[llvm-bugs] [Bug 35399] New: lzcnt causes erroneous code generation (x86_64)

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=35399

            Bug ID: 35399
           Summary: lzcnt causes erroneous code generation (x86_64)
           Product: clang
           Version: 5.0
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: normal
          Priority: P
         Component: -New Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: michel.bernert at gmail.com
                CC: llvm-bugs at lists.llvm.org

Created attachment 19461
  --> https://bugs.llvm.org/attachment.cgi?id=19461&action=edit
Code compiled with : clang++ lzcnt.cpp -Wall -pedantic -O2 -mlzcnt -S

The use of lzcnt has an unexpected behavior, when using the -mlzcnt flag with
optimizations.

Compiling :

uint64_t test64_2(uint64_t x, bool& b1, bool& b2)
{
    uint64_t a = (x == 0 ? 64 : __builtin_clzll(x));
    b1 = (a != 64);
    b2 = !b1;
    return a;
}

Generates :

test64_2(unsigned long, bool&, bool&): # @test64_2(unsigned long, bool&, bool&)
  lzcntq %rdi, %rax
  setae (%rsi)
  setae (%rdx)
  retq

Which sets the two bools to the same value, even though 'b2 = !b1;' and I can't
see any undefined behavior.
This concerns all 16, 32 and 64 bits versions of lzcnt. Omitting -mlzcnt
produces the expected behavior for __builtin_clz versions, using bsr.

-- 
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/20171123/a75b01c4/attachment-0001.html>