Skip to main content

Accera - An optimizing cross-compiler for compute-intensive code

Project description

Accera logo

PyPI package version Python versions MIT License

Problem at Hand

Writing highly optimized compute-intensive code in a traditional programming language is strenuous and time-consuming. Not only does it require advanced engineering skills such as fluency in Assembly language, but a deep understanding of computer architecture is also indispensable. Manual optimization of even the simplest numerical algorithms demands a significant engineering effort. Needless to say, a highly optimized numerical code is often prone to bugs, lacks readability, and offers little to no usability. Code maintenance becomes a nightmare resulting in the reimplementation of the same logic every time an architecture level change is introduced.

Accera: An Optimized Solution

Accera is a compiler that enables you to experiment with loop optimizations without hand-writing Assembly code. With Accera, these problems and impediments can be addressed in an optimized way. It is available as a Python library and supports cross-compiling to a wide range of processor targets.

Accera has THREE primary goals:

  • Performance: To guarantee the fastest implementation for any compute-intensive algorithm.
  • Readability: To ensure effective implementation of algorithms without sacrificing the readability of code.
  • Writability: To provide a user-friendly programming model, designed for agility and maintainability.

Install

To install for Linux, macOS, or Windows (requires Python 3.7-3.10):

pip install accera

See the Install Instructions for more details on installing pre-built Python 3 packages and how to build Accera from the source.

Quickstart

In this example, we will:

  • Implement matrix multiplication with a ReLU activation (matmul + ReLU), commonly used in machine learning algorithms.
  • Generate two implementations: a naive algorithm and loop-based transformations.
  • Compare the execution time of both implementations.

Run in your browser

Binder

No installation is required. This will launch a Jupyter notebook with the quickstart example running in the cloud.

Run on your machine

  1. Create a Python 3 script called quickstart.py:

    import accera as acc
    
    # define placeholder inputs/output
    A = acc.Array(role=acc.Role.INPUT, shape=(512, 512))
    B = acc.Array(role=acc.Role.INPUT, shape=(512, 512))
    C = acc.Array(role=acc.Role.INPUT_OUTPUT, shape=(512, 512))
    
    # implement the logic for matmul and relu
    matmul = acc.Nest(shape=(512, 512, 512))
    i1, j1, k1 = matmul.get_indices()
    @matmul.iteration_logic
    def _():
        C[i1, j1] += A[i1, k1] * B[k1, j1]
    
    relu = acc.Nest(shape=(512, 512))
    i2, j2 = relu.get_indices()
    @relu.iteration_logic
    def _():
        C[i2, j2] = acc.max(C[i2, j2], 0.0)
    
    package = acc.Package()
    
    # fuse the i and j indices of matmul and relu, add to the package
    schedule = acc.fuse(matmul.create_schedule(), relu.create_schedule(), partial=2)
    package.add(schedule, args=(A, B, C), base_name="matmul_relu_fusion_naive")
    
    # transform the schedule, add to the package
    i, j, f, k = schedule.get_indices()
    ii, jj = schedule.tile({
        i: 16,
        j: 16
    }) # loop tiling
    schedule.reorder(j, i, f, k, jj, ii) # loop reordering
    plan = schedule.create_plan()
    plan.unroll(ii) # loop unrolling
    package.add(plan, args=(A, B, C), base_name="matmul_relu_fusion_transformed")
    
    # build a dynamically-linked package (a .dll or .so) that exports both functions
    print(package.build(name="hello_accera", format=acc.Package.Format.HAT_DYNAMIC))
    
  2. Ensure that you have a compiler in your PATH:

    • Windows: Install Microsoft Visual Studio and run vcvars64.bat to setup the command prompt.
    • Linux/macOS: Install gcc

    Don't have a compiler handy? We recommend trying Accera in your browser instead Binder

  3. Install Accera:

    pip install accera
    
  4. Generate the library that implements two versions of matmul + ReLU:

    python quickstart.py
    
  5. To consume and compare the library functions, create a file called benchmark.py in the same location:

    import hatlib as hat
    import numpy as np
    
    # load the package
    _, functions = hat.load("hello_accera.hat")
    
    # call one of the functions with test inputs
    A_test = np.random.rand(512, 512).astype(np.float32)
    B_test = np.random.rand(512, 512).astype(np.float32)
    C_test = np.zeros((512, 512)).astype(np.float32)
    C_numpy = np.maximum(C_test + A_test @ B_test, 0.0)
    
    matmul_relu = functions["matmul_relu_fusion_transformed"]
    matmul_relu(A_test, B_test, C_test)
    
    # check correctness
    np.testing.assert_allclose(C_test, C_numpy, atol=1e-3)
    
    # benchmark all functions
    hat.run_benchmark("hello_accera.hat", batch_size=5, min_time_in_sec=5)
    
  6. Run the benchmark to get the execution time results:

    python benchmark.py
    

Next Steps

The Manual is the best introductory resource for the Accera Python programming model.

In particular, the schedule transformations describe how you can experiment with different loop transformations with just a few lines of Python code.

Finally, the .hat format is just a C header file containing the metadata. Learn more about the HAT format and benchmarking.

How it works

In a nutshell, Accera takes the Python code that defines the loop schedule and algorithm while converting it into MLIR intermediate representation (IR). Accera's compiler then takes this IR through a series of MLIR pipelines to perform transformations. The result is a binary library with a C header file. The library implements the algorithms that are defined in Python and it is compatible with the target.

To peek into the stages of IR transformation that Accera does, try replacing format=acc.Package.Format.HAT_DYNAMIC with format=acc.Package.Format.MLIR_DYNAMIC in quickstart.py, re-run the script, and search the _tmp subfolder for the intermediate *.mlir files. We plan to document these IR constructs in the future.

Documentation

Get familiar with Accera's concepts and Python constructs in the Documentation page.

Tutorials

Step-by-step examples are available on the Tutorials page. We're working on adding more complementary examples and tutorials.

Contributions

Accera is a research platform-in-progress that can certainly benefit from your contributions. We would love your feedback, recommendations, and feature requests. Not to mention that we are excited to answer your questions. Let’s collaborate! Please file a Github issue or send us a pull request. Please review the Microsoft Code of Conduct to learn more.

Credits

Accera is built using several open source libraries, including: LLVM, pybind11, toml++, tomlkit, vcpkg, pyyaml, and HAT. For testing, we used numpy and catch2.

License

This project is released under the MIT License.

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distributions

accera-1.2.29-cp310-cp310-win_amd64.whl (24.4 MB view hashes)

Uploaded CPython 3.10 Windows x86-64

accera-1.2.29-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (31.1 MB view hashes)

Uploaded CPython 3.10 manylinux: glibc 2.17+ x86-64

accera-1.2.29-cp39-cp39-win_amd64.whl (24.4 MB view hashes)

Uploaded CPython 3.9 Windows x86-64

accera-1.2.29-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (31.1 MB view hashes)

Uploaded CPython 3.9 manylinux: glibc 2.17+ x86-64

accera-1.2.29-cp38-cp38-win_amd64.whl (24.4 MB view hashes)

Uploaded CPython 3.8 Windows x86-64

accera-1.2.29-cp38-cp38-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (31.1 MB view hashes)

Uploaded CPython 3.8 manylinux: glibc 2.17+ x86-64

accera-1.2.29-cp37-cp37m-win_amd64.whl (24.4 MB view hashes)

Uploaded CPython 3.7m Windows x86-64

accera-1.2.29-cp37-cp37m-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (31.1 MB view hashes)

Uploaded CPython 3.7m manylinux: glibc 2.17+ x86-64

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page