diff --git a/src/finn/core/tensor.py b/src/finn/core/tensor.py
index 34e6479c39f539816fbae29aee61bcd78070c467..257078ad183dc9df561971c6e3d70677e4b5282e 100644
--- a/src/finn/core/tensor.py
+++ b/src/finn/core/tensor.py
@@ -24,88 +24,95 @@
# (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
from enum import Enum
+import numpy as np
+
+
class DataType(Enum):
- FLOAT32 = 0
- BINARY = 1
- BIPOLAR = 2
- UINT2 = 3
- UINT3 = 4
- UINT4 = 5
- UINT8 = 6
- UINT16 = 7
- UINT32 = 8
- INT2 = 9
- INT3 = 10
- INT4 = 11
- INT8 = 12
- INT16 = 13
- INT32 = 14
-
- def bitwidth(self):
- """Returns the number of bits required for this DataType."""
-
- if self.name.startswith("UINT"):
- return int(self.name.strip("UINT"))
- elif self.name.startswith("INT"):
- return int(self.name.strip("INT"))
- elif "FLOAT" in self.name:
- return int(self.name.strip("FLOAT"))
- elif self.name in ["BINARY", "BIPOLAR"]:
- return 1
- else:
- raise Exception("Unrecognized data type: %s" % self.name)
-
- def min(self):
- """Returns the smallest possible value allowed by this DataType."""
-
- if self.name.startswith("UINT") or self.name == "BINARY":
- return 0
- elif self.name.startswith("INT"):
- return -(2 ** (self.bitwidth() - 1))
- elif self.name == "FLOAT32":
- return np.finfo(np.float32).min
- elif self.name == "BIPOLAR":
- return -1
- else:
- raise Exception("Unrecognized data type: %s" % self.name)
-
- def max(self):
- """Returns the largest possible value allowed by this DataType."""
-
- if self.name.startswith("UINT"):
- return (2 ** (self.bitwidth())) - 1
- elif self.name == "BINARY":
- return +1
- elif self.name.startswith("INT"):
- return (2 ** (self.bitwidth() - 1)) - 1
- elif self.name == "FLOAT32":
- return np.finfo(np.float32).max
- elif self.name == "BIPOLAR":
- return +1
- else:
- raise Exception("Unrecognized data type: %s" % self.name)
-
- def allowed(self, value):
- """Check whether given value is allowed for this DataType.
+ FLOAT32 = 0
+ BINARY = 1
+ BIPOLAR = 2
+ UINT2 = 3
+ UINT3 = 4
+ UINT4 = 5
+ UINT8 = 6
+ UINT16 = 7
+ UINT32 = 8
+ INT2 = 9
+ INT3 = 10
+ INT4 = 11
+ INT8 = 12
+ INT16 = 13
+ INT32 = 14
+
+ def bitwidth(self):
+ """Returns the number of bits required for this DataType."""
+
+ if self.name.startswith("UINT"):
+ return int(self.name.strip("UINT"))
+ elif self.name.startswith("INT"):
+ return int(self.name.strip("INT"))
+ elif "FLOAT" in self.name:
+ return int(self.name.strip("FLOAT"))
+ elif self.name in ["BINARY", "BIPOLAR"]:
+ return 1
+ else:
+ raise Exception("Unrecognized data type: %s" % self.name)
+
+ def min(self):
+ """Returns the smallest possible value allowed by this DataType."""
+
+ if self.name.startswith("UINT") or self.name == "BINARY":
+ return 0
+ elif self.name.startswith("INT"):
+ return -(2 ** (self.bitwidth() - 1))
+ elif self.name == "FLOAT32":
+ return np.finfo(np.float32).min
+ elif self.name == "BIPOLAR":
+ return -1
+ else:
+ raise Exception("Unrecognized data type: %s" % self.name)
+
+ def max(self):
+ """Returns the largest possible value allowed by this DataType."""
+
+ if self.name.startswith("UINT"):
+ return (2 ** (self.bitwidth())) - 1
+ elif self.name == "BINARY":
+ return +1
+ elif self.name.startswith("INT"):
+ return (2 ** (self.bitwidth() - 1)) - 1
+ elif self.name == "FLOAT32":
+ return np.finfo(np.float32).max
+ elif self.name == "BIPOLAR":
+ return +1
+ else:
+ raise Exception("Unrecognized data type: %s" % self.name)
+
+ def allowed(self, value):
+ """Check whether given value is allowed for this DataType.
value (float32): value to be checked"""
- if "FLOAT" in self.name:
- return True
- elif "INT" in self.name:
- return (self.min() <= value) and (value <= self.max()) and float(value).is_integer()
- elif self.name == "BINARY":
- return value in [0, 1]
- elif self.name == "BIPOLAR":
- return value in [-1, +1]
- else:
- raise Exception("Unrecognized data type: %s" % self.name)
+ if "FLOAT" in self.name:
+ return True
+ elif "INT" in self.name:
+ return (
+ (self.min() <= value)
+ and (value <= self.max())
+ and float(value).is_integer()
+ )
+ elif self.name == "BINARY":
+ return value in [0, 1]
+ elif self.name == "BIPOLAR":
+ return value in [-1, +1]
+ else:
+ raise Exception("Unrecognized data type: %s" % self.name)
+
class Tensor(object):
- """A multidimensional array of numbers of given datatype.
+ """A multidimensional array of numbers of given datatype.
Attributes:
dtype (DataType): Element data type for this Tensor
@@ -114,7 +121,7 @@ class Tensor(object):
["N", "C", "H", "W"]
"""
- def __init__(self, dtype, data, dim_names = []):
- self.dtype = dtype
- self.data = data
- self.dim_names = dim_names
+ def __init__(self, dtype, data, dim_names=[]):
+ self.dtype = dtype
+ self.data = data
+ self.dim_names = dim_names
diff --git a/src/finn/onnx_exec.py b/src/finn/onnx_exec.py
index ffb1cccffe50c66aa10fb4cc8a664893fa152d4c..c18d6130c9bccdb3d46cb1424e2a97610e8dc98c 100644
--- a/src/finn/onnx_exec.py
+++ b/src/finn/onnx_exec.py
@@ -24,52 +24,55 @@
# (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 onnx
import onnx.helper as helper
-import numpy as np
-from functools import reduce
-from onnx import numpy_helper as np_helper
import onnxruntime as rt
-
+from onnx import numpy_helper as np_helper
model = onnx.load_model("model.onnx")
graph = model.graph
+
def valueinfo_to_tensor(vi):
- """Creates an all-zeroes numpy tensor from a ValueInfoProto."""
+ """Creates an all-zeroes numpy tensor from a ValueInfoProto."""
+
+ dims = [x.dim_value for x in vi.type.tensor_type.shape.dim]
+ return np.zeros(
+ dims, dtype=onnx.mapping.TENSOR_TYPE_TO_NP_TYPE[vi.type.tensor_type.elem_type]
+ )
- dims = [x.dim_value for x in vi.type.tensor_type.shape.dim]
- return np.zeros(dims, dtype = onnx.mapping.TENSOR_TYPE_TO_NP_TYPE[vi.type.tensor_type.elem_type])
def execute_node(node, context, graph):
- """Call onnxruntime to execute a single node. Input/output provided via context."""
+ """Call onnxruntime to execute a single node. Input/output provided via context."""
+
+ # onnxruntime unfortunately does not implement run_node as defined by ONNX,
+ # it can only execute entire models -- so we create a model which solely
+ # consists of our current node.
+ node_inputs = list(filter(lambda x: x.name in node.input, graph.input))
+ node_inputs += list(filter(lambda x: x.name in node.input, graph.value_info))
+ node_outputs = list(filter(lambda x: x.name in node.output, graph.output))
+ node_outputs += list(filter(lambda x: x.name in node.output, graph.value_info))
+ node_graph = helper.make_graph(
+ nodes=[node], name="single-node-exec", inputs=node_inputs, outputs=node_outputs
+ )
+ node_model = helper.make_model(node_graph)
+ input_dict = dict()
+ for inp in node.input:
+ input_dict[inp] = context[inp]
+ print("Input shape for %s: %s" % (inp, context[inp].shape))
+ sess = rt.InferenceSession(node_model.SerializeToString())
+ output_list = sess.run(None, input_dict)
+ for output_ind in range(len(node.output)):
+ outp = node.output[output_ind]
+ if output_list[output_ind].shape != context[outp].shape:
+ raise Exception(
+ "Output shapes disagree after node execution: found %s vs expected %s"
+ % (str(output_list[output_ind].shape.shape), str(context[outp].shape))
+ )
+ context[outp] = output_list[output_ind]
+ print("Output shape for %s: %s" % (outp, context[outp].shape))
- # onnxruntime unfortunately does not implement run_node as defined by ONNX,
- # it can only execute entire models -- so we create a model which solely
- # consists of our current node.
- node_inputs = list(filter(lambda x: x.name in node.input, graph.input))
- node_inputs += list(filter(lambda x: x.name in node.input, graph.value_info))
- node_outputs = list(filter(lambda x: x.name in node.output, graph.output))
- node_outputs += (list(filter(lambda x: x.name in node.output, graph.value_info)))
- node_graph = helper.make_graph(
- nodes = [node],
- name = "single-node-exec",
- inputs = node_inputs,
- outputs = node_outputs
- )
- node_model = helper.make_model(node_graph)
- input_dict = dict()
- for inp in node.input:
- input_dict[inp] = context[inp]
- print("Input shape for %s: %s" % (inp, context[inp].shape))
- sess = rt.InferenceSession(node_model.SerializeToString())
- output_list = sess.run(None, input_dict)
- for output_ind in range(len(node.output)):
- outp = node.output[output_ind]
- if output_list[output_ind].shape != context[outp].shape:
- raise Exception("Output shapes disagree after node execution: found %s vs expected %s" % (str(output_list[output_ind].shape.shape), str(context[outp].shape)))
- context[outp] = output_list[output_ind]
- print("Output shape for %s: %s" % (outp, context[outp].shape))
# first, we need to make sure that every variable required by the graph has
# some buffer associated with it. this includes graph inputs (which includes
@@ -79,32 +82,32 @@ def execute_node(node, context, graph):
execution_context = dict()
# make empty tensors for all the graph inputs and outputs
for vi in graph.input:
- new_tensor = valueinfo_to_tensor(vi)
- execution_context[vi.name] = new_tensor
+ new_tensor = valueinfo_to_tensor(vi)
+ execution_context[vi.name] = new_tensor
for vi in graph.output:
- new_tensor = valueinfo_to_tensor(vi)
- execution_context[vi.name] = new_tensor
+ new_tensor = valueinfo_to_tensor(vi)
+ execution_context[vi.name] = new_tensor
# make empty tensors for all intermediate buffers
# TODO are we guaranteed to have the .value_info filled?
# do we need to call ONNX shape inference first?
for vi in graph.value_info:
- new_tensor = valueinfo_to_tensor(vi)
- execution_context[vi.name] = new_tensor
+ new_tensor = valueinfo_to_tensor(vi)
+ execution_context[vi.name] = new_tensor
# fill in the constants provided by the initializers (TensorProto to npy)
for t in graph.initializer:
- execution_context[t.name] = np_helper.to_array(t)
+ execution_context[t.name] = np_helper.to_array(t)
# now call each node in the graph nodes list
# we can simply walk down the list since the ONNX spec guarantees that it is
# topologically sorted
all_used_ops = set()
for node in graph.node:
- print("Node name: %s Type: %s" % (node.name, node.op_type))
- all_used_ops.add(node.op_type)
- print("Input(s): " + str(node.input))
- print("Output(s): " + str(node.output))
- print("Attribute(s): " + str(node.attribute))
- execute_node(node, execution_context, graph)
+ print("Node name: %s Type: %s" % (node.name, node.op_type))
+ all_used_ops.add(node.op_type)
+ print("Input(s): " + str(node.input))
+ print("Output(s): " + str(node.output))
+ print("Attribute(s): " + str(node.attribute))
+ execute_node(node, execution_context, graph)
print("Final output(s): ")
print(execution_context[graph.output[0].name])
diff --git a/tests/test_datatypes.py b/tests/test_datatypes.py
index 9ebd2f1eef195bfacd9f47ae2c924fb4123b38bf..06b2b3597c7cc28b662534b6e4c0bde62bed4ca7 100644
--- a/tests/test_datatypes.py
+++ b/tests/test_datatypes.py
@@ -1,36 +1,35 @@
# -*- coding: utf-8 -*-
-import pytest
-
import finn.core.tensor as ten
+
def test_datatypes():
- assert ten.DataType.BIPOLAR.allowed(-1) == True
- assert ten.DataType.BIPOLAR.allowed(0) == False
- assert ten.DataType.BINARY.allowed(-1) == False
- assert ten.DataType.BINARY.allowed(-1) == True
- assert ten.DataType.UINT2.allowed(2) == True
- assert ten.DataType.UINT2.allowed(10) == False
- assert ten.DataType.UINT3.allowed(5) == True
- assert ten.DataType.UINT3.allowed(-7) == False
- assert ten.DataType.UINT4.allowed(15) == True
- assert ten.DataType.UINT4.allowed(150) == False
- assert ten.DataType.UINT8.allowed(150) == True
- assert ten.DataType.UINT8.allowed(777) == False
- assert ten.DataType.UINT16.allowed(14500) == True
- assert ten.DataType.UINT16.allowed(-1) == False
- assert ten.DataType.UINT32.allowed(2 ** 10) == True
- assert ten.DataType.UINT32.allowed(-1) == False
- assert ten.DataType.INT2.allowed(2) == True
- assert ten.DataType.INT2.allowed(-10) == False
- assert ten.DataType.INT3.allowed(5) == False
- assert ten.DataType.INT3.allowed(-2) == True
- assert ten.DataType.INT4.allowed(15) == False
- assert ten.DataType.INT4.allowed(-5) == True
- assert ten.DataType.INT8.allowed(150) == False
- assert ten.DataType.INT8.allowed(-127) == True
- assert ten.DataType.INT16.allowed(-1.04) == False
- assert ten.DataType.INT16.allowed(-7777) == True
- assert ten.DataType.INT32.allowed(7.77) == False
- assert ten.DataType.INT32.allowed(-5) == True
- assert ten.DataType.INT32.allowed(5) == True
+ assert ten.DataType.BIPOLAR.allowed(-1) is True
+ assert ten.DataType.BIPOLAR.allowed(0) is False
+ assert ten.DataType.BINARY.allowed(-1) is False
+ assert ten.DataType.BINARY.allowed(-1) is True
+ assert ten.DataType.UINT2.allowed(2) is True
+ assert ten.DataType.UINT2.allowed(10) is False
+ assert ten.DataType.UINT3.allowed(5) is True
+ assert ten.DataType.UINT3.allowed(-7) is False
+ assert ten.DataType.UINT4.allowed(15) is True
+ assert ten.DataType.UINT4.allowed(150) is False
+ assert ten.DataType.UINT8.allowed(150) is True
+ assert ten.DataType.UINT8.allowed(777) is False
+ assert ten.DataType.UINT16.allowed(14500) is True
+ assert ten.DataType.UINT16.allowed(-1) is False
+ assert ten.DataType.UINT32.allowed(2 ** 10) is True
+ assert ten.DataType.UINT32.allowed(-1) is False
+ assert ten.DataType.INT2.allowed(2) is True
+ assert ten.DataType.INT2.allowed(-10) is False
+ assert ten.DataType.INT3.allowed(5) is False
+ assert ten.DataType.INT3.allowed(-2) is True
+ assert ten.DataType.INT4.allowed(15) is False
+ assert ten.DataType.INT4.allowed(-5) is True
+ assert ten.DataType.INT8.allowed(150) is False
+ assert ten.DataType.INT8.allowed(-127) is True
+ assert ten.DataType.INT16.allowed(-1.04) is False
+ assert ten.DataType.INT16.allowed(-7777) is True
+ assert ten.DataType.INT32.allowed(7.77) is False
+ assert ten.DataType.INT32.allowed(-5) is True
+ assert ten.DataType.INT32.allowed(5) is True