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: conversions between vectors


On 22 Dec 2001, Aldy Hernandez wrote:

> > Does Dan's patch include docs?  Let's first get in docs for vectors
> > corresponding to valarrays (effectively explaining those parts of
> > valarrays for C programmers), stating what is and is not currently
> 
> ok, i don't understand c++ (or valarrays) sufficiently to be able to
> write those docs.  for that matter, i don't agree we should be
> patterning vectors after valarrays just because it's a well known
> interface.  but if this is the route agreed by everyone, then i suggest
> someone knowledgeable with valarrays document the corresponding C
> extensions.
> 
> would you be willing to write the docs?

I'm not familiar enough with valarrays.  Someone else should do those
docs.

> this is something i want to do for 3.2.  for that matter i want to
> implement the whole shebang of sythesizeable vector operations.  if you
> define vector size to be 16, and you only have vectors of 8bytes, then
> the optab code should be able to use a pair of V2SIs to do V4SI
> operations.  this should provide a basis for auto-vectorization work.

Yes, the operations should in general be synthesised if the CPU doesn't
have operations for the particular sort of vector.  (Including converting
to loops over scalars if necessary.)

-- 
Joseph S. Myers
jsm28@cam.ac.uk


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