Dmitry Babokin

> If I may suggest an improvement (not sure whether this is the topic here): > > I believe if the loader is not able to find an ISA implementation...

What target CPU do you care about? I.e. what CPU would be good proxy performance-wise to the hardware that you care about?

WASM support is intended to support 128 bit SIMD extension. But unfortunately the contributor who committed initial support didn't have time to complete it. We also don't have resources for...

Seems you are right. Here's the official status of WebAssembly features support: https://webassembly.org/roadmap Safari is the only browser not supporting [Fixed-width SIMD](https://github.com/WebAssembly/simd/blob/master/proposals/simd/SIMD.md) extension at the moment. They seem to be...

> It's not just Safari. Yes, sure, Safari in this context means desktop Safari + all web browsers on iPhone. > What about not reverting to a vector of size...

Seems placeholder declarations are missing in `builtins/target-wasm-i32x4.ll`. [rcph_decl()](https://github.com/ispc/ispc/blob/575929afa9e9ae0cc44e766b385bf1d70f7dc938/builtins/util.m4#L8104) macro call and some other should be in `builtins/target-wasm-i32x4.ll`. Note that WASM is not fully functional target yet. @aschrein did great job...

Note that we have ``memory_barrier()``, which acts as ``mfence()``. We can add ``load_barrier()`` and ``store_barrier()`` (these name are more in line with ``memory_barrier()`` naming).

Just in case the link becomes unavailable, here's the code: ```c // Copyright Epic Games, Inc. All Rights Reserved. struct FVector4 { float V; }; struct WideFVector4 { float V[programCount];...

What you need to happen, is to trigger generation of "pshufb" instruction somehow (in ISPC, C++, or using any other tool/language). Using this instruction you can efficiently rearrange bytes in...

That's an interesting way of doing meta-programming. Thanks for linking Jinja!