Integrated: 8302167: Avoid allocating register in fast_lock()

Roman Kennke rkennke at openjdk.org
Fri Feb 10 18:15:53 UTC 2023


On Thu, 9 Feb 2023 18:09:08 GMT, Roman Kennke <rkennke at openjdk.org> wrote:

> In x86's C2MacroAssembler::fast_lock(), we are never really using the cx1Reg register, except in the DiagnoseSyncOnValueBasedClasses path, where we can just as well use the scrReg register.
> 
> What's worse, in x86_32.ad, we don't allocate the cx1 register and pass noreg, so the DiagnoseSyncOnValueBasedClasses would be broken on x86_32.
> 
> We can simply avoid allocating that extra register (which may help performance under register pressure), and use scr in DiagnoseSyncOnValueBasedClasses.
> 
> Testing:
>  - [x] tier1
>  - [ ] tier2

This pull request has now been integrated.

Changeset: db1b48ef
Author:    Roman Kennke <rkennke at openjdk.org>
URL:       https://git.openjdk.org/jdk/commit/db1b48ef3bb4f8f0fbb6879200c0655b7fe006eb
Stats:     5 lines in 2 files changed: 1 ins; 0 del; 4 mod

8302167: Avoid allocating register in fast_lock()

Reviewed-by: kvn, kdnilsen

-------------

PR: https://git.openjdk.org/jdk/pull/12499


More information about the hotspot-compiler-dev mailing list