[Mesa-dev] [PATCH] r600g: Adjust pipe format when decompressing depth in BE

[Oded Gabbay]

On Fri, Mar 4, 2016 at 12:16 PM, Marek Olšák <maraeo at gmail.com> wrote:
> I think this is not the right place to do this. It looks like
> r600_translate_colorformat and/or r600_colorformat_endian_swap and
> r600_translate_texformat should be adjusted instead.
>
> Marek

Adjusted how ?
You want me to check specifically for this format there ?
I'll take a look if it can be done and I'll get back to you.

Oded

>
> On Thu, Mar 3, 2016 at 4:47 PM, Oded Gabbay <oded.gabbay at gmail.com> wrote:
>> The following is what happens when trying to do glReadPixels() with
>> GL_DEPTH_COMPONENT:
>>
>> 1. When decompressing depth in BE, the GPU performs a swap of the depth
>>    value when it is written to the memory-mapped buffer.
>>
>> 2. When the pipe format is PIPE_FORMAT_Z24_UNORM_S8_UINT, the values
>>    are unpakeded using unpack_uint_z_Z24_UNORM_X8_UINT() into the staging
>>    buffer that is passed back to glReadPixels().
>>
>> 3. unpack_uint_z_Z24_UNORM_X8_UINT() expects to unpack values that are
>>    packed according to Z24_UNORM_X8_UINT format.
>>
>> The combination of the above makes the values that are returned to the
>> user erroneous, because the values inside the memory-mapped buffer are,
>> in reality, in the packed form of X8_UINT_Z24_UNORM.
>>
>> llvmpipe/softpipe don't have this problem in big-endian because they
>> call util_pack64_z_stencil() when doing the CLEAR operation, which is not
>> a valid solution when a real GPU is doing that operation.
>>
>> This patch fix this problem by checking for these specific conditions
>> when doing the map memory (r600_texture_transfer_map). In case the
>> conditions match, the code will adjust the pipe format of the cbsurf
>> before creating the surface, and later, when the unpack function will be
>> called to copy the values, unpack_uint_z_X8_UINT_Z24_UNORM() will be
>> called instead of unpack_uint_z_Z24_UNORM_X8_UINT.
>>
>> This patch fix gl-1.0-readpixsanity (check_depth and check_stencil).
>>
>> Signed-off-by: Oded Gabbay <oded.gabbay at gmail.com>
>> Cc: "11.1 11.2" <mesa-stable at lists.freedesktop.org>
>> ---
>>  src/gallium/drivers/r600/r600_blit.c | 7 +++++++
>>  1 file changed, 7 insertions(+)
>>
>> diff --git a/src/gallium/drivers/r600/r600_blit.c b/src/gallium/drivers/r600/r600_blit.c
>> index c52d5a9..5cbc24f 100644
>> --- a/src/gallium/drivers/r600/r600_blit.c
>> +++ b/src/gallium/drivers/r600/r600_blit.c
>> @@ -173,6 +173,13 @@ static void r600_blit_decompress_depth(struct pipe_context *ctx,
>>                                 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
>>
>>                                 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
>> +                               /* In BE machine, the GPU returns a swapped result of the depth
>> +                                * decompression, that doesn't comply with gallium packed formats,
>> +                                * so we need to adjust the format of cbsurf */
>> +                               if (R600_BIG_ENDIAN &&
>> +                                                       surf_tmpl.format == PIPE_FORMAT_Z24_UNORM_S8_UINT)
>> +                                       surf_tmpl.format = PIPE_FORMAT_S8_UINT_Z24_UNORM;
>> +
>>                                 cbsurf = ctx->create_surface(ctx,
>>                                                 &flushed_depth_texture->resource.b.b, &surf_tmpl);
>>
>> --
>> 2.5.0
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev