Make texture aniso basic self tests less sensitive
authorAlex Walters <alex.walters@imgtec.com>
Mon, 14 Oct 2019 16:32:12 +0000 (17:32 +0100)
committerAlexander Galazin <Alexander.Galazin@arm.com>
Thu, 17 Oct 2019 15:13:31 +0000 (11:13 -0400)
commit22655d8e1abbb8e6b0fba9d3a6cd68b1a1709028
treea702a5a44f15f24e664147bf6512708186c80fee
parent009074d449a71d6a385ab6814b64b6a1cba5c4ae
Make texture aniso basic self tests less sensitive

The tests cases first check with a fuzzy diff that the images are more
or less the same.

The second diff then checks with a standard image comparison + tolerance
that the images are not identical.

Some of IMG's implementations produce a closer image than the test is
currently expecting - we need to lower the tolerance from 0.05 to 0.02
in order for the diff algorithm to detect that the images are
different 'enough'.

Though there is no actual spec requirement, it is 'generally going to
be the case', for any implementation, that the images will have some
amount of difference. It is potentially useful to still have this test
inplace to check againsty aniso being accidently disabled in the tested
implementation, rather than remove it altogether.

Affects:
dEQP-VK.texture.filtering_anisotropy.basic.*

Components: Vulkan

VK-GL-CTS Issue: 2047

Change-Id: Ia9148e12172ac1c4757f5007e57b3a2b3a3434eb
external/vulkancts/modules/vulkan/texture/vktTextureFilteringAnisotropyTests.cpp