[Mesa-dev] [PATCH v3] mesa: readpixels add support for GL_HALF_FLOAT

[Lin, Johnson]

Guy,

Then how can we deal with
android.view.cts.PixelCopyTest #TestWindowProducerCopyToRGBA16F
It is blocked by this error check..



-----Original Message-----
From: Palli, Tapani 
Sent: Tuesday, April 3, 2018 1:42 PM
To: Eric Anholt <eric at anholt.net>; Lin, Johnson <johnson.lin at intel.com>; mesa-dev at lists.freedesktop.org
Subject: Re: [Mesa-dev] [PATCH v3] mesa: readpixels add support for GL_HALF_FLOAT

Hi Eric;

On 03/30/2018 11:25 PM, Eric Anholt wrote:
> Lin Johnson <johnson.lin at intel.com> writes:
> 
>> Ext_color_buffer_half_float is using type GL_HALF_FLOAT and data_type 
>> GL_FLOAT. This fix Android CTS test android.view.cts.PixelCopyTest
>> v2: remove commtens of Ext_color_buffer_half_float.
>> As Ext_color_buffer__float can use type GL_HALF_FLOAT
>>
>> Reviewed-by:   Palli, Tapani <tapani.palli at intel.com>
>> Signed-off-by: Lin Johnson <johnson.lin at intel.com>
> 
> I've been looking into CTS issues on VC5, and I think this patch 
> should be reverted.

After some inspection of the packed_pixels test I do agree, will revert this. It's too bad dEQP does not have similar test. It bothers me a bit though that 'reading pixels' section has hints that type could be GL_HALF_FLOAT but no spec seems to add this capability, maybe that is then only possible via IMPLEMENTATION_COLOR_READ_FORMAT?


> The issue is that the text above that in the spec disallows this case.
>  From GLES 3.0.2:
> 
>      Only two combinations of format and type are accepted in most
>      cases. The first varies depending on the format of the currently
>      bound rendering surface. For normalized fixed-point rendering
>      surfaces, the combination format RGBA and type UNSIGNED_BYTE is
>      accepted. For signed integer rendering surfaces, the combina- tion
>      format RGBA_INTEGER and type INT is accepted. For unsigned integer
>      rendering surfaces, the combination format RGBA_INTEGER and type
>      UNSIGNED_INT is accepted.
> 
>      [...]
> 
>      ReadPixels generates an INVALID_OPERATION error if the combination
>      of format and type is unsupported.
> 
> and this spec adds on to that first paragraph:
> 
>      For floating-point rendering surfaces, the combination
>      format RGBA and type FLOAT is accepted.
> 
> The second allowed combo is:
> 
>      The second is an implementation-chosen format from among those
>      defined in table 3.2, excluding formats DEPTH_COMPONENT and
>      DEPTH_STENCIL . The values of format and type for this format may be
>      determined by calling Get- Integerv with the symbolic constants
>      IMPLEMENTATION_COLOR_READ_FORMAT and IMPLEMENTATION_COLOR_READ_TYPE
>      , respectively.
> 
> The failing cases are in the CTS's packed_pixels testsuite, such as:
> 
> KHR-GLES3.packed_pixels.pbo_rectangle.r11f_g11f_b10f
> 
> 

// Tapani