Viewing Issue Advanced Details
ID Category [?] Severity [?] Reproducibility Date Submitted Last Update
00241 Speed Major Have not tried Jan 27, 2008, 23:26 Aug 8, 2008, 17:21
Tester Tafoid View Status Public Platform
Assigned To Resolution Won't fix OS
Status [?] Closed Driver
Version 0.122 Fixed in Version Build
Fixed in Git Commit Github Pull Request #
Summary 00241: All sets in namcos1.c: Major speed drop.
Description Simply stated, there is a larger than usual drop in performance for this driver. Strangely all sets, save one [puzlclub], when tested with MAMEDEV releases (0.120 and 0.121), had decreased performance of approx 28-36%!

There is the complete list of my testing of all sets in NAMCOS1.C. for MAME 0.119 through 0.122.

When investigating the source of the problem, the huge drop is due to something in the release of 0.120u2. This specific release had no direct changes of http://mameload.mameworld.info/namcos1.txt that I could note. The only item of interest I saw that all sets had in common were some sweeping changes that effected many games with M6809 processors / HD63701. Sorting out the initial problems (crashing drivers/missing sound) for these sets was done for the next U release.

Unless it were a driver specific change, which improved or further enhanced the emulation... I personally believe a 28-36% drop in 1 particular driver is due to an emulation mistake or oversight with new code. While not functionally a "bug" in the output - it still deserves investigation.

Can anyone confirm/deny or otherwise explain to me why this large of a performance issue is found in this 1 driver which did not get a direct update?

Follow-up by Atari Ace: I looked into this and discovered the following. The HD63701 core now has timers that are disabled by default. If a particular value is written to the internal registers (see case 0x10 in m6803_internal_registers_w), the timers activate, and if the period is short enough MAME will have to run in smaller timeslices (in my testing I saw it go from about 40 cycles to about 15 cycles per timeslice), thus degrading performance. In the case of puzlclub, the timers never activate so the performance doesn't degrade.
Steps To Reproduce
Additional Information
Github Commit
Flags
Regression Version
Affected Sets / Systems All sets in namcos1.c
Attached Files
 
Relationships
There are no relationship linked to this issue.
Notes
7
User avatar
No.01986
M.A.S.H.
Senior Tester
Aug 6, 2008, 23:48
The NamcoSystem1 games uses 3x M6809 CPUs and they need the M6800 CPU,
see source drivers\namcos1: #include "cpu/m6800/m6800.h".
The M6800 CPU was updated in MAME 0.120u2.
I compiled the old MAME 0.120u2 again and use the old cpu/m6800/m6800.c
code from 0.120u1, and the slowdown gone!
It was definitely the changes in the M6800 core.
User avatar
No.01988
robiza
Developer
Aug 7, 2008, 08:14
original bug say "had decreased performance of approx 28-36%!" and not "slowdown"

i don't understand!
what do you mean with the word "slowdown"? have you a video?
in my build i can't see any slowdow
User avatar
No.01991
Haze
Senior Tester
Aug 7, 2008, 15:52
I don't see any slowdown either, and proposing an old version of a file as a 'fix' is just stupid.

There is no bug.
User avatar
No.01992
Haze
Senior Tester
Aug 7, 2008, 15:52
Mame running slower is not a bug.
User avatar
No.01994
Tafoid
Administrator
Aug 7, 2008, 16:17
This bug was Resolved as not fixable - please do not change status. It was investigated already and reason was given for the decreased performance and Resolved (per Legends and Guidelines). Closed is reserved Bugs which have been Invalidated, Duplicated, Suspended or Unable to be reproduced. None of those apply here.
User avatar
No.01996
Haze
Senior Tester
Aug 8, 2008, 11:38
edited on: Aug 8, 2008, 12:29
It's not 'not fixable', it's simply not a bug at all, that's why I closed it. Performance 'issues' are not bugs. It should be closed.

'not fixable' are bugs in the original games etc. which we really can't fix, because they're not *our* bugs... any other bug is fixable (even if the fix is complex)

In this case functionality was added to the CPU core which resulted in an overall slowdown of the driver. This happens all the time, just look how much faster everything is in .36 . If we took reports for every slowdown and marked them as 'not fixable' there would be reports for every driver in MAME.

I made some significant changes to the Kaneko Supernova driver recently which resulted in a massive slowdown, probably cut the speed in half if not worse, but there's no report for that, nor should there be one.

So why should it be marked as not fixable, there's no bug to fix...
User avatar
No.02001
Fujix
Administrator
Aug 8, 2008, 17:21
I talked with Tafoid about it and cleared inconsistency.

I originally assumed all invalid bugs should be closed.
Bugs in the original games will be BTANB.
Accordingly this report should be closed with the

I'm closing some bugs with the inconsistency.