tree 4d3f6d1e85c0a1c422c1213dae0b85b0edfbb1ea
parent 9a3e04748ed3ce97554c6355b26bf90f8bf53539
author Lionel Landwerlin <lionel.g.landwerlin@intel.com> 1476489663 +0100
committer Pyry Haulos <phaulos@google.com> 1477372406 -0400

copy_and_blit.blit_image: loosen threshold of float/float blits

In commit bd6f90c07f4bd0e0667ada50e134b479eeab78c5 :

  blit_image: compute validation threshold based on I/O formats

We tried to introduce an error threshold for unorm/integer formats and
removed all the precomputed thresholds for floating point formats. Also
introducing a bug where the threshold was ceiled to 1.0 (making a lot of
tests pass no matter the results).

In commit 161314f0ff255dfe47e280dd4945292295dab1ed :

  copy_and_blit.blit_image: fix threshold computation

We fixed the threshold to computation for unorm/integer formats but now all
threshold would be computed on the assumption that the resolution of numbers
is based of the number of bits in the fractional part of the number's
representation (which is incorrect for floating point formats).

This commit brings back the precomputed thresholds for floating point
formats, while still computing a threshold based on the number of bits in
the fractional part of the unorm formats.

Affected tests:
- dEQP-VK.api.copy_and_blit.blit_image.all_formats.*float*float*

Change-Id: I52034a58d145a8e046fb33bcc6becb9e83774d73