[Mesa-dev] nvfx-next in Mesa

[Xavier Chantry]

On Tue, Jun 8, 2010 at 8:49 PM, Xavier Chantry <chantry.xavier at gmail.com> wrote:
> On Tue, Jun 8, 2010 at 8:36 PM, Corbin Simpson
> <mostawesomedude at gmail.com> wrote:
>> Luca (and anybody else, especially nouveau people),
>>
>> Does anybody want to step up and look at the nvfx-next branches? In
>> IRC, we're looking over some of the parts that got merged to master
>> (util/u_surfaces and util/u_dirty_surfaces) but not seeing the usage.
>> It would be pretty awesome if the rest of the branches could get
>> merged so they don't bitrot.
>>
>> ~ C.
>>
>
> Last news I had on April 21th :
> 4:58 PM Luca: the temporary code somehow breaks nv3x
> 4:59 PM me: so that blocks the rest, I thought the rest could be
> merged without that
>  Luca: perhaps
>
> Which referred to this commit :
> commit fac5e2d54f7b6f6eac9b96ad14263ebc4ca93832
> Author: Luca Barbieri <luca at luca-barbieri.com>
> Date:   Sun Apr 18 15:22:43 2010 +0200
>
>    nvfx: rewrite render temporaries code, also affecting 2D and resource code
>
>    This commit rewrites the render temporaries code to make it actually robust
>    and try to make it perform decently.
>
>    It also significantly affects how miptrees and the 2D code work.
>
>
>
> I could try to test again this week that merging origin/nvfx-next-6b
> without the above commit does not break anything on nv35.
> It sure fixed a lot of things, especially all the 2D work.
>

So some testing I made today with piglit and games.
mesa-7.8.1-1455-g5acee66 : 95/174
mesa-7.8.1-1466-g3c0120d : 115/174
mesa-7.8.1-1468-g81bdeee : 114/174

The breakage between 3c0120d and 81bdeee is texturing/lodbias and is
caused by the above commit :
nvfx: rewrite render temporaries code, also affecting 2D and resource code
This same commit also broke the following games : etracer, glest,
neverball, foobillard (when using hwtnl, which is the default).

The testing work Luca and I made trying to resolve this issue can be
seen there :
http://repo.or.cz/w/mesa/mesa-lb.git/shortlog/refs/heads/nvfx-next-6b-nv30

If anyone wants to rebase the whole nvfx-next-6b branch without
temporaries on top of master, I will happily test that as well.


Some interesting piglit changes between old master
mesa-7.8.1-1455-g5acee66 and new master mesa-7.8.1-2941-g9ef6d34 :
1) polygonOffset fails
polygonOffset:  FAIL rgba8, db, z24, s8, win+pmap, id 33
	Actual MRD is too large (may waste depth-buffer range)

	Ideal  MRD at near plane is 5.82077e-11 (nominally 0 bits)
	Actual MRD at near plane is 5.96046e-08 (nominally 1 bit)
	Ideal  MRD at infinity is 5.86733e-08 (nominally 1 bit)
	Actual MRD at infinity is 1.19209e-07 (nominally 2 bits)
2) fp-long-alu fails
Testing: alu_depth = 3077
Probe at (16,16)
  Expected: 0.312500 0.000000 0.750000 0.000000
  Observed: 0.000000 0.000000 0.000000 0.000000
3-4) texEnv and levelclamp pass