On Wed, Apr 29, 2015 at 11:50 AM, Eric Botcazou <ebotca...@adacore.com> wrote:
> Historically the pragma Inline_Always of GNAT had been implemented in the FE
> because the RTL inliner and then the Tree inliner weren't invoked at -O0 or
> powerful enough to inline some constructs. But this approach had drawbacks,
> especially wrt debug info. These restrictions were gradually lifted and now
> the pragma is entirely piggybacked on the Tree inliner.
>
> This went mostly OK, except for a few cases where intrisinc operations that
> used to be reasonably handled at -O0 now generate awful code, the typical
> example being a modulus or division instrinsic by a power-of-2 generating a
> fully-fledged modulus or division instruction instead of a simple shift.
>
> Therefore the attached patch implements anonymous constant propagation in the
> inliner to fix the code quality regression.
>
> Tested on x86_64-suse-linux, OK for the mainline?
Hmm, special-casing this in the inliner looks odd to me. ISTR the inliner
already propagates constant parameters to immediate uses, so I guess
you run into the casting case you handle specially.
But then if any constant propagation is ok at -O0 why not perform full-fledged
constant propagation (with !DECL_IGNORED_P vars as a propagation barrier)
as a regular SSA pass? That is, if you'd had a Inline_Always function like
int foo (int i)
{
return (i + 1) / i;
}
you'd not get the desired result as the i + 1 stmt wouldn't be folded
and thus the (i + 1) / i stmt wouldn't either.
That said - why does RTL optimization not save you here anyway?
After all, it is responsible for turning divisions into shifts.
Soo - if you use the existing CCP pass and make surely_varying_stmt_p
return true for defs of !DECL_IGNORED_P stmts at -O0 (strictly needed to make
setting vars in gdb possible) and somehow arrange for ccp to run at -O0...
It might help other obscure cases with inline-asms to behave consistently
with -O0 vs. -On as well.
Richard.
> 2015-04-29 Eric Botcazou <ebotca...@adacore.com>
>
> * tree-inline.c (remap_gimple_op_r): Do anonymous constant
> propagation.
> (copy_bb): Fold conversions of constants immediately.
>
>
> 2015-04-29 Eric Botcazou <ebotca...@adacore.com>
>
> * gnat.dg/inline12.adb: New test.
>
>
> --
> Eric Botcazou
