Wed Mar 17, 2021 1:53 pm
I'd actually love to see how the M1 Mac does GZDoom in on stress maps, so im all eyes on any benchmarks
Wed Mar 17, 2021 1:56 pm
One quick test shows Frozen Time at 59 fps in 1980x1080 with the Vulkan backend. That's a really impressive result.
Sat Mar 20, 2021 2:34 am
Does the SoftPoly renderer work on the M1 Mac?
Sat Mar 20, 2021 2:39 am
I see no reason why it shouldn't. Of course it has no SSE support there. But what's the point anyway? It doesn't really have any advantage over Vulkan/OpenGL.
Sat Mar 20, 2021 3:44 am
I am just asking since there was a report around a month ago or so claiming that SoftPoly crashes on macOS, not to mention the renderer crashing on Raspberry Pi 4 but otherwise working on the Cortex-A53 CPUs.
Sat Mar 20, 2021 4:05 am
Find someone with the hardware and willingness to fix the crash. Or giveaway that hardware to someone who can (and want to) fix it. Otherwise, just wait.
Sat Mar 20, 2021 5:09 am
I'm almost fairly sure that if it can be fixed in Linux on the Raspberry Pi, the fix will work for Apple M1 too, if it's the same crash.
That being said, yeah it isn't the highest priority right now since OpenGL/Vulkan should be working with it. And I very much doubt someone is just going to "give away" a M1 macbook. (Though honestly I wouldn't mind having one
What I find odd though, is that it doesn't crash in Windows on the same hardware, at least it didn't last I checked.
Sat Mar 20, 2021 5:14 am
I don't know if it will fix the issue but Beloko mentioned this to me some time ago:
if using newer Clang compilers there is a VERY important flag you need to set for the compiler which fixes some alignment issues (The optimiser creates 'bad' 64bit load/save instructions, which when unaligned cause a trap)
Add this to compile flags:
Sat Mar 20, 2021 5:26 am
Does debug target crash? If not, it's more like our code needs to be fixed instead of enabling unaligned access.
Also, de-optimization should be done for particular code fragment but not for whole code base.
Sat Mar 20, 2021 7:42 am
So it's a Clang bug after all? I don't think I compiled GZDoom as a release build on my Amlogic S912 hardware so I will need to check the Softpoly renderer again both with GCC and Clang when I compile it as a release build.
Sat Mar 20, 2021 10:17 am
Just for the record, if someone gives Rachael a M1I'll make sure helping her track down whatever unaligned store there is.
The slightly cheaper approach is for someone to get the call stack for the crash and post it here. That will reveal what code is causing the crash and what buffer it is writing to when doing so.
Sat Mar 20, 2021 12:59 pm
I recently compiled GZDoom with Clang for ARM64 as a release build and successfully ran it with the SoftPoly backend and couldn't notice any crashes on my Amlogic S912. I am writing this off as a bug on ARM64 CPUs implementing revision 8.2 and later.
Tue Sep 21, 2021 10:18 am
Would it not be easier to register this as a git submodule in the gzdoom git project to clear up this confusion? It doesn't seem like that much of a stretch given that other ports are doing similar things for third-party libs.
Tue Sep 21, 2021 10:34 am
Tue Sep 21, 2021 10:52 am
Submodules are such a major nuisance that it's not a viable option.
Powered by phpBB © phpBB Group.
phpBB Mobile / SEO by Artodia.