Jim Huang

> I guess I’d have to reach out to Brian Greenstone about this. The last point isn’t even true anymore with Nanosaur staying „Mac only code“, correct? Besides the above...

By the way, shall we support `__ARM_FEATURE_SIMD32`? [ARM C language extensions specification](https://developer.arm.com/documentation/ihi0053/d/?lang=en) describes the `__ARM_FEATURE_SIMD32` macro as follows: > `__ARM_FEATURE_SIMD32` is defined to 1 if the 32-bit SIMD instructions are...

> I can see only two options: > 1. function templates that take the immediate as a template parameter (i.e. how the old implementation worked) > 2. lambda functions that...

> Here's `_mm_shuffle_ps_default` as an example. I have included both template and lambda implementations for demonstration purposes, but we only need to pick one. Assume `SSE2NEON_RETURN` is defined as `return`...

> Notably, the DECL macro above is only required by the template implementation. It is possible to get rid of the CALL macro too with the lambda implementation, but it...

> However, for SSE2NEON, we need ARMv7/non-crypto-ext counterparts. The portable implementations: * QEMU [target/i386/ops_sse.h](https://github.com/qemu/qemu/blob/master/target/i386/ops_sse.h): Check `helper_aesdec` and `helper_aesdeclast`. * [WebAssembly/design Issue #1433](https://github.com/WebAssembly/design/issues/1433)

See also [sse4.2: added the implementation for mm_cmpestra](https://github.com/simd-everywhere/simde/pull/295).

Thank @Cuda-Chen for contributing!

> I'm missing intrinsic `_mm_cmpestri`. Recently, [SSE2NEON](https://github.com/DLTcollab/sse2neon) implements both `_mm_cmpestri` and `_mm_cmpestrm`.

@oucs638, would you rework this patch?