[LLVMbugs] [Bug 3754] New: FunctionAttrs pass marks function with MallocInst as readnone

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=3754

           Summary: FunctionAttrs pass marks function with MallocInst as
                    readnone
           Product: libraries
           Version: trunk
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Interprocedural Optimizations
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: nicholas at mxc.ca
                CC: llvmbugs at cs.uiuc.edu


$ cat m3.ll
define i32 @foo_malloc(i32 %n) nounwind noinline {
entry:
        %0 = icmp eq i32 %n, 0          ; <i1> [#uses=1]
        %iftmp.0.0 = select i1 %0, i32 1, i32 %n                ; <i32>
[#uses=1]
        %1 = malloc i8, i32 %iftmp.0.0          ; <i8*> [#uses=1]
        %2 = ptrtoint i8* %1 to i32             ; <i32> [#uses=1]
        ret i32 %2
}
$ llvm-as < m3.ll | opt -functionattrs -stats -disable-output
===-------------------------------------------------------------------------===
                          ... Statistics Collected ...
===-------------------------------------------------------------------------===

1 functionattrs - Number of functions marked readnone


That's wrong as can be seen from this C program:

  #include <stdlib.h>
  int foo_malloc(unsigned n) __attribute__((noinline));
  int foo_malloc(unsigned n) {
    return malloc(n ? n : 1);
  }
  int main(void) {
    void *x, *y;
    x = foo_malloc(10);
    y = foo_malloc(10);
    return x == y;
  }

The malloc instruction does write to memory, but it's sorta a secret special
"new memory" pointer that no other pointer in LLVM can see. :) I'm filing this
bug instead of just fixing it because I'm not sure what the fix out to be.

Make FunctionAttrs treat MallocInst specially? Set
MallocInst->mayWriteToMemory() to true?


-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.