[llvm-bugs] [Bug 33578] New: Optimization introduces error when right shifting by integer width.

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=33578

            Bug ID: 33578
           Summary: Optimization introduces error when right shifting by
                    integer width.
           Product: clang
           Version: 3.9
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: C++11
          Assignee: unassignedclangbugs at nondot.org
          Reporter: d.horne at danielhorne.co.uk
                CC: dgregor at apple.com, llvm-bugs at lists.llvm.org

I have the following inline, to simulate a rotl instruction:

static uint64_t rotateLeft(uint64_t x, int n)
{
    return ((x << n) | (x >> (64 - n)));
}

It produces the expected results on MSVC, GCC, and clang++ with optimisation
switched off, but somehow produces 4294967295 in the n=0 case with -O3
optimisations. The results for n>0 look ok.

I then made the following tests:

        printf(" 6 << 0 = %u \n", 6<<0);
        printf(" 6 >> (64-0) = %u \n", 6>>(64-0));
        printf(" 6 >> 64 = %u \n", 6>>64);
        printf("((6 << 0) | (6 >> (64 - 0))) = %u \n", ((6 << 0) | (6 >> (64 -
0))));
        printf(" rotateLeft(6,0) = %u \n", rotateLeft(6,0));

which produced:

 6 << 0 = 6 
 6 >> (64-0) = 24579104 
 6 >> 64 = 24579104 
((6 << 0) | (6 >> (64 - 0))) = 4294967295 

indicating the problem shows up when right shifting by 64..

According to the c++11 spec, right shifts should produce "the integral part of
the quotient of E1 / 2^E2", which in this case is zero, leading me to consider
this a compliance bug.

-- 
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/20170623/7ca8cc07/attachment.html>