?

Log in

airlied
airlied
:.:..:.
May 2017
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
28 29 30 31

airlied [userpic]
how close to conformant is radv?

I spent some time staring into the results of the VK-GL-CTS test suite on radv, which contains the Vulkan 1.0 conformance tests.

In order to be conformant you have to pass all the tests on the mustpass list for the Vulkan version you want to conform to, from the branch of the test suite for that version.

The latest CTS tests for 1.0 is the vulkan-cts-1.0.2 branch, and the mustpass list is in external/vulkancts/mustpass/1.0.2/vk-default.txt

Using some WIP radv patches in my github radv-wip-conform branch and the 1.0.2 test suite, today's results are on my Tonga GPU:

Test run totals:
Passed: 82551/150950 (54.7%)
Failed: 0/150950 (0.0%)
Not supported: 68397/150950 (45.3%)
Warnings: 2/150950 (0.0%)

That is pretty conformant (in fact it would pass as-is). However I need to clean up the patches in the branch and maybe figure out how to do some bits properly without hacks (particularly some semaphore wait tweaks), but that is most of the work done.

Thanks again to Bas and all other radv contributors.