Subject: [linux-audio-dev] The intel C compiler
From: Steve Harris (S.W.Harris_AT_ecs.soton.ac.uk)
Date: Mon Feb 18 2002 - 21:37:59 EET
I've been using the intel C compiler for a few months now, but I was
unable to comment on it quantatively before due to licensing restrictions.
It appears to be free to download now (without support), without those
restrictions, so I guess it's OK now.
I've found that for the majority of my plugins it makes them significantly
faster, eg. 10-30%, a few become in the order of twice as fast and 1
became about 30% slower.
The vectorisation is impressive, but tends to err on the side of caution.
The actual numerical output is different what you get from gcc with
heavy optimiations, but only slightly.
The diagnostics are very good, and the error messages are, if anything,
better than gcc's.
As I mentioned in the SIMD thread, it can produce multi instruction set
code, but it won't work with LADSPA without recompiling all the hosts, and
I don't think its a good idea anyway.
I do plan to distrubite i786 RPM binaries of my plugins built with the
compiler, and the autoconf will detect it and build, but the compiler is
*very* hard to get working if you aren't using redhat 6.2 or 7.[12].
On one hand the compiler is fundamentally evil and goes against everything
I believe in ;), but on the other it lets me run annother 3 or 4 plugins
with no cost. I will ofcourse continue to ship i686 binaries too, and you
can compile the source yourself with whatever you want.
- Steve
This archive was generated by hypermail 2b28 : Mon Feb 18 2002 - 21:28:20 EET