Subject: Re: [linux-audio-dev] Float to int conversions
From: Erik de Castro Lopo (erikd_AT_mega-nerd.com)
Date: Thu Nov 01 2001 - 21:06:46 EET
It really blew my mind when "Daniel V." <dvenkita_AT_rand.ss.ecel.uwa.edu.au> said:
> A quick question: have any tests of this code been made with non-x86
> architectures (specifically ppc)? Or did they not have the problem to
> begin with?
I'm stil trying to get access to PPC. Linux on PPC should have lrint/lrintf
so thats not a big issue. I'm much more interested in gettig access to
MacOS.
I did run this on UltraSparc II and could not get any speed improvement in
going from cast -> integer pipeline operations. I still haven't figured out
whether this is because the UltraSparc II has a shallow pipleine wrt i386
or whether the UltraSparc has a specific assembler instruction for float
to int casts. I should also try this on UltraSparc III which I has a deeper
pipeline than USII.
Erik
-- +-----------------------------------------------------------+ Erik de Castro Lopo nospam_AT_mega-nerd.com (Yes it's valid) +-----------------------------------------------------------+ Linux: the only OS that makes you feel guilty when you reboot -- Kenneth Crudup in comp.os.linux.misc
This archive was generated by hypermail 2b28 : Thu Nov 01 2001 - 21:02:16 EET