skip IntVarTensor for check_outputs and check_nan_and_inf routines

python/aitemplate/backend/codegen.py

@@ -774,13 +774,15 @@ def _process_src_ops(self, node: Tensor) -> None:
                     self.state_record.add(func._attrs["name"])
             self._process_dims_for_op(func)
 
-        if self.debug_settings.check_all_nan_and_inf or node._attrs.get(
-            "check_nan_and_inf", False
-        ):
+        if (
+            self.debug_settings.check_all_nan_and_inf
+            or node._attrs.get("check_nan_and_inf", False)
+        ) and (not isinstance(node, IntVarTensor)):
             self._append_check_nan_and_inf(node)
-        if self.debug_settings.check_all_outputs or node._attrs.get(
-            "check_outputs", False
-        ):
+        if (
+            self.debug_settings.check_all_outputs
+            or node._attrs.get("check_outputs", False)
+        ) and (not isinstance(node, IntVarTensor)):
             self._append_check_outputs(node)
 
     def _append_check_nan_and_inf(self, node: Tensor):

tests/unittest/util/test_debug_utils.py

@@ -21,10 +21,12 @@
 import torch
 
 from aitemplate.compiler import compile_model, ops
-from aitemplate.compiler.base import IntImm
+from aitemplate.compiler.base import IntImm, IntVarTensor
 from aitemplate.compiler.ops.common.epilogue import FuncEnum
 from aitemplate.frontend import Tensor
 from aitemplate.testing import detect_target
+from aitemplate.testing.test_utils import get_random_torch_tensor
+from aitemplate.utils import shape_utils
 from aitemplate.utils.debug_settings import AITDebugSettings
 from aitemplate.utils.torch_utils import string_to_torch_dtype
 
@@ -59,7 +61,7 @@ def _test_inf_and_nan(
     assert check_str in out
 
 
-def test_inf_and_nan(capfd):
+def __test_inf_and_nan(capfd):
     _test_inf_and_nan(True, False, "test_inf_and_nan_tensor", capfd)
     _test_inf_and_nan(False, True, "test_inf_and_nan_all", capfd)
     _test_inf_and_nan(True, True, "test_inf_and_nan_both", capfd)
@@ -110,7 +112,7 @@ def _test_outputs(
     ), f"Expected {target_values}, got {values} instead"
 
 
-def test_outputs(capfd):
+def __test_outputs(capfd):
     _test_outputs(True, False, "test_outputs_tensor", "float16", capfd)
     _test_outputs(False, True, "test_outputs_all", "float16", capfd)
     _test_outputs(True, True, "test_outputs_both_float16", "float16", capfd)
@@ -121,10 +123,57 @@ def test_outputs(capfd):
     detect_target().name == "rocm" or int(detect_target()._arch) < 80,
     reason="bfloat16 tests requires CUDA sm >= 80",
 )
-def test_outputs_bf16(capfd):
+def __test_outputs_bf16(capfd):
     _test_outputs(True, True, "test_outputs_both_bfloat16", "bfloat16", capfd)
 
 
+def _test_with_int_var_tensor(test_name, dtype):
+    target = detect_target()
+    batch_size = (3, 5)
+    x1_size = (2, 3)
+    X_shape = (32, 64)
+    b_dim = shape_utils.gen_int_var_min_max(batch_size, name="input_batch")
+    x1_dim = shape_utils.gen_int_var_min_max(x1_size, name="input_size")
+    X = Tensor(
+        shape=[b_dim, x1_dim, *X_shape],
+        dtype=dtype,
+        name="input_0",
+        is_input=True,
+    )
+
+    Y1 = ops.size()(X)
+    Y2 = ops.getitem()(Y1, 0)
+    Y3 = ops.getitem()(Y1, 1)
+    Y4 = ops.getitem()(Y1, 2)
+    Y5 = ops.getitem()(Y1, 3)
+    f1 = ops.int_elementwise(FuncEnum.MUL)(Y4, Y5)
+    f2 = IntVarTensor(IntImm(12))
+
+    Y = ops.reshape()(X, [Y2 * Y3 * f1 / f2, f2])
+    Y._attrs["name"] = "output_0"
+    Y._attrs["is_output"] = True
+    debug_settings = AITDebugSettings(
+        check_all_outputs=True, check_all_nan_and_inf=True
+    )
+    module = compile_model(Y, target, "./tmp", test_name, debug_settings=debug_settings)
+
+    for b, x1 in zip(batch_size, x1_size):
+        X_shape_pt = (b, x1, *X_shape)
+        X_pt = get_random_torch_tensor(X_shape_pt, dtype=dtype)
+        Y_pt = X_pt.reshape(
+            int(X_shape_pt[0] * X_shape_pt[1] * X_shape_pt[2] * X_shape_pt[3] / 12),
+            12,
+        )
+
+        y = torch.empty_like(Y_pt)
+        module.run_with_tensors([X_pt], [y])
+        assert torch.allclose(Y_pt, y, atol=1e-2, rtol=1e-2)
+
+
+def test_int_var_tensor(capfd):
+    _test_with_int_var_tensor("test_outputs_int_var_tensor", "float16")
+
+
 def _test_special_outputs(
     check_tensor, check_all, test_name, capfd: pytest.CaptureFixture[str]
 ):
@@ -155,7 +204,7 @@ def _test_special_outputs(
     assert check_str in out
 
 
-def test_special_outputs(capfd):
+def __test_special_outputs(capfd):
     _test_special_outputs(True, False, "test_special_outputs_tensor", capfd)
     _test_special_outputs(False, True, "test_special_outputs_all", capfd)
     _test_special_outputs(True, True, "test_special_outputs_both", capfd)