[PATCH] D32324: CodeGen: BlockPlacement: Increase tail duplication size for O3.

[Kyle Butt via Phabricator via llvm-commits]

iteratee created this revision.
Herald added a subscriber: nemanjai.

At https://reviews.llvm.org/owners/package/3/ we are more willing to increase size if we believe it will improve
performance. The current threshold for tail-duplication of 2 instructions is
conservative, and can be relaxed at https://reviews.llvm.org/owners/package/3/.

Benchmark results:
llvm test-suite:
6% improvement in aha, due to duplication of loop latch
3% improvement in hexxagon for similar reasons.

2% slowdown in lpbench. Seems related, but couldn't completely diagnose.

Internal google benchmark:
Produces 4% improvement on internal google protocol buffer serialization
benchmarks.


Repository:
  rL LLVM

https://reviews.llvm.org/D32324

Files:
  lib/CodeGen/MachineBlockPlacement.cpp
  test/CodeGen/PowerPC/tail-dup-layout.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D32324.96051.patch
Type: text/x-patch
Size: 5436 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170420/a38c1a21/attachment.bin>