Re: [eigen] FLENS C++ expression template Library has excellent documentation

[Ilya Baran <baran37@xxxxxxxxx>]

Hi, Benoit,

On Sat, Apr 18, 2009 at 12:06 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
> 2009/4/18 Ilya Baran <baran37@xxxxxxxxx>:
>> Hi, Benoit,
>>
>> Wow, you're looking far ahead :)
>>
>> While fully automatic optimized implementation is their "ultimate
>> goal,"  I think their methods already allow much simplified manual
>> implementation.  Look, for example, at their Cholesky code:
>>
>> http://z.cs.utexas.edu/wiki/LA.wiki/Chol_l/FLAMEC/BlkVar3
>
> OK, I'm not sure I understand. This header seems to be implementing
> block cholesky calling a lot of convenience functions helping it with
> that. So FLAME is that, a set of helper functions to write block
> algorithms?

It looks like it's both the helper functions and the implemented
algorithms themselves.

> If yes, then Eigen can be improved to allow that kind of thing without
> much effort, I mean our Block expression allows to do part of that
> already, also our comma-initializer combined with xprs, so i think
> it's easy to add the rest (like repartitioning). I understand that
> they had to add a lot of C functions on top of BLAS to deal with
> blocks, but we don't need to, thanks to expressions.

Yeah, that's all I meant: they picked the right set of convenience
functions that are easy to replicate in Eigen and that make the
algorithms look clean, simple, and similar.

> I really mean to implement Block LLt or partial LU soon,  it can be
> written cleanly with Eigen blocks.
>
> What I was more thinking about when I said 2015, was automatic
> transformation of non-blocked algorithms into blocked algorithms. But
> maybe I misunderstood!

I think they were talking about automatically transforming math into
code, not non-blocked into blocked: a non-blocked algorithm that's
heavily optimized to the hardware (for small matrices) makes a very
good base case, but a poor starting point for an automatic
transformation.

   -Ilya