in the recent past I spent some time on this issue triing to change smsq/e source code to use some 68020+ instructions.
I had on this issue some mail echange with Marcel and George (and Thierry....) to try to understand why, only the specific function of pan and scroll in the 16 bit driver, are so slow for the q60 compared to the equivalent emulated QPC.
The specific files "iod_con2_16_mblock_asm" make a massive use of move.l instruction in ciclic program loops and I am convinced that the use of MOVE16 could improve dramatically the performances results. Sadly the use of MOVE16 is not a simple replacement of move but need a complete rework of the logic cause the registers addressing changes
Due my (still) poor programmer capacity I was not able to produce a working -r files, I renew the request of some help, obviuously from this changes will benefit also the Sgold card users.
Ciao
Fabrizio
----- Original Message ----- From: "Phoebus R. Dokos ("Φοίβος Ρ. Ντόκος")" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Sunday, September 26, 2004 4:53 AM
Subject: Re: [ql-users] QXL: SMSQ
On Sun, 26 Sep 2004 04:22:05 +0200, Marcel Kilgus <[EMAIL PROTECTED]> wrote:
Phoebus R. Dokos wrote:This assumption is based on what exactly?On how much faster 68020+ instructions work performing tasks that need a series of 68000 instructions to normally perform.
So this assumes that all or most tasks within SMSQ/E can be done with fewer 20+ instructions. I'm not sure about that.
Not all of course but select few (like MOVE16 for example) and especially the math ones could be significantly be sped up (if you even take into account the FPU part of 040 and 060 there is a significant boost up to be had)...That is possible but not necessarily; if the emulated instruction would need more native instructions to be emulated for example....
Of course that applies on real machines only because on PC based 68K emulators that would be completely pointless
Actually emulators would gain far more than the native hardware if your assumption is right and the instruction count could be significantly reduced.
Speed up on a well written and mature emulator core like QPC's for example can be achieved by the extra processing power without the hassle of potentially re-writing parts of the emulation core and maybe in the process introducing some new bugs (which probably doesn't apply in your situation given the known quality of your software :-) but still it is possible)
Phoebus
_______________________________________________
QL-Users Mailing List
http://www.quanta.org.uk/mailing.htm
_______________________________________________ QL-Users Mailing List http://www.quanta.org.uk/mailing.htm
