Re: [eigen] nesting

[Hauke Heibel <hauke.heibel@xxxxxxxxxxxxxx>]

This could be a solution:

template <typename T>
struct ei_ref_selector
{
  typedef typename ei_meta_if<
    sizeof(T)<=8, // tiny expression storing e.g. two references
    T,
    T const&
  >::ret type;
};

I am failing to add this code since it requires T to be completely
defined. When this is not working, the only alternatives I am seeing
are
a) to specialize ei_nested or ei_ref_selector for all expressions
b) add traits to expressions stating whether they are 'light weight'
or not and use them in ei_ref_selector
c) going back to copying everything by value

I am still convinced that c) is not a good alternative - not only
because we were required to bring back NestByValue but also because I
am sure that it is less error prone and ignoring for a moment all the
redundant copies we are seeing, there was a measurable performance
increase when I first tested the code with various expressions
(obviously not too many involving products).

- Hauke

On Thu, Feb 4, 2010 at 4:01 PM, Hauke Heibel
<hauke.heibel@xxxxxxxxxxxxxx> wrote:
> Well, specializing the product is a bad idea. Actually, we need to
> deduce expressions that have products nested (or heavy weight objects
> in general), those must then be nested by reference in order to
> prevent copies.
>
> - Hauke
>
> On Thu, Feb 4, 2010 at 2:35 PM, Hauke Heibel
> <hauke.heibel@xxxxxxxxxxxxxx> wrote:
>> Hi,
>>
>> While looking into the "performance degradation" issue from the forum
>> I found out that it is due to temporaries - as Benoit already guessed.
>>
>> I am a little bit afraid, that what I once proposed, namely copying
>> expressions by value, is now backfiring. The reason is that initially
>> I assumed expressions to be tiny little objects with close to no copy
>> costs. The issue is related to those expressions holding temporaries.
>> Copying them (e.g. a product expression) means copying all the data
>> including the temporary and that will happen as many times as we nest
>> expressions.
>>
>> The only solution I can think about at the moment is the
>> specialization of ei_nested for those types and to go back to nesting
>> by reference for these heavy weight guys.
>>
>> - Hauke
>>
>