[Mesa-dev] [PATCH] anv/device: Add limits for InterpolationOffset

[Anuj Phogat]

On Fri, Jul 29, 2016 at 2:30 PM, Anuj Phogat <anuj.phogat at gmail.com> wrote:
>
>
> On Fri, Jul 29, 2016 at 12:32 PM, Anuj Phogat <anuj.phogat at gmail.com> wrote:
>>
>>
>>
>> On Thu, Jul 28, 2016 at 6:31 PM, Jason Ekstrand <jason at jlekstrand.net>
>> wrote:
>>>
>>> On Jul 28, 2016 7:37 PM, "Anuj Phogat" <anuj.phogat at gmail.com> wrote:
>>> >
>>> > Fixes the vulkan cts regression in test
>>> > dEQP-VK.api.info.device.properties
>>> >
>>> > Cc: Mark Janes <mark.a.janes at intel.com>
>>> > Cc: Jason Ekstrand <jason at jlekstrand.net>
>>> > Signed-off-by: Anuj Phogat <anuj.phogat at gmail.com>
>>> > ---
>>> >  src/intel/vulkan/anv_device.c | 6 +++---
>>> >  1 file changed, 3 insertions(+), 3 deletions(-)
>>> >
>>> > diff --git a/src/intel/vulkan/anv_device.c
>>> > b/src/intel/vulkan/anv_device.c
>>> > index c2c5153..9fd9683 100644
>>> > --- a/src/intel/vulkan/anv_device.c
>>> > +++ b/src/intel/vulkan/anv_device.c
>>> > @@ -524,9 +524,9 @@ void anv_GetPhysicalDeviceProperties(
>>> >        .maxTexelOffset                           = 7,
>>> >        .minTexelGatherOffset                     = -8,
>>> >        .maxTexelGatherOffset                     = 7,
>>> > -      .minInterpolationOffset                   = 0, /* FIXME */
>>> > -      .maxInterpolationOffset                   = 0, /* FIXME */
>>> > -      .subPixelInterpolationOffsetBits          = 0, /* FIXME */
>>> > +      .minInterpolationOffset                   = -0.5,
>>> > +      .maxInterpolationOffset                   = 0.5,
>
> Correction: maxInterpolationOffset = 0.4375
> From Vulkan 1.0 spec:
> "
> If subPixelInterpolationOffsetBits is 4, this provides increments of (1/24)
> = 0.0625, and thus the range of supported interpolation offsets would be
> [-0.5, 0.4375].
> "
Jason, other than above correction, is there anything else you want me
to address in V2 of this patch ?

>>>
>>> > +      .subPixelInterpolationOffsetBits          = 4,
>>>
>>> I looked and neither gen7_pipeline.c nor gen8_pipeline.c set the subpixel
>>> precision so we get a default of enum value 0 which is 8 bits.  We should
>>> probably be explicitly setting it to something.  I think 4 is probably fine
>>> but we should be seeing it.
>>
>> I think you're talking about "Vertex Sub Pixel Precision Select" in
>> 3DSTATE_SF. Yes, we are setting a default value of zero which is 8 bits in
>> both OpenGL and Vulkan. This meets the minimum requirement of 4 bits for
>> GL_SUBPIXEL_BITS. Yes, it'll be nicer to make it more obvious what we are
>> setting here.
>>
>> The value we are defining in this patch is subPixelInterpolationOffsetBits
>> (GL_FRAGMENT_INTERPOLATION_OFFSET_BITS in OpenGL). I found no option of
>> setting it on Intel Hardware. I just replicated what we do in i965 and made
>> the CTS test happy again.
>>
>> Looks unrelated but there is this another thing 'viewportSubPixelBits'
>> which is initialized to 13 in Vulkan but 0 (required min. value) in i965.
>> Again found no option of setting on Intel hardware.So, a similar case to
>> subPixelInterpolationOffsetBits.
>>>
>>> As a side note, I think we could probably combine 3DSTATE_SF and
>>> 3DSTATE_RASTER setup between gen7 and gen8 (and move it to
>>> genX_pipeline_until.h).  We'd probably have to do some #define trick like we
>>> did for 3DSTATE_SBE and 3DSTATE_SBE_SWIZ because the two packets are
>>> combined on gen7 and split on gen8.
>>
>> I'll look in to it.
>>
>>>
>>> >        .maxFramebufferWidth                      = (1 << 14),
>>> >        .maxFramebufferHeight                     = (1 << 14),
>>> >        .maxFramebufferLayers                     = (1 << 10),
>>> > --
>>> > 2.5.5
>>> >
>>
>>
>