[Mesa-dev] Piglit test for 972e995b1 (ARB_draw_elements_base_vertex)?

[Marek Olšák]

On Thu, Oct 11, 2012 at 1:54 AM, Eric Anholt <eric at anholt.net> wrote:
> Matt Turner <mattst88 at gmail.com> writes:
>
>> Hi,
>>
>> I notice that Mesa's 972e995b1 commit has disabled some code since
>> 2009. I re-enabled it and didn't see any changes in piglit results.
>> Should this still be disabled, and if so could you write a piglit test
>> or give enough information for me to do it?
>>
>> ARB_draw_elements_base_vertex doesn't seem to be in good shape on
>> i965. draw-elements-base-vertex-neg-user_varrays crashes, and
>> draw-elements-base-vertex-user_varrays and
>> draw-elements-instanced-base-vertex-user_varrays fail.
>
> The summary of these failures is:
>
> When using base_vertex, we need to load non-vbo vertex data from
> min_index/max_index offset by basevertex.  I've got a branch to do so,
> except that neg-user_varrays still fails due to what I think is overflow
> issues in my pointer math.
>
> (This is an extension it would have been very nice to not introduce in
> legacy GL)

You mean the intel drivers only, don't you? I consider the "legacy GL"
to be 99% of all use cases we get today, that is, the most important
ones to all our users. Hell, Mesa hadn't even had non-legacy contexts
in an official release until two days ago.

We pass all the piglit tests you mentioned on r300g, r600g, and any
other driver that doesn't expose PIPE_CAP_USER_VERTEX_BUFFERS, because
gallium takes care of that in the shared code. It's not so hard to
make the tests pass, in fact, it was quite easy.

Marek