Re: [PATCH][RFC] New function for checking misalignment

[Revital1 Eres <ERES at il dot ibm dot com>]

Hello,

> > This patch adds a new target hook for checking if unaligned access
> > with a certain misalignment factor is supported by the target.
> >
> > Tested on powerpc64 and x86.
> > OK for mainline? Comments are welcome.
>
> The bool parameter to the targethook needs documenting.  In fact
> it sounds redundant - the misalignment parameter should be enough
> here, no?  And yes, the hook needs documenting in doc/tm.texi.

I think that if the misalignment is unknown we need to have the packed
field in order to determine whether it can be supported.
I'll fix the rest of the comments and post a new version.

Thanks,
Revital


>
> Thanks,
> Richard.
>
> > Thanks to Michael for providing initial feedback for the patch.
> >
> > Thanks,
> > Revital
> >
> > ? ? ? ?* targhooks.c (default_builtin_support_vector_misalignment):
> > ? ? ? ?New builtin function.
> > ? ? ? ?* targhooks.h (default_builtin_support_vector_misalignment):
> > ? ? ? ?Declare.
> > ? ? ? ?* target.h (builtin_support_vector_misalignment):
> > ? ? ? ?New field in struct gcc_target.
> > ? ? ? ?* tree-vect-data-refs.c (vect_supportable_dr_alignment): Call
> > ? ? ? ?new builtin function.
> > ? ? ? ?* target-def.h (TARGET_SUPPORT_VECTOR_MISALIGNMENT):
> > ? ? ? ?Define.
> > ? ? ? ?* config/rs6000/rs6000.c
> > ? ? ? ?(rs6000_builtin_support_vector_misalignment): New function.
> > ? ? ? ?(TARGET_SUPPORT_VECTOR_MISALIGNMENT): Define.
> >
> >
> > (See attached file: misalign_2_9.txt)