[llvm-commits] AMDIL scheduling override patch

[Villmow, Micah]

From: llvm-commits-bounces at cs.uiuc.edu [mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Andrew Trick
Sent: Friday, September 02, 2011 4:31 PM
To: Villmow, Micah
Cc: llvm-commits
Subject: Re: [llvm-commits] AMDIL scheduling override patch

On Sep 2, 2011, at 9:46 AM, Villmow, Micah wrote:


Ping...

From: llvm-commits-bounces at cs.uiuc.edu<mailto:llvm-commits-bounces at cs.uiuc.edu> [mailto:llvm-commits-bounces at cs.uiuc.edu] On Behalf Of Villmow, Micah
Sent: Friday, August 26, 2011 6:23 PM
To: llvm-commits
Subject: [llvm-commits] AMDIL scheduling override patch

There are cases where LLVM scheduling makes assumptions that perform horrendously bad on certain types of instructions.
This patch adds a hook into the Scheduling algorithm that lets the backend override any checks that might occur when determining the node priority.

This is one of the few hooks that the AMDIL backend relies on that is currently not in the public LLVM tree.
_______________________________________________
llvm-commits mailing list
llvm-commits at cs.uiuc.edu<mailto:llvm-commits at cs.uiuc.edu>
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

Hi Micah,

The scheduler design goes to some trouble to make getPriority reasonably efficient. Introducing a virtual call in all cases (not guarded by any flags) that won't be used by most targets seems unfortunate. I think we can spare a bit in SUnit though, since the concept at least makes sense. Then you can move your target hook to BuildSchedUnits, set the flag there, and do the flag check in getNodePriority.

Thanks,
-Andy
[Villmow, Micah] Thanks, let me see if doing it there is possible.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20110902/e067d3e5/attachment.html>