Bug 2195

Created attachment 1614 [details] [review]
this works for me

please test.

feel free to commit after removing the remainig questions.

It works here (radeon 7000, 1002:5159) although it's a bit slower than before.

The texture of the ground is missing in ut2003_demo with this patch.
But it works well with r200 driver in sw-tnl and hw-tcl mode, and works well
with old radeon driver.
It seems that everything thats rendered in the tcl stage is ok (has texture),
but everything thats rendered "below" has no texture.
This is strange since its almost the same code as in r200 driver.

could someone confirm?

any hints?

(In reply to comment #3)
> The texture of the ground is missing in ut2003_demo with this patch.
> But it works well with r200 driver in sw-tnl and hw-tcl mode, and works well
> with old radeon driver.
What do you mean with "missing" - is it completely gone, or can you still see
about the right color? Is this only dm-antalus? If the answer to both is yes,
that's probably a problem with projected textures. On r200, projected textures
didn't work at all (and ground texture in dm-antalus was correct), and when I
played around with that I got projected textures working (for instance tree
shadow) but the ground texture was more or less gone - I guess wrong q texture
coordinate. Eric fixed that then together with some texgen fixes. The issue on
r100 you're seeing might be along similar lines.

This patch causes segfaults with the rasterization fallbacks, which seem to be
identical to these of the r200 driver (#2516).

Created attachment 2162 [details] [review]
mesa_radeon_tvertex_20050320.diff.txt

I re-checked ut2003_demo and it works well, at least as well as with the old
driver.
Maybe the error I saw was with the cube patch
https://bugs.freedesktop.org/show_bug.cgi?id=2241 but cubemapping disabled
because that patch (still) shows the missing-texture error.

Created attachment 2520 [details] [review]
mesa_radeon_tvertex_20050423.diff.txt

Created attachment 2683 [details] [review]
mesa_radeon_tvertex_20050515.diff.txt

I've tested this on ut2004, ut, q3, and tcl_mode=0 projtex, and it seemed to be
fine.  There's a 25k reduction in the stripped binary size.  However performance
is also reduced somewhat (tcl_mode=0 quake3, 800x600 demofours on 1ghz p3 and
rv200 card):

x pre-tvertex
+ post-tvertex
+--------------------------------------------------------------------------+
|    +                                                                    x|
|+ + +       +                                                          x x|
||___A____|                                                             |AM|
+--------------------------------------------------------------------------+
    N           Min           Max        Median           Avg        Stddev
x   5            79          79.1          79.1         79.08    0.04472136
+   5          75.4            76          75.6         75.62    0.22803509
Difference at 95.0% confidence
        -3.46 +/- 0.239647
        -4.37532% +/- 0.303043%
        (Student's t, pooled s = 0.164317)

Do we want to apply the patch in this case?  It would increase maintainability,
definitely.

Committed, thanks!  Also, you really oughta be a committer :-)

Mass version move, cvs -> git