This no longer uses a script to generate code for every possible type
conversion or resampler. This caused a bloat in binary size and and compile
times. Now we use a handful of more generic functions and assume staying in
the CPU cache is the most important thing anyhow.
This shrinks the size of the final build (in this case: macOS X amd64, -Os to
optimize for size) by 15%. When compiling on a single core, build times drop
by about 15% too (although the previous cost was largely hidden by multicore
builds).
--HG--
extra : amend_source : ce9deadd24e237eb83d2ef900c493f25c7cf3a80
- Use the SDL_AUDIO_MASK_DATATYPE bit when selecting an implementation
where it matters. Previously two existing AUDIO_F32 cases had been
written, but were unreachable.
- Add AUDIO_F32 case for SDL_ConvertSurround_4.
- Fix incorrect pointer arithmetic causing the 2 to 6 channel
conversion for 4 byte audio formats to read and write beyond the end
of the buffer.
This lets us change things like this...
if (Failed) {
SDL_SetError("We failed");
return -1;
}
...into this...
if (Failed) {
return SDL_SetError("We failed");
}
Fixes Bugzilla #1778.
Markovtsev Vadim 2011-01-18 22:00:16 PST
SDL_audiocvt.c:
static void SDLCALL
SDL_ConvertStereo(SDL_AudioCVT * cvt, SDL_AudioFormat format):
#define dup_chans_1_to_2(type) \
{ \
const type *src = (const type *) (cvt->buf + cvt->len_cvt); \
type *dst = (type *) (cvt->buf + cvt->len_cvt * 2); \
for (i = cvt->len_cvt / 2; i; --i, --src) { \
const type val = *src; \
dst -= 2; \
dst[0] = dst[1] = val; \
} \
}
Pay attention to cvt->len_cvt / 2. 2 is the sizeof(Uint16), hovewer, below we
see that the conversion function supports Uint8 and Uint32:
switch (SDL_AUDIO_BITSIZE(format)) {
case 8:
dup_chans_1_to_2(Uint8);
break;
case 16:
dup_chans_1_to_2(Uint16);
break;
case 32:
dup_chans_1_to_2(Uint32);
break;
}
If type is Uint32, src will be decreased twice as it should be, memory being
written before the cvt->buf. If type is Uint8, the conversion will not be
complete. I suggest to change that define to
#define dup_chans_1_to_2(type) \
{ \
const type *src = (const type *) (cvt->buf + cvt->len_cvt); \
type *dst = (type *) (cvt->buf + cvt->len_cvt * 2); \
for (i = cvt->len_cvt / sizeof(type); i; --i, --src) { \
const type val = *src; \
dst -= 2; \
dst[0] = dst[1] = val; \
} \
}
I tested that and now it's working fine. I did not consider the similar defines
in functions nearby.
* Some math functions become intrinsic in release mode, so we need to
convert all the math functions into SDL math functions, like we did
with the stdlib functions.
* Constant initializers of 8-bit values become calls to memset() in
release mode, but memset() itself is an intrinsic when explicitly
called. So we'll just explicitly call memset() in those cases.
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%403474
FIXME: this is a memory leak. We don't have an SDL_FreeAudioCVT() yet.
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%403342
Bring SDL to iPhone and iPod Touch
by Holmes Futrell, mentored by Sam Lantinga
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%403222
Audio Ideas - Resampling and Pitch Shifting
by Aaron Wishnick, mentored by Ryan C. Gordon
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%403165
bitshift an Sint64, but it can't find this function since we don't use the
C runtime on Windows.
Division doesn't have this problem, though. Strange.
Thanks, Suzuki Masahiro.
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%402242
I might be on crack here.
It looks like SDL_ConvertMono() in src/audio/SDL_audiocvt.c adds the left and
right channels of a stereo stream together, and clamps the new mono channel if
it would overflow.
Shouldn't it be dividing by 2 to average the two sample points instead of
clamping? Otherwise the mono sample point's volume doubles in the conversion.
This would also make the conversion faster, as it replaces two branches per
sample frame with a bitwise shift.
--ryan.
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%402119
Notable changes:
- Converters between types are autogenerated. Instead of making multiple
passes over the data with seperate filters for endianess, size, signedness,
etc, converting between data types is always one specialized filter. This
simplifies SDL_BuildAudioCVT(), which otherwise had a million edge cases
with the new types, and makes the actually conversions more CPU cache
friendly. Left a stub for adding specific optimized versions of these
routines (SSE/MMX/Altivec, assembler, etc)
- Autogenerated converters are built by SDL/src/audio/sdlgenaudiocvt.pl. This
does not need to be run unless tweaking the code, and thus doesn't need
integration into the build system.
- Went through all the drivers and tried to weed out all the "Uint16"
references that are better specified with the new SDL_AudioFormat typedef.
- Cleaned out a bunch of hardcoded bitwise magic numbers and replaced them
with new SDL_AUDIO_* macros.
- Added initial float32 and int32 support code. Theoretically, existing
drivers will push these through converters to get the data they want to
feed to the hardware.
Still TODO:
- Optimize and debug new converters.
- Update the CoreAudio backend to accept float32 data directly.
- Other backends, too?
- SDL_LoadWAV() needs to be updated to support int32 and float32 .wav files
(both of which exist and can be generated by 'sox' for testing purposes).
- Update the mixer to handle new datatypes.
- Optionally update SDL_sound and SDL_mixer, etc.
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%402029
The current SVN trunk is missing the SDLCALL specifier at numerous locations.
It has to be added for all (possibly user provided) callbacks.
I stumbled over this while creating a makefile for the OpenWatcom compiler for
Win32.
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%401728
Make sure every source file includes SDL_config.h, so the proper system
headers are chosen.
--HG--
extra : convert_revision : svn%3Ac70aab31-4412-0410-b14c-859654838e24/trunk%401406
