Re: arch-specific template code

[Paolo Carlini <paolo dot carlini at oracle dot com>]

Hi,

On 08/28/2012 01:51 PM, Ulrich Drepper wrote:
> On Mon, Aug 27, 2012 at 6:26 PM, Paolo Carlini <paolo.carlini@oracle.com> wrote:
> > My personal opinion is that a concrete example, small, but meaningful and
> > rather self contained, would help. To be honest, at this stage, isn't clear
> > to me which kind of arch-specific optimizations you are thinking about.
> Here is a first example.  Note that for now I just added the code in
> the middle of random.tcc.  This is an implementation for the
> normal_distribution<double>::__generate<> function using SSE3.  The
> resulting code runs about 25% faster.  There is really no way to use
> the function for any other architecture because it heavily depends on
> the x86 intrinsics and hence the x86 instructions.  But there is no
> reason why there couldn't be functions with the same interface but
> completely different implementation for other archs.  PPC has Altivec,
> Arm has Neon.
Good, good. I don't see any problem with integrating the code more or 
less as-is, we have only to figure out a suited, neat, scheme for the 
includes. Involving the cpu subdirectory as you suggested seems indeed a 
nice idea, but probably doing everything with a single header per cpu 
will not scale well in the future, I suppose better adding a whole 
subdirectory of specializations for each cpu, one file per for each std 
header. And obviously use a generic fall back for the generic cpu which 
essentially has just empty headers. I think something quite 
straightforward should do.

Paolo.