John IV |
IV/Play, MAME, MAMEUI
|
|
|
Reged: 09/22/03
|
Posts: 1981
|
Loc: Washington, USA
|
|
Send PM
|
|
|
Couple of data points for .197 vs. .196 benches.
#375902 - 04/30/18 01:30 AM
|
|
|
Was re-doing some of the bench runs and noticed a couple of big deltas beyond the regular testing variance noise.
http://www.mameui.info/Bench.htm
Road Blasters & Hard Drivin' dropped 50% or so. Rival Schools gained 32%. Somewhat academic w/ newer processors but mildly interesting.
Note: I also added vf2 and sushibar going forward.
|
|
|
|
Re: Couple of data points for .197 vs. .196 benches.
[Re: John IV]
#375903 - 04/30/18 01:35 AM
|
|
|
> Was re-doing some of the bench runs and noticed a couple of big deltas beyond the > regular testing variance noise. > > http://www.mameui.info/Bench.htm > > Road Blasters & Hard Drivin' dropped 50% or so. Rival Schools gained 32%. Somewhat > academic w/ newer processors but mildly interesting. > > Note: I also added vf2 and sushibar going forward.
that hard drivin' one is unexpected
if that's across the whole driver it probably pushes racedrivpan below 100%
unless it's an anomaly it needs investigating for sure
not sure where that rival schools boost would come from either, I'm not aware of any significant PSX optimizations, might be worth making sure it's still working properly.
Edited by Haze (04/30/18 01:36 AM)
|
|
|
|
Re: Couple of data points for .197 vs. .196 benches.
[Re: Haze]
#375905 - 04/30/18 03:01 AM
|
|
|
> not sure where that rival schools boost would come from either, I'm not aware of any > significant PSX optimizations, might be worth making sure it's still working > properly.
I changed some stuff around in src/emu to make CPU cores run slightly more efficiently when the debugger isn't being used. It gives a small improvement on things like Neo-Geo, but it's more noticeable on stuff with higher instructions per second like PSX.
I don't know what would have caused the drop in the other games.
|
|
|
John IV |
IV/Play, MAME, MAMEUI
|
|
|
Reged: 09/22/03
|
Posts: 1981
|
Loc: Washington, USA
|
|
Send PM
|
|
|
Re: Couple of data points for .197 vs. .196 benches.
[Re: Haze]
#375907 - 04/30/18 04:40 AM
|
|
|
Big drop in racedrivpan as suspected.
.196 mame racedrivpan -bench 90 = 214.15% .197 mame racedrivpan -bench 90 = 96.98%
In-game drops to ~85% during attract screen. 4.5Ghz i7-6700 / 32GB DDR 3200
|
|
|
Tafoid![Administrator Administrator](//www.mameworld.info/ubbthreads/images/adm.gif) |
I keep on testing.. testing.. testing... into the future!
|
|
|
Reged: 04/19/06
|
Posts: 3138
|
Loc: USA
|
|
Send PM
|
|
|
Re: Couple of data points for .197 vs. .196 benches.
[Re: John IV]
#375909 - 04/30/18 08:15 AM
|
|
|
> Was re-doing some of the bench runs and noticed a couple of big deltas beyond the > regular testing variance noise. > > http://www.mameui.info/Bench.htm > > Road Blasters & Hard Drivin' dropped 50% or so. Rival Schools gained 32%. Somewhat > academic w/ newer processors but mildly interesting. > > Note: I also added vf2 and sushibar going forward.
- The slowdown for the Atari games is pretty much across the board for AtariSys1 hardware and later or wherever the new ADC0809 A/D Converter device was added/implemented (April 3-4 timeframe). I'm sure something can be done to improve things there. - rvschool performance increase is what Vas mentioned, adjustment of Debugger hooks (March 27-28 timeframe).
|
|
|
|
Re: Couple of data points for .197 vs. .196 benches.
[Re: Tafoid]
#375910 - 04/30/18 09:14 AM
|
|
|
> > Was re-doing some of the bench runs and noticed a couple of big deltas beyond the > > regular testing variance noise. > > > > http://www.mameui.info/Bench.htm > > > > Road Blasters & Hard Drivin' dropped 50% or so. Rival Schools gained 32%. Somewhat > > academic w/ newer processors but mildly interesting. > > > > Note: I also added vf2 and sushibar going forward. > > - The slowdown for the Atari games is pretty much across the board for AtariSys1 > hardware and later or wherever the new ADC0809 A/D Converter device was > added/implemented (April 3-4 timeframe). I'm sure something can be done to improve > things there.
Hopefully, as it's taken one of the more demanding cases significantly below 100% (from 200%) on the very top-end hardware you can buy. That kind of performance regression isn't acceptable.
|
|
|
|
Re: Couple of data points for .197 vs. .196 benches.
[Re: Tafoid]
#375911 - 04/30/18 11:57 AM
|
|
|
> - rvschool performance increase is what Vas mentioned, adjustment of Debugger hooks > (March 27-28 timeframe).
MooglyGuy deserves some credit for this, since he noticed the hotspots while experimenting with aggressive ARM optimisations.
|
|
|
|
Atari perf regressions fixed *nt*
[Re: John IV]
#375912 - 04/30/18 12:16 PM
|
|
|
> Was re-doing some of the bench runs and noticed a couple of big deltas beyond the > regular testing variance noise. > > http://www.mameui.info/Bench.htm > > Road Blasters & Hard Drivin' dropped 50% or so. Rival Schools gained 32%. Somewhat > academic w/ newer processors but mildly interesting. > > Note: I also added vf2 and sushibar going forward.
|
|
|
John IV |
IV/Play, MAME, MAMEUI
|
|
|
Reged: 09/22/03
|
Posts: 1981
|
Loc: Washington, USA
|
|
Send PM
|
|
|
Re: Atari perf regressions fixed *nt*
[Re: MooglyGuy]
#375943 - 05/01/18 04:20 AM
|
|
|
Thanks MG:
mame racedrivpan -bench 90 = 214.15% mame roadblst -bench 90 = 1096% mame harddriv -bench 90 = 712%
MAME v0.197 (mame0197-145-g7414a8e270)
|
|
|