Petr Kobalicek
Petr Kobalicek
Should be fixed by 2f79eb23ab0b00c5a00535c4f2477c15e626dfa4
I will check it out. In general it should compile, but without JIT. I'm not sure now whether JIT has to be explicitly disabled or whether cmake is able to...
@jhmaloney Hi, can you try again with the recent master? I tried it on RPI4 with GCC/Clang and I was able to fix some issues, not sure it fixes all...
I'm not sure what the issue is - I tried with GCC 10, which is in Ubuntu, and that worked without any problem. Maybe it needs some target specific -m...
Well, I think this is an issue in Blend2D in the end - it should either set `-mneon` in `CMakeLists.txt` or it should check for neon in SIMD helpers. I...
Yeah, I think `CXXFLAGS="-mneon" cmake ...` should work without even editing CMakeLists - I'm not sure how to detect the target platform in cmake yet to be able to do...
BTW are you using 64-bit ARM? Because in 64-bit mode NEON should be implied (maybe with another name ASIMD). I think `-mneon` is only required on 32-bit ARM/Thumb
I see, btw I'm testing Blend2D on 64-bit RPI 4 (8GB RAM model), but I have installed Ubuntu on that machine (not raspbian). I still think that this is an...
Blend2D is using premultiplied 32-bit ARGB pixel format - in little endian it's BGRA in memory - I think your cv::imwrite() uses a different format.
To update this: There is some work that has been already merged and should solve basic issues with NaNs, but this is not thoroughly tested. Keeping open, we need fuzzing...