[Bug middle-end/68360] GCC bitfield processing code is very inefficient

[ebotcazou at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68360

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ebotcazou at gcc dot gnu.org

--- Comment #4 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
> Here's another example where GCC generates suboptimal code for bitfield
> initialization on x64: https://godbolt.org/g/WeqAr5.

Again the code is better for example on SPARC:

        sll     %o1, 30, %o1
        sethi   %hi(-1073741824), %g1
        andn    %o2, %g1, %o2
        or      %o1, %o2, %o1
        jmp     %o7+8
         st     %o1, [%o0]

It is very bad on x86 because, when get_best_mode is invoked as such:

get_best_mode (bitsize=2, bitpos=0, bitregion_start=0, bitregion_end=31, 
    align=32, largest_mode=SImode, volatilep=false)
    at /home/eric/svn/gcc/gcc/stor-layout.c:2800

it returns QImode.  If it returned SImode as on SPARC, the code would be:

        sall    $2, %edx
        andl    $3, %esi
        orl     %edx, %esi
        movl    %esi, (%rdi)
        ret