Re: [eigen] Intermediate Packet Storage

[Joel Holdsworth <joel@xxxxxxxxxxxxxxxxxxx>]

> I'm no ARM expert, but I assume {d{2k}-d{2k+1}} is an alias for `q{k}`

Correct.

> > As I mentioned, my project requires GCC 5. I would be interested to 
> > know if newer versions of ARM GCC have the same issue - but there 
> > seems to be some issue with Eigen on newer versions, because godbolt 
> > is giving me errors.
>
> Yes, I have no idea about what causes this -- maybe some ARM expert can 
> chip in.

Thinking about it, I suppose the question is how to get an ARM SIMD 
(e.g. int32x4_t) in to/out of a POD int32_t[4] array.

At the moment, Eigen implements this in NEON pload/ploudu using these 
intrinsics:

http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0472k/chr1359125041077.html

And for storing:

http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.dui0472k/chr1359125041077.html

So, if Eigen uses these intrinsics, I guess it's no surprise that this 
results in the intructions we see. The problem is that I don't know how 
else to load/store an int32x4_t into POD without using these intrinsics.

Perhaps there is a way.


> Are you able to implement the above (or something similar) with pure 
> intrinsics?

Yes - the intermediate would be stored in a float32x4_t register in each 
of these cases.

> If that is not possible, I'd see no way at all to do this with that
> compiler. If it is possible, I'd see some hope in implementing the
> previously mentioned Meta-Packets.

I guess one thought would be the syntax inspired by std::placeholders, 
and boost::lambda:

using Eigen = E;
y = x.bind((E::_1 > 0.0f).select(E::sin(E::_1), E::cos(E::_1));

And if there were combined with std::tie inspired syntax you could do 
multi-assignment one-liners:

E::tie(x, y) = E::tie(E::cos(a), E::cos(b))
    .bind(E::_1 + E::_2, E::_1 - E::2);

Just a thought!


Thanks for you thoughts.
Joel