[llvm-bugs] [Bug 48599] New: -memcpyopt moves alloca struct load past stackrestore

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=48599

            Bug ID: 48599
           Summary: -memcpyopt moves alloca struct load past stackrestore
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: default_357-feep at yahoo.de
                CC: llvm-bugs at lists.llvm.org

See https://godbolt.org/z/38nqh6

Consider the following IR:

declare i8* @llvm.stacksave()
declare void @llvm.stackrestore(i8*)
declare i8* @xmalloc(i64)
define i8* @bug() {
  %stack = tail call i8* @llvm.stacksave()
  %alloc = alloca { i64 }, align 8

  %v0 = insertvalue { i64 } undef, i64 1, 0
  store { i64 } %v0, { i64 }* %alloc, align 8

  ; load before restoring stack
  %loaded = load { i64 }, { i64 }* %alloc, align 8

  tail call void @llvm.stackrestore(i8* %stack)

  %mem = tail call i8* @xmalloc(i64 8)
  %mem.i64x1 = bitcast i8* %mem to { i64 }*

  ; loading %alloc here would no longer be valid!
  store { i64 } %loaded, { i64 }* %mem.i64x1, align 8
  ret i8* %mem
}

When passed to `opt -memcpyopt` on trunk of Dec-25-2020, this "optimizes" the
load followed by store into a memcpy. However, the memcpy is placed after the
stackrestore, leading to it reading stack memory that is no longer valid.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20201225/7b142cd4/attachment.html>