[llvm-dev] RFC: An Extension Mechanism for Parallel Compilers Based on LLVM
Adve, Vikram Sadanand via llvm-dev
llvm-dev at lists.llvm.org
Mon Oct 22 21:35:14 PDT 2018
This is not focused on OpenMP-type systems – we aim to support a wide range of languages. Tapir supports Cilk, for example. HPVM is flexible enough to support both general-purpose languages like OpenMP, OpenCL and CUDA, as well as domain-specific ones like TensorFlow.
Halide and Tiramisu are more domain-specific, and could in principle be hosted on top of this mechanism (although there may not be any reason to).
The point about “vastly more targets” seems like a misunderstanding: the RFC puts no restrictions on how many targets any parallel compiler hosted on top of the mechanism could target (e.g., HPVM can target as many as Halide or Tiramisu).
Date: Tue, 16 Oct 2018 10:47:33 -0600
From: chris nuernberger via llvm-dev <llvm-dev at lists.llvm.org>
To: llvm-dev at lists.llvm.org
Subject: Re: [llvm-dev] RFC: An Extension Mechanism for Parallel
Compilers Based on LLVM
This appears this is an incremental extension mechanism for openMP-type
systems.
I was wondering if the authors are familiar with halide (halide-lang.org)
or the recent Tiramisu paper (https://arxiv.org/pdf/1804.10694.pdf).
These compilers can target a vastly wider range of hardware architectures
one of which could be openMP and get amazing performance results.
While I think it is most likely out of scope for current work it may be
helpful to keep these more specialized compilers in mind.
Chris
--Vikram Adve
// Interim Head, Department of Computer Science
// Donald B. Gillies Professor of Computer Science
// University of Illinois at Urbana-Champaign
// Admin Assistant: Amy Simons - aoboyle at illinois.edu
// Google Hangouts: vikram.s.adve at gmail.com || Skype: vikramsadve
// Research page: http://vikram.cs.illinois.edu <http://vikram.cs.illinois.edu/>
// 10 Commandments for Email Survival: http://www.emailcharter.org/
More information about the llvm-dev
mailing list