It seems that they've made some progress: https://www.youtube.com/watch?v=EeEVrx7rH8w
Do you think it could be playable at some point?
Do you think it could be playable at some point?
Dolphin on the pi 4?
|
08-25-2019, 09:57 AM
It seems that they've made some progress: https://www.youtube.com/watch?v=EeEVrx7rH8w
Do you think it could be playable at some point? 08-26-2019, 12:39 AM
I can't see it getting full speed honestly. That said though if they could that would be amazing.
08-26-2019, 02:02 AM
(08-25-2019, 09:57 AM)quetzal Wrote: It seems that they've made some progress: https://www.youtube.com/watch?v=EeEVrx7rH8w I'm actually quite impressed by this performance, I expected worse. 08-26-2019, 03:13 AM
I've managed to compile Mesa with -Dgallium-drivers=vc4,v3d,swrast,kmsro for Buster arm64 and changed the one line to enable EXT_buffer_storage. Unfortunately this results in a segmentation fault. If I revert that line and install the recompiled Mesa 19.3, it can run Luigi's Mansion as before.
I compiled with -DCMAKE_BUILD_TYPE=RelWithDebInfo and got stack trace info. I cannot run gdb live--it results in an unrelated crash in the CPU thread, independent of everything else--so I had to save and load a core dump. This seems to remove some info like thread names but retains plenty of other details. It's crashing in CopyEFBToCacheEntry(). This gist shows the backtrace along with local variables from the two stack frames that could be informative. Any insights or suggestions for next steps would be appreciated. The root cause could lie in Dolphin, Mesa v3d, or both, and with more understanding we could apply appropriate fixes or workarounds. Is there any unit test to ensure EXT_buffer_storage is working correctly? Would a FIFO log help here, and is it even possible to capture one on a segfaulting run? (08-23-2019, 04:50 PM)JosJuice Wrote:(08-23-2019, 09:04 AM)jdonald Wrote: My initial thought was that for __arm__ || __aarch64__ we could have PreferGLES default to True, otherwise False. Fair point. I had only considered the TX2, but apparently TX1 supports only up to GLES 3.1. Looks like it won't be possible to statically determine a good default. 08-26-2019, 06:42 AM
(08-26-2019, 03:13 AM)jdonald Wrote: I've managed to compile Mesa with -Dgallium-drivers=vc4,v3d,swrast,kmsro for Buster arm64 and changed the one line to enable EXT_buffer_storage. Unfortunately this results in a segmentation fault. If I revert that line and install the recompiled Mesa 19.3, it can run Luigi's Mansion as before. I attempted that as well but after cloning mesa I realised I had no clue what flags to set to get it to compile D: Might be worth getting that pi engineers input for a possible solution. 08-26-2019, 12:32 PM
I used these guides for cross-compiling and testing Mesa:
* https://github.com/anholt/mesa/wiki/Buil...sa-for-VC4 * https://github.com/anholt/mesa/wiki/Rasp...ss-compile For targeting 64-bit you need to change all instances of arm-linux-gnueabihf to aarch-linux-gnu, and :armhf to :arm64. I think the most annoying part is that prerequisites differ on Buster (x11proto-dri2-dev may no longer exist, some additional wayland development libraries required) compared to what's on the wiki. You get to figure that parts out as meson initially complains about missing packages. Also, on the Pi 3B+ it was enough to specify -Dgallium-drivers=vc4 but some users on the forums found out the hard way that it now needs all of vc4,v3d,kmsro (08-26-2019, 06:42 AM)bomblord Wrote: Might be worth getting that pi engineers input for a possible solution. Good idea. Have called out 6by9 in the EXT_buffer_storage thread on the Pi forums. 08-26-2019, 07:38 PM
And yet another impressive video: https://www.youtube.com/watch?v=1M-TetK1MtQ
Super Mario Sunshine running at stable 30FPS :O (08-26-2019, 07:38 PM)quetzal Wrote: And yet another impressive video: https://www.youtube.com/watch?v=1M-TetK1MtQ Not really... Once that video gets into gameplay, the speed doesn't go over 85% and often stays around 75%. And judging by how the goop collision doesn't work correctly, they most be using the inaccurate settings that speed up Super Mario Sunshine significantly (or maybe the drivers just don't support EFB copies correctly?) EDIT: Ah yes, upon a closer look they deleted the whole Sys/GameSettings/ folder. That would cause the Super Mario Sunshine inaccuracies. (Please never delete that whole folder.)
speed dolphin up massively, just delete all the files Dolphin needs to actually emulate your games properly lmao
We should add a "fast" button to dolphin. It's just the null backend and nothing gets fed to the JIT. Solid 60 FPS baby! What do you mean you want to actually play your games? 08-27-2019, 03:50 AM
(08-27-2019, 03:10 AM)Helios Wrote: speed dolphin up massively, just delete all the files Dolphin needs to actually emulate your games properly lmao You have my full permission and support. Just to see people's reactions. |
« Next Oldest | Next Newest »
|