Example of SIMD-like behavior using intrinsics?

[qpwo]

It is discussed here and here, but I'm no c++ wiz and I can't find an example in vlang, so:

Can someone provide a minimal example of broadcasting an operation over an array in vlang?

How could I achieve something like [1, 2, 3, 4] .* 2 (in whatever syntax)?

[JalonSolov]

If you mean "multiply all the elements by 2", that would be [1, 2, 3, 4].map(it * 2).

Otherwise, I don't know what you meant by .* 2

[JalonSolov]

Like I said... V doesn't currently support it. It is definitely something that would have to be added to the compiler.

Nobody has done any work on it so far, and I haven't seen it mentioned in any priority lists so far. Doesn't mean it won't be handled, just no estimates on when work might start.

[qpwo]

Cool, fyi SIMD seems to be 10x faster in ideal cases such as dot product.

I'm not sure (1) whether you'd want it be done explicitly in syntax or have the v compiler do it, or (2) if gcc or clang will automatically do this already somehow or (3) how valued matrix math use cases (video processing, ML, etc) are by the v community.

But if vector operations / SIMD were supported then it would make V a very compelling language for such cases.

[qpwo]

Unfortunately I would have no idea where to begin to help out

[JalonSolov]

While you're waiting for SIMD support, you may want to check these links. :-)

https://github.com/vlang/vsl
https://github.com/vlang/vtl

[dumblob]

@qpwo yes it is technically possible (and not as difficult to implement compared to other language stuff which has understandably (much) higher priority).

OTOH you can achieve partial SIMD parallelization already now just by using the vsl and vtl libraries @JalonSolov suggested (as these libraries are partially written in a way which allows the compiler to get a bit wild about aggressive optimizations incl. SIMD).

There is also blas V module/library and other wrappers to known highly optimized math/scientific/vector/matrix/3D/... libraries etc. So I think you have several things to try out and chances are they will serve your purpose as well (or close to that) as if you used direct SIMD intrinsics in V 😉 (from my experience such low-level use of SIMD is very unintuitive and without enough experience there is a high chance there will be either no gains or even slight slow down compared to traditional V code without intrinsics!).

[Wajinn]

While you're waiting for SIMD support, you may want to check these links. --JalonSolov

https://github.com/vlang/vsl
https://github.com/vlang/vtl

There is also blas V module/library and other wrappers to know... --dumblob

Such good suggestions, maybe needs to make sure it's not buried in the replies and possibly overlooked.