Re: [jamboost] Intel toolset (Was: Intel compiler questions)

From: Vladimir Prus (ghost_at_[hidden])
Date: 2004-05-13 04:45:33

Michael Stevens wrote:
> On Wednesday 12 May 2004 09:15, Vladimir Prus wrote:
> > Let me start with some nit-picking ;-)
>
> No problem I've attached a new version called "icclinux.jam"
>
> > 1. How the version is to be specified? As "80" or as "8.0". At least all
> > other toolset use the latter variant.
>
> Am I missing something here? You can use either when calling init. Up to
> the user what he/she prefers

With your previous code, the version was used to "guess" the installation 
path. So it seemed that you need to use the same spelling as in path.

> > 2. My copy is installed at /opt/intel_cc_80/bin/icc and you toolset
> > assumes different path. Is that the "new" scheme you mention above?
>
> Yep this is the new naming scheme.
>
> > 3. The path passes to the 'init' rule is required to end in slash, while
> > other toolsets are happy either way.
>
> Path was really a freeform prefix!. It could be used for compiler wrappers
> (things like distcc etc).
>
> I have simply dropped this from the new version. There are nothing else
> that requires the directory location so this seems to be the robusted way
> to go. The single name argument should be the name used to invoke the
> compiler.

I think that's fine appoach.

> Automatically finding the tool always fails. With every new compiler
> version things change! Witness all the confused message regarding the BBv1
> msvc

Well, V2 msvc *seems* to guess it right, well, until folks started to try the 
free 7.1 version ;-) Anyway, this can be added later.

> > 4. Finally, I think the the first declared version should become default
> > one.
>
> This happens by magic anyway. Not sure why this is but it works!!

The magic is that the "version" subfeature is not optional. So it defaults to 
the first declarated value.

> > Here, the compiler emits the following warning:
> >
> > icc: Command line warning: ignoring option '-O'; no argument required
> >
> > it happens only for "optimization=space", where the command line looks
> > like:
>
> Oh. Looks like icc doesn't support -Os. In fact it doesn't seem to have any
> flags to enable space optimisation. I've added code so -O1 is used instead.

Great. I've rerun the tests on my box and now there are only two failures -- 
"expansion" and "railsys" (this was not run previously). Both fail due to 
problems with the test itself, so the toolset works perfectly.

I'll commit it as soon as we finish discussion about single Intel toolset vs. 
two toolsets.

- Volodya