[PATCH] D20873: [TTI] Refine default cost for interleaved load groups with gaps

[Matthew Simpson via llvm-commits]

mssimpso updated this revision to Diff 60049.
mssimpso added a comment.

Simplified ceiling computation.

Hi Silviu,

Other than a 9% improvement in spec2000/art, I didn't observe any non-noise performance differences in the test suite, spec2000, or spec2006. Some loops in spec2000/art access only a few elements of a large struct. This patch enables those loops to be vectorized, even though we don't map the accesses to ldN instructions.


http://reviews.llvm.org/D20873

Files:
  include/llvm/CodeGen/BasicTTIImpl.h
  test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D20873.60049.patch
Type: text/x-patch
Size: 5044 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160608/27339109/attachment.bin>