TOS RAM test routine ?

[exxos]

Runs the same speed as before, but yes, CACR is now set to 0x3111 (both caches enabled). This seems odd to me (why data cache when testing memory?) but it's like it under TOS3.06 and I suppose if you're scanning everything once the cache will miss all the time anyway.

So this ought to be faster on the real hardware.

Just posted results in my blog in fact..No difference basically..

This is what I put in ..

Code: Select all

/*  exxos patch to turn on instruction cache*/
#if TOSVERSION >= 0x200
        move.l    #$00003111,d0
        D0_CACR
#endif

[exxos]

So I guess it should have been #$00000001 then @Badwolf ? I can change it. But I don't think it would make any odds. If 306 is doing it that way then probably does not make any odds at all anyway. I think it would be happier just having it as instruction only.. I assume it would change to both enabled later on but no idea.

[exxos]

I found this code which seems a little odd as well. Though possibly this could be a different memory test routine..

The whole thing is frustrating.. Does anyone know how to contact any of the Hatari developers to see if they would build me a custom version with the 24bit thing removed ?

Code: Select all

#if (TOSVERSION < 0x300) | TP_50
memchk0:
#endif
/* 306: unused */
        adda.l    d1,a0         /* a0 -> memory to check */
        clr.w     d0            /* zap pattern seed */
        lea.l     504(a0),a1    /* a1 -> ending address */
memchk1:
        cmp.w     (a0)+,d0      /* match? */
        bne.s     memchk2       /* (no -- return NE) */
        add.w     #$FA54,d0     /* yes -- bump pattern */
        cmpa.l    a0,a1         /* matched entire pattern? */
        bne.s     memchk1       /* (no) */
memchk2:
        jmp       (a4)          /* "return" to caller */

[exxos]

Finally found a version of Hatari which works. 2.0.0.

So for whatever reason TOS306 has 1min 20seconds for 64MB and ST536 TOS takes 2mins 25 seconds. So with this config now I have something I can work with.

EDIT:

So 206 turns on 24bit addressing and changing the header to 306 it crashes FFS!.

Capture.PNG (221.71 KiB) Viewed 1017 times

So I give up. It is bad enough trying to fight with the TOS sources without having to fight with third-party programs at the same time.

[Badwolf]

So I guess it should have been #$00000001 then @Badwolf ? I can change it. But I don't think it would make any odds. If 306 is doing it that way then probably does not make any odds at all anyway. I think it would be happier just having it as instruction only.. I assume it would change to both enabled later on but no idea.

#$1 would, if I understand correctly, only enable the instruction cache, yes.

But I've verified that TOS3.06 sets 3111, so it should be safe.

BW

[exxos]

But I've verified that TOS3.06 sets 3111, so it should be safe.

Thanks.

The only other odd thing was a '/4' in what looks to the the TTram test routine for block size. So give this one a try when'ya go a moment.

4.zip

(127.41 KiB) Downloaded 31 times

[exxos]

Hatari 1.9.0. I set 64 in the cfg file. It refused to show the TTRam bar until I restarted it just.

Capture.PNG (42.78 KiB) Viewed 991 times

1.PNG (47.25 KiB) Viewed 991 times

So putting 10 in TTram size gives me this

Capture.PNG (38.59 KiB) Viewed 990 times

Does anyone know if there was more " nightly builds" of Hatari 1.9.x ?? I mean it jumps to version 2.0.0 which definitely starts screwing with the 24bit options.. But maybe there is a later version of 1.9.x kicking about which may solve some problems before it breaks other things ?

[exxos]

There is another possibility.. That for some reason TOS306 might only be detecting RAM as opposed to actually testing it There is a switch for that in the code, but seems bad not to do a little diagnostic test on RAM. but maybe just has to be because it takes a long time

[stephen_usher]

It does do a memory test. I've seen the X's

[exxos]

It does do a memory test. I've seen the X's

It might write only zeros and not do any other tests. There's nothing else to try unfortunately.