[PATCH][AARCH64] Fixed fused multiply add/sub patterns

[Ana Pazos]

I slip the patch in 2 and checked them in.

We need both, cortex-a53 has one MAC unit.

 

Thanks,

Ana.

 

From: Jiangning Liu [mailto:liujiangning1 at gmail.com] 
Sent: Sunday, December 22, 2013 11:15 PM
To: Ana Pazos
Cc: llvm-commits at cs.uiuc.edu for LLVM; Tim Northover
Subject: Re: [PATCH][AARCH64] Fixed fused multiply add/sub patterns

 

Hi Ana,

 

The fix for fnmadd/fnmsub is OK for me.

 

But for single_use change, I'm not sure if the following claim is correct,
and it depends on hardware implementation, right?

AArch64 targets most probably have one MAC pipe and this will be a
performance issue as well.

So I think at least it would be better to get them decoupled and upstreamed
with two separate patches. If we find any mistake in future, it would be
easy for us to roll back as well. Also this optimization could probably
depends on different aarch64 target.

 

Thanks,

-Jiangning

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131223/d4ab0bea/attachment.html>