[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: altas 3.1b

Hi Clint!

R Clint Whaley <rwhaley@cs.utk.edu> writes:

> Camm,
> >Need this line at line 200 in gemvtune:
> >
> >	 *lda= *lda < *N ? *N : *lda;
> My line 200 is:
>          *N = atoi(args[++i]);
> and adding the line you suggest, if I understand correctly, would make
> lda incorrect for the notrans case . . .

You are, of course, right here, and if this were to be implemented,
similar logic for m in no transpose would have to be added.  Please
excuse my hasty comments here -- I'm just trying to report issues as
they come up as quickly as possible for the purposes of productivity,
and do not suppose that the fixes I suggest should be the final ones
implemented, if indeed any need to be implemented at all.

> The expectation is that the user, if he does not accept the defaults
> for M, N, lda, will set them correctly (the defaults are correct).

A little confused here -- there doesn't seem to be a command line
switch to set the lda, only the m and n.  Or do you mean that the
knowledgeable user would edit the source?  Well, of course that's what
I did, but only after spending 30min to an hour trying to figure out
why my kernel segfaulted when I ran xdmv with -m 2000 -n 2000.  The
other testers (i.e. xsl2blastst, etc.) won't allow lda to be less than
the number of rows(columns) input by the user, no?

> If you want to guard against error here, you could think about adding
> something like:
>    if (*cta == 'N' && *lda < Mmax(1,*M))
>    {
>       fprintf(stderr, "Illegal lda=%d, M=%d\n", *lda, *M);
>       exit(0);
>    }
>    else if (*lda < Mmax(1,*N))
>    {
>       fprintf(stderr, "Illegal lda=%d, M=%d\n", *lda, *M);
>       exit(0);
>    }
> to line 213.

Good suggestion!  Will do.

> Cheers,
> Clint

Thanks, as always!

Camm Maguire			     			camm@enhanced.com
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah