[LLVMdev] [Polly] Move Polly's execution later

[Sebastian Pop]

Star Tan wrote:
> I have not yet found out why Polly produces such wrong code.

IMO this is because Polly does not recognize a given number of patterns that
occur after the scalar opts have transformed the code.  There are a given number
of assumptions that we took for granted and these do not hold down the pipeline.

> However, maybe I should also investigate other points where we can move
> Polly. Do you have any suggestion?

Yep, try to move Polly just after function inlining, that's better than the
current place, and will extend the loop coverage in many cases.

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