About

The PyTimeloop project is a combination of Python bindings to the Timeloop library as well as TimeloopFE. The Timeloop Front-End (TimeloopFE) is a Python front-end interface to the Timeloop infrastructure, which allows users to model tensor accelerators and explore the vast space of architectures, workloads, and mappings.

TimeloopFE provides a rich Python interface, error checking, and automation tools. With closely-aligned Python and YAML interfaces, TimeloopFE is designed to enable easy design space exploration and automation.

Documentation

Documentation for the full framework is available at timeloop.csail.mit.edu. Documentation for TimeloopFE is available at accelergy-project.github.io/timeloopfe/index.html.

Dependencies

Since building PyTimeloop requires building Timeloop, dependencies of Timeloop are also required.

// Timeloop dependencies
scons
libconfig++-dev
libboost-dev
libboost-iostreams-dev
libboost-serialization-dev
libyaml-cpp-dev
libncurses-dev
libtinfo-dev
libgpm-dev

// PyTimeloop dependencies
cmake

Moreover, PyTimeloop requires the installation of islpy with Barvinok support enabled.

Installation

First, ensure that Timeloop and Accelergy are installed following the Timeloop+Accelergy install instructions.

Recrusively clone the repo

 git clone --recurse-submodules git@github.com:Accelergy-Project/timeloop-python.git

Update the git submodules using

$ git submodule update

Point to your Timeloop source and built libraries:

export TIMELOOP_INCLUDE_PATH=your/timeloop/path/include
export TIMELOOP_LIB_PATH=directory/with/timeloop/libs

Note that Timeloop needs to be built to generate dynamic libraries.

Then, install PyTimeloop by running

$ pip3 install -e .

If you ran pip3 install -e . recently, the build/ directory has to be cleaned by running rm -rf build.

Tutorials and Examples

Tutorials and examples available in the Timeloop and Accelergy exercises repository. In this repository, examples can be found in the workspace/baseline_designs directory and tutorials can be found in the workspace/exercises directory.

Minimal Usage

TimeloopFE interface provides two primary functions: - Input file gathering & error checking - Python interface for design space exploration

import pytimeloop.timeloopfe.v4 as tl
from joblib import Parallel, delayed

# Basic setup. Gathers input files, checks for errors
spec = tl.Specification.from_yaml_files(
  "your_input_file.yaml", "your_other_input_file.yaml"
)
# Call Timeloop mapper
tl.call_mapper(spec, output_dir="your_output_dir")
# Call Accelergy verbose
tl.call_accelergy_verbose(spec, output_dir="your_output_dir")

# Multiprocessed design space exploration
def run_mapper_with_spec(buf_size: int):
  spec = tl.Specification.from_yaml_files(
    "your_input_file.yaml", "your_other_input_file.yaml"
  )
  spec.architecture.find("my_buffer").attributes.depth = buf_size
  return tl.call_mapper(spec, output_dir=f"outputs_bufsize={buf_size}")

buf_sizes = [1024, 2048, 4096, 8192, 16384]
results = Parallel(n_jobs=8)(
  delayed(run_mapper_with_spec)(buf_size) for buf_size in buf_sizes
)

Please visit the Timeloop and Accelergy exercises repository for more examples and tutorials.

Migrating from Standalone TimeloopFE

If you have installed TimeloopFE that is not included in pytimeloop and want to migrate to this one, you can simply import pytimeloop.timeloopfe instead of import timeloopfe. Submodule structures are unchanged.

General Debugging Information

1. When debugging the C++ bindings, it may be faster to build by calling cmake directly instead of pip3 install

Citation

Please cite the following:

• A. Parashar, P. Raina, Y. S. Shao, Y.-H. Chen, V. A. Ying, A. Mukkara, R. Venkatesan, B. Khailany, S. W. Keckler, and J. Emer, “Timeloop: A systematic approach to DNN accelerator evaluation,” in 2019 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), 2019, pp. 304–315.
• M. Horeni, P. Taheri, P. Tsai, A. Parashar, J. Emer, and S. Joshi, “Ruby: Improving hardware efficiency for tensor algebra accelerators through imperfect factorization,” in 2022 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), 2022, pp. 254–266.
• Y. N. Wu, P.-A. Tsai, A. Parashar, V. Sze, and J. S. Emer, “Sparseloop: An analytical, energy-focused design space exploration methodology for sparse tensor accelerators,” in 2021 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), 2021, pp. 232–234.
• Y. N. Wu, J. S. Emer, and V. Sze, “Accelergy: An architecture-level energy estimation methodology for accelerator designs,” in 2019 IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2019, pp. 1–8.
• T. Andrulis, J. S. Emer, and V. Sze, “CiMLoop: A flexible, accurate, and fast compute-in-memory modeling tool,” in 2024 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), 2024.

Or use the following BibTeX:

@inproceedings{timeloop,
  author      = {Parashar, Angshuman and Raina, Priyanka and Shao, Yakun Sophia and  Chen, Yu-Hsin and Ying, Victor A and Mukkara, Anurag and Venkatesan, Rangharajan and Khailany, Brucek and Keckler, Stephen W and Emer, Joel},
  booktitle   = {2019 IEEE international symposium on performance analysis of systems and software (ISPASS)}, pages={304--315}, year={2019},
  title       = {Timeloop: A systematic approach to dnn accelerator evaluation},
  year        = {2019},
}
@inproceedings{ruby,
  author      = {M. Horeni and P. Taheri and P. Tsai and A. Parashar and J. Emer and S. Joshi},
  booktitle   = {2022 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)},
  title       = {Ruby: Improving Hardware Efficiency for Tensor Algebra Accelerators Through Imperfect Factorization},
  year        = {2022},
}
@inproceedings{sparseloop,
  author      = {Wu, Yannan N. and Tsai, Po-An, and Parashar, Angshuman and Sze, Vivienne and Emer, Joel S.},
  booktitle   = {{ ACM/IEEE International Symposium on Microarchitecture (MICRO)}},
  title       = {{Sparseloop: An Analytical Approach To Sparse Tensor Accelerator Modeling }},
  year        = {{2022}}
}
@inproceedings{accelergy,
  author      = {Wu, Yannan Nellie and Emer, Joel S and Sze, Vivienne},
  booktitle   = {2019 IEEE/ACM International Conference on Computer-Aided Design (ICCAD)},
  title       = {Accelergy: An architecture-level energy estimation methodology for accelerator designs},
  year        = {2019},
}
@inproceedings{cimloop,
  author      = {Andrulis, Tanner and Emer, Joel S. and Sze, Vivienne},
  booktitle   = {2024 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)}, 
  title       = {{CiMLoop}: A Flexible, Accurate, and Fast Compute-In-Memory Modeling Tool}, 
  year        = {2024},
}