RFR: 8364305: Support AVX10 saturating floating point conversion instructions [v18]

[Mohamed Issa]

On Fri, 26 Sep 2025 21:32:34 GMT, Vladimir Ivanov <vlivanov at openjdk.org> wrote:

> I haven't thoroughly reviewed the patch, but what caught my eye is that avx10 and avx10_2 are used interchangeably which adds confusion. My recollection is that AVX10.1 is equivalent to AVX512 set of capabilities. Can we uniformly refer to AVX10.2 as AVX10 in the code base then?

In the future, we could get AVX10.3 which would be superset of AVX10.2. So, these new floating point conversion instructions would apply to AVX10.3 as well. With that in mind, I think it's useful to have the generic AVX10 label as an umbrella and then distinguish between sub-versions when strictly necessary.

Is the main issue that each AVX10 reference (e.g., `C2_MacroAssembler::vector_castF2X_avx10`) isn't descriptive enough about which minimum sub-version is in use at a quick glance? Or are other concerns that should be addressed?

-------------

PR Comment: https://git.openjdk.org/jdk/pull/26919#issuecomment-3340824400