[Mesa-dev] [PATCH] radeonsi: fix dvec[34] attributes sourced from current attribute state

Ilia Mirkin imirkin at alum.mit.edu
Thu Mar 23 17:34:54 UTC 2017


Ah I see. So we still get 2 vertex attributes, good. This was just
tripping some assumptions in your code about how attributes are laid
out in a constant vertex buffer. I guess I should double-check nouveau
as well, but I think we should be good. I always forget about the
disconnect between attributes and buffers :)

On Thu, Mar 23, 2017 at 1:29 PM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> On 23.03.2017 18:19, Ilia Mirkin wrote:
>>
>> Wait, you can have constant attribs that are dvec4's? Those don't get
>> split up into multiple constant attribs (2x dvec2, as it would for a
>> regular VBO)? Or am I misunderstanding the circumstances?
>
>
> The draw call will have one vertex buffer for each GL attribute, but the
> attribute is split into two vertex elements, i.e. it occupies two slots in
> the vertex array.
>
> E.g. the CTS has a test where there are 9 vertex buffers and I believe 15
> attributes at the Gallium interface level.
>
> Cheers,
> Nicolai
>
>
>>
>> On Thu, Mar 23, 2017 at 1:14 PM, Nicolai Hähnle <nhaehnle at gmail.com>
>> wrote:
>>>
>>> From: Nicolai Hähnle <nicolai.haehnle at amd.com>
>>>
>>> The state tracker no longer uploads those attributes for us,
>>> so we must conservatively upload the size of the largest
>>> attribute, which is a dvec4.
>>>
>>> Fixes a regression of GL45-CTS.gpu_shader_fp64.varyings and
>>> GL45-CTS.vertex_attrib_64bit.limits_test.
>>>
>>> Fixes: 9b91e0b54cc2 ("radeonsi: allow unaligned vertex buffer offsets and
>>> strides on CIK-VI")
>>> ---
>>>  src/gallium/drivers/radeonsi/si_state.c | 7 ++++---
>>>  1 file changed, 4 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/src/gallium/drivers/radeonsi/si_state.c
>>> b/src/gallium/drivers/radeonsi/si_state.c
>>> index 6948a74..0ee4af3 100644
>>> --- a/src/gallium/drivers/radeonsi/si_state.c
>>> +++ b/src/gallium/drivers/radeonsi/si_state.c
>>> @@ -3525,26 +3525,27 @@ static void si_set_vertex_buffers(struct
>>> pipe_context *ctx,
>>>
>>>         if (buffers) {
>>>                 for (i = 0; i < count; i++) {
>>>                         const struct pipe_vertex_buffer *src = buffers +
>>> i;
>>>                         struct pipe_vertex_buffer *dsti = dst + i;
>>>
>>>                         if (unlikely(src->user_buffer)) {
>>>                                 /* Zero-stride attribs only. */
>>>                                 assert(src->stride == 0);
>>>
>>> -                               /* Assume the attrib has 4 dwords like
>>> the vbo
>>> -                                * module. This is also a good upper
>>> bound.
>>> +                               /* Assume that the user_buffer comes from
>>> +                                * gl_current_attrib, which implies it
>>> has
>>> +                                * 4 * 8 bytes (for dvec4 attributes).
>>>                                  *
>>>                                  * Use const_uploader to upload into VRAM
>>> directly.
>>>                                  */
>>> -                               u_upload_data(sctx->b.b.const_uploader,
>>> 0, 16, 16,
>>> +                               u_upload_data(sctx->b.b.const_uploader,
>>> 0, 32, 32,
>>>                                               src->user_buffer,
>>>                                               &dsti->buffer_offset,
>>>                                               &dsti->buffer);
>>>                                 dsti->stride = 0;
>>>                         } else {
>>>                                 struct pipe_resource *buf = src->buffer;
>>>
>>>                                 pipe_resource_reference(&dsti->buffer,
>>> buf);
>>>                                 dsti->buffer_offset = src->buffer_offset;
>>>                                 dsti->stride = src->stride;
>>> --
>>> 2.9.3
>>>
>>> _______________________________________________
>>> mesa-dev mailing list
>>> mesa-dev at lists.freedesktop.org
>>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
>
>
> --
> Lerne, wie die Welt wirklich ist,
> Aber vergiss niemals, wie sie sein sollte.


More information about the mesa-dev mailing list