Relintai/sdl2_frt
1
0
Fork 0
mirror of https://github.com/Relintai/sdl2_frt.git synced 2024-12-29 20:27:12 +01:00
Code Issues Packages Projects Releases Wiki Activity
9a291c1e59
sdl2_frt/src/audio/SDL_audiocvt.c

1674 lines
57 KiB
C
Raw Normal View History

Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
/*
Simple DirectMedia Layer
Updated copyright for 2017
2017-01-02 03:33:28 +01:00
Copyright (C) 1997-2017 Sam Lantinga <slouken@libsdl.org>
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
This software is provided 'as-is', without any express or implied
warranty. In no event will the authors be held liable for any damages
arising from the use of this software.
Permission is granted to anyone to use this software for any purpose,
including commercial applications, and to alter it and redistribute it
freely, subject to the following restrictions:
1. The origin of this software must not be misrepresented; you must not
claim that you wrote the original software. If you use this software
in a product, an acknowledgment in the product documentation would be
appreciated but is not required.
2. Altered source versions must be plainly marked as such, and must not be
misrepresented as being the original software.
3. This notice may not be removed or altered from any source distribution.
*/
#include "../SDL_internal.h"
/* Functions for audio drivers to perform runtime conversion of audio format */
Added a note about adjusting channel weights when converting to fewer channels
2017-10-20 23:51:22 +02:00
/* FIXME: Channel weights when converting from more channels to fewer may need to be adjusted, see https://msdn.microsoft.com/en-us/library/windows/desktop/ff819070(v=vs.85).aspx
*/
audio: better docs on conversion APIs, error if not init'd (thanks, Simon!). Fixes Bugzilla #3662.
2017-08-18 22:52:19 +02:00
#include "SDL.h"
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
#include "SDL_audio.h"
#include "SDL_audio_c.h"
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
#include "SDL_loadso.h"
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
#include "SDL_assert.h"
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
#include "../SDL_dataqueue.h"
audio: Added SSE3 implementation of SDL_ConvertStereoToMono().
2017-01-23 06:57:19 +01:00
#include "SDL_cpuinfo.h"
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
#define DEBUG_AUDIOSTREAM 0
audio: Wired up new SSE code to build system.
2017-01-23 07:05:44 +01:00
#ifdef __SSE3__
#define HAVE_SSE3_INTRINSICS 1
audio: Added SSE3 implementation of SDL_ConvertStereoToMono().
2017-01-23 06:57:19 +01:00
#endif
#if HAVE_SSE3_INTRINSICS
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Convert from stereo to mono. Average left and right. */
audio: Added SSE3 implementation of SDL_ConvertStereoToMono().
2017-01-23 06:57:19 +01:00
static void SDLCALL
SDL_ConvertStereoToMono_SSE3(SDL_AudioCVT * cvt, SDL_AudioFormat format)
{
float *dst = (float *) cvt->buf;
const float *src = dst;
int i = cvt->len_cvt / 8;
LOG_DEBUG_CONVERT("stereo", "mono (using SSE3)");
SDL_assert(format == AUDIO_F32SYS);
/* We can only do this if dst is aligned to 16 bytes; since src is the
same pointer and it moves by 2, it can't be forcibly aligned. */
if ((((size_t) dst) & 15) == 0) {
/* Aligned! Do SSE blocks as long as we have 16 bytes available. */
const __m128 divby2 = _mm_set1_ps(0.5f);
while (i >= 4) { /* 4 * float32 */
_mm_store_ps(dst, _mm_mul_ps(_mm_hadd_ps(_mm_load_ps(src), _mm_load_ps(src+4)), divby2));
i -= 4; src += 8; dst += 4;
}
}
/* Finish off any leftovers with scalar operations. */
while (i) {
*dst = (src[0] + src[1]) * 0.5f;
dst++; i--; src += 2;
}
cvt->len_cvt /= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
#endif
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Convert from stereo to mono. Average left and right. */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
static void SDLCALL
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
SDL_ConvertStereoToMono(SDL_AudioCVT * cvt, SDL_AudioFormat format)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
float *dst = (float *) cvt->buf;
const float *src = dst;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
int i;
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
LOG_DEBUG_CONVERT("stereo", "mono");
SDL_assert(format == AUDIO_F32SYS);
for (i = cvt->len_cvt / 8; i; --i, src += 2) {
audio: don't cast to double in SDL_ConvertStereoToMono(). It's expensive and (hopefully) unnecessary. If this becomes an overflow problem, we could multiply both values by 0.5f before adding them, but let's see if we can get by without the extra multiplication first.
2017-01-23 02:18:59 +01:00
*(dst++) = (src[0] + src[1]) * 0.5f;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
cvt->len_cvt /= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Convert from 5.1 to stereo. Average left and right, distribute center, discard LFE. */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
static void SDLCALL
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
SDL_Convert51ToStereo(SDL_AudioCVT * cvt, SDL_AudioFormat format)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
float *dst = (float *) cvt->buf;
const float *src = dst;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
int i;
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
LOG_DEBUG_CONVERT("5.1", "stereo");
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
SDL_assert(format == AUDIO_F32SYS);
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* SDL's 5.1 layout: FL+FR+FC+LFE+BL+BR */
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
for (i = cvt->len_cvt / (sizeof (float) * 6); i; --i, src += 6, dst += 2) {
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
const float front_center_distributed = src[2] * 0.5f;
dst[0] = (src[0] + front_center_distributed + src[4]) / 2.5f; /* left */
dst[1] = (src[1] + front_center_distributed + src[5]) / 2.5f; /* right */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
cvt->len_cvt /= 3;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Convert from quad to stereo. Average left and right. */
static void SDLCALL
SDL_ConvertQuadToStereo(SDL_AudioCVT * cvt, SDL_AudioFormat format)
{
float *dst = (float *) cvt->buf;
const float *src = dst;
int i;
LOG_DEBUG_CONVERT("quad", "stereo");
SDL_assert(format == AUDIO_F32SYS);
for (i = cvt->len_cvt / (sizeof (float) * 4); i; --i, src += 4, dst += 2) {
dst[0] = (src[0] + src[2]) * 0.5f; /* left */
dst[1] = (src[1] + src[3]) * 0.5f; /* right */
}
cvt->len_cvt /= 3;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
/* Convert from 7.1 to 5.1. Distribute sides across front and back. */
static void SDLCALL
SDL_Convert71To51(SDL_AudioCVT * cvt, SDL_AudioFormat format)
{
float *dst = (float *) cvt->buf;
const float *src = dst;
int i;
LOG_DEBUG_CONVERT("7.1", "5.1");
SDL_assert(format == AUDIO_F32SYS);
for (i = cvt->len_cvt / (sizeof (float) * 8); i; --i, src += 8, dst += 6) {
const float surround_left_distributed = src[6] * 0.5f;
const float surround_right_distributed = src[7] * 0.5f;
dst[0] = (src[0] + surround_left_distributed) / 1.5f; /* FL */
dst[1] = (src[1] + surround_right_distributed) / 1.5f; /* FR */
dst[2] = src[2] / 1.5f; /* CC */
dst[3] = src[3] / 1.5f; /* LFE */
dst[4] = (src[4] + surround_left_distributed) / 1.5f; /* BL */
dst[5] = (src[5] + surround_right_distributed) / 1.5f; /* BR */
}
cvt->len_cvt /= 8;
cvt->len_cvt *= 6;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
/* Convert from 5.1 to quad. Distribute center across front, discard LFE. */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
static void SDLCALL
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
SDL_Convert51ToQuad(SDL_AudioCVT * cvt, SDL_AudioFormat format)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
float *dst = (float *) cvt->buf;
const float *src = dst;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
int i;
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
LOG_DEBUG_CONVERT("5.1", "quad");
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
SDL_assert(format == AUDIO_F32SYS);
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* SDL's 4.0 layout: FL+FR+BL+BR */
/* SDL's 5.1 layout: FL+FR+FC+LFE+BL+BR */
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
for (i = cvt->len_cvt / (sizeof (float) * 6); i; --i, src += 6, dst += 4) {
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
const float front_center_distributed = src[2] * 0.5f;
dst[0] = (src[0] + front_center_distributed) / 1.5f; /* FL */
dst[1] = (src[1] + front_center_distributed) / 1.5f; /* FR */
dst[2] = src[4] / 1.5f; /* BL */
dst[3] = src[5] / 1.5f; /* BR */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
cvt->len_cvt /= 6;
cvt->len_cvt *= 4;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Upmix mono to stereo (by duplication) */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
static void SDLCALL
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
SDL_ConvertMonoToStereo(SDL_AudioCVT * cvt, SDL_AudioFormat format)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
const float *src = (const float *) (cvt->buf + cvt->len_cvt);
float *dst = (float *) (cvt->buf + cvt->len_cvt * 2);
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
int i;
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
LOG_DEBUG_CONVERT("mono", "stereo");
SDL_assert(format == AUDIO_F32SYS);
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
for (i = cvt->len_cvt / sizeof (float); i; --i) {
src--;
dst -= 2;
dst[0] = dst[1] = *src;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
cvt->len_cvt *= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Upmix stereo to a pseudo-5.1 stream */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
static void SDLCALL
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
SDL_ConvertStereoTo51(SDL_AudioCVT * cvt, SDL_AudioFormat format)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
int i;
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
float lf, rf, ce;
const float *src = (const float *) (cvt->buf + cvt->len_cvt);
float *dst = (float *) (cvt->buf + cvt->len_cvt * 3);
LOG_DEBUG_CONVERT("stereo", "5.1");
SDL_assert(format == AUDIO_F32SYS);
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
for (i = cvt->len_cvt / (sizeof(float) * 2); i; --i) {
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
dst -= 6;
src -= 2;
lf = src[0];
rf = src[1];
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
ce = (lf + rf) * 0.5f;
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* !!! FIXME: FL and FR may clip */
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
dst[0] = lf + (lf - ce); /* FL */
dst[1] = rf + (rf - ce); /* FR */
dst[2] = ce; /* FC */
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
dst[3] = 0; /* LFE (only meant for special LFE effects) */
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
dst[4] = lf; /* BL */
dst[5] = rf; /* BR */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
cvt->len_cvt *= 3;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Upmix quad to a pseudo-5.1 stream */
static void SDLCALL
SDL_ConvertQuadTo51(SDL_AudioCVT * cvt, SDL_AudioFormat format)
{
int i;
float lf, rf, lb, rb, ce;
const float *src = (const float *) (cvt->buf + cvt->len_cvt);
float *dst = (float *) (cvt->buf + cvt->len_cvt * 3 / 2);
LOG_DEBUG_CONVERT("quad", "5.1");
SDL_assert(format == AUDIO_F32SYS);
SDL_assert(cvt->len_cvt % (sizeof(float) * 4) == 0);
for (i = cvt->len_cvt / (sizeof(float) * 4); i; --i) {
dst -= 6;
src -= 4;
lf = src[0];
rf = src[1];
lb = src[2];
rb = src[3];
ce = (lf + rf) * 0.5f;
/* !!! FIXME: FL and FR may clip */
dst[0] = lf + (lf - ce); /* FL */
dst[1] = rf + (rf - ce); /* FR */
dst[2] = ce; /* FC */
dst[3] = 0; /* LFE (only meant for special LFE effects) */
dst[4] = lb; /* BL */
dst[5] = rb; /* BR */
}
cvt->len_cvt = cvt->len_cvt * 3 / 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
/* Upmix stereo to a pseudo-4.0 stream (by duplication) */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
static void SDLCALL
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
SDL_ConvertStereoToQuad(SDL_AudioCVT * cvt, SDL_AudioFormat format)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
const float *src = (const float *) (cvt->buf + cvt->len_cvt);
float *dst = (float *) (cvt->buf + cvt->len_cvt * 2);
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
float lf, rf;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
int i;
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
LOG_DEBUG_CONVERT("stereo", "quad");
SDL_assert(format == AUDIO_F32SYS);
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
for (i = cvt->len_cvt / (sizeof(float) * 2); i; --i) {
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
dst -= 4;
src -= 2;
lf = src[0];
rf = src[1];
audio: Improvements in channel conversion code.
2017-01-08 22:18:49 +01:00
dst[0] = lf; /* FL */
dst[1] = rf; /* FR */
dst[2] = lf; /* BL */
dst[3] = rf; /* BR */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
cvt->len_cvt *= 2;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Upmix 5.1 to 7.1 */
static void SDLCALL
SDL_Convert51To71(SDL_AudioCVT * cvt, SDL_AudioFormat format)
{
float lf, rf, lb, rb, ls, rs;
int i;
const float *src = (const float *) (cvt->buf + cvt->len_cvt);
float *dst = (float *) (cvt->buf + cvt->len_cvt * 4 / 3);
LOG_DEBUG_CONVERT("5.1", "7.1");
SDL_assert(format == AUDIO_F32SYS);
SDL_assert(cvt->len_cvt % (sizeof(float) * 6) == 0);
for (i = cvt->len_cvt / (sizeof(float) * 6); i; --i) {
dst -= 8;
src -= 6;
lf = src[0];
rf = src[1];
lb = src[4];
rb = src[5];
ls = (lf + lb) * 0.5f;
rs = (rf + rb) * 0.5f;
/* !!! FIXME: these four may clip */
lf += lf - ls;
rf += rf - ls;
lb += lb - ls;
rb += rb - ls;
dst[3] = src[3]; /* LFE */
dst[2] = src[2]; /* FC */
dst[7] = rs; /* SR */
dst[6] = ls; /* SL */
dst[5] = rb; /* BR */
dst[4] = lb; /* BL */
dst[1] = rf; /* FR */
dst[0] = lf; /* FL */
}
cvt->len_cvt = cvt->len_cvt * 4 / 3;
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index] (cvt, format);
}
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
/* SDL's resampler uses a "bandlimited interpolation" algorithm:
https://ccrma.stanford.edu/~jos/resample/ */
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
#define RESAMPLER_ZERO_CROSSINGS 5
#define RESAMPLER_BITS_PER_SAMPLE 16
#define RESAMPLER_SAMPLES_PER_ZERO_CROSSING (1 << ((RESAMPLER_BITS_PER_SAMPLE / 2) + 1))
#define RESAMPLER_FILTER_SIZE ((RESAMPLER_SAMPLES_PER_ZERO_CROSSING * RESAMPLER_ZERO_CROSSINGS) + 1)
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
/* This is a "modified" bessel function, so you can't use POSIX j0() */
static double
bessel(const double x)
{
const double xdiv2 = x / 2.0;
double i0 = 1.0f;
double f = 1.0f;
int i = 1;
while (SDL_TRUE) {
const double diff = SDL_pow(xdiv2, i * 2) / SDL_pow(f, 2);
if (diff < 1.0e-21f) {
break;
audio: Make the simple resampler operate in-place. This allows us to avoid an extra copy, allocate less memory and reduce cache pressure. On the downside: we have to do a lot of tapdancing to resample the buffer in reverse when the output is growing.
2017-01-23 05:48:15 +01:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
i0 += diff;
i++;
f *= (double) i;
}
audio: Make the simple resampler operate in-place. This allows us to avoid an extra copy, allocate less memory and reduce cache pressure. On the downside: we have to do a lot of tapdancing to resample the buffer in reverse when the output is growing.
2017-01-23 05:48:15 +01:00
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
return i0;
}
/* build kaiser table with cardinal sine applied to it, and array of differences between elements. */
static void
kaiser_and_sinc(float *table, float *diffs, const int tablelen, const double beta)
{
const int lenm1 = tablelen - 1;
const int lenm1div2 = lenm1 / 2;
int i;
table[0] = 1.0f;
for (i = 1; i < tablelen; i++) {
const double kaiser = bessel(beta * SDL_sqrt(1.0 - SDL_pow(((i - lenm1) / 2.0) / lenm1div2, 2.0))) / bessel(beta);
table[tablelen - i] = (float) kaiser;
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
for (i = 1; i < tablelen; i++) {
const float x = (((float) i) / ((float) RESAMPLER_SAMPLES_PER_ZERO_CROSSING)) * ((float) M_PI);
table[i] *= SDL_sinf(x) / x;
diffs[i - 1] = table[i] - table[i - 1];
}
diffs[lenm1] = 0.0f;
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
static SDL_SpinLock ResampleFilterSpinlock = 0;
static float *ResamplerFilter = NULL;
static float *ResamplerFilterDifference = NULL;
int
SDL_PrepareResampleFilter(void)
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
{
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
SDL_AtomicLock(&ResampleFilterSpinlock);
if (!ResamplerFilter) {
/* if dB > 50, beta=(0.1102 * (dB - 8.7)), according to Matlab. */
const double dB = 80.0;
const double beta = 0.1102 * (dB - 8.7);
const size_t alloclen = RESAMPLER_FILTER_SIZE * sizeof (float);
ResamplerFilter = (float *) SDL_malloc(alloclen);
if (!ResamplerFilter) {
SDL_AtomicUnlock(&ResampleFilterSpinlock);
return SDL_OutOfMemory();
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
ResamplerFilterDifference = (float *) SDL_malloc(alloclen);
if (!ResamplerFilterDifference) {
SDL_free(ResamplerFilter);
ResamplerFilter = NULL;
SDL_AtomicUnlock(&ResampleFilterSpinlock);
return SDL_OutOfMemory();
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
kaiser_and_sinc(ResamplerFilter, ResamplerFilterDifference, RESAMPLER_FILTER_SIZE, beta);
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
SDL_AtomicUnlock(&ResampleFilterSpinlock);
return 0;
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
void
SDL_FreeResampleFilter(void)
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
{
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
SDL_free(ResamplerFilter);
SDL_free(ResamplerFilterDifference);
ResamplerFilter = NULL;
ResamplerFilterDifference = NULL;
}
Some patches to make SDL compile with armcc (ARM's C compiler).
2017-03-02 19:33:04 +01:00
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
static int
ResamplerPadding(const int inrate, const int outrate)
{
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
if (inrate == outrate) {
return 0;
} else if (inrate > outrate) {
return (int) SDL_ceil(((float) (RESAMPLER_SAMPLES_PER_ZERO_CROSSING * inrate) / ((float) outrate)));
}
return RESAMPLER_SAMPLES_PER_ZERO_CROSSING;
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
}
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
/* lpadding and rpadding are expected to be buffers of (ResamplePadding(inrate, outrate) * chans * sizeof (float)) bytes. */
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
static int
SDL_ResampleAudio(const int chans, const int inrate, const int outrate,
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
const float *lpadding, const float *rpadding,
const float *inbuf, const int inbuflen,
float *outbuf, const int outbuflen)
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
{
audio: Moved unchanging variable out of loop.
2017-10-11 08:31:58 +02:00
const double finrate = (double) inrate;
audio: Moved the resampler state up to double precision. Fixes more buffer overflows.
2017-10-11 17:51:14 +02:00
const double outtimeincr = 1.0 / ((float) outrate);
const double ratio = ((float) outrate) / ((float) inrate);
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
const int paddinglen = ResamplerPadding(inrate, outrate);
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
const int framelen = chans * (int)sizeof (float);
const int inframes = inbuflen / framelen;
const int wantedoutframes = (int) ((inbuflen / framelen) * ratio); /* outbuflen isn't total to write, it's total available. */
const int maxoutframes = outbuflen / framelen;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
const int outframes = SDL_min(wantedoutframes, maxoutframes);
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
float *dst = outbuf;
audio: Moved the resampler state up to double precision. Fixes more buffer overflows.
2017-10-11 17:51:14 +02:00
double outtime = 0.0;
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
int i, j, chan;
for (i = 0; i < outframes; i++) {
const int srcindex = (int) (outtime * inrate);
audio: Moved the resampler state up to double precision. Fixes more buffer overflows.
2017-10-11 17:51:14 +02:00
const double intime = ((double) srcindex) / finrate;
const double innexttime = ((double) (srcindex + 1)) / finrate;
const double interpolation1 = 1.0 - ((innexttime - outtime) / (innexttime - intime));
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
const int filterindex1 = (int) (interpolation1 * RESAMPLER_SAMPLES_PER_ZERO_CROSSING);
audio: Moved the resampler state up to double precision. Fixes more buffer overflows.
2017-10-11 17:51:14 +02:00
const double interpolation2 = 1.0 - interpolation1;
audio: Fixed compiler warning on Visual Studio.
2017-09-23 04:28:21 +02:00
const int filterindex2 = (int) (interpolation2 * RESAMPLER_SAMPLES_PER_ZERO_CROSSING);
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
for (chan = 0; chan < chans; chan++) {
float outsample = 0.0f;
/* do this twice to calculate the sample, once for the "left wing" and then same for the right. */
/* !!! FIXME: do both wings in one loop */
for (j = 0; (filterindex1 + (j * RESAMPLER_SAMPLES_PER_ZERO_CROSSING)) < RESAMPLER_FILTER_SIZE; j++) {
const int srcframe = srcindex - j;
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
/* !!! FIXME: we can bubble this conditional out of here by doing a pre loop. */
const float insample = (srcframe < 0) ? lpadding[((paddinglen + srcframe) * chans) + chan] : inbuf[(srcframe * chans) + chan];
Fixed compiler warning
2017-10-12 22:55:35 +02:00
outsample += (float)(insample * (ResamplerFilter[filterindex1 + (j * RESAMPLER_SAMPLES_PER_ZERO_CROSSING)] + (interpolation1 * ResamplerFilterDifference[filterindex1 + (j * RESAMPLER_SAMPLES_PER_ZERO_CROSSING)])));
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
}
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
for (j = 0; (filterindex2 + (j * RESAMPLER_SAMPLES_PER_ZERO_CROSSING)) < RESAMPLER_FILTER_SIZE; j++) {
const int srcframe = srcindex + 1 + j;
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
/* !!! FIXME: we can bubble this conditional out of here by doing a post loop. */
const float insample = (srcframe >= inframes) ? rpadding[((srcframe - inframes) * chans) + chan] : inbuf[(srcframe * chans) + chan];
Fixed compiler warning
2017-10-12 22:55:35 +02:00
outsample += (float)(insample * (ResamplerFilter[filterindex2 + (j * RESAMPLER_SAMPLES_PER_ZERO_CROSSING)] + (interpolation2 * ResamplerFilterDifference[filterindex2 + (j * RESAMPLER_SAMPLES_PER_ZERO_CROSSING)])));
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
}
*(dst++) = outsample;
}
audio: Turns out the accumulation errors sound better. :/ Moving to double fixed the overflows, but using "time = i * incr" instead of "time += incr" causes clicks in the output.
2017-10-11 18:07:43 +02:00
outtime += outtimeincr;
audio: Special case for resampling stereo AUDIO_S16SYS audio data. This is a fairly common case, so we avoid the conversion to/from float here.
2017-01-23 02:27:48 +01:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
return outframes * chans * sizeof (float);
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
int
SDL_ConvertAudio(SDL_AudioCVT * cvt)
{
/* !!! FIXME: (cvt) should be const; stack-copy it here. */
/* !!! FIXME: (actually, we can't...len_cvt needs to be updated. Grr.) */
/* Make sure there's data to convert */
if (cvt->buf == NULL) {
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
return SDL_SetError("No buffer allocated for conversion");
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
/* Return okay if no conversion is necessary */
cvt->len_cvt = cvt->len;
if (cvt->filters[0] == NULL) {
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
return 0;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
/* Set up the conversion and go! */
cvt->filter_index = 0;
cvt->filters[0] (cvt, cvt->src_format);
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
return 0;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
static void SDLCALL
SDL_Convert_Byteswap(SDL_AudioCVT *cvt, SDL_AudioFormat format)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
Fixed Windows build
2016-11-05 09:52:28 +01:00
#if DEBUG_CONVERT
printf("Converting byte order\n");
#endif
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
switch (SDL_AUDIO_BITSIZE(format)) {
#define CASESWAP(b) \
case b: { \
Uint##b *ptr = (Uint##b *) cvt->buf; \
int i; \
for (i = cvt->len_cvt / sizeof (*ptr); i; --i, ++ptr) { \
*ptr = SDL_Swap##b(*ptr); \
} \
break; \
}
CASESWAP(16);
CASESWAP(32);
CASESWAP(64);
#undef CASESWAP
default: SDL_assert(!"unhandled byteswap datatype!"); break;
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
if (cvt->filters[++cvt->filter_index]) {
/* flip endian flag for data. */
if (format & SDL_AUDIO_MASK_ENDIAN) {
format &= ~SDL_AUDIO_MASK_ENDIAN;
} else {
format |= SDL_AUDIO_MASK_ENDIAN;
}
cvt->filters[cvt->filter_index](cvt, format);
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
static int
SDL_AddAudioCVTFilter(SDL_AudioCVT *cvt, const SDL_AudioFilter filter)
{
if (cvt->filter_index >= SDL_AUDIOCVT_MAX_FILTERS) {
return SDL_SetError("Too many filters needed for conversion, exceeded maximum of %d", SDL_AUDIOCVT_MAX_FILTERS);
}
if (filter == NULL) {
return SDL_SetError("Audio filter pointer is NULL");
}
cvt->filters[cvt->filter_index++] = filter;
cvt->filters[cvt->filter_index] = NULL; /* Moving terminator */
return 0;
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
static int
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
SDL_BuildAudioTypeCVTToFloat(SDL_AudioCVT *cvt, const SDL_AudioFormat src_fmt)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
int retval = 0; /* 0 == no conversion necessary. */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
if ((SDL_AUDIO_ISBIGENDIAN(src_fmt) != 0) == (SDL_BYTEORDER == SDL_LIL_ENDIAN)) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, SDL_Convert_Byteswap) < 0) {
return -1;
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
retval = 1; /* added a converter. */
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
if (!SDL_AUDIO_ISFLOAT(src_fmt)) {
Patched to compile on C89 compilers.
2016-11-05 08:53:59 +01:00
const Uint16 src_bitsize = SDL_AUDIO_BITSIZE(src_fmt);
const Uint16 dst_bitsize = 32;
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
SDL_AudioFilter filter = NULL;
Patched to compile on C89 compilers.
2016-11-05 08:53:59 +01:00
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
switch (src_fmt & ~SDL_AUDIO_MASK_ENDIAN) {
case AUDIO_S8: filter = SDL_Convert_S8_to_F32; break;
case AUDIO_U8: filter = SDL_Convert_U8_to_F32; break;
case AUDIO_S16: filter = SDL_Convert_S16_to_F32; break;
Fixed audio conversion for unsigned 16 bit data.
2016-11-07 21:10:01 +01:00
case AUDIO_U16: filter = SDL_Convert_U16_to_F32; break;
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
case AUDIO_S32: filter = SDL_Convert_S32_to_F32; break;
default: SDL_assert(!"Unexpected audio format!"); break;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
if (!filter) {
audio: better docs on conversion APIs, error if not init'd (thanks, Simon!). Fixes Bugzilla #3662.
2017-08-18 22:52:19 +02:00
return SDL_SetError("No conversion from source format to float available");
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
}
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, filter) < 0) {
return -1;
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
if (src_bitsize < dst_bitsize) {
const int mult = (dst_bitsize / src_bitsize);
cvt->len_mult *= mult;
cvt->len_ratio *= mult;
} else if (src_bitsize > dst_bitsize) {
cvt->len_ratio /= (src_bitsize / dst_bitsize);
}
Patched to compile on C89 compilers.
2016-11-05 08:53:59 +01:00
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
retval = 1; /* added a converter. */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
return retval;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
static int
SDL_BuildAudioTypeCVTFromFloat(SDL_AudioCVT *cvt, const SDL_AudioFormat dst_fmt)
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
{
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
int retval = 0; /* 0 == no conversion necessary. */
if (!SDL_AUDIO_ISFLOAT(dst_fmt)) {
Also patched to compile on C89 compilers.
2016-11-05 08:56:55 +01:00
const Uint16 dst_bitsize = SDL_AUDIO_BITSIZE(dst_fmt);
const Uint16 src_bitsize = 32;
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
SDL_AudioFilter filter = NULL;
switch (dst_fmt & ~SDL_AUDIO_MASK_ENDIAN) {
case AUDIO_S8: filter = SDL_Convert_F32_to_S8; break;
case AUDIO_U8: filter = SDL_Convert_F32_to_U8; break;
case AUDIO_S16: filter = SDL_Convert_F32_to_S16; break;
Fixed audio conversion for unsigned 16 bit data.
2016-11-07 21:10:01 +01:00
case AUDIO_U16: filter = SDL_Convert_F32_to_U16; break;
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
case AUDIO_S32: filter = SDL_Convert_F32_to_S32; break;
default: SDL_assert(!"Unexpected audio format!"); break;
}
if (!filter) {
audio: better docs on conversion APIs, error if not init'd (thanks, Simon!). Fixes Bugzilla #3662.
2017-08-18 22:52:19 +02:00
return SDL_SetError("No conversion from float to destination format available");
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, filter) < 0) {
return -1;
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
if (src_bitsize < dst_bitsize) {
const int mult = (dst_bitsize / src_bitsize);
cvt->len_mult *= mult;
cvt->len_ratio *= mult;
} else if (src_bitsize > dst_bitsize) {
cvt->len_ratio /= (src_bitsize / dst_bitsize);
}
retval = 1; /* added a converter. */
}
if ((SDL_AUDIO_ISBIGENDIAN(dst_fmt) != 0) == (SDL_BYTEORDER == SDL_LIL_ENDIAN)) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, SDL_Convert_Byteswap) < 0) {
return -1;
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
retval = 1; /* added a converter. */
}
return retval;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
static void
SDL_ResampleCVT(SDL_AudioCVT *cvt, const int chans, const SDL_AudioFormat format)
{
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
/* !!! FIXME in 2.1: there are ten slots in the filter list, and the theoretical maximum we use is six (seven with NULL terminator).
!!! FIXME in 2.1: We need to store data for this resampler, because the cvt structure doesn't store the original sample rates,
!!! FIXME in 2.1: so we steal the ninth and tenth slot. :( */
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
const int inrate = (int) (size_t) cvt->filters[SDL_AUDIOCVT_MAX_FILTERS-1];
const int outrate = (int) (size_t) cvt->filters[SDL_AUDIOCVT_MAX_FILTERS];
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
const float *src = (const float *) cvt->buf;
const int srclen = cvt->len_cvt;
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
/*float *dst = (float *) cvt->buf;
const int dstlen = (cvt->len * cvt->len_mult);*/
/* !!! FIXME: remove this if we can get the resampler to work in-place again. */
float *dst = (float *) (cvt->buf + srclen);
const int dstlen = (cvt->len * cvt->len_mult) - srclen;
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
const int paddingsamples = (ResamplerPadding(inrate, outrate) * chans);
Fixed memory leak when HAVE_ALLOCA isn't defined
2017-09-22 17:51:45 +02:00
float *padding;
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
SDL_assert(format == AUDIO_F32SYS);
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
/* we keep no streaming state here, so pad with silence on both ends. */
audio: Don't stack-allocate resampler padding. (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.
2017-10-11 04:18:46 +02:00
padding = (float *) SDL_calloc(paddingsamples, sizeof (float));
Fixed memory leak when HAVE_ALLOCA isn't defined
2017-09-22 17:51:45 +02:00
if (!padding) {
SDL_OutOfMemory();
return;
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
cvt->len_cvt = SDL_ResampleAudio(chans, inrate, outrate, padding, padding, src, srclen, dst, dstlen);
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
audio: Don't stack-allocate resampler padding. (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.
2017-10-11 04:18:46 +02:00
SDL_free(padding);
Fixed memory leak when HAVE_ALLOCA isn't defined
2017-09-22 17:51:45 +02:00
audio: SDL_ResampleCVT() should use memmove instead of memcpy. This copy can overlap. Fixes Bugzilla #3849.
2017-10-11 04:31:02 +02:00
SDL_memmove(cvt->buf, dst, cvt->len_cvt); /* !!! FIXME: remove this if we can get the resampler to work in-place again. */
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
if (cvt->filters[++cvt->filter_index]) {
cvt->filters[cvt->filter_index](cvt, format);
}
}
/* !!! FIXME: We only have this macro salsa because SDL_AudioCVT doesn't
!!! FIXME: store channel info, so we have to have function entry
!!! FIXME: points for each supported channel count and multiple
!!! FIXME: vs arbitrary. When we rev the ABI, clean this up. */
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
#define RESAMPLER_FUNCS(chans) \
static void SDLCALL \
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
SDL_ResampleCVT_c##chans(SDL_AudioCVT *cvt, SDL_AudioFormat format) { \
SDL_ResampleCVT(cvt, chans, format); \
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
}
RESAMPLER_FUNCS(1)
RESAMPLER_FUNCS(2)
RESAMPLER_FUNCS(4)
RESAMPLER_FUNCS(6)
RESAMPLER_FUNCS(8)
#undef RESAMPLER_FUNCS
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
static SDL_AudioFilter
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
ChooseCVTResampler(const int dst_channels)
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
{
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
switch (dst_channels) {
case 1: return SDL_ResampleCVT_c1;
case 2: return SDL_ResampleCVT_c2;
case 4: return SDL_ResampleCVT_c4;
case 6: return SDL_ResampleCVT_c6;
case 8: return SDL_ResampleCVT_c8;
default: break;
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
}
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
return NULL;
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
}
static int
SDL_BuildAudioResampleCVT(SDL_AudioCVT * cvt, const int dst_channels,
const int src_rate, const int dst_rate)
{
SDL_AudioFilter filter;
if (src_rate == dst_rate) {
return 0; /* no conversion necessary. */
}
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
filter = ChooseCVTResampler(dst_channels);
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
if (filter == NULL) {
return SDL_SetError("No conversion available for these rates");
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
if (SDL_PrepareResampleFilter() < 0) {
return -1;
}
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
/* Update (cvt) with filter details... */
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, filter) < 0) {
return -1;
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
/* !!! FIXME in 2.1: there are ten slots in the filter list, and the theoretical maximum we use is six (seven with NULL terminator).
!!! FIXME in 2.1: We need to store data for this resampler, because the cvt structure doesn't store the original sample rates,
!!! FIXME in 2.1: so we steal the ninth and tenth slot. :( */
if (cvt->filter_index >= (SDL_AUDIOCVT_MAX_FILTERS-2)) {
return SDL_SetError("Too many filters needed for conversion, exceeded maximum of %d", SDL_AUDIOCVT_MAX_FILTERS-2);
}
cvt->filters[SDL_AUDIOCVT_MAX_FILTERS-1] = (SDL_AudioFilter) (size_t) src_rate;
cvt->filters[SDL_AUDIOCVT_MAX_FILTERS] = (SDL_AudioFilter) (size_t) dst_rate;
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
if (src_rate < dst_rate) {
const double mult = ((double) dst_rate) / ((double) src_rate);
cvt->len_mult *= (int) SDL_ceil(mult);
cvt->len_ratio *= mult;
} else {
cvt->len_ratio /= ((double) src_rate) / ((double) dst_rate);
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
/* !!! FIXME: remove this if we can get the resampler to work in-place again. */
/* the buffer is big enough to hold the destination now, but
we need it large enough to hold a separate scratch buffer. */
cvt->len_mult *= 2;
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
return 1; /* added a converter. */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
audio: Converter now checks a strict list of channels and formats we support.
2017-06-13 03:35:24 +02:00
static SDL_bool
SDL_SupportedAudioFormat(const SDL_AudioFormat fmt)
{
switch (fmt) {
case AUDIO_U8:
case AUDIO_S8:
case AUDIO_U16LSB:
case AUDIO_S16LSB:
case AUDIO_U16MSB:
case AUDIO_S16MSB:
case AUDIO_S32LSB:
case AUDIO_S32MSB:
case AUDIO_F32LSB:
case AUDIO_F32MSB:
return SDL_TRUE; /* supported. */
default:
break;
}
return SDL_FALSE; /* unsupported. */
}
static SDL_bool
SDL_SupportedChannelCount(const int channels)
{
switch (channels) {
case 1: /* mono */
case 2: /* stereo */
case 4: /* quad */
case 6: /* 5.1 */
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
case 8: /* 7.1 */
return SDL_TRUE; /* supported. */
audio: Converter now checks a strict list of channels and formats we support.
2017-06-13 03:35:24 +02:00
default:
break;
}
return SDL_FALSE; /* unsupported. */
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
/* Creates a set of audio filters to convert from one format to another.
audio: better docs on conversion APIs, error if not init'd (thanks, Simon!). Fixes Bugzilla #3662.
2017-08-18 22:52:19 +02:00
Returns 0 if no conversion is needed, 1 if the audio filter is set up,
or -1 if an error like invalid parameter, unsupported format, etc. occurred.
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
*/
int
SDL_BuildAudioCVT(SDL_AudioCVT * cvt,
SDL_AudioFormat src_fmt, Uint8 src_channels, int src_rate,
SDL_AudioFormat dst_fmt, Uint8 dst_channels, int dst_rate)
{
/* Sanity check target pointer */
if (cvt == NULL) {
return SDL_InvalidParamError("cvt");
}
Don't do any audio conversion if none is necessary
2017-01-06 08:53:46 +01:00
/* Make sure we zero out the audio conversion before error checking */
SDL_zerop(cvt);
audio: Converter now checks a strict list of channels and formats we support.
2017-06-13 03:35:24 +02:00
if (!SDL_SupportedAudioFormat(src_fmt)) {
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
return SDL_SetError("Invalid source format");
audio: Converter now checks a strict list of channels and formats we support.
2017-06-13 03:35:24 +02:00
} else if (!SDL_SupportedAudioFormat(dst_fmt)) {
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
return SDL_SetError("Invalid destination format");
audio: Converter now checks a strict list of channels and formats we support.
2017-06-13 03:35:24 +02:00
} else if (!SDL_SupportedChannelCount(src_channels)) {
return SDL_SetError("Invalid source channels");
} else if (!SDL_SupportedChannelCount(dst_channels)) {
return SDL_SetError("Invalid destination channels");
} else if (src_rate == 0) {
return SDL_SetError("Source rate is zero");
} else if (dst_rate == 0) {
return SDL_SetError("Destination rate is zero");
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Fixed Windows build
2016-11-05 09:52:28 +01:00
#if DEBUG_CONVERT
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
printf("Build format %04x->%04x, channels %u->%u, rate %d->%d\n",
src_fmt, dst_fmt, src_channels, dst_channels, src_rate, dst_rate);
#endif
/* Start off with no conversion necessary */
cvt->src_format = src_fmt;
cvt->dst_format = dst_fmt;
cvt->needed = 0;
cvt->filter_index = 0;
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
SDL_zero(cvt->filters);
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
cvt->len_mult = 1;
cvt->len_ratio = 1.0;
cvt->rate_incr = ((double) dst_rate) / ((double) src_rate);
Fixed bug 3662 - Error message when using the audio conversion setup without an initialized audio subsystem is a bit vague 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.
2017-08-29 06:42:39 +02:00
/* Make sure we've chosen audio conversion functions (MMX, scalar, etc.) */
SDL_ChooseAudioConverters();
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
/* Type conversion goes like this now:
- byteswap to CPU native format first if necessary.
- convert to native Float32 if necessary.
- resample and change channel count if necessary.
- convert back to native format.
- byteswap back to foreign format if necessary.
The expectation is we can process data faster in float32
(possibly with SIMD), and making several passes over the same
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
buffer is likely to be CPU cache-friendly, avoiding the
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
biggest performance hit in modern times. Previously we had
(script-generated) custom converters for every data type and
it was a bloat on SDL compile times and final library size. */
Don't do any audio conversion if none is necessary
2017-01-06 08:53:46 +01:00
/* see if we can skip float conversion entirely. */
if (src_rate == dst_rate && src_channels == dst_channels) {
if (src_fmt == dst_fmt) {
return 0;
}
/* just a byteswap needed? */
if ((src_fmt & ~SDL_AUDIO_MASK_ENDIAN) == (dst_fmt & ~SDL_AUDIO_MASK_ENDIAN)) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, SDL_Convert_Byteswap) < 0) {
return -1;
}
Don't do any audio conversion if none is necessary
2017-01-06 08:53:46 +01:00
cvt->needed = 1;
return 1;
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
/* Convert data types, if necessary. Updates (cvt). */
Don't do any audio conversion if none is necessary
2017-01-06 08:53:46 +01:00
if (SDL_BuildAudioTypeCVTToFloat(cvt, src_fmt) < 0) {
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
return -1; /* shouldn't happen, but just in case... */
}
/* Channel conversion */
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
if (src_channels < dst_channels) {
/* Upmixing */
/* Mono -> Stereo [-> ...] */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
if ((src_channels == 1) && (dst_channels > 1)) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, SDL_ConvertMonoToStereo) < 0) {
return -1;
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
cvt->len_mult *= 2;
src_channels = 2;
cvt->len_ratio *= 2;
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* [Mono ->] Stereo -> 5.1 [-> 7.1] */
if ((src_channels == 2) && (dst_channels >= 6)) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, SDL_ConvertStereoTo51) < 0) {
return -1;
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
src_channels = 6;
cvt->len_mult *= 3;
cvt->len_ratio *= 3;
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* Quad -> 5.1 [-> 7.1] */
if ((src_channels == 4) && (dst_channels >= 6)) {
if (SDL_AddAudioCVTFilter(cvt, SDL_ConvertQuadTo51) < 0) {
return -1;
}
src_channels = 6;
cvt->len_mult = (cvt->len_mult * 3 + 1) / 2;
cvt->len_ratio *= 1.5;
}
/* [[Mono ->] Stereo ->] 5.1 -> 7.1 */
if ((src_channels == 6) && (dst_channels == 8)) {
if (SDL_AddAudioCVTFilter(cvt, SDL_Convert51To71) < 0) {
return -1;
}
src_channels = 8;
cvt->len_mult = (cvt->len_mult * 4 + 2) / 3;
/* Should be numerically exact with every valid input to this
function */
cvt->len_ratio = cvt->len_ratio * 4 / 3;
}
/* [Mono ->] Stereo -> Quad */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
if ((src_channels == 2) && (dst_channels == 4)) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, SDL_ConvertStereoToQuad) < 0) {
return -1;
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
src_channels = 4;
cvt->len_mult *= 2;
cvt->len_ratio *= 2;
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
} else if (src_channels > dst_channels) {
/* Downmixing */
/* 7.1 -> 5.1 [-> Stereo [-> Mono]] */
/* 7.1 -> 5.1 [-> Quad] */
if ((src_channels == 8) && (dst_channels <= 6)) {
if (SDL_AddAudioCVTFilter(cvt, SDL_Convert71To51) < 0) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
return -1;
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
src_channels = 6;
cvt->len_ratio *= 0.75;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* [7.1 ->] 5.1 -> Stereo [-> Mono] */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
if ((src_channels == 6) && (dst_channels <= 2)) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, SDL_Convert51ToStereo) < 0) {
return -1;
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
src_channels = 2;
cvt->len_ratio /= 3;
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* 5.1 -> Quad */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
if ((src_channels == 6) && (dst_channels == 4)) {
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, SDL_Convert51ToQuad) < 0) {
return -1;
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
src_channels = 4;
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
cvt->len_ratio = cvt->len_ratio * 2 / 3;
}
/* Quad -> Stereo [-> Mono] */
if ((src_channels == 4) && (dst_channels <= 2)) {
if (SDL_AddAudioCVTFilter(cvt, SDL_ConvertQuadToStereo) < 0) {
return -1;
}
src_channels = 2;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
cvt->len_ratio /= 2;
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
/* [... ->] Stereo -> Mono */
if ((src_channels == 2) && (dst_channels == 1)) {
audio: Added SSE3 implementation of SDL_ConvertStereoToMono().
2017-01-23 06:57:19 +01:00
SDL_AudioFilter filter = NULL;
#if HAVE_SSE3_INTRINSICS
if (SDL_HasSSE3()) {
filter = SDL_ConvertStereoToMono_SSE3;
}
#endif
if (!filter) {
filter = SDL_ConvertStereoToMono;
}
Fixed bug 3668 - Overflow of SDL_AudioCVT.filters with some downmixes Simon Hug There's a chance that an audio conversion from many channels to a few can use more than 9 audio filters. SDL_AudioCVT has 10 SDL_AudioFilter pointers of which one has to be the terminating NULL pointer. The SDL code has no checks for this limit. If it overflows there can be stack or heap corruption or a call to 0xa. Attached patch adds a function that checks for this limit and throws an error if it is reached. Also adds some documentation. Test parameters that trigger this issue: AUDIO_U16MSB with 224 channels at 46359 Hz V AUDIO_S16MSB with 6 channels at 27463 Hz The fuzzer program I uploaded in bug 3667 has more of them.
2017-06-13 01:39:15 +02:00
if (SDL_AddAudioCVTFilter(cvt, filter) < 0) {
return -1;
}
audio: Added SSE3 implementation of SDL_ConvertStereoToMono().
2017-01-23 06:57:19 +01:00
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
src_channels = 1;
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
cvt->len_ratio /= 2;
}
}
audio: A whole bunch of improvements to audio conversion (thanks, Solra!). "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.
2017-08-29 06:41:45 +02:00
if (src_channels != dst_channels) {
/* All combinations of supported channel counts should have been
handled by now, but let's be defensive */
return SDL_SetError("Invalid channel combination");
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
/* Do rate conversion, if necessary. Updates (cvt). */
Don't do any audio conversion if none is necessary
2017-01-06 08:53:46 +01:00
if (SDL_BuildAudioResampleCVT(cvt, dst_channels, src_rate, dst_rate) < 0) {
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
return -1; /* shouldn't happen, but just in case... */
}
audio: More effort to improve and simplify audio resamplers.
2017-01-06 01:12:20 +01:00
/* Move to final data type. */
Don't do any audio conversion if none is necessary
2017-01-06 08:53:46 +01:00
if (SDL_BuildAudioTypeCVTFromFloat(cvt, dst_fmt) < 0) {
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
return -1; /* shouldn't happen, but just in case... */
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
}
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
cvt->needed = (cvt->filter_index != 0);
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
return (cvt->needed);
}
audio: allow stereo Sint16 resampling fast path in SDL_AudioStream. This currently favors libsamplerate over the fast path (quality over speed), but I'm not sure that's the correct approach, as there may be surprising changes in performance metrics depending on what packages are available on a user's system. That being said, currently, the only thing with access to SDL_AudioStream is an SDL audio device's thread, and it might be mostly idle otherwise, so maybe this is generally good.
2017-01-24 06:08:24 +01:00
typedef int (*SDL_ResampleAudioStreamFunc)(SDL_AudioStream *stream, const void *inbuf, const int inbuflen, void *outbuf, const int outbuflen);
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
typedef void (*SDL_ResetAudioStreamResamplerFunc)(SDL_AudioStream *stream);
typedef void (*SDL_CleanupAudioStreamResamplerFunc)(SDL_AudioStream *stream);
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
Added audio stream conversion functions: SDL_NewAudioStream SDL_AudioStreamPut SDL_AudioStreamGet SDL_AudioStreamAvailable SDL_AudioStreamClear SDL_FreeAudioStream
2017-10-19 00:54:05 +02:00
struct _SDL_AudioStream
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
{
SDL_AudioCVT cvt_before_resampling;
SDL_AudioCVT cvt_after_resampling;
SDL_DataQueue *queue;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
SDL_bool first_run;
Added a staging buffer to the audio stream so that we can accumulate small amounts of data if needed when resampling
2017-10-19 04:26:36 +02:00
Uint8 *staging_buffer;
int staging_buffer_size;
int staging_buffer_filled;
audio: Make sure SDL_AudioStream's work buffer is 16-byte aligned, for SIMD. Note the giantic FIXME, though!
2017-01-24 06:51:33 +01:00
Uint8 *work_buffer_base; /* maybe unaligned pointer from SDL_realloc(). */
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
int work_buffer_len;
int src_sample_frame_size;
SDL_AudioFormat src_format;
Uint8 src_channels;
int src_rate;
int dst_sample_frame_size;
SDL_AudioFormat dst_format;
Uint8 dst_channels;
int dst_rate;
double rate_incr;
Uint8 pre_resample_channels;
int packetlen;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
int resampler_padding_samples;
float *resampler_padding;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
void *resampler_state;
SDL_ResampleAudioStreamFunc resampler_func;
SDL_ResetAudioStreamResamplerFunc reset_resampler_func;
SDL_CleanupAudioStreamResamplerFunc cleanup_resampler_func;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
};
audio: libsamplerate can't resample in-place; make space for a copy if needed.
2017-01-25 02:30:48 +01:00
static Uint8 *
EnsureStreamBufferSize(SDL_AudioStream *stream, const int newlen)
{
Uint8 *ptr;
size_t offset;
if (stream->work_buffer_len >= newlen) {
ptr = stream->work_buffer_base;
} else {
ptr = (Uint8 *) SDL_realloc(stream->work_buffer_base, newlen + 32);
if (!ptr) {
SDL_OutOfMemory();
return NULL;
}
/* Make sure we're aligned to 16 bytes for SIMD code. */
stream->work_buffer_base = ptr;
stream->work_buffer_len = newlen;
}
offset = ((size_t) ptr) & 15;
return offset ? ptr + (16 - offset) : ptr;
}
Added configure and cmake support for libsamplerate
2017-01-07 05:43:53 +01:00
#ifdef HAVE_LIBSAMPLERATE_H
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
static int
audio: allow stereo Sint16 resampling fast path in SDL_AudioStream. This currently favors libsamplerate over the fast path (quality over speed), but I'm not sure that's the correct approach, as there may be surprising changes in performance metrics depending on what packages are available on a user's system. That being said, currently, the only thing with access to SDL_AudioStream is an SDL audio device's thread, and it might be mostly idle otherwise, so maybe this is generally good.
2017-01-24 06:08:24 +01:00
SDL_ResampleAudioStream_SRC(SDL_AudioStream *stream, const void *_inbuf, const int inbuflen, void *_outbuf, const int outbuflen)
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
{
audio: allow stereo Sint16 resampling fast path in SDL_AudioStream. This currently favors libsamplerate over the fast path (quality over speed), but I'm not sure that's the correct approach, as there may be surprising changes in performance metrics depending on what packages are available on a user's system. That being said, currently, the only thing with access to SDL_AudioStream is an SDL audio device's thread, and it might be mostly idle otherwise, so maybe this is generally good.
2017-01-24 06:08:24 +01:00
const float *inbuf = (const float *) _inbuf;
float *outbuf = (float *) _outbuf;
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
const int framelen = sizeof(float) * stream->pre_resample_channels;
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
SRC_STATE *state = (SRC_STATE *)stream->resampler_state;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
SRC_DATA data;
int result;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
SDL_assert(inbuf != ((const float *) outbuf)); /* SDL_AudioStreamPut() shouldn't allow in-place resamples. */
audio: libsamplerate can't resample in-place; make space for a copy if needed.
2017-01-25 02:30:48 +01:00
Added configure and cmake support for libsamplerate
2017-01-07 05:43:53 +01:00
data.data_in = (float *)inbuf; /* Older versions of libsamplerate had a non-const pointer, but didn't write to it */
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
data.input_frames = inbuflen / framelen;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
data.input_frames_used = 0;
data.data_out = outbuf;
audio: Replaced older resamplers in SDL_AudioCVT with the new ones.
2017-01-09 12:00:58 +01:00
data.output_frames = outbuflen / framelen;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
data.end_of_input = 0;
data.src_ratio = stream->rate_incr;
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
result = SRC_src_process(state, &data);
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
if (result != 0) {
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
SDL_SetError("src_process() failed: %s", SRC_src_strerror(result));
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
return 0;
}
/* If this fails, we need to store them off somewhere */
SDL_assert(data.input_frames_used == data.input_frames);
return data.output_frames_gen * (sizeof(float) * stream->pre_resample_channels);
}
static void
SDL_ResetAudioStreamResampler_SRC(SDL_AudioStream *stream)
{
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
SRC_src_reset((SRC_STATE *)stream->resampler_state);
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
}
static void
SDL_CleanupAudioStreamResampler_SRC(SDL_AudioStream *stream)
{
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
SRC_STATE *state = (SRC_STATE *)stream->resampler_state;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
if (state) {
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
SRC_src_delete(state);
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
}
stream->resampler_state = NULL;
stream->resampler_func = NULL;
stream->reset_resampler_func = NULL;
stream->cleanup_resampler_func = NULL;
}
static SDL_bool
SetupLibSampleRateResampling(SDL_AudioStream *stream)
{
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
int result = 0;
SRC_STATE *state = NULL;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
if (SRC_available) {
audio: Offer a hint for libsamplerate quality/speed tradeoff. This defaults to the internal SDL resampler, since that's the likely default without a system-wide install of libsamplerate, but those that need more can tweak this.
2017-01-24 21:52:22 +01:00
state = SRC_src_new(SRC_converter, stream->pre_resample_channels, &result);
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
if (!state) {
SDL_SetError("src_new() failed: %s", SRC_src_strerror(result));
}
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
}
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
if (!state) {
SDL_CleanupAudioStreamResampler_SRC(stream);
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
return SDL_FALSE;
}
stream->resampler_state = state;
stream->resampler_func = SDL_ResampleAudioStream_SRC;
stream->reset_resampler_func = SDL_ResetAudioStreamResampler_SRC;
stream->cleanup_resampler_func = SDL_CleanupAudioStreamResampler_SRC;
return SDL_TRUE;
}
Added configure and cmake support for libsamplerate
2017-01-07 05:43:53 +01:00
#endif /* HAVE_LIBSAMPLERATE_H */
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
audio: libsamplerate loading now happens once at init time.
2017-01-08 20:18:03 +01:00
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
static int
audio: allow stereo Sint16 resampling fast path in SDL_AudioStream. This currently favors libsamplerate over the fast path (quality over speed), but I'm not sure that's the correct approach, as there may be surprising changes in performance metrics depending on what packages are available on a user's system. That being said, currently, the only thing with access to SDL_AudioStream is an SDL audio device's thread, and it might be mostly idle otherwise, so maybe this is generally good.
2017-01-24 06:08:24 +01:00
SDL_ResampleAudioStream(SDL_AudioStream *stream, const void *_inbuf, const int inbuflen, void *_outbuf, const int outbuflen)
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
{
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
const Uint8 *inbufend = ((const Uint8 *) _inbuf) + inbuflen;
audio: allow stereo Sint16 resampling fast path in SDL_AudioStream. This currently favors libsamplerate over the fast path (quality over speed), but I'm not sure that's the correct approach, as there may be surprising changes in performance metrics depending on what packages are available on a user's system. That being said, currently, the only thing with access to SDL_AudioStream is an SDL audio device's thread, and it might be mostly idle otherwise, so maybe this is generally good.
2017-01-24 06:08:24 +01:00
const float *inbuf = (const float *) _inbuf;
float *outbuf = (float *) _outbuf;
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
const int chans = (int) stream->pre_resample_channels;
const int inrate = stream->src_rate;
const int outrate = stream->dst_rate;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
const int paddingsamples = stream->resampler_padding_samples;
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
const int paddingbytes = paddingsamples * sizeof (float);
float *lpadding = (float *) stream->resampler_state;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
const float *rpadding = (const float *) inbufend; /* we set this up so there are valid padding samples at the end of the input buffer. */
audio: Make sure audio stream resampling doesn't overflow buffers.
2017-10-11 08:03:05 +02:00
const int cpy = SDL_min(inbuflen, paddingbytes);
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
int retval;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
SDL_assert(inbuf != ((const float *) outbuf)); /* SDL_AudioStreamPut() shouldn't allow in-place resamples. */
Fixed memory leak when HAVE_ALLOCA isn't defined
2017-09-22 17:51:45 +02:00
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
retval = SDL_ResampleAudio(chans, inrate, outrate, lpadding, rpadding, inbuf, inbuflen, outbuf, outbuflen);
audio: allow stereo Sint16 resampling fast path in SDL_AudioStream. This currently favors libsamplerate over the fast path (quality over speed), but I'm not sure that's the correct approach, as there may be surprising changes in performance metrics depending on what packages are available on a user's system. That being said, currently, the only thing with access to SDL_AudioStream is an SDL audio device's thread, and it might be mostly idle otherwise, so maybe this is generally good.
2017-01-24 06:08:24 +01:00
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
/* update our left padding with end of current input, for next run. */
audio: Make sure audio stream resampling doesn't overflow buffers.
2017-10-11 08:03:05 +02:00
SDL_memcpy((lpadding + paddingsamples) - (cpy / sizeof (float)), inbufend - cpy, cpy);
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
return retval;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
}
static void
SDL_ResetAudioStreamResampler(SDL_AudioStream *stream)
{
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
/* set all the padding to silence. */
const int len = stream->resampler_padding_samples;
audio: Stream resampling now saves some samples from previous run for padding. 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.
2017-09-22 13:42:24 +02:00
SDL_memset(stream->resampler_state, '\0', len * sizeof (float));
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
}
static void
SDL_CleanupAudioStreamResampler(SDL_AudioStream *stream)
{
SDL_free(stream->resampler_state);
}
Fixed coding style on a function signature.
2017-01-08 20:17:09 +01:00
SDL_AudioStream *
SDL_NewAudioStream(const SDL_AudioFormat src_format,
const Uint8 src_channels,
const int src_rate,
const SDL_AudioFormat dst_format,
const Uint8 dst_channels,
const int dst_rate)
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
{
const int packetlen = 4096; /* !!! FIXME: good enough for now. */
Uint8 pre_resample_channels;
SDL_AudioStream *retval;
retval = (SDL_AudioStream *) SDL_calloc(1, sizeof (SDL_AudioStream));
if (!retval) {
return NULL;
}
/* If increasing channels, do it after resampling, since we'd just
do more work to resample duplicate channels. If we're decreasing, do
it first so we resample the interpolated data instead of interpolating
the resampled data (!!! FIXME: decide if that works in practice, though!). */
pre_resample_channels = SDL_min(src_channels, dst_channels);
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
retval->first_run = SDL_TRUE;
audio: SDL_AudioStream's *_sample_frame_size should be in bytes, not bits. Fixes failures where SDL_AudioStreamGet() incorrectly thinks it got a partial sample frame request.
2017-02-13 22:56:41 +01:00
retval->src_sample_frame_size = (SDL_AUDIO_BITSIZE(src_format) / 8) * src_channels;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
retval->src_format = src_format;
retval->src_channels = src_channels;
retval->src_rate = src_rate;
audio: SDL_AudioStream's *_sample_frame_size should be in bytes, not bits. Fixes failures where SDL_AudioStreamGet() incorrectly thinks it got a partial sample frame request.
2017-02-13 22:56:41 +01:00
retval->dst_sample_frame_size = (SDL_AUDIO_BITSIZE(dst_format) / 8) * dst_channels;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
retval->dst_format = dst_format;
retval->dst_channels = dst_channels;
retval->dst_rate = dst_rate;
retval->pre_resample_channels = pre_resample_channels;
retval->packetlen = packetlen;
retval->rate_incr = ((double) dst_rate) / ((double) src_rate);
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
retval->resampler_padding_samples = ResamplerPadding(retval->src_rate, retval->dst_rate) * pre_resample_channels;
retval->resampler_padding = (float *) SDL_calloc(retval->resampler_padding_samples, sizeof (float));
if (retval->resampler_padding == NULL) {
SDL_FreeAudioStream(retval);
SDL_OutOfMemory();
return NULL;
}
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
Added a staging buffer to the audio stream so that we can accumulate small amounts of data if needed when resampling
2017-10-19 04:26:36 +02:00
retval->staging_buffer_size = ((retval->resampler_padding_samples / retval->pre_resample_channels) * retval->src_sample_frame_size);
if (retval->staging_buffer_size > 0) {
retval->staging_buffer = (Uint8 *) SDL_malloc(retval->staging_buffer_size);
Check correct variable for malloc() results.
2017-10-19 05:49:46 +02:00
if (retval->staging_buffer == NULL) {
Added a staging buffer to the audio stream so that we can accumulate small amounts of data if needed when resampling
2017-10-19 04:26:36 +02:00
SDL_FreeAudioStream(retval);
SDL_OutOfMemory();
return NULL;
}
}
/* Not resampling? It's an easy conversion (and maybe not even that!) */
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
if (src_rate == dst_rate) {
retval->cvt_before_resampling.needed = SDL_FALSE;
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
if (SDL_BuildAudioCVT(&retval->cvt_after_resampling, src_format, src_channels, dst_rate, dst_format, dst_channels, dst_rate) < 0) {
SDL_FreeAudioStream(retval);
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
return NULL; /* SDL_BuildAudioCVT should have called SDL_SetError. */
}
} else {
/* Don't resample at first. Just get us to Float32 format. */
/* !!! FIXME: convert to int32 on devices without hardware float. */
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
if (SDL_BuildAudioCVT(&retval->cvt_before_resampling, src_format, src_channels, src_rate, AUDIO_F32SYS, pre_resample_channels, src_rate) < 0) {
SDL_FreeAudioStream(retval);
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
return NULL; /* SDL_BuildAudioCVT should have called SDL_SetError. */
}
Added configure and cmake support for libsamplerate
2017-01-07 05:43:53 +01:00
#ifdef HAVE_LIBSAMPLERATE_H
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
SetupLibSampleRateResampling(retval);
#endif
if (!retval->resampler_func) {
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
retval->resampler_state = SDL_calloc(retval->resampler_padding_samples, sizeof (float));
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
if (!retval->resampler_state) {
SDL_FreeAudioStream(retval);
SDL_OutOfMemory();
return NULL;
}
audio: Replaced the resampler. Again. 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.
2017-09-21 08:51:14 +02:00
if (SDL_PrepareResampleFilter() < 0) {
SDL_free(retval->resampler_state);
retval->resampler_state = NULL;
SDL_FreeAudioStream(retval);
return NULL;
}
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
retval->resampler_func = SDL_ResampleAudioStream;
retval->reset_resampler_func = SDL_ResetAudioStreamResampler;
retval->cleanup_resampler_func = SDL_CleanupAudioStreamResampler;
}
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
/* Convert us to the final format after resampling. */
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
if (SDL_BuildAudioCVT(&retval->cvt_after_resampling, AUDIO_F32SYS, pre_resample_channels, dst_rate, dst_format, dst_channels, dst_rate) < 0) {
SDL_FreeAudioStream(retval);
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
return NULL; /* SDL_BuildAudioCVT should have called SDL_SetError. */
}
}
retval->queue = SDL_NewDataQueue(packetlen, packetlen * 2);
if (!retval->queue) {
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
SDL_FreeAudioStream(retval);
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
return NULL; /* SDL_NewDataQueue should have called SDL_SetError. */
}
return retval;
}
Added a staging buffer to the audio stream so that we can accumulate small amounts of data if needed when resampling
2017-10-19 04:26:36 +02:00
static int
audio: Added SDL_AudioStreamFlush().
2017-10-20 00:05:42 +02:00
SDL_AudioStreamPutInternal(SDL_AudioStream *stream, const void *buf, int len, int *maxputbytes)
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
{
Added audio stream conversion functions: SDL_NewAudioStream SDL_AudioStreamPut SDL_AudioStreamGet SDL_AudioStreamAvailable SDL_AudioStreamClear SDL_FreeAudioStream
2017-10-19 00:54:05 +02:00
int buflen = len;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
int workbuflen;
Uint8 *workbuf;
Uint8 *resamplebuf = NULL;
int resamplebuflen = 0;
audio: Fixed check for minimum audio stream put size.
2017-10-11 07:37:11 +02:00
int neededpaddingbytes;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
int paddingbytes;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
audio: Make sure SDL_AudioStream's work buffer is 16-byte aligned, for SIMD. Note the giantic FIXME, though!
2017-01-24 06:51:33 +01:00
/* !!! FIXME: several converters can take advantage of SIMD, but only
!!! FIXME: if the data is aligned to 16 bytes. EnsureStreamBufferSize()
!!! FIXME: guarantees the buffer will align, but the
!!! FIXME: converters will iterate over the data backwards if
!!! FIXME: the output grows, and this means we won't align if buflen
!!! FIXME: isn't a multiple of 16. In these cases, we should chop off
!!! FIXME: a few samples at the end and convert them separately. */
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
/* no padding prepended on first run. */
audio: Fixed check for minimum audio stream put size.
2017-10-11 07:37:11 +02:00
neededpaddingbytes = stream->resampler_padding_samples * sizeof (float);
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
paddingbytes = stream->first_run ? 0 : neededpaddingbytes;
stream->first_run = SDL_FALSE;
/* Make sure the work buffer can hold all the data we need at once... */
workbuflen = buflen;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
if (stream->cvt_before_resampling.needed) {
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
workbuflen *= stream->cvt_before_resampling.len_mult;
}
if (stream->dst_rate != stream->src_rate) {
/* resamples can't happen in place, so make space for second buf. */
const int framesize = stream->pre_resample_channels * sizeof (float);
const int frames = workbuflen / framesize;
resamplebuflen = ((int) SDL_ceil(frames * stream->rate_incr)) * framesize;
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: will resample %d bytes to %d (ratio=%.6f)\n", workbuflen, resamplebuflen, stream->rate_incr);
#endif
workbuflen += resamplebuflen;
}
if (stream->cvt_after_resampling.needed) {
/* !!! FIXME: buffer might be big enough already? */
workbuflen *= stream->cvt_after_resampling.len_mult;
}
workbuflen += neededpaddingbytes;
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: Putting %d bytes of preconverted audio, need %d byte work buffer\n", buflen, workbuflen);
#endif
workbuf = EnsureStreamBufferSize(stream, workbuflen);
if (!workbuf) {
return -1; /* probably out of memory. */
}
resamplebuf = workbuf; /* default if not resampling. */
SDL_memcpy(workbuf + paddingbytes, buf, buflen);
if (stream->cvt_before_resampling.needed) {
stream->cvt_before_resampling.buf = workbuf + paddingbytes;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
stream->cvt_before_resampling.len = buflen;
if (SDL_ConvertAudio(&stream->cvt_before_resampling) == -1) {
return -1; /* uhoh! */
}
buflen = stream->cvt_before_resampling.len_cvt;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: After initial conversion we have %d bytes\n", buflen);
#endif
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
}
if (stream->dst_rate != stream->src_rate) {
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
/* save off some samples at the end; they are used for padding now so
the resampler is coherent and then used at the start of the next
put operation. Prepend last put operation's padding, too. */
/* prepend prior put's padding. :P */
if (paddingbytes) {
SDL_memcpy(workbuf, stream->resampler_padding, paddingbytes);
buflen += paddingbytes;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
}
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
/* save off the data at the end for the next run. */
SDL_memcpy(stream->resampler_padding, workbuf + (buflen - neededpaddingbytes), neededpaddingbytes);
resamplebuf = workbuf + buflen; /* skip to second piece of workbuf. */
audio: Make sure audio stream resampling doesn't overflow buffers.
2017-10-11 08:03:05 +02:00
SDL_assert(buflen >= neededpaddingbytes);
if (buflen > neededpaddingbytes) {
buflen = stream->resampler_func(stream, workbuf, buflen - neededpaddingbytes, resamplebuf, resamplebuflen);
} else {
buflen = 0;
}
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: After resampling we have %d bytes\n", buflen);
#endif
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
}
audio: Make sure audio stream resampling doesn't overflow buffers.
2017-10-11 08:03:05 +02:00
if (stream->cvt_after_resampling.needed && (buflen > 0)) {
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
stream->cvt_after_resampling.buf = resamplebuf;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
stream->cvt_after_resampling.len = buflen;
if (SDL_ConvertAudio(&stream->cvt_after_resampling) == -1) {
return -1; /* uhoh! */
}
buflen = stream->cvt_after_resampling.len_cvt;
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: After final conversion we have %d bytes\n", buflen);
#endif
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
}
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: Final output is %d bytes\n", buflen);
#endif
audio: Added SDL_AudioStreamFlush().
2017-10-20 00:05:42 +02:00
if (maxputbytes) {
const int maxbytes = *maxputbytes;
if (buflen > maxbytes)
buflen = maxbytes;
*maxputbytes -= buflen;
}
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
/* resamplebuf holds the final output, even if we didn't resample. */
audio: Make sure audio stream resampling doesn't overflow buffers.
2017-10-11 08:03:05 +02:00
return buflen ? SDL_WriteToDataQueue(stream->queue, resamplebuf, buflen) : 0;
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
}
Added a staging buffer to the audio stream so that we can accumulate small amounts of data if needed when resampling
2017-10-19 04:26:36 +02:00
int
SDL_AudioStreamPut(SDL_AudioStream *stream, const void *buf, int len)
{
/* !!! FIXME: several converters can take advantage of SIMD, but only
!!! FIXME: if the data is aligned to 16 bytes. EnsureStreamBufferSize()
!!! FIXME: guarantees the buffer will align, but the
!!! FIXME: converters will iterate over the data backwards if
!!! FIXME: the output grows, and this means we won't align if buflen
!!! FIXME: isn't a multiple of 16. In these cases, we should chop off
!!! FIXME: a few samples at the end and convert them separately. */
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: wants to put %d preconverted bytes\n", buflen);
#endif
if (!stream) {
return SDL_InvalidParamError("stream");
} else if (!buf) {
return SDL_InvalidParamError("buf");
} else if (len == 0) {
return 0; /* nothing to do. */
} else if ((len % stream->src_sample_frame_size) != 0) {
return SDL_SetError("Can't add partial sample frames");
}
if (!stream->cvt_before_resampling.needed &&
(stream->dst_rate == stream->src_rate) &&
!stream->cvt_after_resampling.needed) {
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: no conversion needed at all, queueing %d bytes.\n", len);
#endif
return SDL_WriteToDataQueue(stream->queue, buf, len);
}
while (len > 0) {
int amount;
/* If we don't have a staging buffer or we're given enough data that
we don't need to store it for later, skip the staging process.
*/
if (!stream->staging_buffer_filled && len >= stream->staging_buffer_size) {
audio: Added SDL_AudioStreamFlush().
2017-10-20 00:05:42 +02:00
return SDL_AudioStreamPutInternal(stream, buf, len, NULL);
Added a staging buffer to the audio stream so that we can accumulate small amounts of data if needed when resampling
2017-10-19 04:26:36 +02:00
}
/* If there's not enough data to fill the staging buffer, just save it */
if ((stream->staging_buffer_filled + len) < stream->staging_buffer_size) {
SDL_memcpy(stream->staging_buffer + stream->staging_buffer_filled, buf, len);
stream->staging_buffer_filled += len;
return 0;
}
/* Fill the staging buffer, process it, and continue */
amount = (stream->staging_buffer_size - stream->staging_buffer_filled);
SDL_assert(amount > 0);
SDL_memcpy(stream->staging_buffer + stream->staging_buffer_filled, buf, amount);
stream->staging_buffer_filled = 0;
audio: Added SDL_AudioStreamFlush().
2017-10-20 00:05:42 +02:00
if (SDL_AudioStreamPutInternal(stream, stream->staging_buffer, stream->staging_buffer_size, NULL) < 0) {
Added a staging buffer to the audio stream so that we can accumulate small amounts of data if needed when resampling
2017-10-19 04:26:36 +02:00
return -1;
}
buf = (void *)((Uint8 *)buf + amount);
len -= amount;
}
return 0;
}
audio: Added SDL_AudioStreamFlush().
2017-10-20 00:05:42 +02:00
int SDL_AudioStreamFlush(SDL_AudioStream *stream)
{
if (!stream) {
return SDL_InvalidParamError("stream");
}
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: flushing! staging_buffer_filled=%d bytes\n", stream->staging_buffer_filled);
#endif
/* shouldn't use a staging buffer if we're not resampling. */
SDL_assert((stream->dst_rate != stream->src_rate) || (stream->staging_buffer_filled == 0));
if (stream->staging_buffer_filled > 0) {
/* push the staging buffer + silence. We need to flush out not just
the staging buffer, but the piece that the stream was saving off
for right-side resampler padding. */
const SDL_bool first_run = stream->first_run;
const int filled = stream->staging_buffer_filled;
int actual_input_frames = filled / stream->src_sample_frame_size;
if (!first_run)
actual_input_frames += stream->resampler_padding_samples / stream->pre_resample_channels;
if (actual_input_frames > 0) { /* don't bother if nothing to flush. */
/* This is how many bytes we're expecting without silence appended. */
int flush_remaining = ((int) SDL_ceil(actual_input_frames * stream->rate_incr)) * stream->dst_sample_frame_size;
#if DEBUG_AUDIOSTREAM
printf("AUDIOSTREAM: flushing with padding to get max %d bytes!\n", flush_remaining);
#endif
SDL_memset(stream->staging_buffer + filled, '\0', stream->staging_buffer_size - filled);
if (SDL_AudioStreamPutInternal(stream, stream->staging_buffer, stream->staging_buffer_size, &flush_remaining) < 0) {
return -1;
}
/* we have flushed out (or initially filled) the pending right-side
resampler padding, but we need to push more silence to guarantee
the staging buffer is fully flushed out, too. */
SDL_memset(stream->staging_buffer, '\0', filled);
if (SDL_AudioStreamPutInternal(stream, stream->staging_buffer, stream->staging_buffer_size, &flush_remaining) < 0) {
return -1;
}
}
}
stream->staging_buffer_filled = 0;
stream->first_run = SDL_TRUE;
return 0;
}
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
/* get converted/resampled data from the stream */
int
Added audio stream conversion functions: SDL_NewAudioStream SDL_AudioStreamPut SDL_AudioStreamGet SDL_AudioStreamAvailable SDL_AudioStreamClear SDL_FreeAudioStream
2017-10-19 00:54:05 +02:00
SDL_AudioStreamGet(SDL_AudioStream *stream, void *buf, int len)
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
{
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
#if DEBUG_AUDIOSTREAM
Added audio stream conversion functions: SDL_NewAudioStream SDL_AudioStreamPut SDL_AudioStreamGet SDL_AudioStreamAvailable SDL_AudioStreamClear SDL_FreeAudioStream
2017-10-19 00:54:05 +02:00
printf("AUDIOSTREAM: want to get %d converted bytes\n", len);
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
#endif
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
if (!stream) {
return SDL_InvalidParamError("stream");
} else if (!buf) {
return SDL_InvalidParamError("buf");
Added audio stream conversion functions: SDL_NewAudioStream SDL_AudioStreamPut SDL_AudioStreamGet SDL_AudioStreamAvailable SDL_AudioStreamClear SDL_FreeAudioStream
2017-10-19 00:54:05 +02:00
} else if (len <= 0) {
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
return 0; /* nothing to do. */
} else if ((len % stream->dst_sample_frame_size) != 0) {
return SDL_SetError("Can't request partial sample frames");
}
audio: Fixed SDL_AudioStreamGet() function parameters. There was a draft of this where it did audio conversion into the final buffer, if there was enough room available past what you asked for, but that interface got removed, so the parameters didn't make sense (and we were using the wrong one in any case, too!).
2017-01-06 07:02:58 +01:00
return (int) SDL_ReadFromDataQueue(stream->queue, buf, len);
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
}
/* number of converted/resampled bytes available */
int
SDL_AudioStreamAvailable(SDL_AudioStream *stream)
{
return stream ? (int) SDL_CountDataQueue(stream->queue) : 0;
}
Added audio stream conversion functions: SDL_NewAudioStream SDL_AudioStreamPut SDL_AudioStreamGet SDL_AudioStreamAvailable SDL_AudioStreamClear SDL_FreeAudioStream
2017-10-19 00:54:05 +02:00
void
SDL_AudioStreamClear(SDL_AudioStream *stream)
{
if (!stream) {
SDL_InvalidParamError("stream");
} else {
SDL_ClearDataQueue(stream->queue, stream->packetlen * 2);
if (stream->reset_resampler_func) {
stream->reset_resampler_func(stream);
}
stream->first_run = SDL_TRUE;
audio: Added SDL_AudioStreamFlush().
2017-10-20 00:05:42 +02:00
stream->staging_buffer_filled = 0;
Added audio stream conversion functions: SDL_NewAudioStream SDL_AudioStreamPut SDL_AudioStreamGet SDL_AudioStreamAvailable SDL_AudioStreamClear SDL_FreeAudioStream
2017-10-19 00:54:05 +02:00
}
}
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
/* dispose of a stream */
void
SDL_FreeAudioStream(SDL_AudioStream *stream)
{
if (stream) {
Added support for using libsamplerate to do audio resampling
2017-01-06 11:16:26 +01:00
if (stream->cleanup_resampler_func) {
stream->cleanup_resampler_func(stream);
}
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
SDL_FreeDataQueue(stream->queue);
Added a staging buffer to the audio stream so that we can accumulate small amounts of data if needed when resampling
2017-10-19 04:26:36 +02:00
SDL_free(stream->staging_buffer);
audio: Make sure SDL_AudioStream's work buffer is 16-byte aligned, for SIMD. Note the giantic FIXME, though!
2017-01-24 06:51:33 +01:00
SDL_free(stream->work_buffer_base);
audio: reworked audio streams to have right-hand resampling padding available. Fixes Bugzilla #3851.
2017-10-10 22:12:56 +02:00
SDL_free(stream->resampler_padding);
audio: Added SDL_AudioStream. Non-power-of-two resampling now works!
2017-01-06 01:29:38 +01:00
SDL_free(stream);
}
}
Fixed crash if initialization of EGL failed but was tried again later. The internal function SDL_EGL_LoadLibrary() did not delete and remove a mostly uninitialized data structure if loading the library first failed. A later try to use EGL then skipped initialization and assumed it was previously successful because the data structure now already existed. This led to at least one crash in the internal function SDL_EGL_ChooseConfig() because a NULL pointer was dereferenced to make a call to eglBindAPI().
2015-06-21 17:33:46 +02:00
/* vi: set ts=4 sw=4 expandtab: */
Reworked audio converter code. 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).
2016-11-05 07:34:38 +01:00
Reference in New Issue Copy Permalink