cometpy 0.1

Comet Domain Specific Compiler as Python package

cometpy

Comet Domain Specific Compiler for NumPy frontends.

Latest Version : v0.1

Requirements:

1. python 3 and above
2. COMET Python package v0.1
3. OS support - Linux and macOS

"COMET" / Domain specific COMpiler for Extreme Targets

The COMET compiler consists of a Domain Specific Language (DSL) for sparse and dense tensor algebra computations, a progressive lowering process to map high-level operations to low-level architectural resources, a series of optimizations performed in the lowering process, and various IR dialects to represent key concepts, operations, and types at each level of the multi-level IR. At each level of the IR stack, COMET performs different optimizations and code transformations. Domain-specific, hardware- agnostic optimizations that rely on high-level semantic information are applied at high-level IRs. These include reformulation of high-level operations in a form that is amenable for execution on heterogeneous devices (e.g., rewriting Tensor contraction operations as Transpose-Transpose-GEMM-Transpose) and automatic parallelization of high-level primitives (e.g., tiling for thread- and task-level parallelism).

More information about the COMET compiler can be found at:

1. COMET's source code - https://github.com/pnnl/COMET
2. COMET’s documentation - https://pnnl-comet.readthedocs.io/en/latest/

Installation and Testing:

1. Getting COMET pyhton package:

   • Install the COMET python package using pip (or) pip3 python package managers as:
   • For macOS users: python3 -m pip install cometpy
   • For Linux users: python3 -m pip install cometpy-lnx

2. Using COMET compiler as a backend to lower and execute test methods:

   • Import the Comet Python package in your python script using - "import cometpy as comet"

   • To compile a test method using COMET, add the decorator "@comet.compile(flags=...)" before the method.

   • The "numpy" keyword in a target method needs to be replace with "comet" e.g. "comet.einsum" , "comet.multiply"

   • Supported "numpy" methods:

     • .einsum()
     • .multiply()

   • For the actual "einsum" computation, follow the convention - "comet.einsum()"