I'm trying to turn on a flag (-inline-generic-switch-cost) introduced in r301649 which improved the inline cost heuristic for switch. In my experiment for LLVM test suite and spec2000/2006 in AArch64, +17.82% performance and -8% code size reduce was observed in spec2000/vertex with O3 LTO. No significant code size / performance regression was found in O3/O2/Os.
Before we turn this flag on by default, I want this change to be tested on other targets as well as for other applications. Can anyone please take a change to test this and share the result in D32653.
Can anyone please take a chance to test the new switch inline cost heuristic (r301649). In my experiment for spec2000/spec2006/llvm-ts in AArch64, significant performance (+17.8%) and size (-8%) benefit was observed in spec2000/vertex with LTO while no significant code size / performance regression was found in O3/O2/Os. I posted D32653 to enable the flag (-inline-generic-switch-cost) for your test. It will be really great if this heuristic is tested in different targets and for other applications.
We are planning to test this for x86/CPU2006 suite. Most like in a week or so.
We ran cpu2006 INT rate for x86. We didnt see any perecptible positive or negative swing in any of the bms.
Thank you very much for the test. I guess you didn’t see any obvious change both in performance and size right? Can you please share about your config?