[Bug c/66552] Missed optimization when shift amount is result of signed modulus

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66552

Martin Sebor <msebor at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization
             Target|x86/generic                 |x86, powerpc64

--- Comment #1 from Martin Sebor <msebor at gcc dot gnu.org> ---
On powerpc64, with -O2, GCC emits the following 

f:
        srawi 9,4,5
        addze 9,9
        slwi 9,9,5
        subf 4,9,4
        srw 3,3,4
        rldicl 3,3,0,32
        blr

while Clang emits what looks like optimal code:

f:
        rlwinm 4, 4, 0, 27, 31
        srw 3, 3, 4
        blr

With the return expression rewritten as suggested (i.e., x >> (n & 31)), GCC
still emits an extra instruction compare to Clang.  I believe this aspect is
tracked in bug 66217.

        rldicl 4,4,0,59
        srw 3,3,4
        rldicl 3,3,0,32
        blr