[LLVMdev] MIScheduler / bundling

[Sergei Larin]

Jonas,

 

 

Ø  As far as I can see, Hexagon does not bundle the VLIW-bundles by calling
bundleWithPred() on MIs of the completed cycle.

 

  You are right as of now. The reason is purely historical. At the point of
VLIW MI scheduler (aka Hexagom MI scheduler) creation bundleWithPreds was
not yet available and most passes after the MI sched could not handle
bundled code. In fact you can see that VLIW scheduler does form _temporary_
bundles, but then simply throws them away. This only preserves order of
instructions until the later bundle formation.

 

  It all will change. The goal is to have bundles formed at the MI sched
time and keeping them throughout the rest of compilation. Note that this
might involve occasional dissolution of a bundle, which might not always be
possible.

 

  If you want to use the process Hexagon uses now, bundle formation takes
place after the second pass scheduling, and if you have specific questions
about it, I can try to answer them.

 

Hope this helps.

 

Sergei Larin

 

---

Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by
The Linux Foundation

 

From: llvmdev-bounces at cs.uiuc.edu [mailto:llvmdev-bounces at cs.uiuc.edu] On
Behalf Of Jonas Paulsson
Sent: Wednesday, February 27, 2013 9:37 AM
To: llvmdev at cs.uiuc.edu
Subject: [LLVMdev] MIScheduler / bundling

 

Hi,

 

I am looking at the Hexagon MI Scheduling and trying to adapt it to my
target. 

 

As far as I can see, Hexagon does not bundle the VLIW-bundles by calling
bundleWithPred() on MIs of the completed cycle.

 

First of all, why is this not done? SlotIndexes seems to have at least some
support for this, by calling getBundleStart() for each MI that is looked up.

 

A follow up question is then, how would I do this, so that the register
allocator can work with bundle intervals?

 

Just calling bundleWithPred() gives

*** Bad machine code: Instruction inside bundle has a slot index ***

 

I would really appreciate if anyone let me know the plan on this,

 

Jonas Paulsson

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130228/c266cffe/attachment.html>