Use the new fast noise LFSR reset by Dag Lem to improve the tests
Versatile Commodore Emulator
Brought to you by:
blackystardust,
gpz
Menu
▾
▴
#1885 Use the new fast noise LFSR reset by Dag Lem to improve the tests
As discussed privately with @gpz we could employ the new fast noise reset routine [1] to make some SID tests faster and more reliable.
Ideally we should include both method, allowing compile time choice, to compare the results.
The tests include at least:
* wb_testsuite/noise_writeback_check.asm
* wf12nsr/wf12nsr.asm
* noisewriteback
Opening this ticket as a reminder, will post patches as time permits.
[1] https://github.com/daglem/reDIP-SID/blob/master/research/noise-reset.a65
Make sure to fetch the program again after reading this, it has been fixed to restore the I flag after calls to CHROUT.
Thanks! I've just exctracted the reset routine part from the source to fit it into the existing tests.
Here's a first patch. It still uses the old method by default but switching to the new one seems to work pretty well.
There are still a few tests that can be converted:
* SID/noisewriteback/noise_writeback_test1.asm
* SID/noisewriteback/noise_writeback_test2.asm
* SID/waveforms/waveforms.asm
One thing to keep in mind is that the new method messes with the frequency registers.
Nice! I think you may be missing a final setting of the test bit in noise-reset_new.asm, i.e. something like this at the end (the first part of point 4 - reset_lfsr, in the original program):
LDA #$08
STA $D412
After this, the clearing of the test bit in the test programs will finalize the reset.
One more thing - set_lfsr_18_22 is timing sensitive, so make sure that no badlines or interrupts can interfere. I'm guessing all the tests already ensure this, but I mention it anyway - better safe than sorry :-)
patch updated, thanks!
I believe you should set the desired frequency in noise_writeback_check.asm and wf12nsr.asm just before clearing the test bit, otherwise the oscillator will start running earlier than without the changes, and you'll be changing the tests ever so slightly.
Just a final small nitpick :-D
In the Makefiles, you could set NEWRESET=0 just once, and then use that as -DNEWRESET=$(NEWRESET) in the recipes.
This would make it simpler to override the setting.
Thanks again for reviewing! Here's the third round.
Even though the result is the same, I think I would still set the frequency immediately before releasing the test bit in wb_testsuite/noise_writeback_check.asm, so that one isn't fooled into thinking that the setting before the noise reset makes any difference.
Other than this nitpick, this looks good to me :-)
Uh right, missed that!
BTW, any plan moving to git? ;)
see https://vice-emu.pokefinder.org/wiki/Move_to_git :)
Nice!
I have tested the patch on a C64, and there is still a minor issue - badlines disrupt the timing of the LFSR clocking in noisewriteback/noise_writeback_check.asm and wb_testsuite/noise_writeback_check.asm
Changing the start of both files to the following fixes the issue:
; disallow interrupts and disable screen to get stable timing
sei
lda #0
sta $d011
sta $d015
raslo:
bit $d011
bpl raslo
rashi:
bit $d011
bmi rashi
I haven't tested wf12nsr.asm, but I think it should be changed like this:
; disallow interrupts and disable screen to get stable timing
sei
jsr clrscreen
lda #5
sta finalresult+1
lda #0
sta $d011
sta $d015
sta currenttest
raslo:
bit $d011
bpl raslo
rashi:
bit $d011
bmi rashi
start
One final thing I noticed when reading through noise_writeback_check.asm is that 9 -> 9 is marked as unstable for the 8580. So perhaps it's better to change the new reset routine to use noise + sawtooth + triangle instead of just noise + triangle?
Updated patch attached.
Right, on the chips I've tested it seems that $b or $f are more reliable. Changed it to N+S+T
applied in r43934