[Mesa-dev] [PATCH v2] radeonsi: Adding tiled to linear conversion functionality

[Nicolai Hähnle]

On 29.11.2017 08:10, Sahu, Satyajit wrote:
> 
> 
> On 11/28/2017 9:33 PM, Nicolai Hähnle wrote:
>> On 28.11.2017 09:34, Sahu, Satyajit wrote:
>> [snip]
>>>>> --- a/src/gallium/targets/dri/dri.sym
>>>>> +++ b/src/gallium/targets/dri/dri.sym
>>>>> @@ -4,6 +4,11 @@
>>>>>            __driDriverGetExtensions*;
>>>>>            nouveau_drm_screen_create;
>>>>>            radeon_drm_winsys_create;
>>>>> +        ac_compute_surface;
>>>>> +        ac_query_gpu_info;
>>>>> +        ac_read_write_tiled_data;
>>>>> +        amdgpu_addr_create;
>>>>> +        amdgpu_addr_destroy;
>>>> NAK.
>>>>
>>>> I told you in internal discussions already that we shouldn't expose
>>>> new public symbols, we're arguably already exposing too much.
>>> We need to expose certain addrlib functions in order get functionalities
>>> like bo create, bo map (get linear data from tiled).
>>>
>>>> Please use the extension mechanism that is already there.
>>> Are you suggesting to add a driExtension to expose addrlib functions?
>> Yes, if it is really needed.
>>
>> By the way, are you sure read_write_tiled is going to work like that?
>> What if the surface has DCC?
> Yes I have tested it on chromiumos, it works fine. When dumped the 
> linear data, the image is proper. One of the chromium test case which 
> does a bit comparison with reference image also passes.

That doesn't really answer the question though :)

Just because one or two tests pass, that doesn't mean the code is 
actually correct.

Cheers,
Nicolai


>>
>> This can be the case for a surface exported by radeonsi with explicit
>> flushing.
>>
>> Cheers,
>> Nicolai
>>
> regards,
> Satyajit


-- 
Lerne, wie die Welt wirklich ist,
Aber vergiss niemals, wie sie sein sollte.