[llvm-bugs] [Bug 28498] New: Passing 0 to __atomic_fetch_add generates mfence instead of lock xadd

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=28498

            Bug ID: 28498
           Summary: Passing 0 to __atomic_fetch_add generates mfence
                    instead of lock xadd
           Product: clang
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: LLVM Codegen
          Assignee: unassignedclangbugs at nondot.org
          Reporter: pierregousseau14 at gmail.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Using clang 3.9 on trunk at r273567
Target: x86_64-unknown-linux-gnu

When passing a '0' constant to '__atomic_fetch_add', an mfence instruction is
generated instead of the usual 'lock xadd' which seems less efficient.

---
#include <stdint.h>

int64_t * g_ptr;

int foo() {
  return __atomic_fetch_add(g_ptr, 0, __ATOMIC_RELAXED);
}
---

clang -O3 -S test.c

Generates:

movq    %rdi, -8(%rsp)
movq    g_ptr(%rip), %rax
mfence
movq    (%rax), %rax
retq

Using gcc:

movq    g_ptr(%rip), %rdx
movq    %rdi, -8(%rsp)
xorl    %eax, %eax
lock xaddq      %rax, (%rdx)
ret

-- 
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/20160711/c9a20b0e/attachment.html>