On Wednesday, 16 July 2025 07:52:56 Pacific Daylight Time Jan Schultke wrote:
> > > Are we really
> > > expecting vendors to create and maintain a decent implementation of
> > > <simd> for their DSPs from scratch?
> >
> > Yes. The basic implementation is where the Multiple in SIMD is just 1
> > where
> > all operations are simply scalar. That suffices to meet the requirements
> > of the header.
>
> That's not what I would call a "decent implementation". That's
> basically malicious compliance.

Is there a possible better implementation for the DSP in the first place? Has
anyone done an analysis to see if <simd> can be implemented decently (at least
in theory) for those DSPs out there? They might be just too different in the
first place.

And how is the situation different from DSP or other unusual architectures with
8-bit bytes? The effort is the same to support <simd>: massive, per
architecture, and provided by those who are interested.

> If you take a good look at what an implementation actually needs to
> support to meet basic user expectations of quality, C++ really has
> become too big for smaller vendors to roll on their own.

Not disputing that. But I am saying it's orthogonal to forcing bytes to be 8
bits. Either the vendor in question has the resources and interest to work on
the particular architecture or they don't. If they do and <simd> *can* be
implemented, they may do it. If they don't, then the architecture will suffer
either way.

-- 
Thiago Macieira - thiago (AT) macieira.info - thiago (AT) kde.org
  Principal Engineer - Intel Platform & System Engineering