Re: How to measure frame rate in fps?

[Eli Zaretskii <eliz@gnu.org>]

> Cc: wyuenho@gmail.com, emacs-devel@gnu.org
> From: Dmitry Gutov <dgutov@yandex.ru>
> Date: Mon, 7 Jun 2021 02:31:50 +0300
> 
> I have one scenario, which I run with toolbar disabled, then enabled, 
> then disabled, then enabled again. With a (benchmark-progn ...) wrapped 
> around a (redisplay) call, to note how long it really takes in practice.
> 
> I'm saying that the buffer, or other window configuration, don't change 
> at those points, so the toolbar likely doesn't need to be re-rendered, 
> and so the extra overhead is surprising. And that the toolbar taking 
> 10-15ms to re-render isn't great either in any case.

But in the profile you showed we didn't see any evidence of
significant CPU usage by functions that redraw the GTK toolbar.  So
I'm not sure the overhead is due to redrawing the toolbar, although it
might be _triggered_ by the presence of the toolbar.

> It's the difference, computed very roughly. Let me show you the contents 
> of the Messages buffer. The times printed are reported by the 
> (benchmark-progn (redisplay)) form:

How do you reconcile this with the 00.00% that GTK functions took?

Can you see what does 'perf' say about functions from gtkutil.c that
are related to toolbar redisplay?