[hatari-devel] Perhaps this evocate something to you...

[Nicolas Pomarède]

Le 19/01/2011 23:37, Laurent Sallafranque a écrit :
> A good candidate for testing 16Mhz vs 32 Mhz is Watership.
>
> It loads quickly and bugs immediately at 16 Mhz (but runs at 32Mhz).
>
> I've taken 2 CPU traces : one at 16 Mzh and one at 32 Mhz and I've made
> a diff of the 2 traces.
>
> Here is the result :
>
> The 2 programs are running the first 3196 instructions the same way.
> Here are the latest instructions ran by both tests :
>
> ...
> 0002E226 4298 CLR.L (A0)+
> 0002E228 51c9 fffc DBF .W D1,#$fffc == $0002E226 (F)
> 0002E226 4298 CLR.L (A0)+
> 0002E228 51c9 fffc DBF .W D1,#$fffc == $0002E226 (F)
> 0002E226 4298 CLR.L (A0)+
> 0002E228 51c9 fffc DBF .W D1,#$fffc == $0002E226 (F)
> 0002E226 4298 CLR.L (A0)+
>
> At line 3197, the 32 Mhz test continues to run the same 2 instructions
> as above.
>
> But the 16 Mhz test does :
>
> 00E03C50 52b9 0000 04ba ADD.L #$00000001,$000004ba
> 00E03C56 e7f9 0000 11ba ROLW.W $000011ba
> 00E03C5C 6a48 BPL.B #$00000048 == $00E03CA6 (F)
> 00E03C5E 48e7 fffe MOVEM.L D0-D7/A0-A6,-(A7)
> 00E03C62 614a BSR.B #$0000004a == $00E03CAE
> 00E03CAE 48e7 c080 MOVEM.L D0-D1/A0,-(A7)
> 00E03CB2 2038 11bc MOVE.L $000011bc,D0
> 00E03CB6 677a BEQ.B #$0000007a == $00E03D32 (T)
> 00E03D32 4cdf 0103 MOVEM.L (A7)+,D0-D1/A0
> 00E03D36 4e75 RTS.L
> 00E03C64 0838 0001 0484 BTST.B #$0001,$00000484
> 00E03C6A 672a BEQ.B #$0000002a == $00E03C96 (F)
> 00E03C6C 4a38 11b1 TST.B $000011b1
> 00E03C70 6724 BEQ.B #$00000024 == $00E03C96 (T)
> 00E03C96 3f38 0442 MOVE.W $00000442,-(A7)
> 00E03C9A 2078 0400 MOVEA.L $00000400,A0
> ...
>
>
> At line 4653, the 32 Mhz test does :
>
> 0002E228 51c9 fffc DBF .W D1,#$fffc == $0002E226 (F)
> 00E03C50 52b9 0000 04ba ADD.L #$00000001,$000004ba
> 00E03C56 e7f9 0000 11ba ROLW.W $000011ba
> 00E03C5C 6a48 BPL.B #$00000048 == $00E03CA6 (T)
> 00E03CA6 11fc 00df fa11 MOVE.B #$df,$fffffa11
> 00E03CAC 4e73 RTE.L
> 0002E226 4298 CLR.L (A0)+
> 0002E228 51c9 fffc DBF .W D1,#$fffc == $0002E226 (F)
> 0002E226 4298 CLR.L (A0)+
> 0002E228 51c9 fffc DBF .W D1,#$fffc == $0002E226 (F)
> 0002E226 4298 CLR.L (A0)+
> 0002E228 51c9 fffc DBF .W D1,#$fffc == $0002E226 (F)
>
>
> Both tests are jumping to address $E03C50, but not after the same amount
> of instructions.
>

Well, the tests are not exactly jumping, they are interrupted by an MFP 
interrupt, which points in the ROM to e03c50.

$4BA is "Raw 200Hz timer", so this is just an interrupt that counts 200 
Hz pulses.

> The test after the first 3 instructions after the "interrupt" is TRUE
> for the 32 Mzh test, but is false for the 16 Mhz test.
>
> The 32 Mhz test continues to CLR / DBF whereas the 16 Mhz test goes into
> completely different code.

At one point, it's normal 32 and 16 Mhz don't return at the same place, 
because the 32 Mhz version is running twice more instructions per cycle, 
so when it's interrupted by the 200 Hz timer, it processed more 
instruction than when it ran at 16 Mhz.

What do you means by "it bugs" ? Crash ? Bomb ? Bad visual result ?


Nicolas