onnxruntime/orttraining/tools/scripts/performance_investigation.py

91 lines
2.8 KiB
Python
Raw Normal View History

import argparse
Adopt linrtunner as the linting tool - take 2 (#15085) ### Description `lintrunner` is a linter runner successfully used by pytorch, onnx and onnx-script. It provides a uniform experience running linters locally and in CI. It supports all major dev systems: Windows, Linux and MacOs. The checks are enforced by the `Python format` workflow. This PR adopts `lintrunner` to onnxruntime and fixed ~2000 flake8 errors in Python code. `lintrunner` now runs all required python lints including `ruff`(replacing `flake8`), `black` and `isort`. Future lints like `clang-format` can be added. Most errors are auto-fixed by `ruff` and the fixes should be considered robust. Lints that are more complicated to fix are applied `# noqa` for now and should be fixed in follow up PRs. ### Notable changes 1. This PR **removed some suboptimal patterns**: - `not xxx in` -> `xxx not in` membership checks - bare excepts (`except:` -> `except Exception`) - unused imports The follow up PR will remove: - `import *` - mutable values as default in function definitions (`def func(a=[])`) - more unused imports - unused local variables 2. Use `ruff` to replace `flake8`. `ruff` is much (40x) faster than flake8 and is more robust. We are using it successfully in onnx and onnx-script. It also supports auto-fixing many flake8 errors. 3. Removed the legacy flake8 ci flow and updated docs. 4. The added workflow supports SARIF code scanning reports on github, example snapshot: ![image](https://user-images.githubusercontent.com/11205048/212598953-d60ce8a9-f242-4fa8-8674-8696b704604a.png) 5. Removed `onnxruntime-python-checks-ci-pipeline` as redundant ### Motivation and Context <!-- - Why is this change required? What problem does it solve? - If it fixes an open issue, please link to the issue here. --> Unified linting experience in CI and local. Replacing https://github.com/microsoft/onnxruntime/pull/14306 --------- Signed-off-by: Justin Chu <justinchu@microsoft.com>
2023-03-24 22:29:03 +00:00
import onnx
parser = argparse.ArgumentParser(description="ONNX file analyzer for performance investigation.")
parser.add_argument("onnx_file", type=str, help="ONNX file to analyze")
args = parser.parse_args()
def process_file(onnx_file):
model = onnx.load(onnx_file)
# Map from output arg to the producer of the output.
output_to_node = {}
for node in model.graph.node:
for o in node.output:
output_to_node[o] = node
aten_ops = []
python_ops = []
memcpu_ops = []
cast_ops = []
msgs = []
for node in model.graph.node:
if "Memcpy" in node.op_type:
memcpu_ops.append(f"{node.op_type} {node.name}")
if node.op_type == "Cast":
cast_ops.append(f"{node.name}")
if node.op_type == "ATen":
for attr in node.attribute:
if attr.name == "operator":
aten_ops.append(f"{node.name}: {attr.s.decode('utf-8')}")
if node.op_type == "PythonOp":
for attr in node.attribute:
if attr.name == "name":
python_ops.append(f"{node.name}: {attr.s.decode('utf-8')}")
# Look for stand-alone Dropout node in *_execution_model_<mode>.onnx graph.
# Examine whether it should be fused with surrounding Add ops into BiasDropout node.
if node.op_type == "Dropout" and len(node.input) == 1:
prev = output_to_node[node.input[0]]
if prev.op_type == "Add":
msgs.append(
f"Examine whether {node.name} should be fused with the leading {prev.name} op into BiasDropout node."
)
# Look for stand-alone Softmax node in *_execution_model_<mode>.onnx graph.
# Examine whether it should be fused with the leading Add ops into BiasSoftmax node.
if node.op_type == "Softmax" and len(node.input) == 1:
prev = output_to_node[node.input[0]]
if prev.op_type == "Add":
msgs.append(
f"Examine whether {node.name} should be fused with the leading {prev.name} op into BiasSoftmax node."
)
if aten_ops:
print("ATen op found:")
for line in aten_ops:
print(line)
print(10 * "-")
if python_ops:
print("PythonOp found:")
for line in python_ops:
print(line)
print(10 * "-")
if memcpu_ops:
print("Memcpu ops found:")
for line in memcpu_ops:
print(line)
print(10 * "-")
if cast_ops:
print("Cast ops found:")
for line in cast_ops:
print(line)
print(10 * "-")
for line in msgs:
print(line)
def main():
process_file(args.onnx_file)
if __name__ == "__main__":
main()