Please refer to original ../resid/README file for general discussion what
ReSID is.
This is ReSID-FP, a fork of ReSID that has been adapted to floating point
numbers. Some SSE assembly is used for vector convolutions when both the CPU
and compiler support it. In addition, some liberties have been taken in the
frequency region > 20 kHz which should be inaudible to humans.
In the emulation front, several changes to the original ReSID (henceforth
classical ReSID) have been made. These changes are listed here:
Waveforms:
- Noise waveform control via test bit is now possible.
(Unknown: how long does it take for the noise bits to fade with test bit on?)
This is used in SounDemoN's Tamaking, Bojojoing, etc, and the patch to
implement it in ReSID is his.
- Waveform 0, the frozen DAC, is emulated, which should make the new 8-bit
sample player routine work.
- Envelope and waveform outputs contain approximation of the imperfect DACs
(Unknown: are there other significant effects that affect the analog waveform
before it goes into filter, which should be modelled?)
- I changed voice DC offsets around for 6581 to better match my R4AR 3789.
Envelope:
- Performance work at envelope. Validation pending, should ensure that the new
code behaves 100% identically to the old one.
Mixer:
- Experimentally, a subtle negative offset is injected into the mixer through
ext-in pin. This part seems, however, physically incorrect and is likely
removed in the future. (The coupling capacitor external to the chip must
eliminate any DC offset, and the ext-in circuit inside the chip has nothing
that could generate offsets. In the meantime, this fix still helps 8580
Netherworld to play more correctly.)
- I removed the mixer_DC very subtle effect on 6581, as it already has 10x
more offsets elsewhere.
Filter:
- 6581 filter output contains approximation of the distortion effect
- 8580 filter output has bp flipped in phase with the other outputs
- 6581 resonance is slightly boosted. Potentially 8580 resonance needs to be
slightly stronger as well, as many songs show a bit more "punch" on the real
chip and one way to get more of that is increasing resonance. 10-20 %
increment is a practical maximum.
The upshot of all this is that for i386/x86-64 hardware, ReSID-FP's more
complicated algorithms may not seem any more expensive than the original ReSID.
For high-quality modes, it is virtually certain that ReSID-FP is actually
faster.
Meanwhile, emulation quality should be improved. If there are bugs, I'd like
to know about them. If the filter sounds wrong, I might be able to improve it,
too.
Here are some problematic songs, to get a feel for what's left to do:
- Markus Mueller: Mechanicus
* the distorted guitar effect is too distorted, but don't know how/why.
- Galway: Wizball
* the initial lead punches through with too much distortion. The "toggle"
between the muffled and more intense sound is too hard, even if similar
things do occur on the chip.
Undoubtedly, many more such examples will be found. However, samplings and such
are really valueable only if they can be made on a chip that I have modelled
for ReSID-FP. In practice I want to know about badly-playing chips, but might
conclude that it actually plays alright. At any event, information about sound
issues is welcome.
alankila@bel.fi
