Increased execution time with TCI in latest git master (was: Re: [PULL 00/46] tcg patch queue)

[Stefan Weil]

Am 05.02.21 um 23:56 schrieb Richard Henderson:

> The following changes since commit d0dddab40e472ba62b5f43f11cc7dba085dabe71:
>
>    Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging 
> (2021-02-05 15:27:02 +0000)
>
> are available in the Git repository at:
>
>    https://gitlab.com/rth7680/qemu.git tags/pull-tcg-20210205
>
> for you to fetch changes up to fb6916dd6ca8bb4b42d44baba9c67ecaf2279577:
>
>    accel: introduce AccelCPUClass extending CPUClass (2021-02-05 10:24:15 -1000)
>
> ----------------------------------------------------------------
> TCGCPUOps cleanups (claudio)
> tcg/s390 compare fix (phil)
> tcg/aarch64 rotli_vec fix
> tcg/tci cleanups and fixes


Hi Richard,

I tested pull-tcg-20210205 on a sparc64 machine.

Timing results for BIOS booting ("./qemu-system-x86_64", no image):

Before pull-tcg-20210205 with my latest patches for TCI:

TCG native: 8 s user time, TCG interpreter: 40 s user time

With pull-tcg-20210205 and 
https://patchwork.kernel.org/project/qemu-devel/patch/20210127123446.2037721-1-sw@weilnetz.de/:

TCG native: 8 s user time, TCG interpreter: 50 s user time

So for this test case TCI was about 5 times slower than native TCG, but 
the latest changes increased the execution time by about 25 %.

I am still searching what caused this detoriation. My first suspect was 
thread local storage, but that wasn't it. Do you have any idea?

Stefan