This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Testsuite tweaks (SPARC)


"David S. Miller" <davem@redhat.com> writes:
> Is the compiler expected to handle vector modes even on platforms that
> don't have something like the altivec vector support the rs6000 backend
> has?
>
> If so, we should fix Sparc.
>
> If not, we should specifically mention the platforms that support
> vector stuff in the testcase since that group is smaller than
> those that don't support vector stuff.

There's some sort of generic vector support but I don't think it works
very well.

This is how I think it _should_ work:  There are no vector modes
defined in generic code.  Targets define exactly the set of vector
modes that the hardware supports.  The user can specify whatever
vector _type_ they want.  If it doesn't match the hardware it gets
BLKmode, and an early lowering pass breaks up the operations into
vector modes that are supported (if possible) or scalar ops (if not).

Needs work all over the compiler to make it happen though.

zw


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]