[PATCH] D47634: [MemCpyOpt] Update pass to also optimize the atomic memory intrinsics (WIP - NOT FOR REVIEW)

[Daniel Neilson via Phabricator via llvm-commits]

dneilson created this revision.
Herald added a subscriber: llvm-commits.

This change updates parts of the memcpyopt pass to also perform optimizations on the atomic
memcpy & memset intrinsics.

Specifically, we update:
 processMemMove - This will change memmoves into memcpys if aliasing allows.
 processMemCpy - Basically exposes opportunities for DSE by changing the source of a memcpy

  if the source was just memcpy/memset into.

Parts of the pass that still need to be updated:
 processMemSet - Merging consecutive memsets/stores into a single memset
 processStore - Merging consecutive load/store pairs into memcpys.

Some care will have to be taken with processMemSet and processStore to ensure
that we don't break the expectations of unordered atomic operations (ex: no merging
an unordered store into a regular memset, etc).


Repository:
  rL LLVM

https://reviews.llvm.org/D47634

Files:
  include/llvm/Transforms/Scalar/MemCpyOptimizer.h
  lib/Transforms/Scalar/MemCpyOptimizer.cpp
  test/Transforms/MemCpyOpt/atomic-memcpy.ll
  test/Transforms/MemCpyOpt/memcpy-to-memset.ll
  test/Transforms/MemCpyOpt/memmove.ll
  test/Transforms/MemCpyOpt/memset-atomicmemcpy-to-2x-memset.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D47634.149459.patch
Type: text/x-patch
Size: 28550 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180601/57d56578/attachment.bin>