Yes, that would be nice but perhaps with an explicit option in the module to enable it with the fallback scalar code with a clear warning when you hit the error that enabling it is possible but with a caveat that it will be sub-optimal.
1 Like
I think the vec2 operator on intel should be _mm_div_pd
3 Likes
I’d like to bump this – it is pretty much impossible to convert anything to SIMD using the built in classes without being able to do a division
2 Likes
I’ve run in to this as well – it basically means I can’t use juce::dsp::SIMDRegister at all, without hacking up JUCE.
FWIW, if one uses sse2neon.h instead of arm_neon.h, you do get _mm_div_ps defined.