Re: [Request for review] fsm-lite

[Helmut Grohne <helmut@subdivi.de>]

Hi Nilesh,

There is no need to Cc me on d-cross@l.d.o.

On Wed, Mar 10, 2021 at 11:57:25PM +0530, Nilesh Patra wrote:
> Looks like it is a violation. Do you think building on amd64 for every
> optimization, i.e. sse4.2 down to the point there's no vectorization a
> good way to go about it?

I'm not exactly sure what you are asking here. It seems like you are
questioning whether disabling optimization is a good idea. That's a good
question with no universal answer. Debian does have an answer ("yes"),
but whether you like that answer depends. Much of the software where
upstream enables vectorization becomes painfully slow when you disable
it.

There also is the sse4.2-support package.

> To make it more clear, I'd simply like pointing you to here[1] for
> example, and the build for amd64 being built in for every optimization.
> Please let me know.

This is a valid way of dealing with the minimum baseline. An even better
way would be runtime detection. You wouldn't build the whole application
for multiple baselines, but merely the performance critical functions
and link all of the functions into a single binary. At runtime, it would
automatically select the appropriate optimization level. Somtimes, the
effort to implement this is not warranted.

Helmut