[llvm-bugs] [Bug 32070] New: Special-case memcpy() after conditional block

[via llvm-bugs]

http://bugs.llvm.org/show_bug.cgi?id=32070

            Bug ID: 32070
           Summary: Special-case memcpy() after conditional block
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: simonhosie77+llvm at gmail.com
                CC: llvm-bugs at lists.llvm.org

Given:

    void f(void *d, void const* s, size_t l, int c) {
      if (__builtin_expect(c != 0, 1)) {
        l = 0;
      }
      __builtin_memcpy(d, s, l);
    }

Compiling this I get a conditional set-register-to-zero and unconditional call
to memcpy().

Ideally we'd just skip over the call when c != 0.  Changing the code to this
makes everything better, but it's not automatic, and in realistic code it's not
something I'd want to do manually:

    void f(void *d, void const* s, size_t l, int c) {
      if (__builtin_expect(c != 0, 1)) {
        l = 0;
        __builtin_memcpy(d, s, l);
      } else {
        __builtin_memcpy(d, s, l);
      }
    }

Generally, finishing a conditional block with additional optimisation
information might produce a candidate for special-case implementation of the
common code that follows it.  For __builtin_memcpy() this would include l being
set to any small constant.

-- 
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/20170225/06f57620/attachment.html>