Bend: a high-level language that runs on GPUs (via HVM2) (github.com)

Article summary

Bend is a high-level, massively parallel programming language that runs on GPUs via the HVM2 runtime, offering features like fast object allocations and support for higher-order functions. It is designed to scale performance with cores and can execute code on multiple threads without explicit parallelism annotations. The language is still in development, with current limitations including lower single-core performance and limited support for operating systems. Bend's performance can be substantial, with examples showing significant speedups compared to other languages.

Main themes

• Parallel programming
• GPU acceleration
• High-level languages
• HVM2 runtime
• Performance optimization

What commenters say

• The ability to parallelize code without explicit annotations is a significant advantage of Bend, making it easier to write efficient programs.
• Some commenters are skeptical about the performance claims, noting that the GPU version may not always be significantly faster than other languages.
• Bend's potential applications in areas like compilers and evolutionary computation are exciting and worth exploring further.
• The lack of a clear license for HVM2 and Bend is a concern, and clarity on this issue is needed to ensure the project's long-term viability.
• The relationship between Bend and Kind2, a proposed type layer, is seen as a key aspect of the project's future development and potential impact.
• Some commenters are interested in using Bend as a library or crate, rather than an external program, to integrate its functionality into their existing workflows.