Don't rely on checking __clang_major__ since it is not comparable
between different vendors. Don't use "#pragma clang attribute" since it
is only available in relatively recent versions, there's no obvious way
to check if it's supported, and just using __attribute__ directly (for
gcc as well) results in simpler code anyway.
This is needed to support CHERI, and thus Arm's experimental Morello
prototype, where pointers are implemented using unforgeable capabilities
that include bounds and permissions metadata to provide fine-grained
spatial and referential memory safety, as well as revocation by sweeping
memory to provide heap temporal memory safety.
On most systems (anything with a flat memory hierarchy rather than using
segment-based addressing), size_t and uintptr_t are the same type.
However, on CHERI, size_t is just an integer offset, whereas uintptr_t
is still a capability as described above. Casting a pointer to size_t
will strip the metadata and validity tag, and casting from size_t to a
pointer will result in a null-derived capability whose validity tag is
not set, and thus cannot be dereferenced without faulting.
The audio and cursor casts were harmless as they intend to stuff an
integer into a pointer, but using uintptr_t is the idiomatic way to do
that and silences our compiler warnings (which our build tool makes
fatal by default as they often indicate real problems). The iconv and
egl casts were true positives as SDL_iconv_t and iconv_t are pointer
types, as is NativeDisplayType on most OSes, so this would have trapped
at run time when using the round-tripped pointers. The gles2 casts were
also harmless; the OpenGL API defines this argument to be a pointer type
(and uses the argument name "pointer"), but it in fact represents an
integer offset, so like audio and cursor the additional idiomatic cast
is needed to silence the warning.
On modern CPUs, there's no penalty for using the unaligned instruction on
aligned memory, but now it can vectorize unaligned data too, which even if
it's not optimal, is still going to be faster than the scalar fallback.
Fixes#4532.
janisozaur
There are many cases which are not able to be handled by SDL's audio conversion routines, including too low (negative) rate, too high rate (impossible to allocate).
This patch aims to report such issues early and handle others in a graceful manner. The "INT32_MAX / RESAMPLER_SAMPLES_PER_ZERO_CROSSING" value is the conservative approach in terms of what can _technically_ be supported, but its value is 4'194'303, or just shy of 4.2MHz. I highly doubt any sane person would use such rates, especially in SDL2, so I would like to drive this limit further down, but would need some assistance to do that, as doing so would have to introduce an arbitrary value. Are you OK with such approach? What would a good value be? Wikipedia (https://en.wikipedia.org/wiki/High-resolution_audio) lists 96kHz as the highest sampling rate in use, even if I quadruple it for a good measure, to 384kHz it's still an order of magnitude lower than 4MHz.
(I thought padding size ranged from 5 frames to ~30 frames (based around
RESAMPLER_ZERO_CROSSINGS, which is 5), but it's actually between 512 and
several thousands (based on RESAMPLER_SAMPLES_PER_ZERO_CROSSING)). It gets
big fast when downsampling.
Previously, the padding was silence, which was a problem when streaming since
you would sample a little bit of this silence between each buffer.
We still need a means to get padding data for the right hand side, but this
patch makes the resampler output more correct.
This time it's using real math from a real whitepaper instead of my previous
amateur, fast-but-low-quality attempt. The new resampler does "bandlimited
interpolation," as described here: https://ccrma.stanford.edu/~jos/resample/
The output appears to sound cleaner, especially at high frequencies, and of
course works with non-power-of-two rate conversions.
There are some obvious optimizations to be done to this still, and there is
other fallout: this doesn't resample a buffer in-place, the 2-channels-Sint16
fast path is gone because this resampler does a _lot_ of floating point math.
There is a nasty hack to make it work with SDL_AudioCVT.
It's possible these issues are solvable, but they aren't solved as of yet.
Still, I hope this effort is slouching in the right direction.
Simon Hug
This issue actually raises the question if this API change (requirement of initialized audio subsystem) is breaking backwards compatibility. I don't see the documentation saying it is needed in 2.0.5.
"Major changes, roughly in order of appearance:
- Use float math everywhere, instead of promoting to double and casting back
all the time.
- Conserve sound energy when downmixing any channel into two other channels.
- Add a QuadToStereo filter. (The previous technique of reusing StereoToMono
never worked, since it assumed an incorrect channel layout for 4.0.)
- Add a 71to51 filter. This removes just under half of the cases the previous
code would silently break in.
- Add a QuadTo51 filter. More silent breakage fixed.
- Add a 51to71 filter, removing another almost-half of the silently broken
cases.
- Add 8 to the list of values SDL_SupportedChannelCount will accept.
- Change SDL_BuildAudioCVT's channel-related logic to handle every case, and
to actually fail if it fails instead of silently corrupting sound data and/or
crashing down the road."
(Note that SDL doesn't otherwise support 7.1 audio yet, but hopefully it will
soon and the 7.1 converters are an important piece of that. --ryan.)
Fixes Bugzilla #3727.
