Enhance LoopVectorizer for vectorizing by 0. Found at least one case …

…from testtopi/tests/python/test_topi_transform.py::test_tile.

src/tir/pass/vectorize_loop.cc

@@ -524,9 +524,13 @@ class LoopVectorizer : public StmtMutator {
       CHECK(is_zero(op->min));
       int lanes = 0;
       bool succ = arith::GetConstInt(op->extent, &lanes);
-      if (!succ || lanes < 1) {
+      if (!succ || lanes < 0) {
         LOG(FATAL) << "Failed to vectorize loop with extent " << op->extent;
       }
+      if (lanes == 0) {
+        // Nothing to run.  This may happen when a tensor has 0-sized dimension.
+        return Stmt();
+      }
       return Vectorizer(op->loop_var, lanes)(op->body);
     } else {
       return StmtMutator::VisitStmt_(op);