[LLVMbugs] [Bug 19183] can't delete empty, terminating, strided loop

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=19183

Richard Smith <richard-llvm at metafoo.co.uk> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|FIXED                       |---

--- Comment #2 from Richard Smith <richard-llvm at metafoo.co.uk> ---
Thanks!

Sadly, this isn't enough to clean up the mess left behind by the vectorizer
(presumably because the vectorization happens too late). This testcase still
leaves behind a do-nothing loop:

#include <memory>
#include <cstring>

long test(int iters, int len, int sticky)
{
  const int block_size = 0x1000;
  const int blocks = len / block_size;
  std::unique_ptr<char[]> source(new char[blocks * block_size]);
  std::unique_ptr<char[]> destination(new char[blocks * block_size]);

  for (int i = 0; i < iters; ++i) {
    char *psrc = source.get();
    char *pdst = destination.get();
    for (int block = 0; block < blocks; ++block) {
      memcpy(psrc, pdst, block_size);
      psrc += block_size;
      pdst += block_size;
      if (sticky) {
        psrc -= block_size;
        pdst -= block_size;
      }
    }
  }

  return long(iters) * blocks * block_size;
}

Running opt a second time on the output of clang -O3 -emit-llvm does remove the
do-nothing loop now, though.

(We should really be able to delete the entire function body, but that's a
separate issue...)

-- 
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/20140325/5d345c94/attachment.html>