[Bug 103743] Nier:Automata - "if" in fragment shader incorrectly evaluated, causes artifacts
https://bugs.freedesktop.org/show_bug.cgi?id=103743
Bug ID: 103743 Summary: Nier:Automata - "if" in fragment shader incorrectly evaluated, causes artifacts Product: Mesa Version: git Hardware: x86-64 (AMD64) OS: Linux (All) Status: NEW Severity: normal Priority: medium Component: Drivers/Gallium/radeonsi Assignee: dri-devel@lists.freedesktop.org Reporter: philip.rebohle@tu-dortmund.de QA Contact: dri-devel@lists.freedesktop.org
Created attachment 135457 --> https://bugs.freedesktop.org/attachment.cgi?id=135457&action=edit Fragment shader with workaround
When compiling Mesa against LLVM >= 5.0, Nier:Automata shows lighting artifacts that are caused by a misbehaving 'if' in a fragment shader.
Apitrace that shows the issue (2.5G, bz2-compressed): https://mega.nz/#!JaBD0IoT!yBekRb7ZmvsSY60mo0E7N6uwaYfp_KfRDlnQWL0-QOo
I attached a copy of the fragment shader that causes the problem, but also includes a workaround. The file name for MESA_SHADER_READ_PATH is FS_feba74122c9590d1522b92bbec52d662ecd99012.glsl.
The offending code is located at line 135:
R1.x = uintBitsToFloat(floatBitsToUint(R0).y >= 0x4u ? 0xffffffffu : 0u); if (bool(floatBitsToUint(R1).x))
The artifacts go away when changing the original 'if' condition to the following:
if (floatBitsToUint(R0).y >= 0x4u)
This happens with both Mesa 17.2 and latest Mesa-git when compiling against LLVM 5.0 or later. This does *not* happen with LLVM 4.0.1. My GPU is an RX 480.
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #1 from Philip Rebohle philip.rebohle@tu-dortmund.de --- Created attachment 135458 --> https://bugs.freedesktop.org/attachment.cgi?id=135458&action=edit Frame 2400 with artifacts
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #2 from Philip Rebohle philip.rebohle@tu-dortmund.de --- Created attachment 135459 --> https://bugs.freedesktop.org/attachment.cgi?id=135459&action=edit Frame 2400 without artifacts
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #3 from Philip Rebohle philip.rebohle@tu-dortmund.de --- PS: While bug 102218 is not directly related to this one, that's another case where two shaders that *should* be functionally equivalent yield different results. In the second screenshot, workarounds for both issues are applied in order to obtain correct rendering.
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #4 from Nicolai Hähnle nhaehnle@gmail.com --- Thanks for the very lucid bug report. I can reproduce this and am investigating.
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #5 from Philip Rebohle philip.rebohle@tu-dortmund.de --- I found another pair of equivalent expressions that produce different results.
Working as expected: if ((floatBitsToUint(R0).y >= 0x4u ? 0xffffffffu : 0u) == 0xffffffffu)
Showing artifacts: if ((floatBitsToUint(R0).y >= 0x4u ? 0xffffffffu : 0u) != 0u)
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #6 from Nicolai Hähnle nhaehnle@gmail.com --- Thanks. To give you an update, this is a super subtle control flow handling bug in LLVM -- and the difference between the original shader and your modification at the LLVM input is merely that the sense of one branch is (correctly) inverted, leading LLVM down a subtly different path in the end.
A sledge-hammer fix is quite simple, but I'm still thinking about how to get a better fix which doesn't pessimize a bunch of cases.
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #7 from Nicolai Hähnle nhaehnle@gmail.com --- Created attachment 135749 --> https://bugs.freedesktop.org/attachment.cgi?id=135749&action=edit patch for related issue
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #8 from Nicolai Hähnle nhaehnle@gmail.com --- Created attachment 135750 --> https://bugs.freedesktop.org/attachment.cgi?id=135750&action=edit patch to fix this issue
Two patches for LLVM attached that fix the issue for me.
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #9 from Philip Rebohle philip.rebohle@tu-dortmund.de --- Thanks for looking into this. While these two patches certainly improve the situation significantly, they don't fix the issue entirely.
The remaining artifacts are hardly noticable in the original apitrace, so I created another one (2.9G, bz2-compressed): https://mega.nz/#!FHxSgZjL!0MzRQhI6W6Lk0beTv9LrPxDWjNTQveDuXVq3Mz2s7-M
Note that my original workaround still works with the LLVM patches applied.
Tested with updated LLVM-git, revision accb337c76e.
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #10 from Philip Rebohle philip.rebohle@tu-dortmund.de --- Created attachment 135780 --> https://bugs.freedesktop.org/attachment.cgi?id=135780&action=edit Frame 2320 with patched LLVM
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #11 from Philip Rebohle philip.rebohle@tu-dortmund.de --- Created attachment 135781 --> https://bugs.freedesktop.org/attachment.cgi?id=135781&action=edit Frame 2800 with patched LLVM
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #12 from Samuel Pitoiset samuel.pitoiset@gmail.com --- Any news on this? The two LLVM patches have been pushed ~2 weeks ago. Can you still reproduce the issue? Thanks!
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #13 from Fabian Maurer dark.shadow4@web.de --- Created attachment 139196 --> https://bugs.freedesktop.org/attachment.cgi?id=139196&action=edit Video showing minor artifacts
While it's a lot better, it's not fully fixed yet. I added a video to show a few artifacts, just look at the top middle area where the lightning jumps around.
Created from this apitrace: http://www.mediafire.com/file/cch4roemrd2gi1f/AutomataArtifacts.7z
https://bugs.freedesktop.org/show_bug.cgi?id=103743
--- Comment #14 from Timothy Arceri t_arceri@yahoo.com.au --- (In reply to Fabian Maurer from comment #13)
Created attachment 139196 [details] Video showing minor artifacts
While it's a lot better, it's not fully fixed yet. I added a video to show a few artifacts, just look at the top middle area where the lightning jumps around.
Created from this apitrace: http://www.mediafire.com/file/cch4roemrd2gi1f/AutomataArtifacts.7z
Those artifacts from the video seem to be gone with recent Mesa (19.1-devel) and LLVM (9.0-devel). Can you confirm if that this is indeed fixed?
https://bugs.freedesktop.org/show_bug.cgi?id=103743
Timothy Arceri t_arceri@yahoo.com.au changed:
What |Removed |Added ---------------------------------------------------------------------------- Resolution|--- |FIXED Status|NEW |RESOLVED
dri-devel@lists.freedesktop.org
-
bugzilla-daemon＠freedesktop.org