Re: [hatari-devel] Is it me or the new hatari release is slow ?

[Laurent Sallafranque <laurent.sallafranque@xxxxxxx>]

> I hardly see how just adding an illegal opcode (that is not used in your test program I guess ?) can slow down the whole cpu emulation ?

I've tested again and again, the rev 5007 is the one.
I think I know what's the problem :
 My 68030 cycles table is static (see falcon_cycle030.h).

If you changed something into table68k, it may have shift the static cycles table.
As my effect use a lot of the same instructions, a change in the timing is probably the answear.

Laurent




Le 01/06/2014 20:50, Nicolas Pomarède a écrit :

Le 01/06/2014 20:34, Laurent Sallafranque a écrit :

OK, I've tested and I get the following :

laurent@OrdiLoloLinux:~/Atari/hatari$ hg bisect --good

La première révision présentant le problème est :

(The first release that shows the problem is :)

changeset:   5007:7a4948830aee
user:        Nicolas Pomarede
date:        Wed Mar 19 20:31:01 2014 +0100
summary:     bset.b d0,8(pc) is not a valid 680xx instruction, but an
illegal one

I'll try again with the release 5006 and 5007 to be sure, but it seems
that this 5007 release is the one that slows a lot the new with my effect.

Hello

I hardly see how just adding an illegal opcode (that is not used in your test program I guess ?) can slow down the whole cpu emulation ?

If you comment/revert the changes in #5007, does it really speed up your program ?

Nicolas