Replace 2A03/FDS audio cores with NSFPlay/Mesen #32
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
nyanpasu64
force-pushed
the
replace-audio-cores
branch
from
373331c
to
be73135
Compare
nyanpasu64
added this to Done
in Port chip emulators to external emulators, remove CChannel
nyanpasu64
moved this from Done
to To Do
in Port chip emulators to external emulators, remove CChannel
nyanpasu64
force-pushed
the
replace-audio-cores
branch
from
be73135
to
8435e4c
Compare
Shift-JIS doesn't have ~ and \, and the ASCII/UTF-8 codepoints for those map to overline and yen. https://stackoverflow.com/questions/56819615/ As a result, we have to postprocess the output of iconv. fd -t f -x iconv -f SHIFT-JIS -t UTF-8 {} -o {} fd -t f -x sed 's/‾/~/g; s/¥/\\/g;' {} -i Additionally, iconv corrupts emu2413.c by turning "OPLL_reset(OPLL" into "OPLL_rePLL_reLL". So I had to revert this change.
nyanpasu64
force-pushed
the
replace-audio-cores
branch
from
8435e4c
to
563c7d6
Compare
Saves CPU in debug builds.
Blip_Buffer::mix_samples() delays all incoming samples by the same amount as Blip_Synth::offset() delays deltas. This is the right choice for audio generated by output-rate synths (used in FamiTracker for VRC7 audio), but causes audio generated by another Blip_Buffer (and already delayed) to be delayed twice. I add a new method, Blip_Buffer::mix_samples_raw(), which does not delay incoming audio, and use it for mixing in filtered FDS audio.
The old code delayed global output by 1 sample, and also delayed CFDS by 1 sample relative to other sound chips.
The constructor with parameters is currently unused.
The nsfplay emulator's APU2 triangle wave has a nonzero offset when reset. This may or may not be accurate. Either way, subtract the offset when the chip resets.
This way, when switching between sound chip emulators with different output levels, we only need to edit the sound chip class to compensate, not any code in CMixer.
nyanpasu64
force-pushed
the
replace-audio-cores
branch
from
563c7d6
to
1a3779d
Compare
nyanpasu64
changed the title
Port chip emulators to external emulators, remove CChannel
automation
moved this from To Do
to Done
Feb 26, 2021
This was referenced Feb 26, 2021
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
rip out the guts, replace with something new, reimplement all the side-channel observability metadata (pitch/volume view, register view, channel meters) that comes with it
hopefully i didn't regress any functionality in the process... it'll be done someday, someday~
Overview
2A03 and FDS audio emulation were replaced with nsfplay and Mesen's cores, respectively.
Both are wrapped in CSoundChip2, a newly added replacement for CSoundChip (which was the wrong abstraction). CSoundChip2 does not use CChannel. CChannel was a mistake, and CChannel plus CMixer's current API (each channel independently streams volume levels and/or deltas to the same Blip_Synth located in CMixer) is hostile to any form of channel mixing other than linear mixing.
CSoundChip2 expects to own a Blip_Synth (and CFDS also owns a Blip_Buffer used for lowpass emulation), and does not hold an aliased pointer to CMixer or Blip_Buffer (instead receiving a borrowed Blip_Buffer& whenever audio needs to be generated).
The mixing interface is still inconsistent; CFDS computes mixing levels internally (encapsulating the implementation detail of the emulation core's output levels), whereas C2A03 lets CMixer compute mixing levels.
Behavioral changes
2A03 noise at the highest pitch may use more CPU than before due to emulation inefficiencies. This is only an issue on debug builds or possibly slow computers (did not test). FDS may use less CPU than before due to cycle-skipping.
FDS emulation changes
FDS sounds different due to the lowpass filter added. Additionally earlier FDS instruments will no longer work; the new sine preset will fix the sound with minimal audible changes on older FamiTracker versions.
TODO In the future, we may add a FDS filter strength slider.
The NSF driver performs a modulator phase reset that in-tracker playback does not, making reset-free modulator envelopes that operate deterministically in NSF export start at random positions in the tracker; modulator phase reset may be added into the tracker into the future.
TODO In the future, we may add a toggle for the FDS's non-monotonic jagged DAC (which scrambles wavetable amplitudes but does not affect their position in time).
Volume meter algorithm changes
On affected chips, the volume meters were rewritten to plot the range of volume levels seen during each engine tick, rather than the largest new volume level seen during a tick (where steady-state chip output is interpreted as 0 volume).
For example, the 2A03 DPCM meter now plots the range of DPCM counter values seen, rather than the largest new DPCM counter value. This more accurately models perceived sample volume:
Note that this assumes all steps are evenly spaced, and does not take into account nonlinearity (output deltas being smaller at high values), but it's closer than before.
2A03 triangle waves will flicker more than before (they were never flicker-free) at very low pitches (below 60 Hz).
Potential issues
The latest commit (373331c) was written on Linux and I did not try compiling it. Hopefully it builds on CI, doesn't have warnings, and works properly (on Wine?)