Mark,
> > > b. I see that you are using /proc/ov511/video* and cpia uses
> > > /proc/cpia/video*. I thought that drivers were supposed to
> > > use /proc/driver/xxx for proc-fs data like this (or did I
> > > dream that?). Or should v4l drivers use /proc/video/* or
> > > /proc/v4l/* ? Is there any standard^W expectation on this?
> >
> > Again, I was just doing what the CPiA driver does, but that
> > doesn't mean
> > it's right, of course. I do agree that they should someplace else,
> > though I don't know where. /proc/v4l seems good as it is
> > consistent with /dev/v4l.
>
> Alan (the v4l gnome) preferred /proc/video (as in /proc/video/ov511):
> > There is no current standard here. I would prefer
> /proc/video/... as it
> > keeps things less cluttered than /proc/randomcardifoundintheshed/...
>
> Can you give me a patch with this also?
I'll take this as a patch to your latest patch if you want to
do it that way.
I'd like to see the cpia driver changed for this also.
> > > c. Is all of the floating point math done at compile
> > > time? It can't stay in there if it's done at run time.
> > > [I just noticed that it was in there previously.]
> >
> > It seems to compile down to integer math one way or another, and it
> > seems to work fine, so I assume it is OK. I was pretty sure
> that basic
> > floating point ops are OK in the kernel, but I am no expert on such
> > things. Should I be saving and restoring the FPU registers or
> > something?
>
> There was just thread on linux-kernel last week about a driver
> using floating point and how it can (theoretically) cause
> random, unexpected occurrances.
> No, you shouldn't be saving/restoring the FPU registers.
> You just shouldn't be using FP in the kernel (at run-time).
> The compile-time stuff that is in ov511 should be OK AFAIK.
>
> > Incidentally, I was planning on using MMX/SSE/3DNOW
> > eventually to speed
> > things up, but is this even safe/doable in the kernel? Alan?
>
> My understanding of MMX/SSE is that they have the same
> requirements as the FPU. Their register set needs to be
> saved/restored.
One suggestion from the kernel mailing list that I liked
on this subject was (from Andi Kleen, [EMAIL PROTECTED]):
> You could compile with -msoft-float, then the linker would complain
> at least.
~Randy
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
