On Tue, Aug 25, 2026 at 03:59:13PM +0200, [email protected] wrote:
> From: Kyrylo Tkachov <[email protected]>
>
> The positive MIN narrowing test checks that a wide vector MIN is replaced
> with a narrow vector MIN and a widening conversion. The current V2SI to
> V2DI case needs target support for the narrow vector MIN. On 32-bit Arm,
> late target lowering can remove the wide vector type and make the final dump
> check pass even when the simplification did not run.
>
> Limit the positive test to AArch64 and x86 targets. Its x86 options and
> int128 requirement select a suitable mode there. I think this test gives too
> much trouble on other targets (arm, sparc) that may not support the narrow
> modes.
>
> The shared-result test checks that the simplification leaves one MIN. Check
> the first forwprop dump, after the rule has run and before target lowering
> can split the vector MIN. This makes the count independent of the target
> vector modes.
>
> Bootstrapped and tested on aarch64-none-linux-gnu.
> Ok for trunk?
> Thanks,
> Kyrill
>
> gcc/testsuite/ChangeLog:
>
> * g++.dg/tree-ssa/vec-narrow-1.C: Restrict the test to AArch64 and
> x86 targets.
> * g++.dg/tree-ssa/vec-narrow-minmax-2.C: Scan before target lowering.
Ok, thanks.
Jakub
