RE: SH optimized software floating point routines

[Joern Rennecke <amylaar at spamcop dot net>]

Quoting "Naveen H. S" <Naveen.S@kpitcummins.com>:

> Hi,
>
> > > you can free to propose a complete and regtested patch for SH
> > > assembly soft fp against trunk.
>
> Please find attached the ported soft float patch "sh_softfloat.patch".
> The original patch was posted at the following link by Joern RENNECKE.
> http://gcc.gnu.org/ml/gcc-patches/2006-09/msg00614.html
>
> The following modifications have been done in the patch.
>
> sched-deps.c :- Hunk was not applied due to modifications in the
> current gcc source.
>
> t-sh :- divsf3, extendsfdf2 and truncdfsf2 routines are not included
> as they resulted in some regressions.
> divsf3 - c-c++-common/torture/complex-sign-mixed-div.c
> extendsfdf2 - gcc.c-torture/execute/conversion.c
> gcc.dg/torture/fp-int-convert-float.c, gcc.dg/pr28796-2.c
> gcc.dg/torture/type-generic-1.c

I've found some bugs in the SH[12] implementation of divsf3 / extendsf2.
What test case fails for truncsf2?

> sh.md :- cbranchsf4, cbranchdf4, cstoresf4, cstoredf4 instruction
> patterns are not included as they are already present in current source.
> Modifying the routines referring patch resulted in build failure.

Without optimized comparisons, conversions and division, the performance
will be heavily compromised.

Could you please test the attached patch.

Attachment: sh-softfp-20100717-1350
Description: Text document