Re: [PATCH] Do not refer directly to gen_lowpart_xxx

["Paolo Bonzini" <paolo dot bonzini at polimi dot it>]

> > bonzini@sola3:~/gcc/obj/gcc$ gdb cc1
> > (gdb) b gen_lowpart
> > "gen_lowpart" is not a function
> > (gdb) b gen_lowpart <TAB>
> > gen_lowpart              gen_lowpart_common       gen_lowpart_general
> > gen_lowpart_SUBREG       gen_lowpart_for_combine
> > gen_lowpart_if_possible
>
> The first symbol displayed is the gen_lowpart function in emit-rtl.c .

Nope.  It is a *pointer to function*.  That's the reason for the error in "b
gen_lowpart".  I renamed the function in emit-rtl.c to gen_lowpart_general,
and gen_lowpart is initialized to gen_lowpart_general.

> The existance of gen_lowpart_for_combine function does not imply that
> all calls to gen_lowpart inside combine.c are calls to
> gen_lowpart_for_combine.

combine.c's only entry point is combine_instructions, which sets gen_lowpart
so that it points gen_lowpart_for_combine.

> It looks more like a coding error, or that you
> can now use both functions side-by-side.

After this patch, everything should go through the gen_lowpart pointer to
function; explicit usage of gen_lowpart_general, gen_lowpart_for_combine or
gen_lowpart_if_possible should be viewed with suspect.  Previously,
combine.c had a cute macro like

    #define gen_lowpart dont_use_gen_lowpart_in_combine_you_dummy

but now it is not needed anymore, just use gen_lowpart.  This is necessary
to move many cunning simplifications in combine.c (that in current HEAD
gen_lowpart_for_combine) to simplify-rtx.c (where they can use the
gen_lowpart pointer to function after this patch).

Paolo