[pl-seminar] LLVM, HPVM, & MLIR

["Mark D. Hill" <markhill@xxxxxxxxxxx>]

Dear Mad-PLers:

As you may know, many believe that many future computer systems will exploit heterogeneity with CPUs augmented with GPUs and other specialized accelerators. E.g., see my blog post: https://www.sigarch.org/accelerator-level-parallelism/. 

Whither the programming and compiling infrastructure for this? To this end LLVM alumni Vikram Adve and Chris Lattner are separately pushing 

* HPVM https://publish.illinois.edu/hpvm-project/  and

* MLIR:  https://arxiv.org/pdf/2002.11054.pdf

Any opinions of these? Are they worthy of discussion some time?

âMark

Mark D. Hill, http://www.cs.wisc.edu/~markhill
John P. Morgridge Prof. and Gene M. Amdahl Prof. of CS 
Computer Sciences Department, University of Wisconsin-Madison
Chair, Computing Community Consortium, http://cra.org/ccc/