[TableGen] Fix instantiating multiclass in foreach

[jyyou.tw]

Summary

If multiclass argument comes from loop varaible and argument is
record type, it will not recognize the type.

This patch ensures that loop variables are resolved correctly.

[Paul-C-Anagnostopoulos]

I presume this change passes all the TableGen, mlir, and clang tests?

llvm/lib/TableGen/TGParser.cpp

3528

Is there a reason why we shouldn't use !CurMultiClass?

llvm/test/TableGen/foreach-multiclass.td

123

Could you add a comment explaining what this is testing, since it is relatively obscure?

[jyyou.tw]

Refine test case

[jyyou.tw]

In D95308#2518218, @Paul-C-Anagnostopoulos wrote:

I presume this change passes all the TableGen, mlir, and clang tests?

Yes, tests are passed.

llvm/lib/TableGen/TGParser.cpp

3528

If defm is placed in foreach and multiclass has foreach, the parsing flow is

parse foreach -> parse defm -> resolve multiclass entries -> resolve foreach of multiclass

In this case, !CurMultiClass is not enough. When resolving foreach of multiclass, Loop.ListValue is resolved as FieldInit only.
It must be further resolved, so I add Loops.empty() to ensure that.

[Paul-C-Anagnostopoulos]

llvm/lib/TableGen/TGParser.cpp

3528

Sorry, I didn't mean just !CurMultiClass. I meant:

!CurMultiClass && Loops.empty()

[jyyou.tw]

Refine condition

[jyyou.tw]

Missing test

[Paul-C-Anagnostopoulos]

Looks good to me.

[jyyou.tw]

If this patch is OK, could someone please commit it? I don't have commit access.

[Paul-C-Anagnostopoulos]

Okay, let me do this one first. I've never committed for someone else, so something might go wrong.

[Paul-C-Anagnostopoulos]

This revision is pushed.

llvm/lib/TableGen/TGParser.cpp

for (unsigned i = 0, e = TArgs.size(); i != e; ++i) {
                            "'");
         }
         Substs.emplace_back(TArgs[i], Default);
       }
     }
 
     Substs.emplace_back(QualifiedNameOfImplicitName(MC), DefmName);
 
-    if (resolve(MC->Entries, Substs, CurMultiClass == nullptr, &NewEntries,
-                &SubClassLoc))
+    if (resolve(MC->Entries, Substs, !CurMultiClass && Loops.empty(),
+                &NewEntries, &SubClassLoc))
       return true;
 
     if (!consume(tgtok::comma))
       break;
 
     if (Lex.getCode() != tgtok::Id)
       return TokError("expected identifier");
 

llvm/test/TableGen/foreach-multiclass.td

 // CHECK-NOT: def F1
 // CHECK-NOT: def F2_0_0
 // CHECK:     def F2_1_0
 // CHECK-NOT: def F2_1_2
 // CHECK:     def F2_2_0
 // CHECK:     def F2_2_1
 // CHECK-NOT: def F2_2_2
 
+// CHECK: def G0
+// CHECK: def H0_G0_0
+
 multiclass A<int x> {
   foreach i = [0, 1] in {
     def NAME#i {
       int sum = !add(x, i);
     }
   }
 }
 
multiclass F<list<int> lst> {
   foreach j = !foldl([]<int>, lst, lhs, x,
                      !if(!lt(x, i), !listconcat(lhs, [x]), lhs)) in
   def _#i#_#j;
 }
 
 defm F0 : F<[]>;
 defm F1 : F<[0]>;
 defm F2 : F<[0, 1, 2]>;
+
+// If multiclass argument comes from loop variable,
+// and field of argument is placed at foreach statement,
+// the record field must be resolved correctly.
+class G {
+  list<int> val = [0];
+}
+
+multiclass H<G g> {
+  foreach n = g.val in
+    def _#g#_#n;
+}
+
+def G0 : G;
+
+foreach g = [G0] in
+  defm H0 : H<g>;