Merge pull request #171 from Xilinx/feature/absorb_mul_into_topk

Feature/absorb mul into topk

src/finn/transformation/streamline/absorb.py

@@ -28,11 +28,13 @@
 
 import numpy as np
 from onnx import helper as oh
+import warnings
 
 from finn.core.datatype import DataType
 from finn.transformation import Transformation
 from finn.util.basic import get_by_name
 from finn.custom_op.registry import getCustomOp
+from finn.transformation.infer_shapes import InferShapes
 from finn.transformation.infer_datatypes import InferDataTypes
 
 
@@ -290,3 +292,38 @@ class AbsorbTransposeIntoMultiThreshold(Transformation):
         if graph_modified:
             model = model.transform(InferDataTypes())
         return (model, graph_modified)
+
+
+class AbsorbScalarMulIntoTopK(Transformation):
+    """Absorb a mul node into a suceeding topk node if the mul is scalar."""
+
+    def apply(self, model):
+        graph = model.graph
+        node_ind = 0
+        graph_modified = False
+        for n in graph.node:
+            node_ind += 1
+            if n.op_type == "TopK":
+                prod = model.find_producer(n.input[0])
+                if prod is not None and prod.op_type == "Mul":
+                    prod_input = prod.input[0]
+                    param_name = prod.input[1]
+                    A = model.get_initializer(param_name)
+                    if A is None:
+                        warnings.warn("Param is not constant, skipping")
+                        continue
+                    if all(x == 1 for x in A.shape) and A > 0:
+                        # if the mul is scalar and positive, we can just delete the
+                        # mul node and rewire the top k node. Because the top k node
+                        # works with probabilities and their relation to each other
+                        # the relation doesn't change if every value is multiplied
+                        # with a scalar
+                        graph.node.remove(prod)
+                        n.input[0] = prod_input
+                        # to avoid error the dataype is set to float32
+                        model.set_tensor_datatype(n.input[0], DataType.FLOAT32)
+                        graph_modified = True
+        if graph_modified:
+            model = model.transform(InferShapes())
+            model = model.transform(InferDataTypes())
+        return (model, graph_modified)

tests/transformation/test_absorb_mul_into_topk.py

+# Copyright (c) 2020, Xilinx
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+# * Redistributions of source code must retain the above copyright notice, this
+#   list of conditions and the following disclaimer.
+#
+# * Redistributions in binary form must reproduce the above copyright notice,
+#   this list of conditions and the following disclaimer in the documentation
+#   and/or other materials provided with the distribution.
+#
+# * Neither the name of FINN nor the names of its
+#   contributors may be used to endorse or promote products derived from
+#   this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+# SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+# CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+# OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+import pytest
+
+import numpy as np
+from onnx import TensorProto, helper
+
+from finn.core.modelwrapper import ModelWrapper
+from finn.transformation.infer_shapes import InferShapes
+from finn.transformation.infer_datatypes import InferDataTypes
+from finn.transformation.general import GiveUniqueNodeNames, GiveReadableTensorNames
+from finn.transformation.insert_topk import InsertTopK
+from finn.transformation.streamline.absorb import AbsorbScalarMulIntoTopK
+import finn.core.onnx_exec as oxe
+
+# parameter to indicate if mul parameter is negative or positive
+@pytest.mark.parametrize("mul_positive", [True, False])
+# parameter to indicate if mul parameter is scalar or not
+@pytest.mark.parametrize("scalar", [True, False])
+def test_absorb_mul_into_topk(mul_positive, scalar):
+    if scalar is True:
+        shape = [1]
+    else:
+        shape = [1, 1, 1, 1000]
+    inp = helper.make_tensor_value_info("inp", TensorProto.FLOAT, [1, 1, 1, 1000])
+    a0 = helper.make_tensor_value_info("a0", TensorProto.FLOAT, shape)
+    outp = helper.make_tensor_value_info("outp", TensorProto.FLOAT, [1, 1, 1, 1000])
+
+    mul_node = helper.make_node("Mul", ["inp", "a0"], ["outp"])
+    mul_graph = helper.make_graph(
+        nodes=[mul_node],
+        name="mul-graph",
+        inputs=[inp],
+        outputs=[outp],
+        value_info=[a0],
+    )
+
+    model = helper.make_model(mul_graph, producer_name="mul_model")
+    model = ModelWrapper(model)
+    # initialize values
+    if mul_positive is True:
+        a0_values = np.random.uniform(low=0.1, high=1, size=tuple(shape)).astype(
+            np.float32
+        )
+    else:
+        a0_values = np.random.uniform(low=-1, high=-0.1, size=tuple(shape)).astype(
+            np.float32
+        )
+    model.set_initializer("a0", a0_values)
+    model = model.transform(InsertTopK())
+    model = model.transform(InferShapes())
+    model = model.transform(InferDataTypes())
+    model = model.transform(GiveUniqueNodeNames())
+    model = model.transform(GiveReadableTensorNames())
+    model_transformed = model.transform(AbsorbScalarMulIntoTopK())
+
+    # compare execution results
+    inp_values = np.random.uniform(low=-10, high=10, size=(1, 1, 1, 1000)).astype(
+        np.float32
+    )
+    idict = {"global_in": inp_values}
+    odict = oxe.execute_onnx(model, idict, True)
+    y_indices = odict["global_out"]
+    y_values = odict["TopK_0_out0"]
+    odict = oxe.execute_onnx(model_transformed, idict, True)
+    y_tr_indices = odict["global_out"]
+    y_tr_values = odict["TopK_0_out0"]
+
+    # the indices stay the same, if the model is transformed or not
+    assert (y_indices == y_tr_indices).all()
+
+    if scalar is True and mul_positive is True:
+        # the values change if the model was transformed
+        assert (y_values != y_tr_values).all()
+
+        # check for new order
+        assert model.graph != model_transformed.graph
+        assert len(model.graph.node) - 1 == len(model_transformed.graph.node)
+        assert model_transformed.graph.node[0].op_type == "TopK"
+
+    else:
+        assert (y_values == y_tr_values).all()
+        assert model.graph == model_transformed.graph