[PATCH 1/2] TableGen: Generate a function for getting operand indices based on their defined names

Tom Stellard tom at stellard.net
Mon Jun 24 19:25:53 PDT 2013


On Mon, Jun 24, 2013 at 02:01:31PM -0700, Sean Silva wrote:
> On Mon, Jun 24, 2013 at 12:22 PM, Tom Stellard <tom at stellard.net> wrote:
> 
> > Ping.
> >
> >
> Oops, sorry.
> 
> +/// \p Operands [out] A map used to generate the OpName enum with operand
> +///     names as its keys and operand enum values as its values.
> +///  \p OperandMap [out] A map for representing the operand name mappings
> for
> +///     each instructions.  This is used to generate the OperandMap table
> as
> +///     well as the getNamedOperandIdx() function.
> 
> These are aligned differently. Also, the prevailing style in the codebase
> seems to be `\param`.
> 
> +  for (unsigned i = 0, e = NumberedInstructions.size(), NumOperands = 0;
> +                                                                  i != e;
> ++i) {
> 
> Could you pull NumOperands onto its own line before the `for` like you did
> for `TableIndex` in another part of the code? I think that's a lot more
> readable since it doesn't hide an unrelated variable inside the `for`
> header.
> 
> +        const std::vector<const CodeGenInstruction *> NumberedInstructions,
> +        std::string Namespace,
> 
> +  std::string Namespace = Target.getInstNamespace();
> 
> Do you need to be making copies here?
> 
> +    for (unsigned i = 0, e = Operands.size(); i != e; ++i)
> 
> I think this will provoke a -Wshadow warning. Best to rename the nested
> loop variables like you did elsewhere (also, it would be nice if you were
> consistent in the naming of the nested loop counters (`j`, `je` vs. `ii`,
> `ie`)).
> 
> Other than that, it looks good to me. Jakob, any other thoughts?
>

Hi Sean,

Here is an updated of the patch which addresses your comments, is this
OK to commit?

-Tom
-------------- next part --------------
>From 6df4fa213999cb42be6ace635014a163fded9eab Mon Sep 17 00:00:00 2001
From: Tom Stellard <thomas.stellard at amd.com>
Date: Thu, 30 May 2013 10:10:50 -0700
Subject: [PATCH] TableGen: Generate a function for getting operand indices
 based on their defined names v6

This patch modifies TableGen to generate a function in
${TARGET}GenInstrInfo.inc called getNamedOperandIdx(), which can be used
to look up indices for operands based on their names.

In order to activate this feature for an instruction, you must set the
UseNamedOperandTable bit.

For example, if you have an instruction like:

def ADD : TargetInstr <(outs GPR:$dst), (ins GPR:$src0, GPR:$src1)>;

You can look up the operand indices using the new function, like this:

Target::getNamedOperandIdx(Target::ADD, Target::OpName::dst)  => 0
Target::getNamedOperandIdx(Target::ADD, Target::OpName::src0) => 1
Target::getNamedOperandIdx(Target::ADD, Target::OpName::src1) => 2

The operand names are case sensitive, so $dst and $DST are considered
different operands.

This change is useful for R600 which has instructions with a large number
of operands, many of which model single bit instruction configuration
values.  These configuration bits are common across most instructions,
but may have a different operand index depending on the instruction type.
It is useful to have a convenient way to look up the operand indices,
so these bits can be generically set on any instruction.

v2:
  - Don't uppercase enum values
  - Use table compresion to reduce function size

v3:
  - Only generate table for instructions with the UseNamedOperandTable
    bit set.

v4:
  - Clean up code and add documentation

v5:
  - More code cleanups
---
 docs/WritingAnLLVMBackend.rst       |  41 ++++++++++++
 include/llvm/Target/Target.td       |   5 ++
 utils/TableGen/InstrInfoEmitter.cpp | 130 ++++++++++++++++++++++++++++++++++++
 3 files changed, 176 insertions(+)

diff --git a/docs/WritingAnLLVMBackend.rst b/docs/WritingAnLLVMBackend.rst
index a03a5e4..9e3cb9e 100644
--- a/docs/WritingAnLLVMBackend.rst
+++ b/docs/WritingAnLLVMBackend.rst
@@ -911,6 +911,47 @@ format instructions will bind the operands to the ``rd``, ``rs1``, and ``rs2``
 fields.  This results in the ``XNORrr`` instruction binding ``$dst``, ``$b``,
 and ``$c`` operands to the ``rd``, ``rs1``, and ``rs2`` fields respectively.
 
+TableGen will also generate a function called getNamedOperandIdx() which
+can be used to look up an operand's index in a MachineInstr based on its
+TableGen name.  Setting the UseNamedOperandTable bit in an instruction's
+TableGen definition will add all of its operands to an enumeration in the
+llvm::XXX:OpName namespace and also add an entry for it into the OperandMap
+table, which can be queried using getNamedOperandIdx()
+
+.. code-block:: llvm
+
+  int DstIndex = SP::getNamedOperandIdx(SP::XNORrr, SP::OpName::dst); // => 0
+  int BIndex = SP::getNamedOperandIdx(SP::XNORrr, SP::OpName::b);     // => 1
+  int CIndex = SP::getNamedOperandIdx(SP::XNORrr, SP::OpName::c);     // => 2
+  int DIndex = SP::getNamedOperandIdx(SP::XNORrr, SP::OpName::d);     // => -1
+
+  ...
+
+The entries in the OpName enum are taken verbatim from the TableGen definitions,
+so operands with lowercase names will have lower case entries in the enum.
+
+To include the getNamedOperandIdx() function in your backend, you will need
+to define a few preprocessor macros in XXXInstrInfo.cpp and XXXInstrInfo.h.
+For example:
+
+XXXInstrInfo.cpp:
+
+.. code-block:: c++ 
+
+  #define GET_INSTRINFO_NAMED_OPS // For getNamedOperandIdx() function
+  #include "XXXGenInstrInfo.inc"
+
+XXXInstrInfo.h:
+
+.. code-block:: c++
+
+  #define GET_INSTRINFO_OPERAND_ENUM // For OpName enum
+  #include "XXXGenInstrInfo.inc"
+
+  namespace XXX {
+    int16_t getNamedOperandIdx(uint16_t Opcode, uint16_t NamedIndex);
+  } // End namespace XXX
+
 Instruction Relation Mapping
 ----------------------------
 
diff --git a/include/llvm/Target/Target.td b/include/llvm/Target/Target.td
index a9644d4..97df64c 100644
--- a/include/llvm/Target/Target.td
+++ b/include/llvm/Target/Target.td
@@ -445,6 +445,11 @@ class Instruction {
   string TwoOperandAliasConstraint = "";
 
   ///@}
+
+  /// UseNamedOperandTable - If set, the operand indices of this instruction
+  /// can be queried via the getNamedOperandIdx() function which is generated
+  /// by TableGen.
+  bit UseNamedOperandTable = 0;
 }
 
 /// PseudoInstExpansion - Expansion information for a pseudo-instruction.
diff --git a/utils/TableGen/InstrInfoEmitter.cpp b/utils/TableGen/InstrInfoEmitter.cpp
index d6020a8..1df8ae5 100644
--- a/utils/TableGen/InstrInfoEmitter.cpp
+++ b/utils/TableGen/InstrInfoEmitter.cpp
@@ -45,11 +45,25 @@ private:
   void emitEnums(raw_ostream &OS);
 
   typedef std::map<std::vector<std::string>, unsigned> OperandInfoMapTy;
+
+  /// The keys of this map are maps which have OpName enum values as their keys
+  /// and instruction operand indices as their values.  The values of this map
+  /// are lists of instruction names.
+  typedef std::map<std::map<unsigned, unsigned>,
+                   std::vector<std::string> > OpNameMapTy;
+  typedef std::map<std::string, unsigned>::iterator StrUintMapIter;
   void emitRecord(const CodeGenInstruction &Inst, unsigned Num,
                   Record *InstrInfo,
                   std::map<std::vector<Record*>, unsigned> &EL,
                   const OperandInfoMapTy &OpInfo,
                   raw_ostream &OS);
+  void initOperandMapData(
+             const std::vector<const CodeGenInstruction *> NumberedInstructions,
+             const std::string &Namespace,
+             std::map<std::string, unsigned> &Operands,
+             OpNameMapTy &OperandMap);
+  void emitOperandNameMappings(raw_ostream &OS, const CodeGenTarget &Target,
+            const std::vector<const CodeGenInstruction*> &NumberedInstructions);
 
   // Operand information.
   void EmitOperandInfo(raw_ostream &OS, OperandInfoMapTy &OperandInfoIDs);
@@ -176,6 +190,120 @@ void InstrInfoEmitter::EmitOperandInfo(raw_ostream &OS,
   }
 }
 
+
+/// Initialize data structures for generating operand name mappings.
+/// 
+/// \param Operands [out] A map used to generate the OpName enum with operand
+///        names as its keys and operand enum values as its values.
+/// \param OperandMap [out] A map for representing the operand name mappings for
+///        each instructions.  This is used to generate the OperandMap table as
+///        well as the getNamedOperandIdx() function.
+void InstrInfoEmitter::initOperandMapData(
+        const std::vector<const CodeGenInstruction *> NumberedInstructions,
+        const std::string &Namespace,
+        std::map<std::string, unsigned> &Operands,
+        OpNameMapTy &OperandMap) {
+
+  unsigned NumOperands = 0;
+  for (unsigned i = 0, e = NumberedInstructions.size(); i != e; ++i) {
+    const CodeGenInstruction *Inst = NumberedInstructions[i];
+    if (!Inst->TheDef->getValueAsBit("UseNamedOperandTable")) {
+      continue;
+    }
+    std::map<unsigned, unsigned> OpList;
+    for (unsigned j = 0, je = Inst->Operands.size(); j != je; ++j) {
+      const CGIOperandList::OperandInfo &Info = Inst->Operands[j];
+      StrUintMapIter I = Operands.find(Info.Name);
+
+      if (I == Operands.end()) {
+        I = Operands.insert(Operands.begin(),
+                    std::pair<std::string, unsigned>(Info.Name, NumOperands++));
+      }
+      OpList[I->second] = Info.MIOperandNo;
+    }
+    OperandMap[OpList].push_back(Namespace + "::" + Inst->TheDef->getName());
+  }
+}
+
+/// Generate a table and function for looking up the indices of operands by
+/// name.
+///
+/// This code generates:
+/// - An enum in the llvm::TargetNamespace::OpName namespace, with one entry
+///   for each operand name.
+/// - A 2-dimensional table called OperandMap for mapping OpName enum values to
+///   operand indices.
+/// - A function called getNamedOperandIdx(uint16_t Opcode, uint16_t NamedIdx)
+///   for looking up the operand index for an instruction, given a value from
+///   OpName enum
+void InstrInfoEmitter::emitOperandNameMappings(raw_ostream &OS,
+           const CodeGenTarget &Target,
+           const std::vector<const CodeGenInstruction*> &NumberedInstructions) {
+
+  const std::string &Namespace = Target.getInstNamespace();
+  std::string OpNameNS = "OpName";
+  // Map of operand names to their enumeration value.  This will be used to
+  // generate the OpName enum.
+  std::map<std::string, unsigned> Operands;
+  OpNameMapTy OperandMap;
+
+  initOperandMapData(NumberedInstructions, Namespace, Operands, OperandMap);
+
+  OS << "#ifdef GET_INSTRINFO_OPERAND_ENUM\n";
+  OS << "#undef GET_INSTRINFO_OPERAND_ENUM\n";
+  OS << "namespace llvm {";
+  OS << "namespace " << Namespace << " {\n";
+  OS << "namespace " << OpNameNS << " { \n";
+  OS << "enum {\n";
+  for (StrUintMapIter i = Operands.begin(), e = Operands.end(); i != e; ++i)
+    OS << "  " << i->first << " = " << i->second << ",\n";
+
+  OS << "OPERAND_LAST";
+  OS << "\n};\n";
+  OS << "} // End namespace OpName\n";
+  OS << "} // End namespace " << Namespace << "\n";
+  OS << "} // End namespace llvm\n";
+  OS << "#endif //GET_INSTRINFO_OPERAND_ENUM\n";
+
+  OS << "#ifdef GET_INSTRINFO_NAMED_OPS\n";
+  OS << "#undef GET_INSTRINFO_NAMED_OPS\n";
+  OS << "namespace llvm {";
+  OS << "namespace " << Namespace << " {\n";
+  OS << "int16_t getNamedOperandIdx(uint16_t Opcode, uint16_t NamedIdx) {\n";
+  OS << "  static const int16_t OperandMap []["<< Operands.size() << "] = {\n";
+  for (OpNameMapTy::iterator i = OperandMap.begin(), e = OperandMap.end();
+                                                     i != e; ++i) {
+    const std::map<unsigned, unsigned> &OpList = i->first;
+    OS << "{";
+
+    // Emit a row of the OperandMap table
+    for (unsigned ii = 0, ie = Operands.size(); ii != ie; ++ii)
+      OS << (OpList.count(ii) == 0 ? -1 : (int)OpList.find(ii)->second) << ", ";
+
+    OS << "},\n";
+  }
+  OS << "};\n";
+
+  OS << "  switch(Opcode) {\n";
+  unsigned TableIndex = 0;
+  for (OpNameMapTy::iterator i = OperandMap.begin(), e = OperandMap.end();
+                                                     i != e; ++i) {
+    std::vector<std::string> &OpcodeList = i->second;
+
+    for (unsigned ii = 0, ie = OpcodeList.size(); ii != ie; ++ii)
+      OS << "  case " << OpcodeList[ii] << ":\n";
+
+    OS << "    return OperandMap[" << TableIndex++ << "][NamedIdx];\n";
+  }
+  OS << "    default: return -1;\n";
+  OS << "  }\n";
+  OS << "}\n";
+  OS << "} // End namespace " << Namespace << "\n";
+  OS << "} // End namespace llvm\n";
+  OS << "#endif //GET_INSTRINFO_NAMED_OPS\n";
+
+}
+
 //===----------------------------------------------------------------------===//
 // Main Output.
 //===----------------------------------------------------------------------===//
@@ -293,6 +421,8 @@ void InstrInfoEmitter::run(raw_ostream &OS) {
   OS << "} // End llvm namespace \n";
 
   OS << "#endif // GET_INSTRINFO_CTOR\n\n";
+
+  emitOperandNameMappings(OS, Target, NumberedInstructions);
 }
 
 void InstrInfoEmitter::emitRecord(const CodeGenInstruction &Inst, unsigned Num,
-- 
1.7.11.4



More information about the llvm-commits mailing list