[Mesa-dev] [PATCH 1/3] i965: solve cubemap negative x/y/z faces buffer offset issue in dEQP.

[Jason Ekstrand]

On Tue, Oct 4, 2016 at 8:55 AM, Tapani Pälli <tapani.palli at intel.com> wrote:

> On 10/04/2016 06:09 PM, Jason Ekstrand wrote:
>
> On Thu, Sep 29, 2016 at 11:27 PM, Xu,Randy <randy.xu at intel.com> wrote:
>
>> Add the miptree level/slice x/y_offset when count the surface offset
>> in brw_emit_surface_state. The surface offset has two parts, one is
>> from mt->offset, which should be 32 aligned in width/height for tiled
>> buffer; another is from mt->level[current_level].slice[current_slice].
>> x/y_offset.
>>
>> This fix will solve 12 deqp failure
>> dEQP-EGL.functional.image.create.gles2_cubemap_negative_*_texture
>>
>> Signed-off-by: Xu,Randy <randy.xu at intel.com>
>> ---
>>  src/mesa/drivers/dri/i965/brw_wm_surface_state.c | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
>> b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
>> index 61a4b94..3a5c573 100644
>> --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
>> +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c
>> @@ -85,7 +85,8 @@ brw_emit_surface_state(struct brw_context *brw,
>>                         unsigned read_domains, unsigned write_domains)
>>  {
>>     const struct surface_state_info ss_info =
>> surface_state_infos[brw->gen];
>> -   uint32_t tile_x = 0, tile_y = 0;
>> +   uint32_t tile_x = mt->level[0].slice[0].x_offset;
>> +   uint32_t tile_y = mt->level[0].slice[0].y_offset;
>>
>
> This isn't correct.  First off, there are some fairly strict restrictions
> on what we can do with tile_x and tile_y and we can't just shove x/y_offset
> in there.  We need to use intel_miptree_get_tile_offsets to get both a byte
> offset and an intratile offset.  Second, we should already be taking slices
> into account for cube maps via base_array_layer where needed.
>
> Unfortunately, I'm not 100% sure what the correct patch is without a bit
> more information about what the test is doing that causes a problem.
>
>
> I did take a brief look and when running the set mentioned above (for
> example with ./deqp-egl --deqp-case=*EGL.functional.
> image.create.gles2_cubemap_negative_*_texture) what happens is that we
> never end up to the part of code calling intel_miptree_get_tile_offsets in
> that function (because surf.dim_layout != dim_layout condition does not
> trigger). This is just what I observed, should we just call
> intel_miptree_get_tile_offsets() unconditionally then?
>

No.  Very much no.  The intel_miptree_get_tile_offsets() stuff is a hack
that lets us convert non-2D things to 2D things and it comes with piles of
restrictions.


>
> --Jason
>
>
>>     uint32_t offset = mt->offset;
>>
>>     struct isl_surf surf;
>> --
>> 2.7.4
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>
>
>
>
> _______________________________________________
> mesa-dev mailing listmesa-dev at lists.freedesktop.orghttps://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20161004/59892be4/attachment.html>