[PATCH] D18237: [SLPVectorizer] Change MinVecRegSize from 128 bits to 64 bits

Wed Mar 16 22:06:09 PDT 2016

JongwonLee created this revision.
JongwonLee added subscribers: llvm-commits, flyingforyou.
Herald added subscribers: mzolotukhin, aemerson.

If the target is AArch64, MinVecRegSize can be 64 (vectorized in double register) not 128 (vectorized in quad register). Thus, we can have more opportunities to vectorize as followings <2 x 32-bit data type>, <4 x 16-bit data type>, <8 x 8-bit data type>.
The exisitng option '-slp-min-reg-size' enables only one kind of MinVecRegSize. With this patch, we can try to vectorize in the range from MaxVecRegSize to MinVecRegSize.

http://reviews.llvm.org/D18237

Files:
  lib/Transforms/Vectorize/SLPVectorizer.cpp
  test/Transforms/SLPVectorizer/AArch64/slp-vectorized-within-64bits.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D18237.50908.patch
Type: text/x-patch
Size: 9071 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160317/f4697417/attachment.bin>