[Bug middle-end/51837] New: Use of result from 64*64->128 bit multiply via __uint128_t not optimized

[svfuerst at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51837

             Bug #: 51837
           Summary: Use of result from 64*64->128 bit multiply via
                    __uint128_t not optimized
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: svfuerst@gmail.com


unsigned long long foo(unsigned long long x, unsigned long long y)
{
    __uint128_t z = (__uint128_t)x * y;
    return z ^ (z >> 64);
}

Compiles into

mov    %rsi, %rax
mul    %rdi
mov    %rax, %r9
mov    %rdx, %rax
xor    %r9, %rax
retq

The final two mov instructions are not needed, and the above is equivalent to

mov    %rsi, %rax
mul    %rdi
xor    %rdx, %rax
retq