From acdaebe2ebe40fd0df16c3862e116a0f87475337 Mon Sep 17 00:00:00 2001
From: Hendrik Borras <hendrikborras@web.de>
Date: Thu, 7 Oct 2021 17:04:31 +0100
Subject: [PATCH] Added support for converting Gemm to MatMul.
---
.../qonnx/convert_qonnx_to_finn.py | 5 +
.../transformation/qonnx/gemm_to_matmul.py | 218 ++++++++++++++++++
2 files changed, 223 insertions(+)
create mode 100644 src/finn/transformation/qonnx/gemm_to_matmul.py
diff --git a/src/finn/transformation/qonnx/convert_qonnx_to_finn.py b/src/finn/transformation/qonnx/convert_qonnx_to_finn.py
index 0c606e36c..f4c69ca53 100644
--- a/src/finn/transformation/qonnx/convert_qonnx_to_finn.py
+++ b/src/finn/transformation/qonnx/convert_qonnx_to_finn.py
@@ -27,12 +27,14 @@
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
from onnx import TensorProto, helper
+from qonnx.transformation.quant_constant_folding import FoldTransposeIntoQuantInit
import finn.core.onnx_exec as oxe
from finn.core.datatype import DataType
from finn.transformation.base import Transformation
from finn.transformation.infer_datatypes import InferDataTypes
from finn.transformation.infer_shapes import InferShapes
+from finn.transformation.qonnx.gemm_to_matmul import GemmToMatMul
from finn.transformation.qonnx.qonnx_activation_handlers import QuantActBaseHandler
from finn.util.basic import get_by_name
@@ -46,6 +48,9 @@ class ConvertQONNXtoFINN(Transformation):
"""
def apply(self, model):
+ # Gemm operations are not supported by FINN, so we convert them to MatMul
+ model = model.transform(GemmToMatMul())
+ model = model.transform(FoldTransposeIntoQuantInit())
# Make sure the datatypes exist, these are required for folding the weights
model = model.transform(InferDataTypes())
# Fold weights
diff --git a/src/finn/transformation/qonnx/gemm_to_matmul.py b/src/finn/transformation/qonnx/gemm_to_matmul.py
new file mode 100644
index 000000000..6debb169f
--- /dev/null
+++ b/src/finn/transformation/qonnx/gemm_to_matmul.py
@@ -0,0 +1,218 @@
+# Copyright (c) 2021, 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 numpy as np
+import warnings
+from onnx import TensorProto, helper
+
+from finn.transformation.base import Transformation
+from finn.util.basic import get_by_name
+
+
+class GemmToMatMul(Transformation):
+ """
+ Converts Gemm op into a MatMul and an Add op.
+ """
+
+ def apply(self, model):
+ graph = model.graph
+ node_ind = 0
+ for n in graph.node:
+ node_ind += 1
+ if n.op_type == "Gemm":
+ running_node_index = node_ind
+ predecessors = model.find_direct_predecessors(n)
+
+ # Transpose A?
+ transA = get_by_name(n.attribute, "transA")
+ if transA is not None and transA.i:
+ # Insert transpose node
+ inp_trans_out = helper.make_tensor_value_info(
+ model.make_new_valueinfo_name(),
+ TensorProto.FLOAT,
+ None,
+ # [1024,1000],
+ )
+ graph.value_info.append(inp_trans_out)
+ inp_trans_node = helper.make_node(
+ "Transpose", [n.input[0]], [inp_trans_out.name]
+ )
+ graph.node.insert(running_node_index, inp_trans_node)
+ running_node_index += 1
+
+ n.input[0] = inp_trans_out.name
+
+ # Transpose B?
+ transB = get_by_name(n.attribute, "transB")
+ if transB is not None and transB.i:
+ # Insert transpose node
+ inp_trans_out = helper.make_tensor_value_info(
+ model.make_new_valueinfo_name(),
+ TensorProto.FLOAT,
+ None,
+ # [1024,1000],
+ )
+ graph.value_info.append(inp_trans_out)
+ inp_trans_node = helper.make_node(
+ "Transpose", [n.input[1]], [inp_trans_out.name]
+ )
+ graph.node.insert(running_node_index, inp_trans_node)
+ running_node_index += 1
+
+ n.input[1] = inp_trans_out.name
+
+ # Insert MatMul: A * B
+ matMul_node = helper.make_node(
+ "MatMul", [n.input[0], n.input[1]], [n.output[0]]
+ )
+ graph.node.insert(running_node_index, matMul_node)
+ matMul_node = graph.node[running_node_index]
+ running_node_index += 1
+
+ # Insert Mul: (A*B) * alpha
+ alpha = get_by_name(n.attribute, "alpha")
+ if alpha is None:
+ alpha = np.array(1.0)
+ else:
+ alpha = np.array(alpha.f)
+ mul_tensor = helper.make_tensor_value_info(
+ model.make_new_valueinfo_name(),
+ TensorProto.FLOAT,
+ None,
+ )
+ graph.value_info.append(mul_tensor)
+ model.set_initializer(mul_tensor.name, alpha)
+
+ act_mul_tensor = helper.make_tensor_value_info(
+ model.make_new_valueinfo_name(),
+ TensorProto.FLOAT,
+ None,
+ )
+ graph.value_info.append(act_mul_tensor)
+ mul_node = helper.make_node(
+ "Mul",
+ [act_mul_tensor.name, mul_tensor.name],
+ [n.output[0]],
+ )
+ graph.node.insert(running_node_index, mul_node)
+ mul_node_main_branch = graph.node[running_node_index]
+ running_node_index += 1
+ matMul_node.output[0] = act_mul_tensor.name
+
+ # Other branch: Insert Mul: beta * C
+ beta = get_by_name(n.attribute, "beta")
+ if alpha is None:
+ beta = np.array(1.0)
+ else:
+ beta = np.array(beta.f)
+ mul_tensor = helper.make_tensor_value_info(
+ model.make_new_valueinfo_name(),
+ TensorProto.FLOAT,
+ None,
+ )
+ graph.value_info.append(mul_tensor)
+ model.set_initializer(mul_tensor.name, beta)
+
+ act_mul_tensor = helper.make_tensor_value_info(
+ model.make_new_valueinfo_name(),
+ TensorProto.FLOAT,
+ None,
+ )
+ graph.value_info.append(act_mul_tensor)
+ mul_node = helper.make_node(
+ "Mul",
+ [act_mul_tensor.name, mul_tensor.name],
+ [n.input[2]],
+ )
+ graph.node.insert(running_node_index, mul_node)
+ running_node_index += 1
+ predecessors[2].output[0] = act_mul_tensor.name
+
+ # Insert Add: ((A*B) * alpha) + (beta * C)
+ act_add_tensor = helper.make_tensor_value_info(
+ model.make_new_valueinfo_name(),
+ TensorProto.FLOAT,
+ None,
+ )
+ graph.value_info.append(act_add_tensor)
+ mul_node_main_branch.output[0] = act_add_tensor.name
+ add_node = helper.make_node(
+ "Add",
+ [act_add_tensor.name, n.input[2]],
+ [n.output[0]],
+ )
+
+ graph.node.insert(running_node_index, add_node)
+ running_node_index += 1
+
+ # Delete Gemm node
+ graph.node.remove(n)
+
+ # Remove potential unity multiplications from alpha and beta attributes
+ model = model.transform(RemoveUnityMul())
+
+ return model, True
+
+ return model, False
+
+
+class RemoveUnityMul(Transformation):
+ """
+ Removes multiplication nodes, which have a unity initializer.
+ """
+
+ def apply(self, model):
+ graph = model.graph
+ node_ind = 0
+ for n in graph.node:
+ node_ind += 1
+ if n.op_type == "Mul":
+ init = model.get_initializer(n.input[1])
+ if init is not None:
+ # Check if all multipliers are unity
+ init = init.flatten()
+ if (init == 1.0).all():
+ predecessors = model.find_direct_predecessors(n)
+ successors = model.find_direct_successors(n)
+ # Check if we reached the top or bottom of the graph
+ if predecessors is not None:
+ for predecessor in predecessors:
+ predecessor.output[0] = n.output[0]
+ graph.node.remove(n)
+ return model, True
+ elif successors is not None:
+ for successor in successors:
+ successor.input[0] = n.input[0]
+ graph.node.remove(n)
+ return model, True
+ else:
+ warnings.warn(
+ f"Can't remove empty unity multiplication node {n}, "
+ f"due to no available successors or predecessors."
+ )
+ return model, False
--
GitLab