
MaxNumOperands = std::max(MaxNumOperands, (*it)->Operands.size());
-
- OS << "bool " << Target.getName() << ClassName
- << "::MatchInstruction("
- << "SmallVectorImpl<" << Target.getName() << "Operand> &Operands, "
- << "MCInst &Inst) {\n";
+
// Emit the static match table; unused classes get initalized to 0 which is
// guaranteed to be InvalidMatchClass.
// order the match kinds appropriately (putting mnemonics last), then we
// should only end up using a few bits for each class, especially the ones
// following the mnemonic.
- OS << " static const struct MatchEntry {\n";
+ OS << "namespace {\n";
+ OS << " struct MatchEntry {\n";
OS << " unsigned Opcode;\n";
+ OS << " const char *Mnemonic;\n";
OS << " ConversionKind ConvertFn;\n";
OS << " MatchClassKind Classes[" << MaxNumOperands << "];\n";
- OS << " } MatchTable[" << Info.Instructions.size() << "] = {\n";
+ OS << " unsigned RequiredFeatures;\n";
+ OS << " };\n\n";
- for (std::vector<InstructionInfo*>::const_iterator it =
- Info.Instructions.begin(), ie = Info.Instructions.end();
+ OS << "// Predicate for searching for an opcode.\n";
+ OS << " struct LessOpcode {\n";
+ OS << " bool operator()(const MatchEntry &LHS, StringRef RHS) {\n";
+ OS << " return StringRef(LHS.Mnemonic) < RHS;\n";
+ OS << " }\n";
+ OS << " bool operator()(StringRef LHS, const MatchEntry &RHS) {\n";
+ OS << " return LHS < StringRef(RHS.Mnemonic);\n";
+ OS << " }\n";
+ OS << " bool operator()(const MatchEntry &LHS, const MatchEntry &RHS) {\n";
+ OS << " return StringRef(LHS.Mnemonic) < StringRef(RHS.Mnemonic);\n";
+ OS << " }\n";
+ OS << " };\n";
+
+ OS << "} // end anonymous namespace.\n\n";
+
+ OS << "static const MatchEntry MatchTable["
+ << Info.Matchables.size() << "] = {\n";
+
+ for (std::vector<MatchableInfo*>::const_iterator it =
+ Info.Matchables.begin(), ie = Info.Matchables.end();
it != ie; ++it) {
- InstructionInfo &II = **it;
+ MatchableInfo &II = **it;
- OS << " { " << Target.getName() << "::" << II.InstrName
- << ", " << II.ConversionFnKind << ", { ";
+ OS << " { " << Target.getName() << "::" << II.InstrName
+ << ", \"" << II.Tokens[0] << "\""
+ << ", " << II.ConversionFnKind << ", { ";
for (unsigned i = 0, e = II.Operands.size(); i != e; ++i) {
- InstructionInfo::Operand &Op = II.Operands[i];
-
+ MatchableInfo::Operand &Op = II.Operands[i];
+
if (i) OS << ", ";
OS << Op.Class->Name;
}
- OS << " } },\n";
+ OS << " }, ";
+
+ // Write the required features mask.
+ if (!II.RequiredFeatures.empty()) {
+ for (unsigned i = 0, e = II.RequiredFeatures.size(); i != e; ++i) {
+ if (i) OS << "|";
+ OS << II.RequiredFeatures[i]->getEnumName();
+ }
+ } else
+ OS << "0";
+
+ OS << "},\n";
}
- OS << " };\n\n";
+ OS << "};\n\n";
+ // Finally, build the match function.
+ OS << Target.getName() << ClassName << "::MatchResultTy "
+ << Target.getName() << ClassName << "::\n"
+ << "MatchInstructionImpl(const SmallVectorImpl<MCParsedAsmOperand*>"
+ << " &Operands,\n";
+ OS << " MCInst &Inst, unsigned &ErrorInfo) {\n";
+
+ // Emit code to get the available features.
+ OS << " // Get the current feature set.\n";
+ OS << " unsigned AvailableFeatures = getAvailableFeatures();\n\n";
+
+ OS << " // Get the instruction mnemonic, which is the first token.\n";
+ OS << " StringRef Mnemonic = ((" << Target.getName()
+ << "Operand*)Operands[0])->getToken();\n\n";
+
+ if (HasMnemonicAliases) {
+ OS << " // Process all MnemonicAliases to remap the mnemonic.\n";
+ OS << " ApplyMnemonicAliases(Mnemonic, AvailableFeatures);\n\n";
+ }
+
// Emit code to compute the class list for this operand vector.
OS << " // Eliminate obvious mismatches.\n";
- OS << " if (Operands.size() > " << MaxNumOperands << ")\n";
- OS << " return true;\n\n";
+ OS << " if (Operands.size() > " << (MaxNumOperands+1) << ") {\n";
+ OS << " ErrorInfo = " << (MaxNumOperands+1) << ";\n";
+ OS << " return Match_InvalidOperand;\n";
+ OS << " }\n\n";
OS << " // Compute the class list for this operand vector.\n";
OS << " MatchClassKind Classes[" << MaxNumOperands << "];\n";
- OS << " for (unsigned i = 0, e = Operands.size(); i != e; ++i) {\n";
- OS << " Classes[i] = ClassifyOperand(Operands[i]);\n\n";
+ OS << " for (unsigned i = 1, e = Operands.size(); i != e; ++i) {\n";
+ OS << " Classes[i-1] = ClassifyOperand(Operands[i]);\n\n";
OS << " // Check for invalid operands before matching.\n";
- OS << " if (Classes[i] == InvalidMatchClass)\n";
- OS << " return true;\n";
+ OS << " if (Classes[i-1] == InvalidMatchClass) {\n";
+ OS << " ErrorInfo = i;\n";
+ OS << " return Match_InvalidOperand;\n";
+ OS << " }\n";
OS << " }\n\n";
OS << " // Mark unused classes.\n";
- OS << " for (unsigned i = Operands.size(), e = " << MaxNumOperands << "; "
+ OS << " for (unsigned i = Operands.size()-1, e = " << MaxNumOperands << "; "
<< "i != e; ++i)\n";
OS << " Classes[i] = InvalidMatchClass;\n\n";
+ OS << " // Some state to try to produce better error messages.\n";
+ OS << " bool HadMatchOtherThanFeatures = false;\n\n";
+ OS << " // Set ErrorInfo to the operand that mismatches if it is \n";
+ OS << " // wrong for all instances of the instruction.\n";
+ OS << " ErrorInfo = ~0U;\n";
+
// Emit code to search the table.
OS << " // Search the table.\n";
- OS << " for (const MatchEntry *it = MatchTable, "
- << "*ie = MatchTable + " << Info.Instructions.size()
- << "; it != ie; ++it) {\n";
- for (unsigned i = 0; i != MaxNumOperands; ++i) {
- OS << " if (Classes[" << i << "] != it->Classes[" << i << "])\n";
- OS << " continue;\n";
- }
+ OS << " std::pair<const MatchEntry*, const MatchEntry*> MnemonicRange =\n";
+ OS << " std::equal_range(MatchTable, MatchTable+"
+ << Info.Matchables.size() << ", Mnemonic, LessOpcode());\n\n";
+
+ OS << " // Return a more specific error code if no mnemonics match.\n";
+ OS << " if (MnemonicRange.first == MnemonicRange.second)\n";
+ OS << " return Match_MnemonicFail;\n\n";
+
+ OS << " for (const MatchEntry *it = MnemonicRange.first, "
+ << "*ie = MnemonicRange.second;\n";
+ OS << " it != ie; ++it) {\n";
+
+ OS << " // equal_range guarantees that instruction mnemonic matches.\n";
+ OS << " assert(Mnemonic == it->Mnemonic);\n";
+
+ // Emit check that the subclasses match.
+ OS << " bool OperandsValid = true;\n";
+ OS << " for (unsigned i = 0; i != " << MaxNumOperands << "; ++i) {\n";
+ OS << " if (IsSubclass(Classes[i], it->Classes[i]))\n";
+ OS << " continue;\n";
+ OS << " // If this operand is broken for all of the instances of this\n";
+ OS << " // mnemonic, keep track of it so we can report loc info.\n";
+ OS << " if (it == MnemonicRange.first || ErrorInfo == i+1)\n";
+ OS << " ErrorInfo = i+1;\n";
+ OS << " else\n";
+ OS << " ErrorInfo = ~0U;";
+ OS << " // Otherwise, just reject this instance of the mnemonic.\n";
+ OS << " OperandsValid = false;\n";
+ OS << " break;\n";
+ OS << " }\n\n";
+
+ OS << " if (!OperandsValid) continue;\n";
+
+ // Emit check that the required features are available.
+ OS << " if ((AvailableFeatures & it->RequiredFeatures) "
+ << "!= it->RequiredFeatures) {\n";
+ OS << " HadMatchOtherThanFeatures = true;\n";
+ OS << " continue;\n";
+ OS << " }\n";
+
OS << "\n";
- OS << " return ConvertToMCInst(it->ConvertFn, Inst, "
- << "it->Opcode, Operands);\n";
+ OS << " ConvertToMCInst(it->ConvertFn, Inst, it->Opcode, Operands);\n";
+
+ // Call the post-processing function, if used.
+ std::string InsnCleanupFn =
+ AsmParser->getValueAsString("AsmParserInstCleanup");
+ if (!InsnCleanupFn.empty())
+ OS << " " << InsnCleanupFn << "(Inst);\n";
+
+ OS << " return Match_Success;\n";
OS << " }\n\n";
- OS << " return true;\n";
+ OS << " // Okay, we had no match. Try to return a useful error code.\n";
+ OS << " if (HadMatchOtherThanFeatures) return Match_MissingFeature;\n";
+ OS << " return Match_InvalidOperand;\n";
OS << "}\n\n";
+
+ OS << "#endif // GET_MATCHER_IMPLEMENTATION\n\n";
}