[15] RFR(S): 8248845: AArch64: stack corruption after spilling vector register

[Andrew Haley]

On 08/07/2020 10:28, Nick Gasson wrote:
> I wonder whether we should only do scheduling on AArch64 for in-order
> CPUs? I tried SPECjvm with/without OptoScheduling on a few different
> AArch64 systems but couldn't get conclusive results either way.

Arm has always been difficult to performance tune because it's an
architecture, not a processor. I didn't test on Arm's own designs at
all for the first few years of AArch64.

We schedule based on a conversation I had with Arm architects, which
basically amounted to "Schedule for in-order cores and the out-of-order
cores will look after themselves." I'd prefer not to disable scheduling
simply because it's buggy; that feels wrong to me. But if it's making
things worse, then we can do so.

-- 
Andrew Haley  (he/him)
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
https://keybase.io/andrewhaley
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671