[RFC][TableGen] Restructure TableGen Source

Lately, while trying to move as much code out of the GlobalISel Combiner Backend, I've been hitting more and more linker issues. It seems like the
library structure was due for an overhaul, so I created #80647

I'm proposing this refactor of the llvm-tblgen source into:
- a "Basic" library, which contains the bare minimum utilities to build `llvm-min-tablegen`
- a "Common" library which contains all of the helpers for TableGen backends. Such helpers can be shared by more than one backend, and even unit tested (e.g. CodeExpander is, maybe we can add more over time)

The backends remain in place and just make use of the new include paths.

Solves #80647

Author: Pierre-vh

llvm/unittests/TableGen/CMakeLists.txt

@@ -15,4 +15,4 @@ add_llvm_unittest(TableGenTests DISABLE_LLVM_LINK_LLVM_DYLIB
   ParserEntryPointTest.cpp
   )
 
-target_link_libraries(TableGenTests PRIVATE LLVMTableGenGlobalISel LLVMTableGen)
+target_link_libraries(TableGenTests PRIVATE LLVMTableGenCommon LLVMTableGen)

llvm/unittests/TableGen/CodeExpanderTest.cpp

@@ -6,8 +6,8 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "GlobalISel/CodeExpander.h"
-#include "GlobalISel/CodeExpansions.h"
+#include "Common/GlobalISel/CodeExpander.h"
+#include "Common/GlobalISel/CodeExpansions.h"
 
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/TableGen/Error.h"

llvm/utils/TableGen/AsmMatcherEmitter.cpp

@@ -95,12 +95,12 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "CodeGenInstAlias.h"
-#include "CodeGenInstruction.h"
-#include "CodeGenRegisters.h"
-#include "CodeGenTarget.h"
-#include "SubtargetFeatureInfo.h"
-#include "Types.h"
+#include "Common/CodeGenInstAlias.h"
+#include "Common/CodeGenInstruction.h"
+#include "Common/CodeGenRegisters.h"
+#include "Common/CodeGenTarget.h"
+#include "Common/SubtargetFeatureInfo.h"
+#include "Common/Types.h"
 #include "llvm/ADT/CachedHashString.h"
 #include "llvm/ADT/PointerUnion.h"
 #include "llvm/ADT/STLExtras.h"

llvm/utils/TableGen/AsmWriterEmitter.cpp

@@ -11,13 +11,13 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "AsmWriterInst.h"
-#include "CodeGenInstAlias.h"
-#include "CodeGenInstruction.h"
-#include "CodeGenRegisters.h"
-#include "CodeGenTarget.h"
-#include "SequenceToOffsetTable.h"
-#include "Types.h"
+#include "Basic/SequenceToOffsetTable.h"
+#include "Common/AsmWriterInst.h"
+#include "Common/CodeGenInstAlias.h"
+#include "Common/CodeGenInstruction.h"
+#include "Common/CodeGenRegisters.h"
+#include "Common/CodeGenTarget.h"
+#include "Common/Types.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/STLExtras.h"
@@ -1262,14 +1262,14 @@ void AsmWriterEmitter::EmitPrintAliasInstruction(raw_ostream &O) {
         << "    break;\n";
     }
     O << "  }\n";
-  }    
+  }
   O << "}\n\n";
 
   if (!MCOpPredicates.empty()) {
     O << "static bool " << Target.getName() << ClassName
       << "ValidateMCOperand(const MCOperand &MCOp,\n"
       << "                  const MCSubtargetInfo &STI,\n"
-      << "                  unsigned PredicateIndex) {\n"      
+      << "                  unsigned PredicateIndex) {\n"
       << "  switch (PredicateIndex) {\n"
       << "  default:\n"
       << "    llvm_unreachable(\"Unknown MCOperandPredicate kind\");\n"

llvm/utils/TableGen/AsmWriterInst.h

@@ -0,0 +1,21 @@
+# The basic TableGen library contains as little dependencies as possible.
+# In particular, it does not depend on vt_gen -> it does not use ValueTypes.
+#
+# This library is the only thing included in `llvm-min-tablegen`.
+
+set(LLVM_LINK_COMPONENTS
+  Support
+  TableGen
+  )
+
+add_llvm_library(LLVMTableGenBasic STATIC OBJECT EXCLUDE_FROM_ALL
+  CodeGenIntrinsics.cpp
+  SDNodeProperties.cpp
+)
+set_target_properties(LLVMTableGenBasic PROPERTIES FOLDER "Tablegenning")
+
+# Users may include its headers as "Basic/*.h"
+target_include_directories(LLVMTableGenBasic
+  INTERFACE
+  $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/..>
+  )

llvm/utils/TableGen/Basic/CMakeLists.txt

@@ -25,12 +25,12 @@ class Record;
 class RecordKeeper;
 
 struct CodeGenIntrinsic {
-  Record *TheDef;             // The actual record defining this intrinsic.
-  std::string Name;           // The name of the LLVM function "llvm.bswap.i32"
-  std::string EnumName;       // The name of the enum "bswap_i32"
+  Record *TheDef;       // The actual record defining this intrinsic.
+  std::string Name;     // The name of the LLVM function "llvm.bswap.i32"
+  std::string EnumName; // The name of the enum "bswap_i32"
   std::string ClangBuiltinName; // Name of the corresponding GCC builtin, or "".
-  std::string MSBuiltinName;  // Name of the corresponding MS builtin, or "".
-  std::string TargetPrefix;   // Target prefix, e.g. "ppc" for t-s intrinsics.
+  std::string MSBuiltinName;    // Name of the corresponding MS builtin, or "".
+  std::string TargetPrefix;     // Target prefix, e.g. "ppc" for t-s intrinsics.
 
   /// This structure holds the return values and parameter values of an
   /// intrinsic. If the number of return values is > 1, then the intrinsic
@@ -136,9 +136,7 @@ struct CodeGenIntrinsic {
 
   void addArgAttribute(unsigned Idx, ArgAttrKind AK, uint64_t V = 0);
 
-  bool hasProperty(enum SDNP Prop) const {
-    return Properties & (1 << Prop);
-  }
+  bool hasProperty(enum SDNP Prop) const { return Properties & (1 << Prop); }
 
   /// Goes through all IntrProperties that have IsDefault
   /// value set and sets the property.
@@ -182,6 +180,6 @@ class CodeGenIntrinsicTable {
     return Intrinsics[Pos];
   }
 };
-}
+} // namespace llvm
 
 #endif

llvm/utils/TableGen/CodeGenIntrinsics.cpp renamed to llvm/utils/TableGen/Basic/CodeGenIntrinsics.cpp

@@ -34,6 +34,6 @@ enum SDNP {
 
 unsigned parseSDPatternOperatorProperties(Record *R);
 
-}
+} // namespace llvm
 
 #endif

llvm/utils/TableGen/CodeGenIntrinsics.h renamed to llvm/utils/TableGen/Basic/CodeGenIntrinsics.h

@@ -44,7 +44,7 @@ static inline void printChar(raw_ostream &OS, char C) {
 ///
 /// @tparam SeqT The sequence container. (vector or string).
 /// @tparam Less A stable comparator for SeqT elements.
-template<typename SeqT, typename Less = std::less<typename SeqT::value_type> >
+template <typename SeqT, typename Less = std::less<typename SeqT::value_type>>
 class SequenceToOffsetTable {
   typedef typename SeqT::value_type ElemT;
 
@@ -53,8 +53,8 @@ class SequenceToOffsetTable {
   struct SeqLess {
     Less L;
     bool operator()(const SeqT &A, const SeqT &B) const {
-      return std::lexicographical_compare(A.rbegin(), A.rend(),
-                                          B.rbegin(), B.rend(), L);
+      return std::lexicographical_compare(A.rbegin(), A.rend(), B.rbegin(),
+                                          B.rend(), L);
     }
   };
 
@@ -153,15 +153,15 @@ class SequenceToOffsetTable {
 
   /// emit - Print out the table as the body of an array initializer.
   /// Use the Print function to print elements.
-  void emit(raw_ostream &OS,
-            void (*Print)(raw_ostream&, ElemT),
+  void emit(raw_ostream &OS, void (*Print)(raw_ostream &, ElemT),
             const char *Term = "0") const {
     assert((empty() || Entries) && "Call layout() before emit()");
     for (typename SeqMap::const_iterator I = Seqs.begin(), E = Seqs.end();
          I != E; ++I) {
       OS << "  /* " << I->second << " */ ";
       for (typename SeqT::const_iterator SI = I->first.begin(),
-             SE = I->first.end(); SI != SE; ++SI) {
+                                         SE = I->first.end();
+           SI != SE; ++SI) {
         Print(OS, *SI);
         OS << ", ";
       }