Implement a high-level emulation of the MP2K audio mixer.

[fleroviux]

No description provided.

[birdybro]

Hi!

I'm looking into what the specific reasoning was for this formula in the apu:

sample[channel] += psg_sample * psg_volume * psg.master[channel] / (28.0 * 0x200);

Hopefully looking to help resolve this issue with the information --> MiSTer-devel/GBA_MiSTer#143

Thank you so much! :)

[fleroviux]

@birdybro psg_volume can take values from 0 to 4 (see psg_volume_tab), psg.master[channel] can take values from 0 to 7. psg_sample iirc has a range of -0x1FC to 0x200 (because each channel has a range of -127 to +128). So dividing by 4 * 7 * 0x200 the mix of the channels is normalized to a roundabout -1.0f to +1.0f range.

However this is for the HQ mixer, which uses floating point math for the final mix. See https://github.com/nba-emu/NanoBoyAdvance/blob/master/src/nba/src/hw/apu/apu.cpp#L186-L196 for the more accurate implementation which has the 28 factor but not the 0x200 factor.

[birdybro]

Thank you! Much appreciated! :)

[birdybro]

After talking with the developer of that core when I noticed something different...

NanoBoyAdvance/src/nba/src/hw/apu/apu.cpp

Line 123 in a30da20

constexpr int psg_volume_tab[4] = { 1, 2, 4, 0 };

This ordering of the division factor is different than no$'s documentation (his would be represented as {4, 2, 1, 0} instead AFAICT). I am going to test making the MiSTer GBA core reflect your division factor since I suspect that's the root cause (and it's the easiest thing to test first lol).

https://github.com/MiSTer-devel/GBA_MiSTer/blob/master/rtl/gba_sound.vhd#L421-L427 - is the area I changed and am going test later...

Edit: nope, not the root cause.

[fleroviux]

@birdybro this is what I'm seeing in GBATEK (see SOUNDCNT_H):

  0-1   R/W  Sound # 1-4 Volume   (0=25%, 1=50%, 2=100%, 3=Prohibited)

[birdybro]

Yeah I was just being stupid and not reading it right. The core is correct. Sorry.