Request for review: Radeon RENDER repeat acceleration for (sm)all POT sizes

[Alex Deucher]

On 10/23/06, Michel Dänzer <michel at tungstengraphics.com> wrote:
> On Sun, 2006-10-22 at 22:11 -0400, Alex Deucher wrote:
> > On 10/22/06, Michel Dänzer <michel at tungstengraphics.com> wrote:
> > >
> > > The attached patch reduces the requirement for pitches to be a multiple
> > > of 32 instead of 64 and thus allows accelerating all POT sizes. It
> > > worked perfectly for me on R200 (M9) but causes lockups on a Radeon 7200
> > > in a Mac Cube. Does anyone spot any obvious problems?
> >
> > Nothing obvious pops out afer a cursory glance.  Does it work on r100
> > on LE?
>
> Haven't got a chance to try that, but I'm sure Dave will. :)

I could give it a shot on rv100 this week.

>
> > Perhaps r100 only likes to be programmed via the mixed OFFSET_PITCH regs
> > rather than the individual regs. Or perhaps if you program the mixed regs
> > you have to stick to them only rather than mixing operations with the
> > individual regs.
>
> Thanks for the ideas, but interestingly, it seems to work even on that
> machine with XAA instead of EXA. I thought I had tried that, but
> apparently not. I don't see what could be the problem with the EXA
> changes though.

very strange.

>
>
> PS: Would be nice if you could trim irrelevant context from your posts,
> such as that below:
>

Normally I do.  I must have missed it when I sent my reply yesterday.

Alex