On Mon, 28 Aug 2023, Adhemerval Zanella Netto wrote:

Hi Richard,

Your patch caused a regression [1] on aarch64-linux-gnu. Would you please investigate? I did a quick analysis and it seems that for test_copy_lane_f32, test_copy_lane_s32, test_copy_lane_u32, gcc is now generating zip1 instead of a ins; which does not seem fully correct.

Let me know if you need any assistance in reproducing these.

The testsuite likely needs adjustments or in case the old code is better for some reason aarch64 vec_perm_const handling needs to special case the insert permutes.

Thanks!

[1] https://ci.linaro.org/job/tcwg_gnu_cross_check_gcc--master-aarch64-build/912...