From 165e7b325c4e8ef941e2051dcc21369f15a0c6f4 Mon Sep 17 00:00:00 2001
From: auphelia <jakobapk@web.de>
Date: Fri, 12 Jun 2020 15:54:21 +0100
Subject: [PATCH] [Test] Add unit test for rounding behaviour in execute
function
---
tests/core/test_basic_onnx_exec.py | 29 +++++++++++++++++++++++++++++
1 file changed, 29 insertions(+)
diff --git a/tests/core/test_basic_onnx_exec.py b/tests/core/test_basic_onnx_exec.py
index a7b6da996..1db8b2cec 100644
--- a/tests/core/test_basic_onnx_exec.py
+++ b/tests/core/test_basic_onnx_exec.py
@@ -35,6 +35,8 @@ import onnx.numpy_helper as np_helper
import finn.core.onnx_exec as oxe
from finn.core.modelwrapper import ModelWrapper
from finn.transformation.infer_shapes import InferShapes
+from finn.core.datatype import DataType
+from finn.util.basic import gen_finn_dt_tensor
def test_mnist_onnx_download_extract_run():
@@ -53,3 +55,30 @@ def test_mnist_onnx_download_extract_run():
assert np.isclose(
np_helper.to_array(output_tensor), output_dict["Plus214_Output_0"], atol=1e-3
).all()
+
+
+def test_onnx_exec_internal_rounding():
+ inp0 = onnx.helper.make_tensor_value_info("inp0", onnx.TensorProto.FLOAT, [2, 2])
+ inp1 = onnx.helper.make_tensor_value_info("inp1", onnx.TensorProto.FLOAT, [1])
+ outp = onnx.helper.make_tensor_value_info("outp", onnx.TensorProto.FLOAT, [2, 2])
+ mul_node = onnx.helper.make_node("Mul", inputs=["inp0", "inp1"], outputs=["outp"],)
+ graph = onnx.helper.make_graph(
+ nodes=[mul_node], name="mul_graph", inputs=[inp0, inp1], outputs=[outp]
+ )
+
+ model = onnx.helper.make_model(graph, producer_name="mul-model")
+ model = ModelWrapper(model)
+ idt = DataType.INT2
+ model.set_tensor_datatype("inp0", idt)
+ model.set_tensor_datatype("inp1", idt)
+ model.transform(InferShapes())
+
+ mul_value = np.asarray([2], dtype=np.float32)
+ inp_int = gen_finn_dt_tensor(idt, [2, 2])
+ scale = np.random.uniform(low=0, high=1, size=(2, 2)).astype(np.float32)
+ inp_rounded = (inp_int * scale) / (scale + 1e-7)
+ input_dict = {"inp0": inp_rounded, "inp1": mul_value}
+ output_dict = oxe.execute_onnx(model, input_dict)
+ produced = output_dict["outp"]
+ expected = np.multiply(inp_int, mul_value)
+ assert (produced == expected).all()
--
GitLab