In C++.
Oups... does it really takes 8secs just to compile ???

C++ version...

IMHO should not have so many 'nice assumptions'... In real life it will require a lot of testing from the calling side

hmmmmm...

Same, it's tragic that the testing framework's It macro makes including <range/v3/all.hpp> impossible. Any brevity you are able to achieve is offset by the gazillion lines of import above it :(

😄 Perfect C++20/23 !!

But you know what : I'm tired with 30mns trying to figure out which include I have to use in rangev3 for each algo/view/.. and line of code !!

Agreed, but I also think that even with those utilities, this task isn't trivial to do correctly. Of course, if you copy everything and turn things into lvalues, it is. But it took me 30 minutes to put together a solution that keeps value categories intact. I assumed there would be tests for that, and also the caching, but apparently not

😃

Thank you for reply + couple of keywords) Im trying to find purpose for "kumite" thing thus bothering people around.. that the spirit, i guess
gotta have some fun with llvm intermediates x)

Of course !!

I was not running for perf...
Use of a rand instead of a mersenne twister, etc.. and even the use of std:experimental::simd etc...

Note on your note: any modern compiler will unroll the loop, and... turn i++ into ++i.. 😀

See my comment on the fork: you will understand why... my old reflex is C-style :-))))) \

Could write it in 6809 8 bit assembly :-)

:-)
You know what ? I'm 60 (yop a gran'pa programmer). Began programming in 1979... with 68000 assembly my boss was a p.. in the ... for ANY SINGLE CPU CYCLE...
Want the whole story: We had to rasterize an A0 plot, black and white. Makes something like 1 or 2 Mpixels. Memory chips were so expensive: my boss gave me a hardware with 256k (yop) of (static) RAM, and a 20Mo (yes, Mo) hard drive and told me: do the shit with what you have :-)
Actually with modern pipelines and cache, had to readapt... Performance is not a matter of cycles, but a matter of data, cache line and prefetch and branch prediction and...

The change to multiplication is brilliant. It's sad that a redundant cast is still needed to avoid the ugly warning

C-style cast makes it 98% close. XD
Your idea with ranges sounds great.