[Translator] Remove global symbol and follow-up fix for tlc-pack#262 (t…

…lc-pack#316)

This PR removes the `global_symbol` linkage added by Relay Translator. It also fixes unaddressed comments of tlc-pack#262.

All tests can pass locally and I believe it is safe to merge this PR directly.

python/tvm/relax/testing/relay_translator.py

@@ -122,17 +122,28 @@ def visit_func(node):
             for arg in args:
                 if arg in var_map:
                     arg_expr = var_map[arg]
-                    if not isinstance(arg_expr.shape, relax.Tuple):
+                    if isinstance(arg_expr.checked_type, relax.DynTensorType):
                         new_args.append(arg_expr)
                         te_inputs.append(tvm.relax.expr.te_tensor(arg_expr))
-                    else:
-                        n_tensor = len(arg_expr.checked_type.fields)
-                        assert isinstance(arg_expr.checked_type, relax.TupleType)
-                        assert len(arg_expr.shape.fields) == n_tensor
+                    elif isinstance(arg_expr.checked_type, relax.TupleType):
+                        assert isinstance(arg_expr.shape, relax.Tuple)
+                        assert len(arg_expr.shape.fields) == len(arg_expr.checked_type.fields)
+                        n_tensor = len(arg_expr.shape.fields)
+                        bound_tuple = bb.lookup_binding(arg_expr)
+                        if isinstance(bound_tuple, relax.Tuple):
+                            assert len(bound_tuple) == n_tensor
                         for i in range(n_tensor):
-                            item = bb.emit(relax.TupleGetItem(arg_expr, i))
+                            if isinstance(bound_tuple, relax.Tuple):
+                                item = bb.emit(bound_tuple[i])
+                            else:
+                                item = bb.emit(relax.TupleGetItem(arg_expr, i))
                             new_args.append(item)
                             te_inputs.append(tvm.relax.expr.te_tensor(item))
+                    else:
+                        raise TypeError(
+                            f"CallTIR argument type being {type(arg_expr.checked_type)} is not "
+                            "supported."
+                        )
 
             op_name = node.op.name
             attrs = node.attrs
@@ -217,6 +228,4 @@ def visit_func(node):
             bb._begin_dataflow_block()
             relay.analysis.post_order_visit(mod["main"], visit_func)
 
-    relax_mod = bb.get()
-    relax_mod["main"] = relax_mod["main"].with_attr("global_symbol", "main")
     return bb.get()

tests/python/relax/test_relay_translator.py

@@ -277,9 +277,8 @@ def tir_matmul(
 def test_translate_tuple_arg():
     x = relay.var("x", shape=(10, 16))
     y = relay.var("y", shape=(10, 16))
-    t = relay.Tuple((x, y))
-    relay_mod = tvm.IRModule.from_expr(relay.Function([x, y], relay.concatenate(t, axis=-1)))
-    relay_vm, relax_vm, relax_mod = translate_and_build_vms(relay_mod)
+    relay_mod = tvm.IRModule.from_expr(relay.Function([x, y], relay.concatenate((x, y), axis=-1)))
+    relax_mod = relay_translator.from_relay(relay_mod["main"], target="llvm")
 
     # Construct the expected module
     bb = relax.BlockBuilder()
@@ -295,9 +294,9 @@ def test_translate_tuple_arg():
     )
     with bb.function("main", [x_relax, y_relax]):
         with bb.dataflow():
-            lv = bb.emit(relax.Tuple((x_relax, y_relax)))
-            lv1 = bb.emit(relax.TupleGetItem(lv, 0))
-            lv2 = bb.emit(relax.TupleGetItem(lv, 1))
+            _ = bb.emit(relax.Tuple((x_relax, y_relax)))
+            lv1 = bb.emit(x_relax)
+            lv2 = bb.emit(y_relax)
             lv3 = bb.emit_te(topi.x86.concatenate, (lv1, lv2), axis=-1)
             gv = bb.emit_output(lv3)
         bb.emit_func_output(gv)