[Qemu-devel] sdl shared buffer performance measurements

[Stefano Stabellini]

Hi all,
I have done some simple tests to measure qemu performances with and
without the shared buffer and sdl.
The tests are really simple: I am just measuring CPU usage and fps on
the host while running the same screensaver on the guest (Windows XP).
The following are the average results:

1)Host 32 bpp, Guest 16 bpp:

SDL shared: CPU 13% average, fps 33
SDL unshared: CPU 16% average, fps 33


2)Host 32 bpp, Guest 24 bpp:

SDL shared: CPU 19% average, fps 33
SDL unshared: CPU 20% average, fps 33

It is seems from the data that sdl has some kind of optimization that
makes qemu perform better converting 16 bpp into 32 bpp compared to the
old vga_template conversion code.
In the 24 bpp case the difference is very small because probably there
isn't much to optimize.
I have to add that I don't think these data are able to fully capture
the advantage given by avoiding the memcpy's, but at least we know that
sdl color conversion routines are not going to make qemu any slower.

Regards,

Stefano