[Mesa-dev] [PATCH 01/12] i965: Change fragment input related bitfields to 64-bit.
Kenneth Graunke
kenneth at whitecape.org
Tue Mar 12 22:55:45 PDT 2013
On 03/12/2013 10:39 PM, Paul Berry wrote:
> On 12 March 2013 22:23, Kenneth Graunke <kenneth at whitecape.org
> <mailto:kenneth at whitecape.org>> wrote:
>
> On 03/11/2013 03:51 PM, Paul Berry wrote:
>
> This patch updates the bitfields brw_context::wm.input_size___masks,
> tracker::size_masks, and brw_wm_prog_key::proj_attrib___mask, all of
> which are indexed by gl_frag_attrib, from 32-bit to 64-bit.
>
> This paves the way for supporting geometry shaders, and for merging
> the gl_frag_attrib and gl_vert_result enums. The combination of
> these
> two will require at least 55 bits in the bitfields.
> ---
> src/mesa/drivers/dri/i965/brw___context.h | 2 +-
> src/mesa/drivers/dri/i965/brw___fs.cpp | 7 ++++---
> src/mesa/drivers/dri/i965/brw___vs_constval.c | 18
> +++++++++---------
> src/mesa/drivers/dri/i965/brw___wm.c | 2 +-
> src/mesa/drivers/dri/i965/brw___wm.h | 2 +-
> 5 files changed, 16 insertions(+), 15 deletions(-)
>
>
> Your patch looks good. However, I hate this code, and so seeing
> your patch inspired me to re-measure whether the optimization is
> actually still worth doing. Turns out, it isn't: it actually makes
> performance slightly -worse- in the game it was supposed to help.
>
> So I've sent out a series to delete all this. Hopefully, if Eric
> agrees, we can just throw it out and you can drop this patch.
>
> Sorry for the trouble!
>
> --Ken
>
>
> Oh, I would be delighted to see this go away. Earlier today I found
> myself wondering what effect geometry shaders are going to have on this
> code--it would sure be nice not to have to worry about it.
Yeah, I found myself wondering that tonight as well, and I really didn't
want to think about it.
Probably no effect, actually, since it shouldn't ever have any effect
when a GLSL vertex shader is active, and that's required to have a
geometry shader active. But..."shouldn't". :) I wouldn't be surprised
if there were bugs.
More information about the mesa-dev
mailing list