This is an archive of the discontinued LLVM Phabricator instance.

[AArch64][ARM] Armv8.2-A: Add the Floating Point Multiplication Variants
AbandonedPublic

Authored by SjoerdMeijer on Jul 26 2018, 1:48 AM.

Diff Detail

Event Timeline

SjoerdMeijer created this revision.Jul 26 2018, 1:48 AM
bogden requested changes to this revision.Jul 26 2018, 2:52 AM

Until we define a separate feature for these instructions, they should be predicated on HasV8_4a rather than HasV8_2A. Until v8.4-A they're optional even for CPUs that have HasFullFP16. Current v8.2-A Cortex CPUs, for example, do not support these instructions.

With that change it's probably fine as a part of the v8.4-A support, but I haven't reviewed the rest of it yet.

This revision now requires changes to proceed.Jul 26 2018, 2:52 AM

Ah, good point, thanks. But I will create a new target feature for this like we normally do (such as e.g. dotprod).

SjoerdMeijer abandoned this revision.Aug 6 2018, 3:22 AM

This work is continued in D50228.