Add a converter from input_pipeline.proto to GViz DataTable format.
PiperOrigin-RevId: 289935716 Change-Id: I1ff7da7a5a527f890ae35cbe6d1075bd1aeee8b7
This commit is contained in:
A. Unique TensorFlower
TensorFlower Gardener
parent
ea924bf44f
commit
bc996ac058
75
tensorflow/python/profiler/input_pipeline_proto_to_gviz.py
Normal file
75
tensorflow/python/profiler/input_pipeline_proto_to_gviz.py
Normal file
@ -0,0 +1,75 @@
|
||||
# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
# ==============================================================================
|
||||
"""For conversion of TF Input Pipeline Analyzer protos to GViz DataTables.
|
||||
|
||||
Usage:
|
||||
gviz_data_table = generate_chart_table(ipa)
|
||||
"""
|
||||
|
||||
from __future__ import absolute_import
|
||||
from __future__ import division
|
||||
from __future__ import google_type_annotations
|
||||
from __future__ import print_function
|
||||
|
||||
import gviz_api
|
||||
|
||||
from tensorflow.core.profiler.protobuf import input_pipeline_pb2
|
||||
|
||||
|
||||
def get_chart_table_args(ipa):
|
||||
"""Creates a gviz DataTable object from an Input Pipeline Analyzer proto.
|
||||
|
||||
Args:
|
||||
ipa: An input_pipeline_pb2.InputPipelineAnalysisResult.
|
||||
|
||||
Returns:
|
||||
Returns a gviz_api.DataTable
|
||||
"""
|
||||
|
||||
table_description = [
|
||||
("stepnum", "string", "Step number"),
|
||||
("deviceComputeTimeMs", "number", "Device compute"),
|
||||
("deviceToDeviceTimeMs", "number", "Device to device"),
|
||||
("hostComputeTimeMs", "number", "Host compute"),
|
||||
("kernelLaunchTimeMs", "number", "Kernel launch"),
|
||||
("infeedTimeMs", "number", "Input"),
|
||||
("outfeedTimeMs", "number", "Output"),
|
||||
("compileTimeMs", "number", "Compilation"),
|
||||
("otherTimeMs", "number", "All others"),
|
||||
]
|
||||
|
||||
data = []
|
||||
for step_details in ipa.step_details:
|
||||
details = input_pipeline_pb2.PerGenericStepDetails()
|
||||
step_details.Unpack(details)
|
||||
row = [
|
||||
str(details.step_number),
|
||||
details.device_compute_ms,
|
||||
details.device_to_device_ms,
|
||||
details.host_compute_ms,
|
||||
details.host_prepare_ms,
|
||||
details.host_wait_input_ms + details.host_to_device_ms,
|
||||
details.output_ms,
|
||||
details.host_compile_ms,
|
||||
details.unknown_time_ms,
|
||||
]
|
||||
data.append(row)
|
||||
|
||||
return (table_description, data, [])
|
||||
|
||||
|
||||
def generate_chart_table(ipa):
|
||||
(table_description, data, custom_properties) = get_chart_table_args(ipa)
|
||||
return gviz_api.DataTable(table_description, data, custom_properties)
|
||||
209
tensorflow/python/profiler/input_pipeline_proto_to_gviz_test.py
Normal file
209
tensorflow/python/profiler/input_pipeline_proto_to_gviz_test.py
Normal file
@ -0,0 +1,209 @@
|
||||
# Copyright 2020 The TensorFlow Authors. All Rights Reserved.
|
||||
#
|
||||
# Licensed under the Apache License, Version 2.0 (the "License");
|
||||
# you may not use this file except in compliance with the License.
|
||||
# You may obtain a copy of the License at
|
||||
#
|
||||
# http://www.apache.org/licenses/LICENSE-2.0
|
||||
#
|
||||
# Unless required by applicable law or agreed to in writing, software
|
||||
# distributed under the License is distributed on an "AS IS" BASIS,
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
# ==============================================================================
|
||||
|
||||
# Lint as: python3
|
||||
"""Tests for input_pipeline_proto_to_gviz."""
|
||||
|
||||
from __future__ import absolute_import
|
||||
from __future__ import division
|
||||
from __future__ import print_function
|
||||
|
||||
import csv
|
||||
import io
|
||||
import enum
|
||||
|
||||
import gviz_api
|
||||
# pylint: disable=g-importing-member
|
||||
from google.protobuf.any_pb2 import Any
|
||||
# pylint: enable=g-importing-member
|
||||
|
||||
# pylint: disable=g-direct-tensorflow-import
|
||||
from tensorflow.core.profiler.protobuf import hardware_types_pb2
|
||||
from tensorflow.core.profiler.protobuf import input_pipeline_pb2
|
||||
from tensorflow.python.platform import test
|
||||
from tensorflow.python.profiler import input_pipeline_proto_to_gviz
|
||||
# pylint: enable=g-direct-tensorflow-import
|
||||
|
||||
|
||||
class MockValues(enum.IntEnum):
|
||||
STEP_NUMBER = 1
|
||||
STEP_TIME_MS = 2
|
||||
UNKNOWN_TIME_MS = 3
|
||||
HOST_WAIT_INPUT_MS = 11
|
||||
HOST_TO_DEVICE_MS = 12
|
||||
OUTPUT_MS = 5
|
||||
DEVICE_COMPUTE_MS = 6
|
||||
DEVICE_TO_DEVICE_MS = 7
|
||||
HOST_COMPUTE_MS = 8
|
||||
HOST_PREPARE_MS = 9
|
||||
HOST_COMPILE_MS = 10
|
||||
|
||||
|
||||
class ProtoToGvizTest(test.TestCase):
|
||||
|
||||
def create_empty_input_pipeline(self):
|
||||
return input_pipeline_pb2.InputPipelineAnalysisResult()
|
||||
|
||||
def create_mock_step_summary(self, base):
|
||||
step_summary = input_pipeline_pb2.StepSummary()
|
||||
step_summary.average = 1 + base
|
||||
step_summary.standard_deviation = 2 + base
|
||||
step_summary.minimum = 3 + base
|
||||
step_summary.maximum = 4 + base
|
||||
return step_summary
|
||||
|
||||
def create_mock_input_pipeline(self):
|
||||
ipa = input_pipeline_pb2.InputPipelineAnalysisResult()
|
||||
ipa.hardware_type = hardware_types_pb2.HardwareType.CPU_ONLY
|
||||
ipa.step_time_summary.CopyFrom(self.create_mock_step_summary(10))
|
||||
ipa.input_percent_summary.CopyFrom(self.create_mock_step_summary(20))
|
||||
|
||||
# Add 3 rows
|
||||
for _ in range(0, 3):
|
||||
step_details = input_pipeline_pb2.PerGenericStepDetails()
|
||||
step_details.step_number = MockValues.STEP_NUMBER
|
||||
step_details.step_time_ms = MockValues.STEP_TIME_MS
|
||||
step_details.unknown_time_ms = MockValues.UNKNOWN_TIME_MS
|
||||
step_details.host_wait_input_ms = MockValues.HOST_WAIT_INPUT_MS
|
||||
step_details.host_to_device_ms = MockValues.HOST_TO_DEVICE_MS
|
||||
step_details.output_ms = MockValues.OUTPUT_MS
|
||||
step_details.device_compute_ms = MockValues.DEVICE_COMPUTE_MS
|
||||
step_details.device_to_device_ms = MockValues.DEVICE_TO_DEVICE_MS
|
||||
step_details.host_compute_ms = MockValues.HOST_COMPUTE_MS
|
||||
step_details.host_prepare_ms = MockValues.HOST_PREPARE_MS
|
||||
step_details.host_compile_ms = MockValues.HOST_COMPILE_MS
|
||||
|
||||
step_details_any = Any()
|
||||
step_details_any.Pack(step_details)
|
||||
ipa.step_details.append(step_details_any)
|
||||
|
||||
input_time_breakdown = input_pipeline_pb2.InputTimeBreakdown()
|
||||
input_time_breakdown.demanded_file_read_us = 1
|
||||
input_time_breakdown.advanced_file_read_us = 2
|
||||
input_time_breakdown.preprocessing_us = 3
|
||||
input_time_breakdown.enqueue_us = 4
|
||||
input_time_breakdown.unclassified_non_enqueue_us = 5
|
||||
ipa.input_time_breakdown.CopyFrom(input_time_breakdown)
|
||||
|
||||
for _ in range(0, 3):
|
||||
input_op_details = input_pipeline_pb2.InputOpDetails()
|
||||
input_op_details.op_name = str(1)
|
||||
input_op_details.count = 2
|
||||
input_op_details.time_in_ms = 3
|
||||
input_op_details.time_in_percent = 4
|
||||
input_op_details.self_time_in_ms = 5
|
||||
input_op_details.self_time_in_percent = 6
|
||||
input_op_details.category = str(7)
|
||||
ipa.input_op_details.append(input_op_details)
|
||||
|
||||
recommendation = input_pipeline_pb2.InputPipelineAnalysisRecommendation()
|
||||
for ss in ["a", "b", "c", "d", "e"]:
|
||||
recommendation.details.append(ss)
|
||||
ipa.recommendation.CopyFrom(recommendation)
|
||||
|
||||
step_time_breakdown = input_pipeline_pb2.GenericStepTimeBreakdown()
|
||||
step_time_breakdown.unknown_time_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(1))
|
||||
step_time_breakdown.host_wait_input_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(9))
|
||||
step_time_breakdown.host_to_device_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(10))
|
||||
step_time_breakdown.input_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(11))
|
||||
step_time_breakdown.output_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(3))
|
||||
step_time_breakdown.device_compute_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(4))
|
||||
step_time_breakdown.device_to_device_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(5))
|
||||
step_time_breakdown.host_compute_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(6))
|
||||
step_time_breakdown.host_prepare_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(7))
|
||||
step_time_breakdown.host_compile_ms_summary.CopyFrom(
|
||||
self.create_mock_step_summary(8))
|
||||
|
||||
step_time_breakdown_any = Any()
|
||||
step_time_breakdown_any.Pack(step_time_breakdown)
|
||||
ipa.step_time_breakdown.CopyFrom(step_time_breakdown_any)
|
||||
|
||||
return ipa
|
||||
|
||||
def test_input_pipeline_empty(self):
|
||||
ipa = self.create_empty_input_pipeline()
|
||||
data_table = input_pipeline_proto_to_gviz.generate_chart_table(ipa)
|
||||
|
||||
self.assertEqual(0, data_table.NumberOfRows(),
|
||||
"Empty table should have 0 rows.")
|
||||
# Input pipeline chart data table has 9 columns.
|
||||
self.assertLen(data_table.columns, 9)
|
||||
|
||||
def test_input_pipeline_simple(self):
|
||||
ipa = self.create_mock_input_pipeline()
|
||||
(table_description, data,
|
||||
custom_properties) = input_pipeline_proto_to_gviz.get_chart_table_args(ipa)
|
||||
data_table = gviz_api.DataTable(table_description, data, custom_properties)
|
||||
|
||||
# Data is a list of 3 rows.
|
||||
self.assertLen(data, 3)
|
||||
self.assertEqual(3, data_table.NumberOfRows(), "Simple table has 3 rows.")
|
||||
# Table descriptor is a list of 9 columns.
|
||||
self.assertLen(table_description, 9)
|
||||
# DataTable also has 9 columns.
|
||||
self.assertLen(data_table.columns, 9)
|
||||
|
||||
csv_file = io.StringIO(data_table.ToCsv())
|
||||
reader = csv.reader(csv_file)
|
||||
|
||||
expected = [
|
||||
str(int(MockValues.STEP_NUMBER)),
|
||||
int(MockValues.DEVICE_COMPUTE_MS),
|
||||
int(MockValues.DEVICE_TO_DEVICE_MS),
|
||||
int(MockValues.HOST_COMPUTE_MS),
|
||||
int(MockValues.HOST_PREPARE_MS),
|
||||
int(MockValues.HOST_WAIT_INPUT_MS) + int(MockValues.HOST_TO_DEVICE_MS),
|
||||
int(MockValues.OUTPUT_MS),
|
||||
int(MockValues.HOST_COMPILE_MS),
|
||||
int(MockValues.UNKNOWN_TIME_MS),
|
||||
]
|
||||
|
||||
for (rr, row_values) in enumerate(reader):
|
||||
if rr == 0:
|
||||
# DataTable columns match schema defined in table_description.
|
||||
for (cc, column_header) in enumerate(row_values):
|
||||
self.assertEqual(table_description[cc][2], column_header)
|
||||
else:
|
||||
for (cc, cell_str) in enumerate(row_values):
|
||||
raw_value = data[rr - 1][cc]
|
||||
value_type = table_description[cc][1]
|
||||
|
||||
# Only number and strings are used in our DataTable schema.
|
||||
self.assertIn(value_type, ["number", "string"])
|
||||
|
||||
# Encode in similar fashion as DataTable.ToCsv().
|
||||
expected_value = gviz_api.DataTable.CoerceValue(raw_value, value_type)
|
||||
self.assertNotIsInstance(expected_value, tuple)
|
||||
self.assertEqual(expected_value, raw_value)
|
||||
self.assertEqual(str(expected_value), cell_str)
|
||||
|
||||
# Check against expected values we have set in our mock table.
|
||||
if isinstance(expected[cc], str):
|
||||
self.assertEqual(expected[cc], cell_str)
|
||||
else:
|
||||
self.assertEqual(str(float(expected[cc])), cell_str)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
test.main()
|
||||
@ -12,7 +12,7 @@
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
# ==============================================================================
|
||||
"""Contains utilities for conversion of TF proto types to GViz types.
|
||||
"""Contains utilities for conversion of TF stats proto types to GViz types.
|
||||
|
||||
Usage:
|
||||
gviz_data_table = generate_chart_table(stats_table)
|
||||
@ -27,7 +27,7 @@ import gviz_api
|
||||
|
||||
|
||||
def get_chart_table_args(stats_table):
|
||||
"""Creates gviz DataTable object from a a TensorFlow stats table.
|
||||
"""Creates gviz DataTable object from a TensorFlow stats table.
|
||||
|
||||
Args:
|
||||
stats_table: A tf_stats_pb2.TfStatsTable.
|
||||
|
||||
Loading…
Reference in New Issue
Block a user