[PATCH] D93360: [LibCallSimplifier] fold memset(malloc(x), 0, x) to calloc(1, x) if memset is intrinsic

[Sanjay Patel via Phabricator via llvm-commits]

spatel added reviewers: xbolva00, lebedev.ri.
spatel added a comment.

Please review the comments in D16337 <https://reviews.llvm.org/D16337>  - especially:
https://reviews.llvm.org/D16337#331117

How does this patch guarantee that memory is not touched between the malloc and the memset? Is the transform safe on this example?

  define float* @store_between_malloc_and_memset(i64 %size) {
  entry:
    %call = tail call i8* @malloc(i64 %size) #1
    %bc1 = bitcast i8* %call to i32*
    store i32 1, i32* %bc1, align 4
    %cmp = icmp eq i8* %call, null
    br i1 %cmp, label %cleanup, label %if.end
  
  if.end:
    %bc = bitcast i8* %call to float*
    %call3 = tail call i8* @memset(i8* %call, i32 0, i64 %size)
    br label %cleanup
  
  cleanup:
    %r = phi float* [ %bc, %if.end ], [ null, %entry ]
    ret float* %r
  }


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D93360/new/

https://reviews.llvm.org/D93360