How to pre-calculate the Protracker mt_PeriodTable

[Siekmanski]

Hi,

How did Karsten Obarski ( coder of the first Amiga SoundTracker ) calculate the period table?
How did Lars Hamre ( coder of Protracker ) calculate the fine_tune stages?

Started coding a Protracker2.3 Replay routine for Windows using the Microsoft Assembler ( Masm )
To shave some bytes of the executable, I want to pre-calculate the "mt_PeriodTable".
Wrote a routine assuming a western equal-tempered scale, but ended up with 37 mismatches.

Calculate_mt_PeriodTable:
    mov         edi,offset PeriodTable+576 ; Start at the second half of the Period Table memory buffer
    movss       xmm0,BasePeriod         ; Best possible period match: 906.9023 -> ( 3546895 / 3911 )
    movss       xmm1,xmm0
    mov         dl,2                    ; Split the Period Table in 2 halves
split_LP:
    mov         dh,8                    ; 8 fine tune stages per half Period Table
next_finetune_step:
    mov         cl,36                   ; 3 octaves ( 3 * 12 notes )
octaves_LP:
    cvtss2si    eax,xmm0                ; Convert period float to 32 bit integer
    mov         [edi],ax                ; Save period value as 16 bit integer
    divss       xmm0,twelfth_root_of_2  ; Calculate next period value in the present octave -> 2^(1/12) = 1.059463
    add         edi,2
    dec         cl
    jnz         octaves_LP
    divss       xmm1,semitone           ; Calculate next fine tune step -> 2^(1/12/8) = 1.007246
    movss       xmm0,xmm1
    dec         dh
    jnz         next_finetune_step
    mov         edi,offset PeriodTable  ; Jump to the start of the Period Table memory buffer
    dec         dl
    jnz         split_LP                ; Code size 70 bytes

Is there some logic I'm missing in the original mt_PeriodTable?
Or do I need to write some reconstruction code to repair the 37 mismatches?

The challenge is to write the smallest piece of code that matches 100% the original "mt_PeriodTable".

[robinsonb5]

Originally Posted by Siekmanski

How did Karsten Obarski ( coder of the first Amiga SoundTracker ) calculate the period table?
How did Lars Hamre ( coder of Protracker ) calculate the fine_tune stages?

Started coding a Protracker2.3 Replay routine for Windows using the Microsoft Assembler ( Masm )
To shave some bytes of the executable, I want to pre-calculate the "mt_PeriodTable".
Wrote a routine assuming a western equal-tempered scale, but ended up with 37 mismatches.

This thread should be some help - but unfortunately, there's no simple solution - the problem appears to be have been floating-point rounding errors in the BASIC implementation used when generating the original period values back in the day.
https://eab.abime.net/showthread.php?t=69675

(I was looking absent-mindedly at your assembly code, wondering why it looked strange, until the penny dropped - it's X86 assembly! In an Amiga forum! Sacrilege! )

[meynaf]

I completely fail to see the point in attempting to shave bytes out of a Windows executable...

[robinsonb5]

Originally Posted by meynaf

I completely fail to see the point in attempting to shave bytes out of a Windows executable...

4K / 64K demos exist in the Windows world too...

[meynaf]

Originally Posted by robinsonb5

4K / 64K demos exist in the Windows world too...

In todays windows world, really ?
Considering the amount of work to do for just a simple 2D screen setup and for just playing a sound, i seriously doubt anything useful can be done with such a small executable size...

[Siekmanski]

Originally Posted by robinsonb5

This thread should be some help - but unfortunately, there's no simple solution - the problem appears to be have been floating-point rounding errors in the BASIC implementation used when generating the original period values back in the day.
https://eab.abime.net/showthread.php?t=69675

(I was looking absent-mindedly at your assembly code, wondering why it looked strange, until the penny dropped - it's X86 assembly! In an Amiga forum! Sacrilege! )

I will admit, sacrilege, but the knowledge of the Amiga trackers is here on this forum I hope.

Thanks for the link to the thread ( missed that thread ), it seems a bit more difficult than I thought.....
The most mismatches are in the Tuning 0 and -8 stages (28).
See if I can create some kind of small period repair code......

[DanScott]

Originally Posted by meynaf

In todays windows world, really ?
Considering the amount of work to do for just a simple 2D screen setup and for just playing a sound, i seriously doubt anything useful can be done with such a small executable size...

WHAT??? have you not seen some of the amazing 4k and 64k productions???

[ Show youtube player ]

[ Show youtube player ]

just a couple....

[Siekmanski]

Originally Posted by meynaf

In todays windows world, really ?
Considering the amount of work to do for just a simple 2D screen setup and for just playing a sound, i seriously doubt anything useful can be done with such a small executable size...

You really don't need to write that much code to set up audio and graphics in todays windows.
DirectSound and Direct3D9 or OpenGL are your friends in the Windows Demos world.

[meynaf]

Originally Posted by Siekmanski

You really don't need to write that much code to set up audio and graphics in todays windows.
DirectSound and Direct3D9 or OpenGL are your friends in the Windows Demos world.

Is some code snippet available somewhere ?
What i've seen so far is much code.

[meynaf]

Originally Posted by DanScott

WHAT??? have you not seen some of the amazing 4k and 64k productions???

No, i haven't.

Originally Posted by DanScott

just a couple....

Neither of them accept to work on my setup.

[StingRay]

Originally Posted by meynaf

Considering the amount of work to do for just a simple 2D screen setup and for just playing a sound, i seriously doubt anything useful can be done with such a small executable size...

1k with sound:
https://www.pouet.net/prod.php?which=32194

[meynaf]

Originally Posted by StingRay

1k with sound:
https://www.pouet.net/prod.php?which=32194

Which, again, doesn't work out of the box. At least not for me.

[DanScott]

and of course, the classic Amiga demo Nexus 7, remade in 8k :P

[ Show youtube player ]

[StingRay]

Originally Posted by meynaf

Which, again, doesn't work out of the box. At least not for me.

And still, size optimised Windows productions have existed for a long time already, if they run on your system or not doesn't change that fact.

[meynaf]

Originally Posted by StingRay

And still, size optimised Windows productions have existed for a long time already, if they run on your system or not doesn't change that fact.

I haven't said otherwise.
It's just that they do only small part of a correct setup and work on a limited amount of machines (and don't multitask properly, etc). It indeed makes it a lot easier to write small code.
I suppose anyway it's inherent to windows boxes that simple things never work properly everywhere.

Doing graphics and sound has never been a problem for me on the miggy but on peecees there is always something that goes wrong. Sigh.

I can't imagine playing protracker (with correct quality) with just a few kb of windows code.
Yet if it is about removing bytes out of the exe, it may be better to use a module packer such as Player 6.1 (P61A). Some probably have a shorter replay routine as well.

[Siekmanski]

I managed to produce a 100% correct mt_PeriodTable with only 144 bytes of code.
That's 12.5% of the original size, 1152 - 144 = 1008 bytes lost.
Think I can make it even smaller.....

[meynaf]

Originally Posted by Siekmanski

I managed to produce a 100% correct mt_PeriodTable with only 144 bytes of code.
That's 12.5% of the original size, 1152 - 144 = 1008 bytes lost.
Think I can make it even smaller.....

Have you considered using fixed point or changing rounding method ?

[Siekmanski]

Yes I have, it's not a rounding error ( tested them all ), round to nearest gives the best possible result.
Even higher resolution ( doubles ) gives the same result.
You end up with 32 periods that are 1 to high and 5 are 1 to low.