[PATCH] Add Load Combine Pass

[Chandler Carruth]

On Thu, May 22, 2014 at 6:14 PM, Michael Spencer <bigcheesegs at gmail.com>wrote:

> I ran test-suite with this pass and got some interesting results. Most
> tests have no changes, but 4 tests showed significant change.
>
> + is faster, - is slower.
>
> | nts.MultiSource/Benchmarks/MiBench/security-sha/security-sha | 44% |
> | nts.MultiSource/Benchmarks/Prolangs-C/agrep/agrep | 12% |
> | nts.MultiSource/Benchmarks/SciMark2-C/scimark2 | -21% |
> | nts.MultiSource/Benchmarks/tramp3d-v4/tramp3d-v4 | -39% | (this drops to
> -20% if we fix the load splicing cost model in DAGCombine)
>
> Full data:
> https://docs.google.com/spreadsheets/d/13_4ZUBQQYXhexrMzVJ5VICNuLzUhovU8lOl0Rr6wG10/edit?usp=sharing
>
> I'd like to commit this disabled by default while the regressions are
> fixed in tree.
>

For full context, Michael chatted with me on IRC about some aspects of the
regressions, and they sound likely to be surprising interactions between
this pass and GVN, the inliner, or SROA. Due to the nature of
pass-interaction bugs, it seems likely that the fixes won't change this
pass substantially, and may require reproducing these issues for other devs
to get help. That's why I suggested getting this into the tree, but
disabling it by default, and then iterating on the remaining regressions
in-tree.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140522/f9eea9b9/attachment.html>