[Ben Boldt]

Has anyone ever seen a hack that utilizes Namco-163 expansion audio?  As much as I search, I have never found such a hack.  I have found quite a few famitracker tunes that are super cool versions of music from games, but I have never seen these inserted into an actual ROM and playable.

I understand that there are 2 obstacles with this:
1.  Make the game's memory mapping work on Namco-163
2.  Change the game's music engine to support Namco-163 expansion audio

Has nobody ever done this successfully?  I think we could open some doors to different sorts of audio hacks if we had at least one example, even if it's a simple one, for example starting from a mapper 0 game to minimize #1 above.

[Disch]

I did this for someone else's Final Fantasy hack that has yet to be released.  I'm not sure if they're actually going to use the music changes I did in the final version though (or if the hack is ever going to be released at all -- this was a year ago).

In fact... I still have that hack on my dropbox:   https://www.dropbox.com/s/ic1q58jztkfq0y4/ffnamcot.zip?dl=0    (The engine is in place, but it doesn't actually play any music because it'd all have to be re-scored and reinserted -- which I left up to the other guy)


The mapper is the easy part... it's working the audio into the music that's difficult.  You basically have to rewrite the entire music engine to really do it effectively.

[Ben Boldt]

The Final Fantasy Reconstructed hack you mention is precisely what got me interested in this.  It seems there is still activity on that project and in fact I just spoke with the guy in charge a couple days ago. (Duke2go?)

I am not an expert with NES software hacking by any means but I have an OK understanding of 6502 assembler.  I like the hardware part of it better but maybe I can help with this!  I have a Namco-163 chip in non-glob form, and my hopes and dreams consist of finding a worthy game and building a cartridge with it.

I am a little unclear on one part of what you said.  You say that "working the audio into the music" is the hard part and you have to rewrite the music engine.  Did you already do this in your Dropbox ROM?  If so, does this boil down to a relatively simple but tedious task of rescoring and reinserting the music?  I may be willing to take on that challenge even if this results only in a normal Final Fantasy with Namco-163 music.  From that point, we could think about adding the changes into Final Fantasy Reconstructed.

[Disch]

I am a little unclear on one part of what you said.  You say that "working the audio into the music" is the hard part and you have to rewrite the music engine.

Whooops.  I meant to say "working the music into the ROM".  I mangled that sentence up  =x

Did you already do this in your Dropbox ROM?  If so, does this boil down to a relatively simple but tedious task of rescoring and reinserting the music?

Yes and Yes.

The Dropbox ROM has the music ENGINE rewritten, as well as the format fully documented.... but no actual song data.  Someone would have to rescore and insert the music -- but hopefully that wouldn't be very difficult as the ROM is expanded and there's tons of space to work with.

I may be willing to take on that challenge even if this results only in a normal Final Fantasy with Namco-163 music.

Feel free.  Download link is already posted.  Let me know if you have any problems.

[Ben Boldt]

Thanks for the info.  I know that there are some Final Fantasy Famitracker tunes out there that use various mappers for expansion audio.  Were any of these ever being considered for this purpose with Namco-163?  I would like to know because I can imagine other people being more creative than me, but I would be happy to give it a try on my own too.

I don't expect this to be an overnight project but maybe it is something I can chip away at little by little and see where it goes.

[Disch]

Thanks for the info.  I know that there are some Final Fantasy Famitracker tunes out there that use various mappers for expansion audio.

Anything made in famitracker or <insert tracker here> is generally unusable.

Famitracker has a music driver that essentially has limitless RAM, ROM and CPU time.  In-game engines need to work under much more strict conditions because they have to run an entire game on top of running the music driver.  Working famitracker into a hack is even more difficult, since it's 99.999% certain that famitracker is going to be using RAM that the game is already using for something else.

So anything from Famitracker would have to be rescored and reinserted by hand -- just like anything else.

Were any of these ever being considered for this purpose with Namco-163?

Probably not, for above reasons.

[Ben Boldt]

I guess I didn't really mean to say that I wanted to drop a tune in from Famitracker.  I understand the need for this to be completely rewritten for your music engine.  I was looking more for a creative source from which to replicate, or if there was something already being considered for this purpose that may even have had some Namcot sound samples implemented.  There is a lot of Final Fantasy music out there that I can probably draw off from though, even if it is not from a tracker.

I have poked at your ROM file a little bit.  I am at the "Hello World" point right now... (I have to start somewhere right?)  I wanted to play a single note first.  Then I could tweak things to make that note sound different, and try all of the different cool features of your music engine.  I have not been able to get that first note yet unfortunately.

Here is what I tried:

First thing's first, I looked at the channel pointer tables for each song (0x3E010 - 0x3E40F).  I notice that these are all 0x0000, which is the special value to disable the channel.  For this 1-note test, I wanted to use only pulse channel 0, so for song #0's pointer table, I inserted this data at 0x3E010:

0000 0000 0000 0000 0100 0000 0000 0000

This means:
 - Disable all channels except for pulse channel 0.
 - Because Namcot channel 0 is disabled, paging will be on the page containing the beginning of song data space (0x3E410).
 - Start Pulse channel 0 data at offset 0x0001, which equals 0x3E411.

I wanted this to be for the song that plays immediately when you start the game ("Final Fantasy Prelude"), which I'm betting is song #0, but instead of making this assumption, I copied the pointer table over all tables in the entire range 0x3E010 - 0x3E40F.  So, all songs should point to the same spot.

Then I went down to offset 0x3E411 and inserted the value 0x60.  Tried the game, no note played.

Then I went back to offset 0x3E411 and tried a few things and ended up with this:

F9 00 FA 87 00 FB 82 08 F6 60 60 B0

Which breaks down to:
 - F9 00: Set tempo to tempo table ID 0, which comes already populated at 0x3DE10 as:
  -- 1E 09 09 09 09 09 09 09 09 09 09 09 09 09 09 09
 - FA 87 00: Following example, sets envelope to fixed volume 7, speed unused.
 - FB 82 08: Set Pulse channel to 50% duty cycle, with a speed of 0x08
 - F6: Set to octave 6 (had tried other values, this is where I ended up.)
 - 60: Play F# with length 1E (ID 0 in selected tempo table)
 - 60: Play F# again, same length
 - B0: Play B, same length


Would you help me figure out why the 3 notes did not play at the end?

[Disch]

If your music data starts at 0x3E411, your pointer would be $0401 (not $0001).  Subtract 0x3E010 from the offset to get the pointer value.

After that fix I did something similar and got beeps and boops:

At offset 0x3E410:

Code: [Select]

F9 00 FA 87 00 FB 82 08 F2 60 60 B0 E0 FD 00 0C
04

Same as you, only with a reasonable octave -- and I followed the final note with a rest, then an infinite loop that jumps back to the rest.

At offset 0x3E020 (the pointer table for the prelude):

Code: [Select]

00 00 00 00 00 00 00 00 00 04 00 00 00 00 00 00

Since my music starts at 0x3E410, my pointer is $0400 -- for pulse 1


Side note:  I tried the 'FF 00' marker to end the song before doing the reset + infinite loop.... and it didn't work -- which was a bit alarming.  I'll have to debug that.

[Ben Boldt]

COOL thanks!

I thought that the channel pointer was relative to the beginning of song data space 0x3E410 for some reason and it seemed kind of strange that the first byte couldn't be used that way.  So now my high-pitch squeaks work!  lol

I'm sure I will have more questions but this should get me going in the right direction for now.

[mrrichard999]

This hack uses that Audio Mapper - http://www.romhacking.net/hacks/2633/

[Ben Boldt]

Thanks for the link, mrrichard999.

I have a couple of new questions now, Disch.

I have inserted this at location 0x3E028:3E029:

00 04

And I have inserted this starting at location 0x3E410:

F9 00 FA 01 08 FB 82 08 F1 A0 00 20 50 A0 F2 00
20 50 A0 F3 00 20 50 A0 F4 00 20 50 A0 50 20 00
F3 A0 50 20 00 F2 A0 50 20 00 F1 A0 50 20 00 70
90 A0 20 70 90 A0 F2 20 70 90 A0 F3 20 70 90 A0
F4 20 70 20 F3 A0 90 70 20 F2 A0 90 70 20 F1 A0
90 70 20 A0 90 FD 01 09 04 

All is well, except the loop at the end does not seem to work.  I wanted this to loop back to location 0x3E419 one time, then continue on afterwards, but instead the game locks up when it hits this.  Did I do the loop command properly?

My other question is:  What happens if I call a subsection from within a subsection, and should this be avoided?

May 19, 2016, 12:33:13 am - (Auto Merged - Double Posts are not allowed before 7 days.)
I have found that it works fine if I make it an infinite loop:

F9 00 FA 01 08 FB 82 08 F1 A0 00 20 50 A0 F2 00
20 50 A0 F3 00 20 50 A0 F4 00 20 50 A0 50 20 00
F3 A0 50 20 00 F2 A0 50 20 00 F1 A0 50 20 00 70
90 A0 20 70 90 A0 F2 20 70 90 A0 F3 20 70 90 A0
F4 20 70 20 F3 A0 90 70 20 F2 A0 90 70 20 F1 A0
90 70 20 A0 90 FD 00 08 04 70 A0 30 50 70 A0 F2
30 50 70 A0 F3 30 50 70 A0 F4 30 50 70 50 30 F3
A0 70 50 30 F2 A0 70 50 30 F1 A0 70 50 30 A0 90
00 50 70 90 F2 00 50 70 90 F3 00 50 70 90 F4 00
50 70 90 70 50 00 F3 90 70 50 00 F2 90 70 50 00
F1 90 70 50 00 60 A0 10 50 60 A0 F2 10 50 60 A0
F3 10 50 60 A0 F4 10 50 60 50 10 F3 A0 60 50 10
F2 A0 60 50 10 F1 A0 60 50 10 A0 80 00 30 70 80
F2 00 30 70 80 F3 00 30 70 80 F4 00 30 70 80 70
30 00 F3 80 70 30 00 F2 80 70 30 00 F1 80 70 30
00 FD 00 08 04

I could hard-code the loop too (i.e. copy / paste all the notes twice) but I would like to use this feature if possible.

Update

I have the full prelude put in now, except for the 1x loop.  Where the 1x loop goes, I have inserted 3 rests.  This is extremely true to the original now, and nothing added. (No Namco-163 audio yet.)

Inserted at 0x3D430 (envelope tables 1 and 2):
0F 0F 0E 0E 0D 0D 0C 0C 0B 0B 0A 0A 09 09 08 08
07 07 06 06 05 05 04 04 03 03 02 02 01 01 00 00
08 08 08 08 07 07 07 07 06 06 06 06 05 05 05 05
04 04 04 04 03 03 03 03 02 02 02 02 01 01 01 01

Inserted at 0x3DE10 (tempo table 0):
09 09 09 09 09 09 09 09 09 09 09 09 09 09 09 01

Inserted at 0x3E020 (channel pointer table for song 1):
00 00 00 00 00 00 00 00 10 04 00 04 00 00 00 00

Inserted at 0x3E410 (song data):
F9 00 FA 02 18 FB 80 00 E0 E0 FE 18 04 00 00 00
F9 00 FA 01 18 FB 81 00 F1 A0 00 20 50 A0 F2 00
20 50 A0 F3 00 20 50 A0 F4 00 20 50 A0 50 20 00
F3 A0 50 20 00 F2 A0 50 20 00 F1 A0 50 20 00 70
90 A0 20 70 90 A0 F2 20 70 90 A0 F3 20 70 90 A0
F4 20 70 20 F3 A0 90 70 20 F2 A0 90 70 20 F1 A0
90 70 20 A0 90 E0 E0 E0 04 70 A0 30 50 70 A0 F2
30 50 70 A0 F3 30 50 70 A0 F4 30 50 70 50 30 F3
A0 70 50 30 F2 A0 70 50 30 F1 A0 70 50 30 A0 90
00 50 70 90 F2 00 50 70 90 F3 00 50 70 90 F4 00
50 70 90 70 50 00 F3 90 70 50 00 F2 90 70 50 00
F1 90 70 50 00 60 A0 10 50 60 A0 F2 10 50 60 A0
F3 10 50 60 A0 F4 10 50 60 50 10 F3 A0 60 50 10
F2 A0 60 50 10 F1 A0 60 50 10 A0 80 00 30 70 80
F2 00 30 70 80 F3 00 30 70 80 F4 00 30 70 80 70
30 00 F3 80 70 30 00 F2 80 70 30 00 F1 80 70 30
00 FD 00 18 04

[Disch]

Sorry for late reply -- was out most of the day.

I'd rather not input this all by hand.  Can you throw a patch (or ROM - whatever... it hardly matters) up on dropbox or something so I can download and debug?

[Ben Boldt]

Okay, I have uploaded it here:

https://sites.google.com/site/benboldt2/files/ff_namcot.nes.zip

Let me know if you have any trouble with this link, I have not used it much lately.

[Disch]

Jeez, I forgot to put a nameless label so it was jumping to a completely different block of code.

- Go to offset 3C289
- It will be $4B
- Change it to $16


That seems to [mostly] fix the looping, although it still sounds like there's a rest there.  I don't know if that's the engine or the score though -- I'll look into it more.


EDIT:

It was the score -- you're looping back to a rest.  Instead of looping to $0408, but you should be looping to $040A.


After those 2 changes (4B->16 and 08->0A), it works and sounds great!  Great job!


EDIT 2:

Sorry, I missed your other question.

Due to RAM limitations, I couldn't make it so you could nest subsections.  So a subsection cannot call another subsection.  If you try it it'll just screw up.  So yeah, that should be avoided.

[Ben Boldt]

I put in your changes and it works great now!  I had intended to loop back to 0x0418 for the 1x loop but your location has the same effect.

I plan to leave this song as-is at this point and work on another version of it that utilizes the Namco-163 expansion audio.  I tried building a Namcot table tonight with some new wave shapes, and then playing some notes, but no success getting the intended wave though yet.  I will start fresh on it again soon.

Edit:

No problem with non-nesting subsections.  I guess I am still a little bit curious about it.  Does the "enter subsection" work like a Goto, and then it stores the return address of the most recent time you use it?  So calling "enter" multiple times would just write over the return address each time?  Here's an example where I think multiple enters could be useful.  Say that you have the Super Mario Bros. NSF where there are copies of all of the level music for when time is running low.  It starts with a sound effect and then plays the same music faster than normal.  For this, you could have a separate song that just plays your sound effect, sets a faster tempo table, than "goto" the data for the normal level music, from which you will never go back.  Then within that "normal level" music data, could you enter and exit subsections as normal, even though you have had that extra enter in the past that you no longer care about?



I kind of hit a wall with getting Namcot audio to work.  I added a Namcot instrument table containing 4 waves:



This table is positioned at 0x3DC10, which is table 0.  I carefully avoided using the last 8 bytes because I was not sure if they had special function based on information here:

http://wiki.nesdev.com/w/index.php/Namco_163_audio

I left the last 8 bytes as zero.  Is this the problem; do I need to research into how these registers work?

I then made sure that the Instrument -> song assignment table contained a 00 for song 1, and it did.  I then used these commands at the beginning of the song data:

F9 00
Use Tempo Table 0.

FA 8F 00
Set Envelope to fixed volume 'F'.  I noticed that this was louder than using example value 87 00.

FB 00 00
Use Namcot Instrument ID 0 within the selected Namcot table.  This is the yellow highlighted wave in the image above.

It does make sound, but not the yellow triangle wave.  If I try the other waves with FB0100, FB0200, or FB0300, it makes no difference.

Output waveform:


Latest ROM file:
https://sites.google.com/site/benboldt2/files/ff_namcot.nes.zip

[Disch]

Does the "enter subsection" work like a Goto, and then it stores the return address of the most recent time you use it?

Yes.  Exactly that.  =)

As for your SMB example... the same could be accomplished by doing an "infinite loop", which also is effectively just a goto.

I kind of hit a wall with getting Namcot audio to work.  I added a Namcot instrument table containing 4 waves:

Ugh... How did I miss all these bugs before!  Sorry!

So I was writing to N163 regs backwards -- writing the length value to where the address should go and vice versa.

3 byte change to fix:

3C5CC:   64 -> 66
3C5D6:   C8 -> 88
3C5D7:   C8 -> 88


Thanks for finding these.  I'm actually surprised I didn't notice these huge bugs.  I swear I remember testing this and having it work  =x

This table is positioned at 0x3DC10, which is table 0.  I carefully avoided using the last 8 bytes because I was not sure if they had special function based on information here:

You can use all of it.  None of it has any special function.  The $60 bytes for instruments is placed at $00-5F in the sound RAM area, which leaves $60-7F for register space.

Is this the problem; do I need to research into how these registers work?

Nope.  The regs are handled by the engine.  You don't need to worry about it.  You just have to understand how to build the instruments -- which you clearly do  =)



Only other thing to mention:  The number of 4-bit samples in the instrument must be divisible by 4.  Or.. in other words.... the byte length of the instrument must be even.  You're currently using 0F for the length of the triangle which is odd and won't work properly.  This is casually mentioned in the doc but is very easy to overlook  =)



Thanks again!  If you run into more problems don't hesitate to let me know!   

[Ben Boldt]

Okay, your magic bytes have gotten my sound waves to play!  But I am still having one issue.

Is each wave limited to 32 samples (i.e. 16 bytes)?  Because if I go over 32 samples, the wave gets truncated.  It works like this:

If I use 32 samples or less, no problem.

If I use 60 samples (cyan wave), I get 60 - 32 = 28 samples, and the frequency is too high.  (I'm betting if it played the whole wave each period, the frequency would be correct.)

If I use 56 samples (purple wave), I get 56 - 32 = 24 samples.  So it looks like the number of samples is truncated to 5 bits.
i.e. 60 & 0b11111 = 28
56 & 0b11111 = 24

Please don't tell me I have to limit to 32 samples!   

Waves adjusted to have sample count divisible by 4:


Instrument 0, Beautiful:


Instrument 1, Beautiful:


Instrument 2, truncated:


Instrument 3, truncated:


Instrument 4, Beautiful:


ROM file updated with new Namcot table and set to play using the cyan wave:
https://sites.google.com/site/benboldt2/files/ff_namcot.nes.zip

[Disch]

Aye-ya.  Nah this is probably me screwing up the instrument length.  You definitely are not limited to 32 samples -- the cap is 256.  I'm probably just chopping off some bits or something.

I'll look at this tomorrow morning.

Thanks again for doing all this and exposing these bugs.  And for the great graphics & binaries!  It makes it so much easier to debug! 
May 22, 2016, 04:22:15 pm - (Auto Merged - Double Posts are not allowed before 7 days.)Soo... I wanted really bad for this to be a error in my code.  But it looks like errors in the emus.

FCEUX and NEStopia both seem to be using old Namco-audio documentation and not the latest that posted on the wiki.  This old documentation DID limit instruments to 32 sample lengths.  I even double-checked and ran Rolling Thunder (J) in each of them (which plays the music at too high a pitch with this old method), and both of them play the music incorrectly.

I didn't verify NEStopia source, but FCEUX source is definitely wrong:

line 146:
https://sourceforge.net/p/fceultra/code/HEAD/tree/fceu/trunk/src/boards/n106.cpp#l146

Code: [Select]

LengthCache[w] = (8 - ((V >> 2) & 7)) << 2;

Here, 'V' is the length reg.  This does the inversion and then left-shifts by 2.  But it does 8- and &7 (instead of 64- and &63)

As such, the largest sample would be when V=0... which will get you a length of 8*4 = 32.


=(

Sorry.  Short of someone updating the emus (which I'm not really willing to do), we're stuck.

[Ben Boldt]

Bummer!  32 samples it shall be.

Moving forward, I will develop using the full waves that I want, and stick in waves squished down to 32 over top of them.  I had planned on building a cartridge with a real Namco-163 chip for this eventually, so maybe that can shed some light on this and/or persuade the emulator authors to do an update!

Anyway, thanks for your help with this.  As I develop these scores, I think that I will need to start a new thread in the personal projects section for sharing updates.

Edit:

Crap, I am running into another problem.  I can only use Namcot channel 0 (I had not tried channels 1 2 or 3 yet until just now.)  I noticed this when trying to add channel 1 in addition to channel 0.

I will provide a new ROM file to demonstrate this soon.  You can reproduce this by taking my latest ROM file, and actually just moving the channel pointer from Namcot 0 to any of the other 3 Namcot channels.  But I will provide a ROM with 2 separate Namcot channels for you soon.

Edit 2:

Here is the ROM file updated with Namco channels 0 and 1 being used, but only channel 0 produces sound:
https://sites.google.com/site/benboldt2/files/ff_namcot.nes.zip

And here is an audio file for what it should sound like if both channels were working (I recorded each through channel 0 and then superimposed them in an audio app):
https://sites.google.com/site/benboldt2/files/Output%201-2.mp3

[Disch]

My weekend is over so I won't have as much free time to debug this.  I'll try to get to it sooner rather than later but it might take me a day or 2.  I'll send you a PM (and reply) when I finally get to it.