[all-commits] [llvm/llvm-project] 186033: [MemCpyOpt] Correctly merge alias scopes during ca...

[modiking via All-commits]

  Branch: refs/heads/master
  Home:   https://github.com/llvm/llvm-project
  Commit: 18603319321a6c1b158800bcc60035ee01549516
      https://github.com/llvm/llvm-project/commit/18603319321a6c1b158800bcc60035ee01549516
  Author: modimo <modimo at fb.com>
  Date:   2020-12-03 (Thu, 03 Dec 2020)

  Changed paths:
    M llvm/include/llvm/Analysis/ScopedNoAliasAA.h
    M llvm/lib/Analysis/ScopedNoAliasAA.cpp
    M llvm/lib/IR/Metadata.cpp
    A llvm/test/Analysis/ScopedNoAliasAA/alias-scope-merging.ll
    M llvm/test/Transforms/GVN/noalias.ll
    M llvm/test/Transforms/InstCombine/fold-phi-load-metadata.ll
    A llvm/test/Transforms/MemCpyOpt/callslot_badaa.ll
    M llvm/test/Transforms/NewGVN/noalias.ll

  Log Message:
  -----------
  [MemCpyOpt] Correctly merge alias scopes during call slot optimization

When MemCpyOpt performs call slot optimization it will concatenate the `alias.scope` metadata between the function call and the memcpy. However, scoped AA relies on the domains in metadata to be maintained in a caller-callee relationship. Naive concatenation breaks this assumption leading to bad AA results.

The fix is to take the intersection of domains then union the scopes within those domains.

The original bug came from a case of rust bad codegen which uses this bad aliasing to perform additional memcpy optimizations. As show in the added test case `%src` got forwarded past its lifetime leading to a dereference of garbage data.

Testing
ninja check-llvm

Reviewed By: jeroen.dobbelaere

Differential Revision: https://reviews.llvm.org/D91576