[Mono-list] Benchmark information needed.

Paolo Molaro lupus@ximian.com
Sat, 17 Aug 2002 12:03:58 +0200


On 08/16/02 Miguel de Icaza wrote:
>    I have this patch that implements spurious jump elimination in the
> Mono runtime, but on my P4 computer I never get any actual improvements
> on the execution time.
> 
>    I would appreciate if people can patch their Mono runtime with this
> patch and run the enclosed test and report back the times as well as the
> machine type they are using.

Note: I changed the outer loop count from 1000 to 100: a shorter test
gives less possible variations due to system activity and the numbers
give interesting information even with shorter run times.

On a PIII 1.13.

test compiled with mcs (before patch):

Elapsed: 00:00:08.0951790
Elapsed: 00:00:08.0977600
Elapsed: 00:00:08.0984560
Elapsed: 00:00:08.1022560

test compiled with mcs (after patch):

Elapsed: 00:00:07.9334740
Elapsed: 00:00:07.9341380
Elapsed: 00:00:07.9408900
Elapsed: 00:00:07.9643370

So the patch gives a 2% speedup.

test compiled with csc (before patch):

Elapsed: 00:00:06.3110210
Elapsed: 00:00:06.3075340
Elapsed: 00:00:06.3037640
Elapsed: 00:00:06.3082570

test compiled with csc (after patch):

Elapsed: 00:00:06.3875170
Elapsed: 00:00:06.3900300
Elapsed: 00:00:06.3892530
Elapsed: 00:00:06.3943290

About 1% slowdown in this case. Note, however, that the code produced by
csc makes the run time faster by about 20%...

It's also interesting to note that the ms jit is not impacted as much by
the mcs code (on a PII 440):

Elapsed: 00:00:08.2718944
Elapsed: 00:00:08.3019376
Elapsed: 00:00:08.2819088

versus the csc-compiled test times:
Elapsed: 00:00:08.1216784
Elapsed: 00:00:08.1216784
Elapsed: 00:00:08.1316928

So, less than 2% slowdown versus the 20% slowdown in the mono jit.

lupus

-- 
-----------------------------------------------------------------
lupus@debian.org                                     debian/rules
lupus@ximian.com                             Monkeys do it better