[LLVMbugs] [Bug 19230] New: Honor __builtin_expect through inlining

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=19230

            Bug ID: 19230
           Summary: Honor __builtin_expect through inlining
           Product: libraries
           Version: trunk
          Hardware: All
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Transformation Utilities
          Assignee: unassignedbugs at nondot.org
          Reporter: wielkiegie at gmail.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Created attachment 12269
  --> http://llvm.org/bugs/attachment.cgi?id=12269&action=edit
Test case

A common usage model of __builtin_expect is to create helper macros called
'likely' and 'unlikely':

#if __has_builtin(__builtin_expect)
#   define likely(x)   __builtin_expect(!!(x), 1)
#   define unlikely(x) __builtin_expect(!!(x), 0)
#else
#   define likely(x)   (x)
#   define unlikely(x) (x)
#endif

However using macros (especially in C++) has a major drawback: short names
pollute namespace and long ones are harder to use.

I wanted to fix this by using inline functions (which can reside in a
namespace) instead of macros:

#if __has_builtin(__builtin_expect)
inline bool likely  (const bool x) { return __builtin_expect(x, 1); }
inline bool unlikely(const bool x) { return __builtin_expect(x, 0); }
#else
inline bool likely  (const bool x) { return x; }
inline bool unlikely(const bool x) { return x; }
#endif

These inline functions above work as expected in GCC but unfortunately not in
LLVM/Clang (tested both with -O3). Apparently, the branch weight metadata from
llvm.expect intrinsic are dropped before the inlining occurs. Doing the
inlining explicitly before any other optimizations fixes this problem.

$ clang++ llvm-expect.cpp -o - -emit-llvm -c -O3 -DUSE_MACROS | llvm-dis | grep
branch_weights
!1 = metadata !{metadata !"branch_weights", i32 64, i32 4}
$ clang++ llvm-expect.cpp -o - -emit-llvm -c -O3 | llvm-dis | grep
branch_weights
$ clang++ llvm-expect.cpp -o - -emit-llvm -c | opt -inline | opt -O3 | llvm-dis
| grep branch_weights
!1 = metadata !{metadata !"branch_weights", i32 64, i32 4}

-- 
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/20140323/ea774a82/attachment.html>