Compiling Radiance with the Intel (Vectorizing) C Compiler

Randolph M. Fritz wrote:

I've got it to compile...but rpict crashes. Anyone tried this before?
Did you get it to work?

I think the answer is "no." The Radiance code doesn't work with that compiler, even with optimization turned off. I would like to chase those bugs down, but I don't think I have time right now.

Sigh. Back to the (slower) gcc.

Meantime, here's a few more buglets the icc found:

- In ccyrgb.c, rgb2ccy has two return statements, and only one of them returns a value.

- In tmapcolrs.c, tmMapPicture, returnErr will reference the unset tms pointer if the arguments are bad. I believe this can be fixed by initializing tms to NULL.

- In tmaptiff.c, tmMapTIFF, same problem.

shouldn't "gcc -O3 -g -Wall " spot all three ?

btw, in 2004 I had gotten as far as getting the Intel compiler and an
Intel book "The software Vectorization Handbook" by Aart J.C. Bik, who
apparently got "the highest Intel award for making (..) SSE easier to
use through automatic vectorization". (there's probably already a
posting on the list on that).
My enquiries into locating industrial funds to look into this had been
stopped then, mostly by Greg stating that setting up vectors takes
longer then gaining cycles. I don't have enough insight into the way
Radiance spends its time and the mystics of automatic vectorizing
compilers to be definitive there. Industry sources stated that anything
below a considerable speed-up-factor (say 2 at least) isn't worth
looking into.

However, between 2004 and 2011, follow-on SSE generations and the
increased use of computing power graphics cards appeared, whose
functions should at least be checked.
If there's someone on the list with some insights and/or experience,
please drop a note.

One of the mantras I do easily agree with is that dissecting runtime
profiles of different type of scenes gotta be worth the dirty work.

IMHO, apart from speed-ups, it'll be sound if Radiance compiled on
different compilers and run.

cheers
Peter

···

On 2011-06-27 14:18:01 -0700, Randolph M. Fritz said:

--
pab-opto, Freiburg, Germany, http://www.pab-opto.de
[see web page to check digital email signature]

Try to run it trough clang's scan-build, that should spot a lot of issues. I can
do that and publish the results if I find the time to do so, but my spare time
is down to nothing these days :frowning:

···

On 06/28/2011 07:25 PM, Randolph M. Fritz wrote:

On 2011-06-27 14:18:01 -0700, Randolph M. Fritz said:

I've got it to compile...but rpict crashes. Anyone tried this before?
Did you get it to work?

I think the answer is "no." The Radiance code doesn't work with that compiler,
even with optimization turned off. I would like to chase those bugs down, but I
don't think I have time right now.

--
Bernd Zeimetz Debian GNU/Linux Developer
http://bzed.de http://www.debian.org
GPG Fingerprint: ECA1 E3F2 8E11 2432 D485 DD95 EB36 171A 6FF9 435F