Tested on SKL.
The d3d12 test suite can be run using the command:
tests/d3d12 or by running make check
d3d12:3758: Test failed: Got 0xbf4c8019, expected 0xbf95bc59 at (0, 0, 0).
d3d12:3818: Test failed: Got 0xff00ff00, expected 0xff0000ff at (0, 0, 0).
d3d12:3818: Test failed: Got 0xff00ff00, expected 0xffff0000 at (0, 0, 0).
For comparision the test passes on RADV.
Instructions for building the Vkd3d test suite:
Source code: git://source.winehq.org/git/vkd3d.git
Updated Spriv headers may be needed:
CPPFLAGS="-I/home/tarceri/git/SPIRV-Headers/include -I/home/tarceri/git/Vulkan-Headers/include" ./configure
Also an undated version of widl (a tool in the wine project) may be needed:
Then pass the new build to configure for Vkd3d e.g.
Created attachment 142718 [details]
Aub dump with simplified test
I've reduced the test to clear & resolve 2 layers only.
The clear color of layer0 ends up in the layer1.
Adding a texture cache invalidate between resolve of layer 0 & 1 fixes the issue. Not quite sure what's going on here...
Created attachment 142719 [details] [review]
The attached patch (which I haven't even compile-tested) should fix the issue with MCS partial resolves. We also should probably stop even doing them in anv (as per Lionel's patch).
Created attachment 142720 [details] [review]
The new patch builds.
(In reply to Jason Ekstrand from comment #3)
> Created attachment 142720 [details] [review] [review]
> Compile-tested patch
> The new patch builds.
Unfortunately, not fixing the issue :(
I'm writing a new crucible test to track that regression.
The following error doesn't happen if I run it on BDW, which does not execute the fast clear path:
d3d12:3506: Test failed: Got 0xbf4c8019, expected 0xbf95bc59 at (0, 0, 0).
So I ran the test disabling fast clear on sRGB on my Kabylake laptop and this error is gone too. It seems there is an issue with fast-clearing on sRGB on gen9.
Looks like Gen9 doesn't support fast clear on single-sampled SRGB buffers.
This MR should fix the last 2 failures reported in the first comment : https://gitlab.freedesktop.org/mesa/mesa/merge_requests/40
Looks like there is an additional issue with sRGB that Samuel is looking at.
(In reply to Lionel Landwerlin from comment #7)
> This MR should fix the last 2 failures reported in the first comment :
> Looks like there is an additional issue with sRGB that Samuel is looking at.
Pushed to master.
I have this series for the SRGB resolve issue : https://gitlab.freedesktop.org/mesa/mesa/merge_requests/50/
A fix landed as ("e2ae5f2f0a anv: don't do partial resolve on layer > 0").
Can we close this as FIXED?
(In reply to Juan A. Suarez from comment #10)
> A fix landed as ("e2ae5f2f0a anv: don't do partial resolve on layer > 0").
> Can we close this as FIXED?
There are 2 issues in this bug, first one is resolved with the commit you mentioned. Second one needs this MR : https://gitlab.freedesktop.org/mesa/mesa/merge_requests/50
Second issue also fixed at : https://gitlab.freedesktop.org/mesa/mesa/commit/add5a2ec92f4b3f7ac8353e5986dc04186a7b6da
Author: Lionel Landwerlin <email@example.com>
Date: Sun Dec 23 17:49:23 2018 +0000
anv: flush fast clear colors into compressed surfaces
Thanks! I can confirm that vkd3d test failures reported in this bug report are fixed. We are left with only 3 test failures related to ordered/unordered comparisons.
(In reply to Józef Kucia from comment #13)
> Thanks! I can confirm that vkd3d test failures reported in this bug report
> are fixed. We are left with only 3 test failures related to
> ordered/unordered comparisons.
Is there a bug for those?
(In reply to Lionel Landwerlin from comment #14)
> Is there a bug for those?