Looks like nouveau works fine when setting PIPE_CAP_USER_*_BUFFERS to
false. I'm not sure why, as the u_vbuf code seems similarly busted --
e.g. it does
map -= vb->stride * min_index;
But... perhaps I'm misreading/misunderstanding the code.When I fudge
the code to force unroll_indices =
Do you think that the 32-bit multiply overflow when adding to the
pointer is the correct behaviour? If it causes radeon/whatever to
fail, I'd argue that they're adjusting their pointer incorrectly as
well. I'll see what's up by setting that cap to 0, and if that causes
the test to break, I'll figur
The draw-elements-base-vertex-neg test passes on Radeon, which uses
the common util/u_vbuf for uploading vertices. I know Nouveau is
probably the only driver which doesn't use it, not counting the swrast
drivers. I'm afraid that your change from fail to pass for Nouveau
will break the test for ever
On Wed, Jan 22, 2014 at 3:27 PM, Marek Olšák wrote:
> Does Nouveau still work if you report PIPE_CAP_USER_VERTEX_BUFFERS = 0?
I'm not in front of a machine with nouveau, so I can't tell you right
now, but I'll test it out later tonight. Out of curiousity though, why
do you ask? Is it related to t
Does Nouveau still work if you report PIPE_CAP_USER_VERTEX_BUFFERS = 0?
Marek
On Wed, Jan 22, 2014 at 3:37 AM, Ilia Mirkin wrote:
> This was discovered as a result of the draw-elements-base-vertex-neg
> piglit test, which passes very negative offsets in, followed up by large
> indices. The nouve
On Wed, Jan 22, 2014 at 12:03 PM, Brian Paul wrote:
> On 01/21/2014 06:37 PM, Ilia Mirkin wrote:
>>
>> This was discovered as a result of the draw-elements-base-vertex-neg
>> piglit test, which passes very negative offsets in, followed up by large
>> indices. The nouveau code correctly adjusts the
On 01/21/2014 06:37 PM, Ilia Mirkin wrote:
This was discovered as a result of the draw-elements-base-vertex-neg
piglit test, which passes very negative offsets in, followed up by large
indices. The nouveau code correctly adjusts the pointer, but the
transfer code needs to do the proper inverse co
This was discovered as a result of the draw-elements-base-vertex-neg
piglit test, which passes very negative offsets in, followed up by large
indices. The nouveau code correctly adjusts the pointer, but the
transfer code needs to do the proper inverse correction. Similarly fix
up the SSE code to do
