Fix Select op with Mul inf input to NaN error (#2035)

* fix select mul inf to nan error

Signed-off-by: Deyu Huang <[email protected]>

* fix pylint

Signed-off-by: Deyu Huang <[email protected]>

* add break condition

Signed-off-by: Deyu Huang <[email protected]>
Co-authored-by: Jay Zhang <[email protected]>
Signed-off-by: Jay Zhang <[email protected]>

Author: hwangdeyu

tests/test_backend.py

@@ -3488,6 +3488,17 @@ def func(x):
             return tf.identity(picks, name=_TFOUTPUT)
         self._run_test_case(func, [_OUTPUT], {_INPUT: x_val})
 
+    @check_opset_min_version(9, "IsNaN")
+    def test_where_ismulinf(self):
+        x_val1 = np.array([np.inf], dtype=np.float32)
+        x_val2 = np.array([0], dtype=np.float32)
+        true_result = np.array([np.inf], dtype=np.float32)
+        def func(x1, x2):
+            mul = tf.multiply(x1, x2)
+            picks = tf.where(x1 < mul, true_result, x2)
+            return tf.identity(picks, name=_TFOUTPUT)
+        self._run_test_case(func, [_OUTPUT], {_INPUT: x_val1, _INPUT1: x_val2})
+
     @check_opset_min_version(9, "Where for strings needs opset 9")
     @skip_tfjs("Technically tf where doesn't support strings and tfjs doesn't like it")
     def test_where_string(self):
@@ -5542,7 +5553,7 @@ def func(x):
         self._run_test_case(func, [_OUTPUT], {_INPUT: x_val0}, rtol=1e-6, atol=1e-4)
         self._run_test_case(func, [_OUTPUT], {_INPUT: x_val}, rtol=1e-6, atol=1e-4)
 
-        x_val = np.random.random(size=[4, 3]).astype(np.float32) * 2048. - 1024
+        x_val = np.random.random(size=[4, 3]).astype(np.float32) * 2048. - 1024.
         x_val[0, 0] = -1024
         x_val[0, 1] = -1023
         x_val[0, 2] = 1024
@@ -5579,7 +5590,7 @@ def func(x):
         self._run_test_case(func, [_OUTPUT], {_INPUT: x_val0}, rtol=1e-6, atol=1e-4)
         self._run_test_case(func, [_OUTPUT], {_INPUT: x_val}, rtol=1e-6, atol=1e-4)
 
-        x_val = np.random.random(size=[4, 3]).astype(np.float32) * 2048. - 1024
+        x_val = np.random.random(size=[4, 3]).astype(np.float32) * 2048. - 1024.
         x_val[0, 0] = -1024
         x_val[0, 1] = -1023
         x_val[0, 2] = 1024

tf2onnx/onnx_opset/controlflow.py

@@ -204,8 +204,15 @@ def version_9(cls, ctx, node, **kwargs):
                 if eq_node.input[0] == eq_node.input[1]:
                     handles_nan = True
             for inp in node.inputs[1:]:
-                if inp.is_const() and np.any(np.isnan(inp.get_tensor_value(as_list=False))):
+                if handles_nan:
+                    break
+                if inp.is_const() and (np.any(np.isnan(inp.get_tensor_value(as_list=False))) or \
+                                       np.any(np.isinf(inp.get_tensor_value(as_list=False)))):
                     handles_nan = True
+                if inp.type == "Mul":
+                    inp0 = inp.inputs[0].is_const() and np.any(np.isinf(inp.inputs[0].get_tensor_value(as_list=False)))
+                    inp1 = inp.inputs[1].is_const() and np.any(np.isinf(inp.inputs[1].get_tensor_value(as_list=False)))
+                    handles_nan = inp0 or inp1
 
         if ctx.get_dtype(node.output[0]) != TensorProto.STRING and not handles_nan:
             # Due to bad ORT implementation, Mul/Add ops are faster than Where op