[PATCH] D13779: [Polly] Full/partial tile separation for vectorization

[Roman via llvm-commits]

gareevroman created this revision.
gareevroman added reviewers: grosser, jdoerfert.
gareevroman added a subscriber: Polly.

We isolate full tiles from partial tiles to be able to, for example, vectorize loops with parametric lower and/or upper bounds.

If we use -polly-vectorizer=stripmine, we can see execution-time improvements: correlation from 1m7361s to 0m5720s (-67.05 %), covariance from 1m5561s to 0m5680s (-63.50 %), ary3 from 2m3201s to 1m2361s (-46.72 %), CrystalMk from 8m5565s to 7m4285s (-13.18 %). However, there is a compile-time regression, for example, for 3mm from 0m6320s to 0m9881s (56.34%), which should be eliminated in future.

http://reviews.llvm.org/D13779

Files:
  lib/Transform/ScheduleOptimizer.cpp
  test/ScheduleOptimizer/full_partial_tile_separation.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D13779.37503.patch
Type: text/x-patch
Size: 10457 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151015/2c06c6d4/attachment.bin>