[Mesa-dev] [PATCH 8/8] etnaviv: fix rs blit for formats with blocksize of 8

[Wladimir J. van der Laan]

On Sat, Oct 21, 2017 at 10:22:09AM +0200, Christian Gmeiner wrote:
> Hi Wladimir
> 
> 2017-10-21 7:28 GMT+02:00 Wladimir J. van der Laan <laanwj at gmail.com>:
> > On Fri, Oct 20, 2017 at 10:21:28PM +0200, Christian Gmeiner wrote:
> >> Fixes piglit/bin/oes_texture_float half
> >>
> >> Signed-off-by: Christian Gmeiner <christian.gmeiner at gmail.com>
> >> ---
> >>  src/gallium/drivers/etnaviv/etnaviv_translate.h | 2 ++
> >>  1 file changed, 2 insertions(+)
> >>
> >> diff --git a/src/gallium/drivers/etnaviv/etnaviv_translate.h b/src/gallium/drivers/etnaviv/etnaviv_translate.h
> >> index 0761251a36..c82a244e6d 100644
> >> --- a/src/gallium/drivers/etnaviv/etnaviv_translate.h
> >> +++ b/src/gallium/drivers/etnaviv/etnaviv_translate.h
> >> @@ -249,6 +249,8 @@ etna_compatible_rs_format(enum pipe_format fmt)
> >>        return PIPE_FORMAT_B4G4R4A4_UNORM;
> >>     case 4:
> >>        return PIPE_FORMAT_B8G8R8A8_UNORM;
> >> +   case 8:
> >> +      return PIPE_FORMAT_B8G8R8A8_UNORM;
> >
> > That's the same as 4. How does the RS know it has to copy 8 bytes per pixel?
> >
> 
> I tried all possible values for case 8 and there are a handful of them
> which helped to pass
> oes_texture_float half piglit. I am not really sure if RS @gc2000 even
> supports 64bit clears/blits (as I never
> have seen them in any cmd stream dumps). Lets see if I can find the
> the format used by RS for 64bit clears/blits.
> In the worst case we need to fall back to a slower blit method. Will
> look into the this today.

OK thanks for the explanation. Unfortunately I'm not sure about that either. I
don't really understand why this makes it pass the test, but it's interesting.

Doesn't seem too easy to fall back on sw here, especially on gc2000 because of
the split-tiled and supertiled formats.

Regards,
Wladimir