[llvm-dev] [cfe-dev] RFC: First-class Matrix type (original) (raw)

JF Bastien via llvm-dev llvm-dev at lists.llvm.org
Fri Oct 12 10:51:10 PDT 2018

On Oct 12, 2018, at 10:36 AM, David Greene via cfe-dev <cfe-dev at lists.llvm.org> wrote:

Adam Nemet via llvm-dev <llvm-dev at lists.llvm.org> writes:

Is this something that we could get if we had better loop transformations (i.e., loop fusion)?

As we both know, aliasing usually makes this pretty difficult. For relatively small matrices run-time memchecks can be costly. Also we need to be able to evaluate these fusion opportunities in the inliner so having first-class type representation makes this cheap. Fortran defines away many of these problems. Would better local restrict support help with C-family languages? There are several old patches from Hal that need some attention.

Agreed these patches would be neat to revive. I think we’d also want someone to pursue wg21.link/n4150 <http://wg21.link/n4150>. However, none of that seems like it should gate matrix support in the IR. -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20181012/b0d167a8/attachment-0001.html>

More information about the llvm-dev mailing list