[Mesa-dev] [PATCH 1/2] radeonsi: simplify gl_FragCoord behavior

[Marek Olšák]

On Fri, Jan 8, 2016 at 4:30 PM, Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> On 08.01.2016 06:48, Marek Olšák wrote:
>>
>> On Fri, Jan 8, 2016 at 12:35 PM, Axel Davy <axel.davy at ens.fr> wrote:
>>>
>>> On 08/01/2016 02:30, Marek Olšák wrote:
>>>>
>>>>
>>>> From: Marek Olšák <marek.olsak at amd.com>
>>>>
>>>> It will become a system value, not an input.
>>>> ---
>>>>    src/gallium/drivers/radeonsi/si_state_shaders.c | 45
>>>> ++++++++++++-------------
>>>>    1 file changed, 22 insertions(+), 23 deletions(-)
>>>>
>>>> diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c
>>>> b/src/gallium/drivers/radeonsi/si_state_shaders.c
>>>> index 64adf69..460dda5 100644
>>>> --- a/src/gallium/drivers/radeonsi/si_state_shaders.c
>>>> +++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
>>>> @@ -399,30 +399,29 @@ static void si_shader_ps(struct si_shader *shader)
>>>>          if (!pm4)
>>>>                  return;
>>>>    -     for (i = 0; i < info->num_inputs; i++) {
>>>> -               switch (info->input_semantic_name[i]) {
>>>> -               case TGSI_SEMANTIC_POSITION:
>>>> -                       /* SPI_BARYC_CNTL.POS_FLOAT_LOCATION
>>>> -                        * Possible vaules:
>>>> -                        * 0 -> Position = pixel center (default)
>>>> -                        * 1 -> Position = pixel centroid
>>>> -                        * 2 -> Position = at sample position
>>>> -                        */
>>>> -                       switch (info->input_interpolate_loc[i]) {
>>>> -                       case TGSI_INTERPOLATE_LOC_CENTROID:
>>>> -                               spi_baryc_cntl |=
>>>> S_0286E0_POS_FLOAT_LOCATION(1);
>>>> -                               break;
>>>> -                       case TGSI_INTERPOLATE_LOC_SAMPLE:
>>>> -                               spi_baryc_cntl |=
>>>> S_0286E0_POS_FLOAT_LOCATION(2);
>>>> -                               break;
>>>> -                       }
>>>> +       /* SPI_BARYC_CNTL.POS_FLOAT_LOCATION
>>>> +        * Possible vaules:
>>>> +        * 0 -> Position = pixel center
>>>> +        * 1 -> Position = pixel centroid
>>>> +        * 2 -> Position = at sample position
>>>> +        *
>>>> +        * From GLSL 4.5 specification, section 7.1:
>>>> +        *   "The variable gl_FragCoord is available as an input
>>>> variable
>>>> from
>>>> +        *    within fragment shaders and it holds the window relative
>>>> coordinates
>>>> +        *    (x, y, z, 1/w) values for the fragment. If multi-sampling,
>>>> this
>>>> +        *    value can be for any location within the pixel, or one of
>>>> the
>>>> +        *    fragment samples. The use of centroid does not further
>>>> restrict
>>>> +        *    this value to be inside the current primitive."
>>>> +        *
>>>> +        * Meaning that centroid has no effect and we can return
>>>> anything
>>>> within
>>>> +        * the pixel. Thus, return the value at sample position, because
>>>> that's
>>>> +        * the most accurate one shaders can get.
>>>> +        */
>>>> +       spi_baryc_cntl |= S_0286E0_POS_FLOAT_LOCATION(2);
>>>
>>>
>>>
>>> When combined with POS_FLOAT_ULC, in which situation could we see
>>> a difference in the POSITION value ?
>>
>>
>> No MSAA: the value is at the pixel center
>> MSAA, no sample shading: the value is at the sample position of any
>> visible sample (compatible with "centroid")
>> MSAA sample shading: the value is at the sample position of the computed
>> sample
>>
>> As far as I know, POS_FLOAT_ULC only offsets the value by a constant
>> (0.5).
>
>
> According to the docs it forces the position into the upper-left corner,
> AFAICT overriding the choice of POS_FLOAT_LOCATION. But that's exactly what
> we want here, isn't it?

I'm not sure. It either truncates the position, or it moves the pixel
center to the upper-left corner, and then adds the sample position,
whose components are in [-0.5, 0.5).

Marek