def _get_base_op_hander_dicts():
"""Returns the base op_hander_dict for all regularizers."""
base_dict = collections.defaultdict(
grouping_op_handler.GroupingOpHandler, {
'ConcatV2':
concat_op_handler.ConcatOpHandler(),
'DepthToSpace':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'DepthwiseConv2dNative':
depthwise_convolution_op_handler.DepthwiseConvolutionOpHandler(),
'ExpandDims':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'RandomUniform':
leaf_op_handler.LeafOpHandler(),
'Reshape':
leaf_op_handler.LeafOpHandler(),
'Shape':
leaf_op_handler.LeafOpHandler(),
'SpaceToDepth':
leaf_op_handler.LeafOpHandler(),
'StridedSlice':
leaf_op_handler.LeafOpHandler(),
'TensorArrayGatherV3':
leaf_op_handler.LeafOpHandler(),
'Transpose':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
})
for resize_method in RESIZE_OP_NAMES:
# Resize* ops, second input might be a tensor which will result in an error.
base_dict[resize_method] = grouping_op_handler.GroupingOpHandler([0])
return base_dict
def _get_base_op_hander_dicts():
return collections.defaultdict(
grouping_op_handler.GroupingOpHandler, {
'ConcatV2':
concat_op_handler.ConcatOpHandler(),
'DepthToSpace':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'DepthwiseConv2dNative':
depthwise_convolution_op_handler.DepthwiseConvolutionOpHandler(),
'ExpandDims':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'RandomUniform':
leaf_op_handler.LeafOpHandler(),
'Reshape':
leaf_op_handler.LeafOpHandler(),
'Shape':
leaf_op_handler.LeafOpHandler(),
'SpaceToDepth':
leaf_op_handler.LeafOpHandler(),
'StridedSlice':
leaf_op_handler.LeafOpHandler(),
'TensorArrayGatherV3':
leaf_op_handler.LeafOpHandler(),
'Transpose':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
})
def test_AssignGroupingOfGroupingConcatNoSlicing(self):
# In this test, the output op (batch norm) has size 6 and is not sliced.
# and that input Conv2Ds are all of size 6, and are grouped.
# Map ops to slices. Batch norm op is composed of multiple slices.
self.op_slice_dict = {
self.relu1_op: [self.relu1_op_slice],
self.relu2_op: [self.relu2_op_slice],
self.relu3_op: [self.relu3_op_slice],
self.concat_op: [self.concat_op_slice],
self.batch_norm_op: [self.batch_norm_op_slice],
}
# Map each slice to a group.
self.op_group_dict = {
self.relu1_op_slice: self.relu1_op_group,
self.relu2_op_slice: self.relu2_op_group,
self.relu3_op_slice: self.relu3_op_group,
self.batch_norm_op_slice: self.batch_norm_op_group
}
# Call handler to assign grouping.
handler = concat_op_handler.ConcatOpHandler()
handler.assign_grouping(self.concat_op, self.mock_op_reg_manager)
# Verify manager looks up OpSlice for ops of interest.
self.mock_op_reg_manager.get_op_slices.assert_has_calls(
# Checking for ops to process.
[mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Initial slice data.
mock.call(self.concat_op),
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Reslicing.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.concat_op),
mock.call(self.batch_norm_op),
# Refreshing slice data.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Group concat op.
mock.call(self.concat_op)])
# Verify manager does not slices the concat op.
self.mock_op_reg_manager.slice_op.assert_not_called()
# Verify manager groups the new slices.
self.mock_op_reg_manager.group_op_slices.assert_called_once_with([
self.concat_op_slice, self.relu1_op_slice, self.relu2_op_slice,
self.relu3_op_slice, self.batch_norm_op_slice
])
def testGetInputOutputOpSlices(self):
# Map ops to slices.
self.op_slice_dict = {
self.relu1_op: [self.relu1_op_slice],
self.relu2_op: [self.relu2_op_slice],
self.relu3_op: [self.relu3_op_slice],
self.concat_op: [self.concat_op_slice],
self.batch_norm_op: [self.batch_norm_op_slice],
}
input_ops = [self.relu1_op, self.relu2_op, self.relu3_op, self.axis_op]
output_ops = [self.batch_norm_op]
expected_input_op_slices = [[
self.relu1_op_slice, self.relu2_op_slice, self.relu3_op_slice
]]
expected_output_op_slices = [[self.batch_norm_op_slice]]
# Instantiate handler.
handler = concat_op_handler.ConcatOpHandler()
self.assertEqual(
(expected_input_op_slices, expected_output_op_slices),
handler._get_input_output_op_slices(input_ops, output_ops,
self.mock_op_reg_manager))
def testImageIsNotZerothOutputOfOp(self):
# Throughout the framework, we assume that the 0th output of each op is the
# only one of interest. One exception that often happens is when the input
# image comes from a queue or from a staging op. Then the image is one of
# the outputs of the dequeue (or staging) op, not necessarily the 0th one.
# Here we test that the BilinearNetworkRegularizer deals correctly with this
# case.
# Create an input op where the image is output number 1, not 0.
# TODO(g1) Move this mechanism to add_concat_model_stub, possibly using
# tf.split to produce an op where the image is not the 0th output image
# (instead of FIFOQueue).
image = add_concat_model_stub.image_stub()
non_image_tensor = tf.zeros(shape=(41,))
queue = tf.FIFOQueue(
capacity=1,
dtypes=(tf.float32,) * 2,
shapes=(non_image_tensor.shape, image.shape))
# Pass the image (output[1]) to the network.
with arg_scope(self._batch_norm_scope()):
output_op = add_concat_model_stub.build_model(queue.dequeue()[1])
# Create OpHandler dict for test.
op_handler_dict = collections.defaultdict(
grouping_op_handler.GroupingOpHandler)
op_handler_dict.update({
'FusedBatchNorm':
batch_norm_source_op_handler.BatchNormSourceOpHandler(0.1),
'Conv2D':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'ConcatV2':
concat_op_handler.ConcatOpHandler(),
})
# Create OpRegularizerManager and NetworkRegularizer for test.
manager = orm.OpRegularizerManager([output_op], op_handler_dict)
calculator = cost_calculator.CostCalculator(
manager, resource_function.flop_function)
# Calculate expected FLOP cost.
expected_alive_conv1 = sum(add_concat_model_stub.expected_alive()['conv1'])
conv1_op = tf.get_default_graph().get_operation_by_name('conv1/Conv2D')
conv1_coeff = resource_function.flop_coeff(conv1_op)
num_channels = 3
expected_cost = conv1_coeff * num_channels * expected_alive_conv1
with self.session():
tf.global_variables_initializer().run()
# Set gamma values to replicate aliveness in add_concat_model_stub.
name_to_var = {v.op.name: v for v in tf.global_variables()}
gamma1 = name_to_var['conv1/BatchNorm/gamma']
gamma1.assign([0, 1, 1, 0, 1, 0, 1]).eval()
gamma4 = name_to_var['conv4/BatchNorm/gamma']
gamma4.assign([0, 1, 0, 1, 1, 0, 1, 0, 0, 1, 0, 0]).eval()
queue.enqueue((non_image_tensor, image)).run()
self.assertEqual(expected_cost,
calculator.get_cost([conv1_op]).eval())
def __init__(self,
ops,
gamma_threshold,
regularizer_decorator: Type[
generic_regularizers.OpRegularizer] = None,
decorator_parameters=None,
force_group=None,
regularizer_blacklist=None):
"""Creates a GammaFlopsRegularizer object.
Args:
ops: A list of tf.Operation. An OpRegularizer will be created for all the
ops in `ops`, and recursively for all ops they depend on via data
dependency. Typically `ops` would contain a single tf.Operation, which
is the output of the network.
gamma_threshold: A float scalar, will be used as a 'gamma_threshold' for
all instances GammaL1Regularizer created by this class.
regularizer_decorator: A class of OpRegularizer decorator to use.
decorator_parameters: A dictionary of parameters to pass to the decorator
factory. To be used only with decorators that requires parameters,
otherwise use None.
force_group: List of regex for ops that should be force-grouped. Each
regex corresponds to a separate group. Use '|' operator to specify
multiple patterns in a single regex. See op_regularizer_manager for more
detail.
regularizer_blacklist: List of regex for ops that should not be
regularized. See op_regularizer_manager for more detail.
"""
source_op_handler = batch_norm_source_op_handler.BatchNormSourceOpHandler(
gamma_threshold)
if regularizer_decorator:
source_op_handler = op_handler_decorator.OpHandlerDecorator(
source_op_handler, regularizer_decorator, decorator_parameters)
op_handler_dict = collections.defaultdict(
grouping_op_handler.GroupingOpHandler)
op_handler_dict.update({
'FusedBatchNorm':
source_op_handler,
'FusedBatchNormV2':
source_op_handler,
'Conv2D':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'ConcatV2':
concat_op_handler.ConcatOpHandler(),
'DepthToSpace':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'DepthwiseConv2dNative':
depthwise_convolution_op_handler.DepthwiseConvolutionOpHandler(),
'MatMul':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'TensorArrayGatherV3':
leaf_op_handler.LeafOpHandler(),
'RandomUniform':
leaf_op_handler.LeafOpHandler(),
'Reshape':
leaf_op_handler.LeafOpHandler(),
'Transpose':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'ExpandDims':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
})
self._manager = orm.OpRegularizerManager(
ops,
op_handler_dict,
force_group=force_group,
regularizer_blacklist=regularizer_blacklist)
self._calculator = cost_calculator.CostCalculator(
self._manager, resource_function.flop_function)
def __init__(self,
ops,
threshold,
l1_fraction=0,
regularizer_decorator: Type[
generic_regularizers.OpRegularizer] = None,
decorator_parameters=None,
force_group=None,
regularizer_blacklist=None,
convert_to_variable=True):
"""Creates a GroupLassoFlopsRegularizer object.
Args:
ops: A list of tf.Operation. An OpRegularizer will be created for all the
ops in `ops`, and recursively for all ops they depend on via data
dependency. Typically `ops` would contain a single tf.Operation, which
is the output of the network.
threshold: A float scalar, will be used as a 'threshold' for all
regularizer instances created by this class.
l1_fraction: Relative weight of L1 in L1 + L2 regularization.
regularizer_decorator: A class of OpRegularizer decorator to use.
decorator_parameters: A dictionary of parameters to pass to the decorator
factory. To be used only with decorators that requires parameters,
otherwise use None.
force_group: List of regex for ops that should be force-grouped. Each
regex corresponds to a separate group. Use '|' operator to specify
multiple patterns in a single regex. See op_regularizer_manager for more
detail.
regularizer_blacklist: List of regex for ops that should not be
regularized. See op_regularizer_manager for more detail.
convert_to_variable: If `True` convert to variable in the
`GroupLassoBaseOpHandler`. If you're graph creates variables outside of
`tf.get_variable`, set to `False`.
"""
conv2d_handler = conv2d_source_op_handler.Conv2DSourceOpHandler(
threshold, l1_fraction, convert_to_variable)
conv2d_transpose_handler = (
conv2d_transpose_source_op_handler.Conv2DTransposeSourceOpHandler(
threshold, l1_fraction, convert_to_variable))
matmul_handler = matmul_source_op_handler.MatMulSourceOpHandler(
threshold, l1_fraction, convert_to_variable)
if regularizer_decorator:
conv2d_handler = op_handler_decorator.OpHandlerDecorator(
conv2d_handler, regularizer_decorator, decorator_parameters)
conv2d_transpose_handler = op_handler_decorator.OpHandlerDecorator(
conv2d_transpose_handler, regularizer_decorator,
decorator_parameters)
matmul_handler = op_handler_decorator.OpHandlerDecorator(
matmul_handler, regularizer_decorator, decorator_parameters)
op_handler_dict = collections.defaultdict(
grouping_op_handler.GroupingOpHandler)
op_handler_dict.update({
'Conv2D':
conv2d_handler,
'Conv2DBackpropInput':
conv2d_transpose_handler,
'ConcatV2':
concat_op_handler.ConcatOpHandler(),
'DepthToSpace':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'DepthwiseConv2dNative':
depthwise_convolution_op_handler.DepthwiseConvolutionOpHandler(),
'MatMul':
matmul_handler,
'RandomUniform':
leaf_op_handler.LeafOpHandler(),
'Reshape':
leaf_op_handler.LeafOpHandler(),
'Shape':
leaf_op_handler.LeafOpHandler(),
'TensorArrayGatherV3':
leaf_op_handler.LeafOpHandler(),
'Transpose':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'StridedSlice':
leaf_op_handler.LeafOpHandler(),
})
self._manager = orm.OpRegularizerManager(
ops,
op_handler_dict,
force_group=force_group,
regularizer_blacklist=regularizer_blacklist)
self._calculator = cost_calculator.CostCalculator(
self._manager, resource_function.flop_function)
def __init__(self,
ops,
gamma_threshold,
hardware,
batch_size=1,
regularizer_decorator: Type[
generic_regularizers.OpRegularizer] = None,
decorator_parameters=None,
force_group=None,
regularizer_blacklist=None) -> None:
"""Creates a GammaLatencyRegularizer object.
Latency cost and regularization loss is calculated for a specified hardware
platform.
Args:
ops: A list of tf.Operation. An OpRegularizer will be created for all
the ops in `ops`, and recursively for all ops they depend on via data
dependency. Typically `ops` would contain a single tf.Operation, which
is the output of the network.
gamma_threshold: A float scalar, will be used as a 'gamma_threshold' for
all instances GammaL1Regularizer created by this class.
hardware: String name of hardware platform to target. Must be a key from
resource_function.PEAK_COMPUTE.
batch_size: Integer batch size to calculate cost/loss for.
regularizer_decorator: A string, the name of the regularizer decorators
to use. Supported decorators are listed in
op_regularizer_decorator.SUPPORTED_DECORATORS.
decorator_parameters: A dictionary of parameters to pass to the decorator
factory. To be used only with decorators that requires parameters,
otherwise use None.
force_group: List of regex for ops that should be force-grouped. Each
regex corresponds to a separate group. Use '|' operator to specify
multiple patterns in a single regex. See op_regularizer_manager for
more detail.
regularizer_blacklist: List of regex for ops that should not be
regularized. See op_regularizer_manager for more detail.
"""
source_op_handler = batch_norm_source_op_handler.BatchNormSourceOpHandler(
gamma_threshold)
if regularizer_decorator:
source_op_handler = op_handler_decorator.OpHandlerDecorator(
source_op_handler, regularizer_decorator, decorator_parameters)
op_handler_dict = collections.defaultdict(
grouping_op_handler.GroupingOpHandler)
op_handler_dict.update({
'FusedBatchNorm':
source_op_handler,
'FusedBatchNormV2':
source_op_handler,
'Conv2D':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'ConcatV2':
concat_op_handler.ConcatOpHandler(),
'DepthToSpace':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'DepthwiseConv2dNative':
depthwise_convolution_op_handler.DepthwiseConvolutionOpHandler(),
'MatMul':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
'TensorArrayGatherV3':
leaf_op_handler.LeafOpHandler(),
'Transpose':
output_non_passthrough_op_handler.OutputNonPassthroughOpHandler(),
})
self._manager = orm.OpRegularizerManager(
ops,
op_handler_dict,
force_group=force_group,
regularizer_blacklist=regularizer_blacklist)
self._calculator = cost_calculator.CostCalculator(
self._manager,
resource_function.latency_function_factory(hardware, batch_size))
self._hardware = hardware
def testAssignGrouping_NoNeighborGroups(self):
# In this test, both the inputs and outputs are missing groups. The concat
# and batch norm are sliced, but grouping does not happen until the inputs
# and outputs are grouped.
# Map ops to slices.
self.op_slice_dict = {
self.relu1_op: [self.relu1_op_slice],
self.relu2_op: [self.relu2_op_slice],
self.relu3_op: [self.relu3_op_slice],
self.concat_op: [self.concat_op_slice],
self.batch_norm_op: [self.batch_norm_op_slice],
}
# No neighbor slices are grouped.
self.op_group_dict = {
self.concat_op_slice_0_5: self.concat_op_group1,
self.concat_op_slice_5_11: self.concat_op_group2,
self.concat_op_slice_11_18: self.concat_op_group3,
}
# Call handler to assign grouping.
handler = concat_op_handler.ConcatOpHandler()
handler.assign_grouping(self.concat_op, self.mock_op_reg_manager)
# Verify manager looks up OpSlice for ops of interest.
self.mock_op_reg_manager.get_op_slices.assert_has_calls(
# Checking for ops to process.
[
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Initial slice data.
mock.call(self.concat_op),
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Reslicing.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
mock.call(self.concat_op),
# Refreshing slice data.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op)
])
# Verify manager slices ops that do not have aligned OpSlice sizes.
self.mock_op_reg_manager.slice_op.assert_has_calls([
mock.call(self.batch_norm_op, [5, 6, 7]),
mock.call(self.concat_op, [5, 6, 7])
])
# Verify manager doesn't group anything.
self.mock_op_reg_manager.group_op_slices.assert_not_called()
# Verify manager adds ops to processing queue.
self.mock_op_reg_manager.process_ops.assert_called_once_with(
[self.batch_norm_op, self.relu1_op, self.relu2_op, self.relu3_op])
self.mock_op_reg_manager.process_ops_last.assert_called_once_with(
[self.concat_op])
def testAssignGrouping_OutputsGrouped(self):
# In this test, only the output ops are grouped. The concat and batch norm
# ops will be sliced according to the input sizes.
# Map ops to slices.
self.op_slice_dict = {
self.relu1_op: [self.relu1_op_slice],
self.relu2_op: [self.relu2_op_slice],
self.relu3_op: [self.relu3_op_slice],
self.concat_op: [self.concat_op_slice],
self.batch_norm_op: [self.batch_norm_op_slice],
}
# Map each slice to a group. Input ops (ReLU) are not grouped.
self.op_group_dict = {
self.concat_op_slice_0_5: self.concat_op_group1,
self.concat_op_slice_5_11: self.concat_op_group2,
self.concat_op_slice_11_18: self.concat_op_group3,
self.batch_norm_op_slice: self.batch_norm_op_group,
self.batch_norm_op_slice_0_5: self.batch_norm_op_group1,
self.batch_norm_op_slice_5_11: self.batch_norm_op_group2,
self.batch_norm_op_slice_11_18: self.batch_norm_op_group3,
}
# Call handler to assign grouping.
handler = concat_op_handler.ConcatOpHandler()
handler.assign_grouping(self.concat_op, self.mock_op_reg_manager)
# Verify manager looks up OpSlice for ops of interest.
self.mock_op_reg_manager.get_op_slices.assert_has_calls(
# Checking for ops to process.
[
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Initial slice data.
mock.call(self.concat_op),
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Reslicing.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
mock.call(self.concat_op),
# Refreshing slice data.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op)
])
# Verify manager slices ops that do not have aligned OpSlice sizes.
self.mock_op_reg_manager.slice_op.assert_has_calls([
mock.call(self.batch_norm_op, [5, 6, 7]),
mock.call(self.concat_op, [5, 6, 7])
])
# Verify manager does not group ops.
self.mock_op_reg_manager.group_op_slices.assert_not_called()
# Verify manager adds ops to processing queue.
self.mock_op_reg_manager.process_ops.assert_called_once_with(
[self.relu1_op, self.relu2_op, self.relu3_op])
self.mock_op_reg_manager.process_ops_last.assert_called_once_with(
[self.concat_op])
def testAssignGrouping_AllNeighborsGrouped_InputSlicesNotAligned(self):
# In this test, the op c2 has size 6 but is split into 2 slices of size 3.
# The concat op (and its output, the batch norm) both have size 18. This
# test verifies that the concat and batch norm are sliced according to the
# sizes of c1, c2, and c3, and takes into account that c2 is also composed
# of multiple slices.
concat_op_slice_0_5 = orm.OpSlice(self.concat_op, orm.Slice(0, 5))
concat_op_slice_5_8 = orm.OpSlice(self.concat_op, orm.Slice(5, 3))
concat_op_slice_8_11 = orm.OpSlice(self.concat_op, orm.Slice(8, 3))
concat_op_slice_11_18 = orm.OpSlice(self.concat_op, orm.Slice(11, 7))
relu2_op_slice_0_3 = orm.OpSlice(self.relu2_op, orm.Slice(0, 3))
relu2_op_slice_3_6 = orm.OpSlice(self.relu2_op, orm.Slice(3, 3))
relu2_op_group1 = orm.OpGroup(
relu2_op_slice_0_3, omit_source_op_slices=[relu2_op_slice_0_3])
relu2_op_group2 = orm.OpGroup(
relu2_op_slice_3_6, omit_source_op_slices=[relu2_op_slice_3_6])
batch_norm_op_slice = orm.OpSlice(self.batch_norm_op, orm.Slice(0, 18))
batch_norm_op_group = orm.OpGroup(
batch_norm_op_slice, omit_source_op_slices=[batch_norm_op_slice])
batch_norm_op_slice_0_5 = orm.OpSlice(self.batch_norm_op,
orm.Slice(0, 5))
batch_norm_op_group1 = orm.OpGroup(
batch_norm_op_slice_0_5,
omit_source_op_slices=[batch_norm_op_slice_0_5])
batch_norm_op_slice_5_8 = orm.OpSlice(self.batch_norm_op,
orm.Slice(5, 3))
batch_norm_op_group2 = orm.OpGroup(
batch_norm_op_slice_5_8,
omit_source_op_slices=[batch_norm_op_slice_5_8])
batch_norm_op_slice_8_11 = orm.OpSlice(self.batch_norm_op,
orm.Slice(8, 3))
batch_norm_op_group3 = orm.OpGroup(
batch_norm_op_slice_8_11,
omit_source_op_slices=[batch_norm_op_slice_8_11])
batch_norm_op_slice_11_18 = orm.OpSlice(self.batch_norm_op,
orm.Slice(11, 7))
batch_norm_op_group4 = orm.OpGroup(
batch_norm_op_slice_11_18,
omit_source_op_slices=[batch_norm_op_slice_11_18])
# Map ops to slices. The op c2 is composed of multiple slices.
self.op_slice_dict = {
self.relu1_op: [self.relu1_op_slice],
self.relu2_op: [relu2_op_slice_0_3, relu2_op_slice_3_6],
self.relu3_op: [self.relu3_op_slice],
self.concat_op: [self.concat_op_slice],
self.batch_norm_op: [batch_norm_op_slice],
}
# Map each slice to a group.
self.op_group_dict = {
self.relu1_op_slice: self.relu1_op_group,
relu2_op_slice_0_3: relu2_op_group1,
relu2_op_slice_3_6: relu2_op_group2,
self.relu3_op_slice: self.relu3_op_group,
batch_norm_op_slice: batch_norm_op_group,
batch_norm_op_slice_0_5: batch_norm_op_group1,
batch_norm_op_slice_5_8: batch_norm_op_group2,
batch_norm_op_slice_8_11: batch_norm_op_group3,
batch_norm_op_slice_11_18: batch_norm_op_group4,
}
# Update op_slice_dict when an op is sliced.
def slice_op(op, _):
if op == self.batch_norm_op:
self.op_slice_dict[self.batch_norm_op] = [
batch_norm_op_slice_0_5, batch_norm_op_slice_5_8,
batch_norm_op_slice_8_11, batch_norm_op_slice_11_18
]
if op == self.concat_op:
self.op_slice_dict[self.concat_op] = [
concat_op_slice_0_5, concat_op_slice_5_8,
concat_op_slice_8_11, concat_op_slice_11_18
]
self.mock_op_reg_manager.slice_op.side_effect = slice_op
# Call handler to assign grouping.
handler = concat_op_handler.ConcatOpHandler()
handler.assign_grouping(self.concat_op, self.mock_op_reg_manager)
# Verify manager looks up OpSlice for ops of interest.
self.mock_op_reg_manager.get_op_slices.assert_has_calls(
# Checking for ops to process.
[
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Initial slice data.
mock.call(self.concat_op),
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Reslicing.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
mock.call(self.concat_op),
# Refreshing slice data.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Group concat op.
mock.call(self.concat_op)
])
# Verify manager slices ops that do not have aligned OpSlice sizes.
self.mock_op_reg_manager.slice_op.assert_has_calls([
mock.call(self.batch_norm_op, [5, 3, 3, 7]),
mock.call(self.concat_op, [5, 3, 3, 7])
])
# Verify manager groups the new slices.
self.mock_op_reg_manager.group_op_slices.assert_has_calls([
mock.call([
concat_op_slice_0_5, self.relu1_op_slice,
batch_norm_op_slice_0_5
]),
mock.call([
concat_op_slice_5_8, relu2_op_slice_0_3,
batch_norm_op_slice_5_8
]),
mock.call([
concat_op_slice_8_11, relu2_op_slice_3_6,
batch_norm_op_slice_8_11
]),
mock.call([
concat_op_slice_11_18, self.relu3_op_slice,
batch_norm_op_slice_11_18
])
])
def testAssignGrouping_AllNeighborsGrouped_OutputSlicesNotAligned(self):
# The output (batch norm) has sizes [9, 4, 5] which are not aligned. This
# test verifies that the concat, batch norm, and Conv2D ops are sliced in
# alignment.
concat_op_slice_0_5 = orm.OpSlice(self.concat_op, orm.Slice(0, 5))
concat_op_slice_5_9 = orm.OpSlice(self.concat_op, orm.Slice(5, 4))
concat_op_slice_9_11 = orm.OpSlice(self.concat_op, orm.Slice(9, 2))
concat_op_slice_11_13 = orm.OpSlice(self.concat_op, orm.Slice(11, 2))
concat_op_slice_13_18 = orm.OpSlice(self.concat_op, orm.Slice(13, 5))
relu2_op_slice_0_4 = orm.OpSlice(self.relu2_op, orm.Slice(0, 4))
relu2_op_slice_4_6 = orm.OpSlice(self.relu2_op, orm.Slice(4, 2))
relu3_op_slice_0_2 = orm.OpSlice(self.relu3_op, orm.Slice(0, 2))
relu3_op_slice_2_7 = orm.OpSlice(self.relu3_op, orm.Slice(2, 5))
batch_norm_op_slice_0_9 = orm.OpSlice(self.batch_norm_op,
orm.Slice(0, 9))
batch_norm_op_group1 = orm.OpGroup(
batch_norm_op_slice_0_9,
omit_source_op_slices=[batch_norm_op_slice_0_9])
batch_norm_op_slice_9_13 = orm.OpSlice(self.batch_norm_op,
orm.Slice(9, 4))
batch_norm_op_group2 = orm.OpGroup(
batch_norm_op_slice_9_13,
omit_source_op_slices=[batch_norm_op_slice_9_13])
batch_norm_op_slice_13_18 = orm.OpSlice(self.batch_norm_op,
orm.Slice(13, 5))
batch_norm_op_group3 = orm.OpGroup(
batch_norm_op_slice_13_18,
omit_source_op_slices=[batch_norm_op_slice_13_18])
batch_norm_op_slice_0_5 = orm.OpSlice(self.batch_norm_op,
orm.Slice(0, 5))
batch_norm_op_group4 = orm.OpGroup(
batch_norm_op_slice_0_5,
omit_source_op_slices=[batch_norm_op_slice_0_5])
batch_norm_op_slice_5_9 = orm.OpSlice(self.batch_norm_op,
orm.Slice(5, 4))
batch_norm_op_group5 = orm.OpGroup(
batch_norm_op_slice_5_9,
omit_source_op_slices=[batch_norm_op_slice_5_9])
batch_norm_op_slice_9_11 = orm.OpSlice(self.batch_norm_op,
orm.Slice(9, 2))
batch_norm_op_group6 = orm.OpGroup(
batch_norm_op_slice_9_11,
omit_source_op_slices=[batch_norm_op_slice_9_11])
batch_norm_op_slice_11_13 = orm.OpSlice(self.batch_norm_op,
orm.Slice(11, 2))
batch_norm_op_group7 = orm.OpGroup(
batch_norm_op_slice_11_13,
omit_source_op_slices=[batch_norm_op_slice_11_13])
batch_norm_op_slice_13_18 = orm.OpSlice(self.batch_norm_op,
orm.Slice(11, 5))
batch_norm_op_group8 = orm.OpGroup(
batch_norm_op_slice_13_18,
omit_source_op_slices=[batch_norm_op_slice_13_18])
# Map ops to slices. Batch norm op is composed of multiple slices.
self.op_slice_dict = {
self.relu1_op: [self.relu1_op_slice],
self.relu2_op: [self.relu2_op_slice],
self.relu3_op: [self.relu3_op_slice],
self.concat_op: [self.concat_op_slice],
self.batch_norm_op: [
batch_norm_op_slice_0_9, batch_norm_op_slice_9_13,
batch_norm_op_slice_13_18
],
}
# Map each slice to a group.
self.op_group_dict = {
self.relu1_op_slice: self.relu1_op_group,
self.relu2_op_slice: self.relu2_op_group,
self.relu3_op_slice: self.relu3_op_group,
batch_norm_op_slice_0_9: batch_norm_op_group1,
batch_norm_op_slice_9_13: batch_norm_op_group2,
batch_norm_op_slice_13_18: batch_norm_op_group3,
batch_norm_op_slice_0_5: batch_norm_op_group4,
batch_norm_op_slice_5_9: batch_norm_op_group5,
batch_norm_op_slice_9_11: batch_norm_op_group6,
batch_norm_op_slice_11_13: batch_norm_op_group7,
batch_norm_op_slice_13_18: batch_norm_op_group8,
}
# Update op_slice_dict when an op is sliced.
def slice_op(op, _):
if op == self.batch_norm_op:
self.op_slice_dict[self.batch_norm_op] = [
batch_norm_op_slice_0_5, batch_norm_op_slice_5_9,
batch_norm_op_slice_9_11, batch_norm_op_slice_11_13,
batch_norm_op_slice_13_18
]
if op == self.concat_op:
self.op_slice_dict[self.concat_op] = [
concat_op_slice_0_5, concat_op_slice_5_9,
concat_op_slice_9_11, concat_op_slice_11_13,
concat_op_slice_13_18
]
if op == self.relu2_op:
self.op_slice_dict[self.relu2_op] = [
relu2_op_slice_0_4, relu2_op_slice_4_6
]
if op == self.relu3_op:
self.op_slice_dict[self.relu3_op] = [
relu3_op_slice_0_2, relu3_op_slice_2_7
]
self.mock_op_reg_manager.slice_op.side_effect = slice_op
# Call handler to assign grouping.
handler = concat_op_handler.ConcatOpHandler()
handler.assign_grouping(self.concat_op, self.mock_op_reg_manager)
# Verify manager looks up OpSlice for ops of interest.
self.mock_op_reg_manager.get_op_slices.assert_has_calls(
# Checking for ops to process.
[
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Initial slice data.
mock.call(self.concat_op),
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Reslicing.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
mock.call(self.concat_op),
# Refreshing slice data.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Group concat op.
mock.call(self.concat_op)
])
# Verify manager slices ops that do not have aligned OpSlice sizes.
self.mock_op_reg_manager.slice_op.assert_has_calls([
mock.call(self.relu2_op, [4, 2]),
mock.call(self.relu3_op, [2, 5]),
mock.call(self.batch_norm_op, [5, 4, 2, 2, 5]),
mock.call(self.concat_op, [5, 4, 2, 2, 5])
])
# Verify manager groups the new slices.
self.mock_op_reg_manager.group_op_slices.assert_has_calls([
mock.call([
concat_op_slice_0_5, self.relu1_op_slice,
batch_norm_op_slice_0_5
]),
mock.call([
concat_op_slice_5_9, relu2_op_slice_0_4,
batch_norm_op_slice_5_9
]),
mock.call([
concat_op_slice_9_11, relu2_op_slice_4_6,
batch_norm_op_slice_9_11
]),
mock.call([
concat_op_slice_11_13, relu3_op_slice_0_2,
batch_norm_op_slice_11_13
]),
mock.call([
concat_op_slice_13_18, relu3_op_slice_2_7,
batch_norm_op_slice_13_18
])
])
def testAssignGrouping_AllNeighborsGrouped_SlicesAligned_SameGroup(self):
# This test verifies that no slicing or grouping occurs.
# Map ops to slices. Batch norm op is composed of multiple slices.
self.op_slice_dict = {
self.relu1_op: [self.relu1_op_slice],
self.relu2_op: [self.relu2_op_slice],
self.relu3_op: [self.relu3_op_slice],
self.concat_op: [
self.concat_op_slice_0_5, self.concat_op_slice_5_11,
self.concat_op_slice_11_18
],
self.batch_norm_op: [
self.batch_norm_op_slice_0_5, self.batch_norm_op_slice_5_11,
self.batch_norm_op_slice_11_18
],
}
# Map each slice to a group. Corresponding op slices have the same group.
self.op_group_dict = {
self.relu1_op_slice: self.batch_norm_op_group1,
self.relu2_op_slice: self.batch_norm_op_group2,
self.relu3_op_slice: self.batch_norm_op_group3,
self.concat_op_slice_0_5: self.batch_norm_op_group1,
self.concat_op_slice_5_11: self.batch_norm_op_group2,
self.concat_op_slice_11_18: self.batch_norm_op_group3,
self.batch_norm_op_slice_0_5: self.batch_norm_op_group1,
self.batch_norm_op_slice_5_11: self.batch_norm_op_group2,
self.batch_norm_op_slice_11_18: self.batch_norm_op_group3,
}
# Call handler to assign grouping.
handler = concat_op_handler.ConcatOpHandler()
handler.assign_grouping(self.concat_op, self.mock_op_reg_manager)
# Verify manager looks up OpSlice for ops of interest.
self.mock_op_reg_manager.get_op_slices.assert_has_calls(
# Checking for ops to process.
[
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Initial slice data.
mock.call(self.concat_op),
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Reslicing.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
mock.call(self.concat_op),
# Refreshing slice data.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Group concat op.
mock.call(self.concat_op)
])
# Verify manager does not slice any ops.
self.mock_op_reg_manager.slice_op.assert_not_called()
# Verify manager does not group any ops.
self.mock_op_reg_manager.group_op_slices.assert_not_called()
def testAssignGrouping_AllNeighborsGrouped_SlicesAligned(self):
# In this test, the output op (batch norm) has size 18 and is sliced into
# sizes [5, 6, 7] which matches the Conv2D sizes which are [5, 6, 7].
# Map ops to slices. Batch norm op is composed of multiple slices.
self.op_slice_dict = {
self.relu1_op: [self.relu1_op_slice],
self.relu2_op: [self.relu2_op_slice],
self.relu3_op: [self.relu3_op_slice],
self.concat_op: [self.concat_op_slice],
self.batch_norm_op: [
self.batch_norm_op_slice_0_5, self.batch_norm_op_slice_5_11,
self.batch_norm_op_slice_11_18
],
}
# Map each slice to a group.
self.op_group_dict = {
self.relu1_op_slice: self.relu1_op_group,
self.relu2_op_slice: self.relu2_op_group,
self.relu3_op_slice: self.relu3_op_group,
self.batch_norm_op_slice_0_5: self.batch_norm_op_group1,
self.batch_norm_op_slice_5_11: self.batch_norm_op_group2,
self.batch_norm_op_slice_11_18: self.batch_norm_op_group3,
}
# Call handler to assign grouping.
handler = concat_op_handler.ConcatOpHandler()
handler.assign_grouping(self.concat_op, self.mock_op_reg_manager)
# Verify manager looks up OpSlice for ops of interest.
self.mock_op_reg_manager.get_op_slices.assert_has_calls(
# Checking for ops to process.
[
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Initial slice data.
mock.call(self.concat_op),
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Reslicing.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
mock.call(self.concat_op),
# Refreshing slice data.
mock.call(self.relu1_op),
mock.call(self.relu2_op),
mock.call(self.relu3_op),
mock.call(self.batch_norm_op),
# Group concat op.
mock.call(self.concat_op)
])
# Verify manager only slices the concat op.
self.mock_op_reg_manager.slice_op.assert_called_once_with(
self.concat_op, [5, 6, 7])
# Verify manager groups the new slices.
self.mock_op_reg_manager.group_op_slices.assert_has_calls([
mock.call([
self.concat_op_slice_0_5, self.relu1_op_slice,
self.batch_norm_op_slice_0_5
]),
mock.call([
self.concat_op_slice_5_11, self.relu2_op_slice,
self.batch_norm_op_slice_5_11
]),
mock.call([
self.concat_op_slice_11_18, self.relu3_op_slice,
self.batch_norm_op_slice_11_18
])
])
