[llvm][TableGen] Add a !initialized predicate to allow testing for ? (llvm#117964)

There are cases (like in an upcoming patch to MLIR's `Property` class)
where the ? value is a useful null value. However, existing predicates
make ti difficult to test if the value in a record one is operating is ?
or not.

This commit adds the !initialized predicate, which is 1 on concrete,
non-? values and 0 on ?.

---------

Co-authored-by: Akshat Oke <[email protected]>

Author: krzysz00

llvm/docs/TableGen/ProgRef.rst

@@ -223,12 +223,13 @@ TableGen provides "bang operators" that have a wide variety of uses:
                : !div         !empty       !eq          !exists      !filter
                : !find        !foldl       !foreach     !ge          !getdagarg
                : !getdagname  !getdagop    !gt          !head        !if
-               : !interleave  !isa         !le          !listconcat  !listflatten
-               : !listremove  !listsplat   !logtwo      !lt          !mul
-               : !ne          !not         !or          !range       !repr
-               : !setdagarg   !setdagname  !setdagop    !shl         !size
-               : !sra         !srl         !strconcat   !sub         !subst
-               : !substr      !tail        !tolower     !toupper     !xor
+               : !initialized !interleave  !isa         !le          !listconcat
+               : !listflatten !listremove  !listsplat   !logtwo      !lt
+               : !mul         !ne          !not         !or          !range
+               : !repr        !setdagarg   !setdagname  !setdagop    !shl
+               : !size        !sra         !srl         !strconcat   !sub
+               : !subst       !substr      !tail        !tolower     !toupper
+               : !xor
 
 The ``!cond`` operator has a slightly different
 syntax compared to other bang operators, so it is defined separately:
@@ -555,7 +556,7 @@ previous case, if the *right-hand-side* operand is an undefined name or a
 global name, it is treated as a verbatim string of characters. The
 left-hand-side operand is treated normally.
 
-Values can have a trailing paste operator, in which case the left-hand-side 
+Values can have a trailing paste operator, in which case the left-hand-side
 operand is concatenated to an empty string.
 
 `Appendix B: Paste Operator Examples`_ presents examples of the behavior of
@@ -1815,6 +1816,10 @@ and non-0 as true.
   ``int``. If the result is not 0, the *then* expression is produced; otherwise
   the *else* expression is produced.
 
+``!initialized(``\ *a*\ ``)``
+  This operator produces 1 if *a* is not the uninitialized value (``?``) and 0
+  otherwise.
+
 ``!interleave(``\ *list*\ ``,`` *delim*\ ``)``
     This operator concatenates the items in the *list*, interleaving the
     *delim* string between each pair, and produces the resulting string.

llvm/include/llvm/TableGen/Record.h

@@ -860,6 +860,7 @@ class UnOpInit : public OpInit, public FoldingSetNode {
     LOG2,
     REPR,
     LISTFLATTEN,
+    INITIALIZED,
   };
 
 private:

llvm/lib/TableGen/Record.cpp

@@ -917,6 +917,13 @@ const Init *UnOpInit::Fold(const Record *CurRec, bool IsFinal) const {
       return NewInit;
     break;
 
+  case INITIALIZED:
+    if (isa<UnsetInit>(LHS))
+      return IntInit::get(RK, 0);
+    if (LHS->isConcrete())
+      return IntInit::get(RK, 1);
+    break;
+
   case NOT:
     if (const auto *LHSi = dyn_cast_or_null<IntInit>(
             LHS->convertInitializerTo(IntRecTy::get(RK))))
@@ -1052,6 +1059,9 @@ std::string UnOpInit::getAsString() const {
   case TOUPPER:
     Result = "!toupper";
     break;
+  case INITIALIZED:
+    Result = "!initialized";
+    break;
   }
   return Result + "(" + LHS->getAsString() + ")";
 }

llvm/lib/TableGen/TGLexer.cpp

@@ -633,6 +633,7 @@ tgtok::TokKind TGLexer::LexExclaim() {
           .Case("listremove", tgtok::XListRemove)
           .Case("range", tgtok::XRange)
           .Case("strconcat", tgtok::XStrConcat)
+          .Case("initialized", tgtok::XInitialized)
           .Case("interleave", tgtok::XInterleave)
           .Case("substr", tgtok::XSubstr)
           .Case("find", tgtok::XFind)

llvm/lib/TableGen/TGLexer.h

@@ -135,6 +135,7 @@ enum TokKind {
   XTail,
   XSize,
   XEmpty,
+  XInitialized,
   XIf,
   XCond,
   XEq,

llvm/lib/TableGen/TGParser.cpp

@@ -969,7 +969,7 @@ const TypedInit *TGParser::ParseSliceElements(Record *CurRec, bool Single) {
 ///   RangePiece ::= INTVAL
 ///   RangePiece ::= INTVAL '...' INTVAL
 ///   RangePiece ::= INTVAL '-' INTVAL
-///   RangePiece ::= INTVAL INTVAL 
+///   RangePiece ::= INTVAL INTVAL
 // The last two forms are deprecated.
 bool TGParser::ParseRangePiece(SmallVectorImpl<unsigned> &Ranges,
                                const TypedInit *FirstItem) {
@@ -1203,7 +1203,8 @@ const Init *TGParser::ParseOperation(Record *CurRec, const RecTy *ItemType) {
   case tgtok::XEmpty:
   case tgtok::XCast:
   case tgtok::XRepr:
-  case tgtok::XGetDagOp: { // Value ::= !unop '(' Value ')'
+  case tgtok::XGetDagOp:
+  case tgtok::XInitialized: { // Value ::= !unop '(' Value ')'
     UnOpInit::UnaryOp Code;
     const RecTy *Type = nullptr;
 
@@ -1291,6 +1292,11 @@ const Init *TGParser::ParseOperation(Record *CurRec, const RecTy *ItemType) {
       }
       Code = UnOpInit::GETDAGOP;
       break;
+    case tgtok::XInitialized:
+      Lex.Lex(); // eat the operation
+      Code = UnOpInit::INITIALIZED;
+      Type = IntRecTy::get(Records);
+      break;
     }
     if (!consume(tgtok::l_paren)) {
       TokError("expected '(' after unary operator");
@@ -1655,8 +1661,8 @@ const Init *TGParser::ParseOperation(Record *CurRec, const RecTy *ItemType) {
               !ArgType->typeIsConvertibleTo(StringRecTy::get(Records)) &&
               !ArgType->typeIsConvertibleTo(RecordRecTy::get(Records, {}))) {
             Error(InitLoc, Twine("expected bit, bits, int, string, or record; "
-                                 "got value of type '") + ArgType->getAsString() + 
-                                 "'");
+                                 "got value of type '") +
+                               ArgType->getAsString() + "'");
             return nullptr;
           }
           break;
@@ -1669,8 +1675,8 @@ const Init *TGParser::ParseOperation(Record *CurRec, const RecTy *ItemType) {
           if (!ArgType->typeIsConvertibleTo(IntRecTy::get(Records)) &&
               !ArgType->typeIsConvertibleTo(StringRecTy::get(Records))) {
             Error(InitLoc, Twine("expected bit, bits, int, or string; "
-                                 "got value of type '") + ArgType->getAsString() + 
-                                 "'");
+                                 "got value of type '") +
+                               ArgType->getAsString() + "'");
             return nullptr;
           }
           break;
@@ -2528,7 +2534,7 @@ const Init *TGParser::ParseOperationForEachFilter(Record *CurRec,
     OutType = RHSt->getType()->getListTy();
   } else if (Operation == tgtok::XFilter) {
     OutType = InEltType->getListTy();
-  }    
+  }
 
   return (TernOpInit::get((Operation == tgtok::XForEach) ? TernOpInit::FOREACH
                                                          : TernOpInit::FILTER,
@@ -3548,7 +3554,7 @@ bool TGParser::ParseBody(Record *CurRec) {
   SMLoc SemiLoc = Lex.getLoc();
   if (consume(tgtok::semi)) {
     PrintError(SemiLoc, "A class or def body should not end with a semicolon");
-    PrintNote("Semicolon ignored; remove to eliminate this error");    
+    PrintNote("Semicolon ignored; remove to eliminate this error");
   }
 
   return false;
@@ -4218,7 +4224,7 @@ bool TGParser::ParseMultiClass() {
     SMLoc SemiLoc = Lex.getLoc();
     if (consume(tgtok::semi)) {
       PrintError(SemiLoc, "A multiclass body should not end with a semicolon");
-      PrintNote("Semicolon ignored; remove to eliminate this error");    
+      PrintNote("Semicolon ignored; remove to eliminate this error");
     }
   }
 

llvm/test/TableGen/initialized.td

@@ -0,0 +1,59 @@
+// RUN: llvm-tblgen %s | FileCheck %s
+
+// CHECK: class F<Y [[ARG:.+]] = ?> {
+// CHECK: string ret = !if(!initialized([[ARG]].str), [[ARG]].str, "N/A");
+// CHECK: }
+
+// CHECK-LABEL: def C
+// CHECK: bit c0 = 0
+// CHECK: bit c1 = 1
+// CHECK: bit c2 = 1
+def C {
+  bit c0 = !initialized(?);
+  bit c1 = !initialized(0);
+  bit c2 = !initialized(1);
+}
+
+class Y {
+  string str = ?;
+}
+
+class F<Y y> {
+  string ret = !if(!initialized(y.str), y.str, "N/A");
+}
+
+def Y0 : Y;
+def Y1 : Y {
+  let str = "foo";
+}
+
+// CHECK-LABEL: def FY0
+// CHECK: string ret = "N/A";
+// CHECK-LABEL: def FY1
+// CHECK: string ret = "foo";
+def FY0 : F<Y0>;
+def FY1 : F<Y1>;
+
+class G<Y y> {
+  list<string> v = [y.str];
+  bit isInit = !initialized(v);
+}
+
+// CHECK-LABEL: def GY0
+// CHECK: isInit = 1
+// CHECK-LABEL: def GY1
+// CHECK: isInit = 1
+def GY0 : G<Y0>;
+def GY1 : G<Y1>;
+
+class Thing;
+def aThing : Thing;
+class Propagate<Thing t> {
+  Thing ret = !if(!initialized(t), t, ?);
+}
+// CHECK-LABEL: def PropagateNothing
+// CHECK: Thing ret = ?
+// CHECK-LABEL: def PropagateThing
+// CHECK: Thing ret = aThing
+def PropagateNothing : Propagate<?>;
+def PropagateThing : Propagate<aThing>;