This doesn't implement full sound support but stubs out most of the
sound code under SDL3, so that we can implement it piecemeal.
Implemented here:
1. Sound device initialization.
2. SVid sound playback.
Does not add a dependency on `SDL_mixer`: SDL3 built-ins
are enough to play SVid audio.
In C++, globals initialization order accross translation units is not
defined. Accessing a global via a function ensures that it is initialized.
This will be needed for #7638, which will statically initialize change
handlers after the Options object has been initialized.
SMK format actually defines frame durations in units that are 0.01ms:
91e732bb69/src/SmackerDecoder.cpp (L329-L334)
That's great because it means we can do everything using integer math.
Each resampler allocates an input buffer and an output buffer.
Since we create ~100 audio stream objects (1 per `sfx_MISC` effect),
the memory cost of these buffers adds up.
Avoids creating a resampler when the audio sampler rate matches the
output sample rate. This is mostly the case when the output sample
rate is 22050.
When using `UNPACKED_MPQS`, avoid all the SDL machinery for reading
files.
This is beneficial not only due to reduced indirection but also because
we can test for the file's existence and get the file size without
opening it, which is much faster.
1. Removes adjustments to `orig/system_palette`.
They're not needed because videos have a separate render loop.
For an in-game movie, the caller sets the palette after the movie.
2. Splits some of `SVidPlayContinue` code into functions
(`UpdatePalette` and `BlitFrame`).
3. Adds a separate config variable for SDL1 SVid video mode.
E.g. on rg99 there isn't enough video RAM for double-buffering
the main game but we can do it for the videos.
libmpq is a much simpler alternative to StormLib for reading MPQ archives.
We use our own fork of libmpq: https://github.com/diasurgical/libmpq
Impact:
* DevilutionX is now a lot more portable. Unlike StormLib, libmpq only
needs platform-specific code for Windows.
* Locks around file access **removed** (instead we duplicate the file descriptor for streamed audio only).
* RAM usage is **300 KiB** lower than StormLib.
* Stripped release linux_x86_64 binary is **32 KiB** smaller.
* Amiga build now hangs instead of crashing.
`std::make_unique<T[]>(size)` always zero-initalizes the array.
C++20 has `std::make_unique_for_overwrite` to avoid that, while
older C++ versions can use the approach applied here.
Previously, we always loaded entire videos in memory before playing them.
`libsmacker` does not provide a streaming API but it does provide a
`FILE *` file pointer API.
We now take advantage of the `FILE *` API by streaming the videos on
platforms that support `fopencookie`.
This means faster startup and less memory usage on these platforms.
This option completely disables all audio handling, including audio
loading code and dependencies.
Dialog text length is estimated to be somewhere between
Cain and Griswold speed.
SDL_mixer can only stream a single music track
SDL_audiolib has unlimited streams.
With this change, we finally have streaming sounds (respecting
sfx_STREAM).
Audio options can now also be set via diablo.ini, which should help us
better diagnose the static noise issues.