Is the trace supposed to crash? Because that replays perfectly fine here.
Haven't managed to run the demo yet, seems to be an enourmous pain in the arse to set up especially since half the winetricks verbs don't seem to exist anymore.
Yeah, idk, the demo crashes inside nvcuda regardless of whether I include that DXVK commit in my build or not. This is with an RTX 4070.
This is either going to require a much simpler / more reliable way to reproduce, or you'll have to debug it on your own since I cannot reproduce this using the trace either.
It crashes when playing back using 570.133.07 and latest DXVK master (732b1664460a81f3fa8e1b289ce3f1a2d76a302a)
The demo is a bit of a pain, so that was why i created the apitrace so setting all that crud up would not be needed. Tested this on both 4070 and 2070 card with the same driver on 2 different systems and it crashes on both. (2 different machines with Ubuntu, although same DXVK compiled binaries).
Now.. downloading the artifact from https://github.com/doitsujin/dxvk/actions/runs/14112349045 did not seem to crash.. I usually compile native on my Ubuntu 24.04 system, so i thought it would work.. but something weird going on there i suppose
https://github.com/doitsujin/dxvk/actions/runs/14112349041
Crashes for me.. (MingW version).. strange that it would not crash for you?
Again, does replaying the trace crash on your end?
info: Memory type mask for sparse resources: 0x2
info: Memory type mask for buffer resources: 0x1f, usage: 0x19ff
info: DXVK: Read 2 valid state cache entries
info: DXVK: Graphics pipeline libraries supported
16390: debug: apitrace: warning: caught exception 0xc0000005
apitrace: warning: caught exception 0xc0000005
16390: error: caught an unhandled exception
wine: Unhandled page fault on read access to FFFFFFFF at address 779DC0CB (thread 012c), starting debugger...
WineDbg attached to pid 0128
0744:fixme:dbghelp:elf_search_auxv can't find symbol in module
Unhandled exception: page fault on read access to 0xffffffff in wow64 32-bit code (0x779dc0cb).
0744:fixme:dbghelp:elf_search_auxv can't find symbol in module
0744:fixme:dbghelp_msc:codeview_snarf Unexpected S_FRAMEPROC -1 (00000000 00000000) 1ef4
0744:fixme:dbghelp_msc:codeview_snarf Unsupported symbol id 1180
0744:fixme:dbghelp_msc:dump 00000000: 0a 00 80 11 38 55 02 00 03 00 00 00 ....8U......
0744:fixme:dbghelp_msc:codeview_snarf Unsupported symbol id 1180
0744:fixme:dbghelp_msc:dump 00000000: 0a 00 80 11 64 67 07 00 02 00 00 00 ....dg......
0744:fixme:dbghelp_msc:codeview_snarf Unsupported symbol id 1180
0744:fixme:dbghelp_msc:dump 00000000: 0a 00 80 11 d9 78 04 00 02 00 00 00 .....x......
When using wine ./apitrace replay SupersonicSled.trace using the apitrace binaries from:
https://github.com/apitrace/apitrace/releases/tag/12.0
Yeah, dunno, works fine even with the artifact here.
Takes a while to get going because the demo is fucking stupid and creates hundreds of D3D11 devices for literally no reason, but it replays fine both on NV and AMD here. This is on 570.86.16 though.
I could try the old 570.86.16 tomorrow to check it makes a difference. Odd that it would work fine when using the artifacts from the msvc build tho?
Can you test this branch: https://github.com/doitsujin/dxvk/tree/no-bcmpeq
I don't see a single reason why this would be broken, but this wouldn't be the first time that mingw-gcc completely ignores alignas.
Can you test this branch: https://github.com/doitsujin/dxvk/tree/no-bcmpeq
I don't see a single reason why this would be broken, but this wouldn't be the first time that mingw-gcc completely ignores
alignas.
Yep, plays nice with that one 👍
Please confirm whether the PR above works too, this is a more correct fix.
Please confirm whether the PR above works too, this is a more correct fix.
Yes, using this input-layout-alignment PR fixes both the apitrace playback and running the demo for me 👍 Thank you 😄
Why this playback worked for you is a bit strange to me tho.. I tests both with 570.86.16 and building dxvk(master) using a docker image with Archlinux, but still same issue for me. So maybe some weird Ubuntu issue that causes this?
PS. If the demo crashes on nvcuda:wine_cuModuleLoadFatBinary you can try to rearrange the LD load order to prioritize 32-bit libs. Eg. LD_LIBRARY_PATH=/lib/i386-linux-gnu/:/lib/x86_64-linux-gnu/:$LD_LIBRARY_PATH (It is some extra loading of nvidia helper library that probably silently drop trying if the first load attempt is wrong ARCH..)
There is a old NVIDIA demo called "SuperSonic Sled" that play a videoclip before entering the "game". This videoclip is a .WMV video.
This clip is in a VC1/WVC1 codec using YUV420p pixel format and WMAPro audio
I created a apitrace capture using
apitrace 12.0 32-bitin windows 11. It can be downloaded here:https://drive.google.com/file/d/1XP0uB9CDjDm6CcevTbyXM8lm6XChYy2M/view?usp=drive_link
This crashes when played with DXVK (master) currently, but i bisected the culprit to:
https://github.com/doitsujin/dxvk/commit/fc2b552d10d09175c5e8d5013e82e3fd3663c1c5
Reverting this commit makes the apitrace play without issue. (I used the same apitrace version in a fresh WINEPREFIX, only installing DXVK to replay)
Playing the video (IntroMovie.wmv that can be found after installing the demo) using MPC-HC from a current K-Lite Codec Pack seems to work fine using d3d11 renderer, aswell as using recent VLC with d3d11 renderer...
If interested in running the actual demo using wine-10.x, it requires a NVIDIA card and seems to need the following additions (using winetricks):
(DXVK needs reverting the above mentioned commit if used with current master)
It seems to use gstreamer for decoding.
The demo itself can be downloaded here:
https://international.download.nvidia.com/downloads/cool_stuff/demos/SetupSupersonicSled.exe