On Mon, 31 Mar 2025, Jakub Jelinek wrote:
> On Mon, Mar 31, 2025 at 11:12:45AM +0200, Richard Biener wrote:
> > The following disables tail recursion optimization when fixed-point
> > types are involved as we cannot generate -1 for all fixed-point
> > types.
> >
> > Bootstrapped and tested on x86_64-unknown-linux-gnu. OK?
> >
> > Thanks,
> > Richard.
> >
> > PR tree-optimization/119532
> > * tree-tailcall.cc (process_assignment): FAIL for fixed-point
> > typed functions.
> >
> > * gcc.dg/torture/pr119532.c: New testcase.
>
> So shouldn't we punt in that case just for the problematic operations
> (i.e. NEGATE_EXPR and MINUS_EXPR)?
> Aren't the others ok as is?
We are associating the expression which at least breaks with
saturating fixed-point. We are also rejecting FP types
without -fassociative-math, not trying to special cases some
maybe working cases.
So, I'm not sure the code behaves correctly even when not
running into the build_minus_one_cst case for _Fract types.
We'd at least need to disable TYPE_SATURATING.
I can change to just disable TYPE_SATURATING and
!ALL_SCALAR_ACCUM_MODE_P (TYPE_MODE (type)) (there's no TYPE-based
test for _Fract it seems?). Would you prefer that?
Association could still affect rounding, and I don't think we
document behavior of -fassociative-math on fixed-point?
Richard.
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.dg/torture/pr119532.c
> > @@ -0,0 +1,14 @@
> > +/* { dg-do compile } */
> > +/* { dg-require-effective-target fixed_point } */
> > +
> > +extern _Fract sinuhk_deg (unsigned short _Accum);
> > +
> > +_Fract cosuhk_deg (unsigned short _Accum deg)
> > +{
> > + unsigned short _Accum _90_deg = 90uhk;
> > + __asm ("" : "+r" (_90_deg));
> > +
> > + return deg <= _90_deg
> > + ? sinuhk_deg (_90_deg - deg)
> > + : -sinuhk_deg (deg - _90_deg);
> > +}
> > diff --git a/gcc/tree-tailcall.cc b/gcc/tree-tailcall.cc
> > index 8ba67522191..8ea1c8b5f99 100644
> > --- a/gcc/tree-tailcall.cc
> > +++ b/gcc/tree-tailcall.cc
> > @@ -361,6 +361,10 @@ process_assignment (gassign *stmt,
> > if (FLOAT_TYPE_P (TREE_TYPE (DECL_RESULT (current_function_decl))))
> > return FAIL;
> >
> > + /* We at least cannot build -1 for all fixed point types. */
> > + if (FIXED_POINT_TYPE_P (TREE_TYPE (DECL_RESULT (current_function_decl))))
> > + return FAIL;
> > +
> > if (rhs_class == GIMPLE_UNARY_RHS
> > && op0 == *ass_var)
> > ;
> > --
> > 2.43.0
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)
