If a SoundFont sets `loopstart == loopend` and then uses loop-offset-modulators to fix up those loops assigning them with a valid position, the sample was previously switched to unlooped mode erroneously.
For the long story, see #1017.
For detuned channels it might be better to use another key for Soundfont sample selection
giving better approximations for the pitch than the original key.
Example: play key 60 on 6370 Hz => use tuned key 64 for sample selection
This feature is only enabled for melodic channels.
For drum channels we always select Soundfont samples by key numbers.
Using prepend both in sffile and defsfont reduces insert complexity
to O(1) and does not affect the final order of the sample list, as
the reversed order of the samples in sffile is reversed again in
defsfont.
This change removes the need for the instsamp hack in instrument
and preset zones. The sampleid and instrument generators are
treated as any other generator and simply passed to the defsfont
import functions. Those read the two generators and use the
index to look up valid samples and instruments.
Both generators are then reset to GEN_UNUSED again, just to make
sure that the rest of FluidSynth doesn't get confused. But that might
not be necessary.
The previous implementation used 0 as end-of-list sentinel
value. But 0 is also the id of the first generator in the
invalid_preset_gen list. This effectively prevented checks
for invalid preset generators.
Also contains some code cleanup to make the check for invalid
instrument generators similar to invalid preset generators.
Fixes#821
The loading of SF2 samples can be parallelized as well, at least for bigger soundfonts like the Stgiga 4GB monster.
The following test is performed with hot-caches:
`time src/fluidsynth -i -a alsa Stgiga\'s\ HiDef\ Soundfont\ \(2019-05-25\).sf2`
Serial version:
```
real 0m15,460s
user 0m14,163s
sys 0m1,192s
```
Parallelized version proposed by this PR:
```
real 0m5,851s
user 0m14,089s
sys 0m1,194s
```
fluid_list_remove() should receive the beginning of a list, so it can adjust the predecessor of the element to be removed. Otherwise the element would remain in the list, which in this case led to a use-after-free afterwards.
Following the discussion about an API to pin and unpin preset samples in the sample cache here:
https://lists.nongnu.org/archive/html/fluid-dev/2020-10/msg00016.html
Short explanation of the change:
Only the default loader currently supports dynamic sample loading, so I thought it might be a good idea to keep the changes for this feature mostly contained in the default loader as well. I've added two new preset notify flags (FLUID_PRESET_PIN and FLUID_PRESET_UNPIN) that are handled by the preset->notify callback and trigger the loading and possibly unloading of the samples.
