Compilers, Languages, and Heterogeneous Computing
[Closed]
Work description
- Define a structured data format (e.g., JSON or YML) to specify the structure of an accelerator (i.e., coprocessor) in terms of resources (i.e., units such as adder, multiplier, etc.), and their physical layout. - Define correspondences between nodes of ONNX operations (e.g., dotprod, add, mul) and primitives (i.e., units) of the accelerator. - Use what was defined in the previous point to implement a mapping step capable of converting input ONNX graphs, expressed in JSON textual format, into graphs containing only nodes representing the accelerator's primitives. This component will contribute to a future complete ONNX compilation process for co-processors.
Academic Qualifications
Enrolment in Licenciatura or MSc in in Informatics Engineering, or similar.
Minimum profile required
- programming experience in C/C++ language
Preference factors
- Javascript experience - experience in manipulating data structures in graph format
Application Period
Since 21 Dec 2023 to 05 Jan 2024
[Closed]
Centre
Telecommunications and Multimedia