[PATCH v2] Support Intel AVX10.2 minmax, vector copy and compare instructions

[Jan Beulich]

On 13.12.2024 09:11, Haochen Jiang wrote:
> From: "Mo, Zewei" <zewei.mo@intel.com>

In your submissions, may I ask that you omit Mo's @intel address from the
recipients list? It has been bouncing for me all the time in recent replies
I sent.

> --- /dev/null
> +++ b/gas/testsuite/gas/i386/avx10_2-256-miscs.s
> @@ -0,0 +1,135 @@
> +# Check 32bit AVX10.2/256 instructions
> +
> +	.arch generic32
> +	.arch .avx10.2/256
> +	.text
> +_start:
> +	.irp m, bf16, d, h, s
> +	vminmaxp\m	$123, %xmm4, %xmm5, %xmm6
> +	vminmaxp\m	$123, %ymm4, %ymm5, %ymm6
> +	vminmaxp\m	$123, 0x10000000(%esp, %esi, 8), %ymm5, %ymm6{%k7}
> +	vminmaxp\m	$123, 4064(%ecx), %ymm5, %ymm6
> +	vminmaxp\m	$123, 0x10000000(%esp, %esi, 8), %xmm5, %xmm6{%k7}
> +	vminmaxp\m	$123, 2032(%ecx), %xmm5, %xmm6
> +	.endr

Elsewhere you indicated that the p is slated to be dropped from the BF16
insns. To limit the impact in case that ends up happening, could you switch
to

	.irp m, pbf16, pd, ph, ps
	vminmax\m	$123, %xmm4, %xmm5, %xmm6
	vminmax\m	$123, %ymm4, %ymm5, %ymm6
	vminmax\m	$123, 0x10000000(%esp, %esi, 8), %ymm5, %ymm6{%k7}
	vminmax\m	$123, 4064(%ecx), %ymm5, %ymm6
	vminmax\m	$123, 0x10000000(%esp, %esi, 8), %xmm5, %xmm6{%k7}
	vminmax\m	$123, 2032(%ecx), %xmm5, %xmm6
	.endr

please? (Same in other parts of the patch then, of course.)

Okay preferably with this adjustment.

Jan