[Coin-discuss] Optimization level for OsiClp, bug in pack1.mps

[Michael Hennebry]

On Mon, 30 May 2005, Laszlo Ladanyi wrote:

> On Mon, 30 May 2005, Michael Hennebry wrote:
>
> > OptLevel= -O3 -g
> > is allowed, but my recollection is that there is
> > good reason for not doing that.
> > I think the makefiles do explicit tests for -g.
>
> Yes, OptLevel="-O3 -g" is allowed (must be quoted, because of the space). In

Quotes are not required for internal spaces.
So far as I know, they don't hurt.

> general, if the OptLevel is -O then the optimization flag used will be the
> highest level of optimization on that platform and compiler (if someone does
> some extensive testing, I'd be happy to modify what the "highest level"
> means). Otherwise the optimization level used will be the content of OptLevel.
> Also, some flags are enabled only if '-g' is not part of OptLevel (like
> -ffast-math); and some flags are set only if none of the optimization flags
> (-O, -O1, etc.) are set (like -DNDEBUG).

-g should not disable optimization flags.
Something like the following would be good.

G= -g
OptLevel= -O3
OptSquash=no
SymbolFind=yes
OptFlags=$(OptLevel) -ffast-math -fomit-frame-pointer
...

The OptSquash=no would keep -O3 and -ffast-math from being edited out.
If -fomit-frame-pointer would prevent the finding of debugging symbols,
the SymbolFind=yes would cause -fomit-frame-pointer to be edited out.

-- 
Mike   hennebry at web.cs.ndsu.NoDak.edu
"There are three kinds of people,
those who can count and those who can't."