[RFC] Tests for vector types?

[RFC] Tests for vector types?

[Michael Snyder]

Say, these new vector types such as exist for Altivec...
are they going to be generally available to GCC users from now on?
Like, say, on i386-Linux?

If so, we should make up some generic tests for them for gdb.

Michael

Re: [RFC] Tests for vector types?

[Aldy Hernandez]

On Mon, May 20, 2002 at 05:04:46PM -0700, Michael Snyder wrote:
> Say, these new vector types such as exist for Altivec...
> are they going to be generally available to GCC users from now on?
> Like, say, on i386-Linux?

they are already in x86: sse, mmx, etc.

> If so, we should make up some generic tests for them for gdb.

unfortunately the tests can't be too generic because they depend
on the architecture having support for the SIMD type.  say, you
can't test V4SI on an architecture that only provides 64bit SIMD
registers (V2SI).

i have some patches i'm polishing up to provide just that-- any
SIMD type (say V4SI) regardless of if the hardware supports it.  this
will make it easier to write generic tests.

aldy

Re: [RFC] Tests for vector types?

[Elena Zannoni]

Aldy Hernandez writes:
 > On Mon, May 20, 2002 at 05:04:46PM -0700, Michael Snyder wrote:
 > > Say, these new vector types such as exist for Altivec...
 > > are they going to be generally available to GCC users from now on?
 > > Like, say, on i386-Linux?
 > 
 > they are already in x86: sse, mmx, etc.
 > 

Yes, but the gdb side needs to be cleaned up a bit, the gdb
implementation of these vectors is different from Altivec, i.e. they
use an old/deprecated representation. 

 > > If so, we should make up some generic tests for them for gdb.
 > 
 > unfortunately the tests can't be too generic because they depend
 > on the architecture having support for the SIMD type.  say, you
 > can't test V4SI on an architecture that only provides 64bit SIMD
 > registers (V2SI).
 > 
 > i have some patches i'm polishing up to provide just that-- any
 > SIMD type (say V4SI) regardless of if the hardware supports it.  this
 > will make it easier to write generic tests.
 > 

Cool.

Elena


 > aldy

Re: [RFC] Tests for vector types?

[Michael Snyder]

Aldy Hernandez wrote:
> 
> On Mon, May 20, 2002 at 05:04:46PM -0700, Michael Snyder wrote:
> > Say, these new vector types such as exist for Altivec...
> > are they going to be generally available to GCC users from now on?
> > Like, say, on i386-Linux?
> 
> they are already in x86: sse, mmx, etc.
> 
> > If so, we should make up some generic tests for them for gdb.
> 
> unfortunately the tests can't be too generic because they depend
> on the architecture having support for the SIMD type.  say, you
> can't test V4SI on an architecture that only provides 64bit SIMD
> registers (V2SI).
> 
> i have some patches i'm polishing up to provide just that-- any
> SIMD type (say V4SI) regardless of if the hardware supports it.  this
> will make it easier to write generic tests.

That's what I mean -- I'm thinking, support for the vector type
really has nothing to do with the architecture.  The vector type
is just an array that is passed by value, rather than by reference.
You don't have to have a special register to do that.

Re: [RFC] Tests for vector types?

[Aldy Hernandez]

> > i have some patches i'm polishing up to provide just that-- any
> > SIMD type (say V4SI) regardless of if the hardware supports it.  this
> > will make it easier to write generic tests.
> 
> That's what I mean -- I'm thinking, support for the vector type
> really has nothing to do with the architecture.  The vector type
> is just an array that is passed by value, rather than by reference.
> You don't have to have a special register to do that.

give me a week or two.  the patches are done but need documentation.

aldy