[PATCH v2] Support Intel AVX10.2 satcvt instructions

[Jan Beulich]

(dropping Mo Zewei's email, as it has been bouncing for me before)

On 03.12.2024 08:18, Haochen Jiang wrote:
> --- /dev/null
> +++ b/gas/testsuite/gas/i386/avx10_2-256-satcvt.s
> @@ -0,0 +1,273 @@
> +# Check 32bit AVX10.2/256 instructions
> +
> +	.arch generic32
> +	.arch .avx10.2/256
> +	.text
> +_start:
> +	.irp s, "", u
> +	vcvtnebf162i\s\()bs	%xmm5, %xmm6

As you're doing this neatly here, ...

> --- /dev/null
> +++ b/gas/testsuite/gas/i386/avx10_2-512-satcvt.s
> @@ -0,0 +1,151 @@
> +# Check 32bit AVX10.2/512 instructions
> +
> +	.arch generic32
> +	.arch .avx10.2/512
> +	.text
> +_start:
> +	.irp s, ibs, iubs

... why isn't this

	.irp s, "", u

too?

Further, what does "s" stand for here? As requested in my v1 comments,

	.irp u, "", u

would lead to more natural to read contents, e.g.

	vcvtnebf162i\u\()bs	%zmm5, %zmm6

(having a 'u' where that or nothing would appear, instead of an 's'
which may be confused with the 's' insn suffix, standing for "saturate"
aiui).

Okay with respective adjustments (also to the 64-bit counterpart).

Jan