[yoshimi] Re: faster or safer optimization options
- From: Ichthyostega <prg@xxxxxxxxxxxxxxx>
- To: yoshimi@xxxxxxxxxxxxx
- Date: Mon, 28 May 2018 17:40:36 +0200
On 28.05.2018 16:26, Nikita Zlobin (Redacted sender cook60020tmp for DMARC)
Btw, imho, cpu-specific cflags don't need to have cpu instruction
flags, which are included automatically by correct -march. Or is it
This was also my understanding, that /with today's compilers/, in such
a general and average case, it should be sufficient to use the correct -march
Especially -march=native is your friend, and often does better than some
"hand optimisation" -- unless you /really/ know what you're doing.
However, I recall that in the 90ties, adding some specific instruction
flags often could work wonders. And probably from this experience
stems the habit to set such flags. This kind of settings can be
surprisingly long-lived and last for 20 years and longer, simply
because no-one takes a close look.
I worried about -fno-math-errno, because gcc man says, that it breaks
behavior for apps, relying to exact conformance to some iso standards.
A quick grep for "ftestexcept" or "FE_" shows that this codebase does
nothing specific in that respect.
The whole point with -fast-math and thus with -Ofast is that we can
just ignore some tiny error here and there. With complex enough
instruments, you can well drive yoshimi beyond the real-time
limit, so that you have to resort to rendering a prepared MIDI,
with Jack free-wheeling mode.
Yoshimi source code is available from either:
Our list archive is at: https://www.freelists.org/archive/yoshimi
To post, email to yoshimi@xxxxxxxxxxxxx
Other related posts: