[LLVMbugs] [Bug 4076] New: bug in inline asm constraint

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=4076

           Summary: bug in inline asm constraint
           Product: libraries
           Version: trunk
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Backend: X86
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: nicholas at mxc.ca
                CC: llvmbugs at cs.uiuc.edu, nlewycky at google.com


Created an attachment (id=2886)
 --> (http://llvm.org/bugs/attachment.cgi?id=2886)
.ll testcase

The attached .ll file demonstrates a problem where LLVM is picking %eax to
serve in a ptr computation even though it's already reserved as another input
to the inline asm.

$ llvm-as < b.ll | llc -disable-fp-elim
[...]
.LBB1_16:       # 
        movl    %esi, %edx
        xorl    %eax, %eax
        movl    20(%ebp), %eax
        #APP
        lock; cmpxchgl %edx,(%eax,%edi,8)
        #NO_APP

Note that cmpxchgl takes %eax as input and output, and it needs to be zero as
per the xorl instruction. The original inline asm statement in C looks like:

inline int32_t NoBarrier_CompareAndSwap(volatile int32_t* ptr,
                                        int32_t old_value,
                                        int32_t new_value) {
   int32_t prev; 
   __asm__ __volatile__("lock; cmpxchgl %1,%2"
                        : "=a" (prev)
                        : "q" (new_value), "m" (*ptr), "0" (old_value)
                        : "memory");
  return prev;
}

which compiles fine as a standalone function. The problem only surfaces with
inlining.


-- 
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.