NSF Export uses unitialized CPU RAM

[MrSaturnBoing]

Version

• Dn-FamiTracker version:
  • 0.5.3
• OS:
  • Microsoft Windows [Version 10.0.19045.6456]

Description

So, when I export NSF file it sounds normal in emulator but when I run it on a real console it produced a short click on a triangle channel (looks like it uses uninit CPU RAM)

Steps to reproduce

1. Open Mesen
2. Open Settings, go to NES -> Emulation
3. Set default power-on state for RAM to "All 1s"

Attachments

[untitled.wav](https://github.com/user-attachments/files/28677330/untitled.wav)

[rom and module.zip](https://github.com/user-attachments/files/28677335/rom.and.module.zip)

[Gumball2415]

can you properly attach the files as links and not code blocks? i cannot access them. the urls themselves are also not accessible.

[HeeminTV]

Yes, half of variables including var_ch_DutyCycle, var_ch_Trigger and var_ch_DutyDefault aren't properly initialized. This always has been a problem when using FT NSF driver for my games.

[Gumball2415]

with regards on console initialization and triangle pop

• the triangle sequencer is always initialized on reset with position 0, which means value 15, the highest level. Dn-FT counteracts this pop by cancelling the first delta output of the triangle channel (see 953a028)
• it is the responsibility of the NSF player, not the NSF itself to initialize the console memory. that being said however, most of the variables used by FTDRV in BSS are initialized in the subroutine ft_music_init, in file init.s

[HeeminTV]

Even though I know the FT NSF driver is made for players and not for games and also know NSF format somehow forces all RAM to be cleared before playing, I think it's practical to intialize all variables in driver anyway.

Not even most of variables. Half of variables aren't intialized.

[VinglesSmi]

Clearing the CPU RAM is quite trivial, even though NSF players are mostly used with emulators, which for the most part fill RAM with all 0s by default. I do believe however that there is a sense of satisfaction when hearing the song play on real hardware.
Analogue behavior doesn't help with that though, and the RAM chips in my consoles have a resting pattern that isn't friendly with ROMs that do not clear memory on boot.
It wouldn't hurt adding code to the .nes exported ROM, it's like... 11 lines of 6502 assembly. I do believe this is an issue that should be addressed, because it does not prevent analogue behavior from affecting the audio playback. Who knows how many people ran into this issue and never figured out it was just the driver reading uninitialized memory?

[Gumball2415]

it is literally defined in the NSF format that players have to initialize memory, not the NSF itself. if the driver or any NSF would re-initialize memory, it would break any hardware players that require the use of zeropage memory or BSS

in the case of the NES ROM export, the wrapper code also includes a basic memory initialization routine; see nes.s for a decompilation of the inserted binary.

[Gumball2415]

1. Open Mesen
2. Open Settings, go to NES -> Emulation
3. Set default power-on state for RAM to "All 1s"

with regards to Mesen, i literally cannot randomize the memory in its NSF player; it always zeroes the memory as of version 2.1.1