Bump LLVM to 889c6f3996769a991a24da957f597e7500d158e7

The biggest change here is to upgrade RefineTypes to the new sparse
dataflow framework.

Smaller changes:
- minor changes to type parsing
- suppress warnings in e2e tests

e2e_testing/torchscript/main.py

@@ -108,5 +108,13 @@ def main():
     failed = report_results(results, xfail_set, args.verbose)
     sys.exit(1 if failed else 0)
 
+def _suppress_warnings():
+    import warnings
+    # Ignore warning due to Python bug:
+    # https://stackoverflow.com/questions/4964101/pep-3118-warning-when-using-ctypes-array-as-numpy-array
+    warnings.filterwarnings("ignore",
+                            message="A builtin ctypes object gave a PEP3118 format string that does not match its itemsize")
+
 if __name__ == '__main__':
+    _suppress_warnings()
     main()

externals/llvm-project

@@ -1 +1 @@
-Subproject commit bc07634b5a762686b818932eb350b4fc84217e67
+Subproject commit 889c6f3996769a991a24da957f597e7500d158e7

lib/Dialect/Torch/IR/TorchDialect.cpp

@@ -63,17 +63,8 @@ struct TorchInlinerInterface : public DialectInlinerInterface {
 Type Torch::parseTorchDialectType(AsmParser &parser) {
   SMLoc typeLoc = parser.getCurrentLocation();
   StringRef mnemonic;
-  if (parser.parseOptionalKeyword(&mnemonic)) {
-    parser.emitError(parser.getCurrentLocation())
-        .append("expected type mnemonic")
-        .attachNote()
-        .append("for types like `!torch.list<int>`, you must omit the "
-                "`!torch.` prefix for the nested types");
-    return Type();
-  }
-
   Type genType;
-  auto parseResult = generatedTypeParser(parser, mnemonic, genType);
+  auto parseResult = generatedTypeParser(parser, &mnemonic, genType);
   if (parseResult.hasValue())
     return genType;
   parser.emitError(typeLoc) << "unknown  type `" << mnemonic << "` in dialect `"