3rdparty: Update cubeb to e495bee
Some checks are pending
🐧 Linux Builds / AppImage (push) Waiting to run
🐧 Linux Builds / Flatpak (push) Waiting to run
🍎 MacOS Builds / Defaults (push) Waiting to run
🖥️ Windows Builds / Lint VS Project Files (push) Waiting to run
🖥️ Windows Builds / SSE4 (push) Blocked by required conditions
🖥️ Windows Builds / AVX2 (push) Blocked by required conditions
🖥️ Windows Builds / CMake (push) Waiting to run

This commit is contained in:
JordanTheToaster 2025-10-01 16:19:27 +01:00 committed by Ty
parent e550cf9b63
commit bc11ff0571
16 changed files with 960 additions and 477 deletions

View File

@ -1,7 +1,117 @@
# libcubeb - Cross-platform Audio I/O Library
[![Build Status](https://github.com/mozilla/cubeb/actions/workflows/build.yml/badge.svg)](https://github.com/mozilla/cubeb/actions/workflows/build.yml)
See INSTALL.md for build instructions.
`libcubeb` is a cross-platform C library for high and low-latency audio input/output. It provides a simple, consistent API for audio playback and recording across multiple platforms and audio backends. It is written in C, C++ and Rust, with a C ABI and [Rust](https://github.com/mozilla/cubeb-rs) bindings. While originally written for use in the Firefox Web browser, a number of other software projects have adopted it.
See [Backend Support](https://github.com/mozilla/cubeb/wiki/Backend-Support) in the wiki for the support level of each backend.
## Features
Licensed under an ISC-style license. See LICENSE for details.
- **Cross-platform support**: Windows, macOS, Linux, Android, and other platforms
- **Versatile**: Optimized for low-latency real-time audio applications, or power efficient higher latency playback
- **A/V sync**: Latency compensated audio clock reporting for easy audio/video synchronization
- **Full-duplex support**: Simultaneous audio input and output, reclocked
- **Device enumeration**: Query available audio devices
- **Audio processing for speech**: Can use VoiceProcessing IO on recent macOS
## Supported Backends & status
| *Backend* | *Support Level* | *Platform version* | *Notes* |
|-------------------|-----------------|--------------------|--------------------------------------------------|
| PulseAudio (Rust) | Tier-1 | | Main Linux desktop backend |
| AudioUnit (Rust) | Tier-1 | | Main macOS backend |
| WASAPI | Tier-1 | Windows >= 7 | Main Windows backend |
| AAudio | Tier-1 | Android >= 8 | Main Android backend for most devices |
| OpenSL | Tier-1 | Android >= 2.3 | Android backend for older devices |
| OSS | Tier-2 | | |
| sndio | Tier-2 | | |
| Sun | Tier-2 | | |
| WinMM | Tier-3 | Windows XP | Was Tier-1, Firefox minimum Windows version 7. |
| AudioTrack | Tier-3 | Android < 2.3 | Was Tier-1, Firefox minimum Android version 4.1. |
| ALSA | Tier-3 | | |
| JACK | Tier-3 | | |
| KAI | Tier-3 | | |
| PulseAudio (C) | Tier-4 | | Was Tier-1, superseded by Rust |
| AudioUnit (C++) | Tier-4 | | Was Tier-1, superseded by Rust |
Tier-1: Actively maintained. Should have CI coverage. Critical for Firefox.
Tier-2: Actively maintained by contributors. CI coverage appreciated.
Tier-3: Maintainers/patches accepted. Status unclear.
Tier-4: Deprecated, obsolete. Scheduled to be removed.
Note that the support level is not a judgement of the relative merits
of a backend, only the current state of support, which is informed
by Firefox's needs, the responsiveness of a backend's
maintainer, and the level of contributions to that backend.
## Building
### Prerequisites
- CMake 3.15 or later
- Non-ancient MSVC, clang or gcc, for compiling both C and C++
- Platform-specific audio libraries (automatically detected)
- Optional but recommended: Rust compiler to compile and link more recent backends for macOS and PulseAudio
### Quick build
```bash
git clone https://github.com/mozilla/cubeb.git
cd cubeb
cmake -B build
cmake --build build
```
### Better build with Rust backends
```bash
git clone --recursive https://github.com/mozilla/cubeb.git
cd cubeb
cmake -B build -DBUILD_RUST_LIBS=ON
cmake --build build
```
### Platform-Specific Notes
**Windows**: Supports Visual Studio 2015+ and MinGW-w64. Use `-G "Visual Studio 16 2019"` or `-G "MinGW Makefiles"`.
**macOS**: Requires Xcode command line tools. Audio frameworks are automatically linked.
**Linux**: Development packages for desired backends:
```bash
# Ubuntu/Debian
sudo apt-get install libpulse-dev libasound2-dev libjack-dev
# Fedora/RHEL
sudo dnf install pulseaudio-libs-devel alsa-lib-devel jack-audio-connection-kit-devel
```
**Android**: Use with Android NDK. AAudio requires API level 26+.
## Testing
Run the test suite:
```bash
cd build
ctest
```
Use the interactive test tool:
```bash
./cubeb-test
```
## License
Licensed under an ISC-style license. See [LICENSE](LICENSE) for details.
## Contributing
Contributions are welcome! Please see the [contribution guidelines](CONTRIBUTING.md) and check the [issue tracker](https://github.com/mozilla/cubeb/issues).
## Links
- [GitHub Repository](https://github.com/mozilla/cubeb)
- [API Documentation](https://mozilla.github.io/cubeb/)

View File

@ -49,6 +49,7 @@ extern "C" {
output_params.channels = 2;
output_params.layout = CUBEB_LAYOUT_UNDEFINED;
output_params.prefs = CUBEB_STREAM_PREF_NONE;
output_params.input_params = CUBEB_INPUT_PROCESSING_PARAM_NONE;
rv = cubeb_get_min_latency(app_ctx, &output_params, &latency_frames);
if (rv != CUBEB_OK) {
@ -62,6 +63,7 @@ extern "C" {
input_params.channels = 1;
input_params.layout = CUBEB_LAYOUT_UNDEFINED;
input_params.prefs = CUBEB_STREAM_PREF_NONE;
input_params.input_params = CUBEB_INPUT_PROCESSING_PARAM_NONE;
cubeb_stream * stm;
rv = cubeb_stream_init(app_ctx, &stm, "Example Stream 1",
@ -193,39 +195,39 @@ typedef uint32_t cubeb_channel_layout;
// Some common layout definitions.
enum {
CUBEB_LAYOUT_UNDEFINED = 0, // Indicate the speaker's layout is undefined.
CUBEB_LAYOUT_MONO = (uint32_t)CHANNEL_FRONT_CENTER,
CUBEB_LAYOUT_MONO_LFE = (uint32_t)CUBEB_LAYOUT_MONO | (uint32_t)CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_STEREO = (uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT,
CUBEB_LAYOUT_STEREO_LFE = (uint32_t)CUBEB_LAYOUT_STEREO | (uint32_t)CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_MONO = CHANNEL_FRONT_CENTER,
CUBEB_LAYOUT_MONO_LFE = CUBEB_LAYOUT_MONO | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_STEREO = CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT,
CUBEB_LAYOUT_STEREO_LFE = CUBEB_LAYOUT_STEREO | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F =
(uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT | (uint32_t)CHANNEL_FRONT_CENTER,
CUBEB_LAYOUT_3F_LFE = (uint32_t)CUBEB_LAYOUT_3F | (uint32_t)CHANNEL_LOW_FREQUENCY,
CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT | CHANNEL_FRONT_CENTER,
CUBEB_LAYOUT_3F_LFE = CUBEB_LAYOUT_3F | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_2F1 =
(uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT | (uint32_t)CHANNEL_BACK_CENTER,
CUBEB_LAYOUT_2F1_LFE = (uint32_t)CUBEB_LAYOUT_2F1 | (uint32_t)CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F1 = (uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT |
(uint32_t)CHANNEL_FRONT_CENTER | (uint32_t)CHANNEL_BACK_CENTER,
CUBEB_LAYOUT_3F1_LFE = (uint32_t)CUBEB_LAYOUT_3F1 | (uint32_t)CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_2F2 = (uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT |
(uint32_t)CHANNEL_SIDE_LEFT | (uint32_t)CHANNEL_SIDE_RIGHT,
CUBEB_LAYOUT_2F2_LFE = (uint32_t)CUBEB_LAYOUT_2F2 | (uint32_t)CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_QUAD = (uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT |
(uint32_t)CHANNEL_BACK_LEFT | (uint32_t)CHANNEL_BACK_RIGHT,
CUBEB_LAYOUT_QUAD_LFE = (uint32_t)CUBEB_LAYOUT_QUAD | (uint32_t)CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F2 = (uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT |
(uint32_t)CHANNEL_FRONT_CENTER | (uint32_t)CHANNEL_SIDE_LEFT |
(uint32_t)CHANNEL_SIDE_RIGHT,
CUBEB_LAYOUT_3F2_LFE = (uint32_t)CUBEB_LAYOUT_3F2 | (uint32_t)CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F2_BACK = (uint32_t)CUBEB_LAYOUT_QUAD | (uint32_t)CHANNEL_FRONT_CENTER,
CUBEB_LAYOUT_3F2_LFE_BACK = (uint32_t)CUBEB_LAYOUT_3F2_BACK | (uint32_t)CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F3R_LFE = (uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT |
(uint32_t)CHANNEL_FRONT_CENTER | (uint32_t)CHANNEL_LOW_FREQUENCY |
(uint32_t)CHANNEL_BACK_CENTER | (uint32_t)CHANNEL_SIDE_LEFT |
(uint32_t)CHANNEL_SIDE_RIGHT,
CUBEB_LAYOUT_3F4_LFE = (uint32_t)CHANNEL_FRONT_LEFT | (uint32_t)CHANNEL_FRONT_RIGHT |
(uint32_t)CHANNEL_FRONT_CENTER | (uint32_t)CHANNEL_LOW_FREQUENCY |
(uint32_t)CHANNEL_BACK_LEFT | (uint32_t)CHANNEL_BACK_RIGHT |
(uint32_t)CHANNEL_SIDE_LEFT | (uint32_t)CHANNEL_SIDE_RIGHT,
CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT | CHANNEL_BACK_CENTER,
CUBEB_LAYOUT_2F1_LFE = CUBEB_LAYOUT_2F1 | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F1 = CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT |
CHANNEL_FRONT_CENTER | CHANNEL_BACK_CENTER,
CUBEB_LAYOUT_3F1_LFE = CUBEB_LAYOUT_3F1 | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_2F2 = CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT |
CHANNEL_SIDE_LEFT | CHANNEL_SIDE_RIGHT,
CUBEB_LAYOUT_2F2_LFE = CUBEB_LAYOUT_2F2 | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_QUAD = CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT |
CHANNEL_BACK_LEFT | CHANNEL_BACK_RIGHT,
CUBEB_LAYOUT_QUAD_LFE = CUBEB_LAYOUT_QUAD | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F2 = CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT |
CHANNEL_FRONT_CENTER | CHANNEL_SIDE_LEFT |
CHANNEL_SIDE_RIGHT,
CUBEB_LAYOUT_3F2_LFE = CUBEB_LAYOUT_3F2 | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F2_BACK = CUBEB_LAYOUT_QUAD | CHANNEL_FRONT_CENTER,
CUBEB_LAYOUT_3F2_LFE_BACK = CUBEB_LAYOUT_3F2_BACK | CHANNEL_LOW_FREQUENCY,
CUBEB_LAYOUT_3F3R_LFE = CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT |
CHANNEL_FRONT_CENTER | CHANNEL_LOW_FREQUENCY |
CHANNEL_BACK_CENTER | CHANNEL_SIDE_LEFT |
CHANNEL_SIDE_RIGHT,
CUBEB_LAYOUT_3F4_LFE = CHANNEL_FRONT_LEFT | CHANNEL_FRONT_RIGHT |
CHANNEL_FRONT_CENTER | CHANNEL_LOW_FREQUENCY |
CHANNEL_BACK_LEFT | CHANNEL_BACK_RIGHT |
CHANNEL_SIDE_LEFT | CHANNEL_SIDE_RIGHT,
};
/** Miscellaneous stream preferences. */
@ -279,7 +281,10 @@ typedef struct {
cubeb_channel_layout
layout; /**< Requested channel layout. This must be consistent with the
provided channels. CUBEB_LAYOUT_UNDEFINED if unknown */
cubeb_stream_prefs prefs; /**< Requested preferences. */
cubeb_stream_prefs prefs; /**< Requested preferences. */
cubeb_input_processing_params input_params; /**< Requested input processing
params. Ignored for output streams. At present, only supported on the
WASAPI backend; others should use cubeb_set_input_processing_params. */
} cubeb_stream_params;
/** Audio device description */
@ -414,6 +419,13 @@ typedef struct {
size_t count; /**< Device count in collection. */
} cubeb_device_collection;
/** Array of compiled backends returned by `cubeb_get_backend_names`. */
typedef struct {
const char * const *
names; /**< Array of strings representing backend names. */
size_t count; /**< Length of the array. */
} cubeb_backend_names;
/** User supplied data callback.
- Calling other cubeb functions from this callback is unsafe.
- The code in the callback should be non-blocking.
@ -454,6 +466,8 @@ typedef void (*cubeb_device_changed_callback)(void * user_ptr);
/**
* User supplied callback called when the underlying device collection changed.
* This callback will be called when devices are added or removed from the
* system, or when the default device changes for the specified device type.
* @param context A pointer to the cubeb context.
* @param user_ptr The pointer passed to
* cubeb_register_device_collection_changed. */
@ -485,17 +499,18 @@ CUBEB_EXPORT int
cubeb_init(cubeb ** context, char const * context_name,
char const * backend_name);
/** Returns a list of backend names which can be supplid to cubeb_init().
Array is null-terminated. */
CUBEB_EXPORT const char**
cubeb_get_backend_names();
/** Get a read-only string identifying this context's current backend.
@param context A pointer to the cubeb context.
@retval Read-only string identifying current backend. */
CUBEB_EXPORT char const *
cubeb_get_backend_id(cubeb * context);
/** Get a read-only array of strings identifying available backends.
These can be passed as `backend_name` parameter to `cubeb_init`.
@retval Struct containing the array with backend names. */
CUBEB_EXPORT cubeb_backend_names
cubeb_get_backend_names();
/** Get the maximum possible number of channels.
@param context A pointer to the cubeb context.
@param max_channels The maximum number of channels.
@ -674,7 +689,7 @@ cubeb_stream_get_current_device(cubeb_stream * stm,
application is accessing audio input. When all inputs are muted they can
prove to the user that the application is not actively capturing any input.
@param stream the stream for which to set input mute state
@param muted whether the input should mute or not
@param mute whether the input should mute or not
@retval CUBEB_OK
@retval CUBEB_ERROR_INVALID_PARAMETER if this stream does not have an input
device
@ -745,14 +760,16 @@ cubeb_device_collection_destroy(cubeb * context,
cubeb_device_collection * collection);
/** Registers a callback which is called when the system detects
a new device or a device is removed.
a new device or a device is removed, or when the default device
changes for the specified device type.
@param context
@param devtype device type to include. Different callbacks and user pointers
can be registered for each devtype. The hybrid devtype
`CUBEB_DEVICE_TYPE_INPUT | CUBEB_DEVICE_TYPE_OUTPUT` is also valid
and will register the provided callback and user pointer in both
sides.
@param callback a function called whenever the system device list changes.
@param callback a function called whenever the system device list changes,
including when default devices change.
Passing NULL allow to unregister a function. You have to unregister
first before you register a new callback.
@param user_ptr pointer to user specified data which will be present in

View File

@ -31,6 +31,10 @@ struct cubeb_stream {
int
pulse_init(cubeb ** context, char const * context_name);
#endif
#if defined(USE_PULSE_RUST)
int
pulse_rust_init(cubeb ** contet, char const * context_name);
#endif
#if defined(USE_JACK)
int
jack_init(cubeb ** context, char const * context_name);
@ -43,6 +47,10 @@ alsa_init(cubeb ** context, char const * context_name);
int
audiounit_init(cubeb ** context, char const * context_name);
#endif
#if defined(USE_AUDIOUNIT_RUST)
int
audiounit_rust_init(cubeb ** contet, char const * context_name);
#endif
#if defined(USE_WINMM)
int
winmm_init(cubeb ** context, char const * context_name);
@ -55,10 +63,30 @@ wasapi_init(cubeb ** context, char const * context_name);
int
sndio_init(cubeb ** context, char const * context_name);
#endif
#if defined(USE_SUN)
int
sun_init(cubeb ** context, char const * context_name);
#endif
#if defined(USE_OPENSL)
int
opensl_init(cubeb ** context, char const * context_name);
#endif
#if defined(USE_OSS)
int
oss_init(cubeb ** context, char const * context_name);
#endif
#if defined(USE_AAUDIO)
int
aaudio_init(cubeb ** context, char const * context_name);
#endif
#if defined(USE_AUDIOTRACK)
int
audiotrack_init(cubeb ** context, char const * context_name);
#endif
#if defined(USE_KAI)
int
kai_init(cubeb ** context, char const * context_name);
#endif
static int
validate_stream_params(cubeb_stream_params * input_stream_params,
@ -123,6 +151,10 @@ cubeb_init(cubeb ** context, char const * context_name,
if (!strcmp(backend_name, "pulse")) {
#if defined(USE_PULSE)
init_oneshot = pulse_init;
#endif
} else if (!strcmp(backend_name, "pulse-rust")) {
#if defined(USE_PULSE_RUST)
init_oneshot = pulse_rust_init;
#endif
} else if (!strcmp(backend_name, "jack")) {
#if defined(USE_JACK)
@ -135,6 +167,10 @@ cubeb_init(cubeb ** context, char const * context_name,
} else if (!strcmp(backend_name, "audiounit")) {
#if defined(USE_AUDIOUNIT)
init_oneshot = audiounit_init;
#endif
} else if (!strcmp(backend_name, "audiounit-rust")) {
#if defined(USE_AUDIOUNIT_RUST)
init_oneshot = audiounit_rust_init;
#endif
} else if (!strcmp(backend_name, "wasapi")) {
#if defined(USE_WASAPI)
@ -147,10 +183,30 @@ cubeb_init(cubeb ** context, char const * context_name,
} else if (!strcmp(backend_name, "sndio")) {
#if defined(USE_SNDIO)
init_oneshot = sndio_init;
#endif
} else if (!strcmp(backend_name, "sun")) {
#if defined(USE_SUN)
init_oneshot = sun_init;
#endif
} else if (!strcmp(backend_name, "opensl")) {
#if defined(USE_OPENSL)
init_oneshot = opensl_init;
#endif
} else if (!strcmp(backend_name, "oss")) {
#if defined(USE_OSS)
init_oneshot = oss_init;
#endif
} else if (!strcmp(backend_name, "aaudio")) {
#if defined(USE_AAUDIO)
init_oneshot = aaudio_init;
#endif
} else if (!strcmp(backend_name, "audiotrack")) {
#if defined(USE_AUDIOTRACK)
init_oneshot = audiotrack_init;
#endif
} else if (!strcmp(backend_name, "kai")) {
#if defined(USE_KAI)
init_oneshot = kai_init;
#endif
} else {
/* Already set */
@ -163,6 +219,9 @@ cubeb_init(cubeb ** context, char const * context_name,
* to override all other choices
*/
init_oneshot,
#if defined(USE_PULSE_RUST)
pulse_rust_init,
#endif
#if defined(USE_PULSE)
pulse_init,
#endif
@ -178,6 +237,9 @@ cubeb_init(cubeb ** context, char const * context_name,
#if defined(USE_OSS)
oss_init,
#endif
#if defined(USE_AUDIOUNIT_RUST)
audiounit_rust_init,
#endif
#if defined(USE_AUDIOUNIT)
audiounit_init,
#endif
@ -189,6 +251,18 @@ cubeb_init(cubeb ** context, char const * context_name,
#endif
#if defined(USE_SUN)
sun_init,
#endif
#if defined(USE_AAUDIO)
aaudio_init,
#endif
#if defined(USE_OPENSL)
opensl_init,
#endif
#if defined(USE_AUDIOTRACK)
audiotrack_init,
#endif
#if defined(USE_KAI)
kai_init,
#endif
};
int i;
@ -214,13 +288,26 @@ cubeb_init(cubeb ** context, char const * context_name,
return CUBEB_ERROR;
}
const char**
char const *
cubeb_get_backend_id(cubeb * context)
{
if (!context) {
return NULL;
}
return context->ops->get_backend_id(context);
}
cubeb_backend_names
cubeb_get_backend_names()
{
static const char* backend_names[] = {
static const char * const backend_names[] = {
#if defined(USE_PULSE)
"pulse",
#endif
#if defined(USE_PULSE_RUST)
"pulse-rust",
#endif
#if defined(USE_JACK)
"jack",
#endif
@ -230,6 +317,9 @@ cubeb_get_backend_names()
#if defined(USE_AUDIOUNIT)
"audiounit",
#endif
#if defined(USE_AUDIOUNIT_RUST)
"audiounit-rust",
#endif
#if defined(USE_WASAPI)
"wasapi",
#endif
@ -239,23 +329,30 @@ cubeb_get_backend_names()
#if defined(USE_SNDIO)
"sndio",
#endif
#if defined(USE_SUN)
"sun",
#endif
#if defined(USE_OPENSL)
"opensl",
#endif
#if defined(USE_OSS)
"oss",
#endif
NULL,
#if defined(USE_AAUDIO)
"aaudio",
#endif
#if defined(USE_AUDIOTRACK)
"audiotrack",
#endif
#if defined(USE_KAI)
"kai",
#endif
};
return backend_names;
}
char const *
cubeb_get_backend_id(cubeb * context)
{
if (!context) {
return NULL;
}
return context->ops->get_backend_id(context);
return (cubeb_backend_names){
.names = backend_names,
.count = NELEMS(backend_names),
};
}
int

View File

@ -213,12 +213,19 @@ struct cubeb_stream {
cubeb_device_changed_callback device_changed_callback = nullptr;
owned_critical_section device_changed_callback_lock;
/* Stream creation parameters */
cubeb_stream_params input_stream_params = {CUBEB_SAMPLE_FLOAT32NE, 0, 0,
cubeb_stream_params input_stream_params = {CUBEB_SAMPLE_FLOAT32NE,
0,
0,
CUBEB_LAYOUT_UNDEFINED,
CUBEB_STREAM_PREF_NONE};
cubeb_stream_params output_stream_params = {CUBEB_SAMPLE_FLOAT32NE, 0, 0,
CUBEB_LAYOUT_UNDEFINED,
CUBEB_STREAM_PREF_NONE};
CUBEB_STREAM_PREF_NONE,
CUBEB_INPUT_PROCESSING_PARAM_NONE};
cubeb_stream_params output_stream_params = {
CUBEB_SAMPLE_FLOAT32NE,
0,
0,
CUBEB_LAYOUT_UNDEFINED,
CUBEB_STREAM_PREF_NONE,
CUBEB_INPUT_PROCESSING_PARAM_NONE};
device_info input_device;
device_info output_device;
/* Format descriptions */

View File

@ -16,8 +16,8 @@
#include <time.h>
#endif
static std::atomic<cubeb_log_level> g_cubeb_log_level;
static std::atomic<cubeb_log_callback> g_cubeb_log_callback;
std::atomic<cubeb_log_level> g_cubeb_log_level;
std::atomic<cubeb_log_callback> g_cubeb_log_callback;
/** The maximum size of a log message, after having been formatted. */
const size_t CUBEB_LOG_MESSAGE_MAX_SIZE = 256;
@ -32,6 +32,133 @@ cubeb_noop_log_callback(char const * /* fmt */, ...)
{
}
/**
* This wraps an inline buffer, that represents a log message, that must be
* null-terminated.
* This class should not use system calls or other potentially blocking code.
*/
class cubeb_log_message {
public:
cubeb_log_message() { *storage = '\0'; }
cubeb_log_message(char const str[CUBEB_LOG_MESSAGE_MAX_SIZE])
{
size_t length = strlen(str);
/* paranoia against malformed message */
assert(length < CUBEB_LOG_MESSAGE_MAX_SIZE);
if (length > CUBEB_LOG_MESSAGE_MAX_SIZE - 1) {
return;
}
PodCopy(storage, str, length);
storage[length] = '\0';
}
char const * get() { return storage; }
private:
char storage[CUBEB_LOG_MESSAGE_MAX_SIZE]{};
};
/** Lock-free asynchronous logger, made so that logging from a
* real-time audio callback does not block the audio thread. */
class cubeb_async_logger {
public:
/* This is thread-safe since C++11 */
static cubeb_async_logger & get()
{
static cubeb_async_logger instance;
return instance;
}
void push(char const str[CUBEB_LOG_MESSAGE_MAX_SIZE])
{
cubeb_log_message msg(str);
auto * owned_queue = msg_queue.load();
// Check if the queue is being deallocated. If not, grab ownership. If yes,
// return, the message won't be logged.
if (!owned_queue ||
!msg_queue.compare_exchange_strong(owned_queue, nullptr)) {
return;
}
owned_queue->enqueue(msg);
// Return ownership.
msg_queue.store(owned_queue);
}
void run()
{
assert(logging_thread.get_id() == std::thread::id());
logging_thread = std::thread([this]() {
CUBEB_REGISTER_THREAD("cubeb_log");
while (!shutdown_thread) {
cubeb_log_message msg;
while (msg_queue_consumer.load()->dequeue(&msg, 1)) {
cubeb_log_internal_no_format(msg.get());
}
std::this_thread::sleep_for(
std::chrono::milliseconds(CUBEB_LOG_BATCH_PRINT_INTERVAL_MS));
}
CUBEB_UNREGISTER_THREAD();
});
}
// Tell the underlying queue the producer thread has changed, so it does not
// assert in debug. This should be called with the thread stopped.
void reset_producer_thread()
{
if (msg_queue) {
msg_queue.load()->reset_thread_ids();
}
}
void start()
{
auto * queue =
new lock_free_queue<cubeb_log_message>(CUBEB_LOG_MESSAGE_QUEUE_DEPTH);
msg_queue.store(queue);
msg_queue_consumer.store(queue);
shutdown_thread = false;
run();
}
void stop()
{
assert(((g_cubeb_log_callback == cubeb_noop_log_callback) ||
!g_cubeb_log_callback) &&
"Only call stop after logging has been disabled.");
shutdown_thread = true;
if (logging_thread.get_id() != std::thread::id()) {
logging_thread.join();
logging_thread = std::thread();
auto * owned_queue = msg_queue.load();
// Check if the queue is being used. If not, grab ownership. If yes,
// try again shortly. At this point, the logging thread has been joined,
// so nothing is going to dequeue.
// If there is a valid pointer here, then the real-time audio thread that
// logs won't attempt to write into the queue, and instead drop the
// message.
while (!msg_queue.compare_exchange_weak(owned_queue, nullptr)) {
}
delete owned_queue;
msg_queue_consumer.store(nullptr);
}
}
private:
cubeb_async_logger() {}
~cubeb_async_logger()
{
assert(logging_thread.get_id() == std::thread::id() &&
(g_cubeb_log_callback == cubeb_noop_log_callback ||
!g_cubeb_log_callback));
if (msg_queue.load()) {
delete msg_queue.load();
}
}
/** This is quite a big data structure, but is only instantiated if the
* asynchronous logger is used. The two pointers point to the same object, but
* the first one can be temporarily null when a message is being enqueued. */
std::atomic<lock_free_queue<cubeb_log_message> *> msg_queue = {nullptr};
std::atomic<lock_free_queue<cubeb_log_message> *> msg_queue_consumer = {
nullptr};
std::atomic<bool> shutdown_thread = {false};
std::thread logging_thread;
};
void
cubeb_log_internal(char const * file, uint32_t line, char const * fmt, ...)
{
@ -49,6 +176,29 @@ cubeb_log_internal_no_format(const char * msg)
g_cubeb_log_callback.load()(msg);
}
void
cubeb_async_log(char const * fmt, ...)
{
// This is going to copy a 256 bytes array around, which is fine.
// We don't want to allocate memory here, because this is made to
// be called from a real-time callback.
va_list args;
va_start(args, fmt);
char msg[CUBEB_LOG_MESSAGE_MAX_SIZE];
vsnprintf(msg, CUBEB_LOG_MESSAGE_MAX_SIZE, fmt, args);
cubeb_async_logger::get().push(msg);
va_end(args);
}
void
cubeb_async_log_reset_threads(void)
{
if (!g_cubeb_log_callback) {
return;
}
cubeb_async_logger::get().reset_producer_thread();
}
void
cubeb_log_set(cubeb_log_level log_level, cubeb_log_callback log_callback)
{
@ -57,8 +207,15 @@ cubeb_log_set(cubeb_log_level log_level, cubeb_log_callback log_callback)
// nullptr, to prevent a TOCTOU race between checking the pointer
if (log_callback && log_level != CUBEB_LOG_DISABLED) {
g_cubeb_log_callback = log_callback;
if (log_level == CUBEB_LOG_VERBOSE) {
cubeb_async_logger::get().start();
}
} else if (!log_callback || CUBEB_LOG_DISABLED) {
g_cubeb_log_callback = cubeb_noop_log_callback;
// This returns once the thread has joined.
// This is safe even if CUBEB_LOG_VERBOSE was not set; the thread will
// simply not be joinable.
cubeb_async_logger::get().stop();
} else {
assert(false && "Incorrect parameters passed to cubeb_log_set");
}

View File

@ -39,7 +39,12 @@ cubeb_log_get_callback(void);
void
cubeb_log_internal_no_format(const char * msg);
void
cubeb_log_internal(const char * filename, uint32_t line, const char * fmt, ...);
cubeb_log_internal(const char * filename, uint32_t line, const char * fmt, ...)
PRINTF_FORMAT(3, 4);
void
cubeb_async_log(const char * fmt, ...) PRINTF_FORMAT(1, 2);
void
cubeb_async_log_reset_threads(void);
#ifdef __cplusplus
}
@ -55,9 +60,16 @@ cubeb_log_internal(const char * filename, uint32_t line, const char * fmt, ...);
} \
} while (0)
#define ALOG_INTERNAL(level, fmt, ...) \
do { \
if (cubeb_log_get_level() >= level && cubeb_log_get_callback()) { \
cubeb_async_log(fmt, ##__VA_ARGS__); \
} \
} while (0)
/* Asynchronous logging macros to log in real-time callbacks. */
/* Should not be used on android due to the use of global/static variables. */
#define ALOGV(msg, ...) LOG_INTERNAL(CUBEB_LOG_VERBOSE, msg, ##__VA_ARGS__)
#define ALOG(msg, ...) LOG_INTERNAL(CUBEB_LOG_NORMAL, msg, ##__VA_ARGS__)
#define ALOGV(msg, ...) ALOG_INTERNAL(CUBEB_LOG_VERBOSE, msg, ##__VA_ARGS__)
#define ALOG(msg, ...) ALOG_INTERNAL(CUBEB_LOG_NORMAL, msg, ##__VA_ARGS__)
#endif // CUBEB_LOG

View File

@ -371,3 +371,9 @@ cubeb_resampler_latency(cubeb_resampler * resampler)
{
return resampler->latency();
}
cubeb_resampler_stats
cubeb_resampler_stats_get(cubeb_resampler * resampler)
{
return resampler->stats();
}

View File

@ -84,6 +84,20 @@ cubeb_resampler_destroy(cubeb_resampler * resampler);
long
cubeb_resampler_latency(cubeb_resampler * resampler);
/**
* Test-only introspection API to ensure that there is no buffering
* buildup when resampling.
*/
typedef struct {
size_t input_input_buffer_size;
size_t input_output_buffer_size;
size_t output_input_buffer_size;
size_t output_output_buffer_size;
} cubeb_resampler_stats;
cubeb_resampler_stats
cubeb_resampler_stats_get(cubeb_resampler * resampler);
#if defined(__cplusplus)
}
#endif

View File

@ -56,6 +56,7 @@ struct cubeb_resampler {
virtual long fill(void * input_buffer, long * input_frames_count,
void * output_buffer, long frames_needed) = 0;
virtual long latency() = 0;
virtual cubeb_resampler_stats stats() = 0;
virtual ~cubeb_resampler() {}
};
@ -86,6 +87,16 @@ public:
virtual long latency() { return 0; }
virtual cubeb_resampler_stats stats()
{
cubeb_resampler_stats stats;
stats.input_input_buffer_size = internal_input_buffer.length();
stats.input_output_buffer_size = 0;
stats.output_input_buffer_size = 0;
stats.output_output_buffer_size = 0;
return stats;
}
void drop_audio_if_needed()
{
uint32_t to_keep = min_buffered_audio_frame(sample_rate);
@ -122,6 +133,20 @@ public:
virtual long fill(void * input_buffer, long * input_frames_count,
void * output_buffer, long output_frames_needed);
virtual cubeb_resampler_stats stats()
{
cubeb_resampler_stats stats = {};
if (input_processor) {
stats.input_input_buffer_size = input_processor->input_buffer_size();
stats.input_output_buffer_size = input_processor->output_buffer_size();
}
if (output_processor) {
stats.output_input_buffer_size = output_processor->input_buffer_size();
stats.output_output_buffer_size = output_processor->output_buffer_size();
}
return stats;
}
virtual long latency()
{
if (input_processor && output_processor) {
@ -280,29 +305,28 @@ public:
}
/** Returns the number of frames to pass in the input of the resampler to have
* exactly `output_frame_count` resampled frames. This can return a number
* slightly bigger than what is strictly necessary, but it guaranteed that the
* number of output frames will be exactly equal. */
* at least `output_frame_count` resampled frames. */
uint32_t input_needed_for_output(int32_t output_frame_count) const
{
assert(output_frame_count >= 0); // Check overflow
int32_t unresampled_frames_left =
samples_to_frames(resampling_in_buffer.length());
int32_t resampled_frames_left =
samples_to_frames(resampling_out_buffer.length());
float input_frames_needed =
(output_frame_count - unresampled_frames_left) * resampling_ratio -
resampled_frames_left;
if (input_frames_needed < 0) {
return 0;
}
return (uint32_t)ceilf(input_frames_needed);
float input_frames_needed_frac =
static_cast<float>(output_frame_count) * resampling_ratio;
// speex_resample()` can be irregular in its consumption of input samples.
// Provide one more frame than the number that would be required with
// regular consumption, to make the speex resampler behave more regularly,
// and so predictably.
auto input_frame_needed =
1 + static_cast<int32_t>(ceilf(input_frames_needed_frac));
input_frame_needed -= std::min(unresampled_frames_left, input_frame_needed);
return input_frame_needed;
}
/** Returns a pointer to the input buffer, that contains empty space for at
* least `frame_count` elements. This is useful so that consumer can directly
* write into the input buffer of the resampler. The pointer returned is
* adjusted so that leftover data are not overwritten.
* least `frame_count` elements. This is useful so that consumer can
* directly write into the input buffer of the resampler. The pointer
* returned is adjusted so that leftover data are not overwritten.
*/
T * input_buffer(size_t frame_count)
{
@ -312,8 +336,8 @@ public:
return resampling_in_buffer.data() + leftover_samples;
}
/** This method works with `input_buffer`, and allows to inform the processor
how much frames have been written in the provided buffer. */
/** This method works with `input_buffer`, and allows to inform the
processor how much frames have been written in the provided buffer. */
void written(size_t written_frames)
{
resampling_in_buffer.set_length(leftover_samples +
@ -331,6 +355,9 @@ public:
}
}
size_t input_buffer_size() const { return resampling_in_buffer.length(); }
size_t output_buffer_size() const { return resampling_out_buffer.length(); }
private:
/** Wrapper for the speex resampling functions to have a typed
* interface. */
@ -359,6 +386,7 @@ private:
output_frame_count);
assert(rv == RESAMPLER_ERR_SUCCESS);
}
/** The state for the speex resampler used internaly. */
SpeexResamplerState * speex_resampler;
/** Source rate / target rate. */
@ -371,8 +399,8 @@ private:
auto_array<T> resampling_out_buffer;
/** Additional latency inserted into the pipeline for synchronisation. */
uint32_t additional_latency;
/** When `input_buffer` is called, this allows tracking the number of samples
that were in the buffer. */
/** When `input_buffer` is called, this allows tracking the number of
samples that were in the buffer. */
uint32_t leftover_samples;
};
@ -417,8 +445,8 @@ public:
return delay_output_buffer.data();
}
/** Get a pointer to the first writable location in the input buffer>
* @parameter frames_needed the number of frames the user needs to write into
* the buffer.
* @parameter frames_needed the number of frames the user needs to write
* into the buffer.
* @returns a pointer to a location in the input buffer where #frames_needed
* can be writen. */
T * input_buffer(uint32_t frames_needed)
@ -428,8 +456,8 @@ public:
frames_to_samples(frames_needed));
return delay_input_buffer.data() + leftover_samples;
}
/** This method works with `input_buffer`, and allows to inform the processor
how much frames have been written in the provided buffer. */
/** This method works with `input_buffer`, and allows to inform the
processor how much frames have been written in the provided buffer. */
void written(size_t frames_written)
{
delay_input_buffer.set_length(leftover_samples +
@ -450,8 +478,8 @@ public:
return to_pop;
}
/** Returns the number of frames one needs to input into the delay line to get
* #frames_needed frames back.
/** Returns the number of frames one needs to input into the delay line to
* get #frames_needed frames back.
* @parameter frames_needed the number of frames one want to write into the
* delay_line
* @returns the number of frames one will get. */
@ -469,19 +497,23 @@ public:
void drop_audio_if_needed()
{
size_t available = samples_to_frames(delay_input_buffer.length());
uint32_t available = samples_to_frames(delay_input_buffer.length());
uint32_t to_keep = min_buffered_audio_frame(sample_rate);
if (available > to_keep) {
ALOGV("Dropping %u frames", available - to_keep);
delay_input_buffer.pop(nullptr, frames_to_samples(available - to_keep));
}
}
size_t input_buffer_size() const { return delay_input_buffer.length(); }
size_t output_buffer_size() const { return delay_output_buffer.length(); }
private:
/** The length, in frames, of this delay line */
uint32_t length;
/** When `input_buffer` is called, this allows tracking the number of samples
that where in the buffer. */
/** When `input_buffer` is called, this allows tracking the number of
samples that where in the buffer. */
uint32_t leftover_samples;
/** The input buffer, where the delay is applied. */
auto_array<T> delay_input_buffer;
@ -511,8 +543,8 @@ cubeb_resampler_create_internal(cubeb_stream * stream,
"need at least one valid parameter pointer.");
/* All the streams we have have a sample rate that matches the target
sample rate, use a no-op resampler, that simply forwards the buffers to the
callback. */
sample rate, use a no-op resampler, that simply forwards the buffers to
the callback. */
if (((input_params && input_params->rate == target_rate) &&
(output_params && output_params->rate == target_rate)) ||
(input_params && !output_params && (input_params->rate == target_rate)) ||

View File

@ -4,8 +4,12 @@
* This program is made available under an ISC-style license. See the
* accompanying file LICENSE for details.
*/
#ifndef _WIN32_WINNT
#define _WIN32_WINNT 0x0603
#endif // !_WIN32_WINNT
#ifndef NOMINMAX
#define NOMINMAX
#endif // !NOMINMAX
#include <algorithm>
#include <atomic>
@ -37,31 +41,6 @@
#include "cubeb_tracing.h"
#include "cubeb_utils.h"
// Some people have reported glitches with IAudioClient3 capture streams:
// http://blog.nirbheek.in/2018/03/low-latency-audio-on-windows-with.html
// https://bugzilla.mozilla.org/show_bug.cgi?id=1590902
#define ALLOW_AUDIO_CLIENT_3_FOR_INPUT 0
// IAudioClient3::GetSharedModeEnginePeriod() seem to return min latencies
// bigger than IAudioClient::GetDevicePeriod(), which is confusing (10ms vs
// 3ms), though the default latency is usually the same and we should use the
// IAudioClient3 function anyway, as it's more correct
#define USE_AUDIO_CLIENT_3_MIN_PERIOD 1
// If this is true, we allow IAudioClient3 the creation of sessions with a
// latency above the default one (usually 10ms).
// Whether we should default this to true or false depend on many things:
// -Does creating a shared IAudioClient3 session (not locked to a format)
// actually forces all the IAudioClient(1) sessions to have the same latency?
// I could find no proof of that.
// -Does creating a shared IAudioClient3 session with a latency >= the default
// one actually improve the latency (as in how late the audio is) at all?
// -Maybe we could expose this as cubeb stream pref
// (e.g. take priority over other apps)?
#define ALLOW_AUDIO_CLIENT_3_LATENCY_OVER_DEFAULT 1
// If this is true and the user specified a target latency >= the IAudioClient3
// max one, then we reject it and fall back to IAudioClient(1). There wouldn't
// be much point in having a low latency if that's not what the user wants.
#define REJECT_AUDIO_CLIENT_3_LATENCY_OVER_MAX 0
// Windows 10 exposes the IAudioClient3 interface to create low-latency streams.
// Copy the interface definition from audioclient.h here to make the code
// simpler and so that we can still access IAudioClient3 via COM if cubeb was
@ -229,11 +208,6 @@ struct auto_stream_ref {
cubeb_stream * stm;
};
using set_mm_thread_characteristics_function =
decltype(&AvSetMmThreadCharacteristicsW);
using revert_mm_thread_characteristics_function =
decltype(&AvRevertMmThreadCharacteristics);
extern cubeb_ops const wasapi_ops;
static com_heap_ptr<wchar_t>
@ -304,8 +278,8 @@ wasapi_enumerate_devices_internal(cubeb * context, cubeb_device_type type,
static int
wasapi_device_collection_destroy(cubeb * ctx,
cubeb_device_collection * collection);
static char const *
wstr_to_utf8(wchar_t const * str);
static std::unique_ptr<char const[]>
wstr_to_utf8(LPCWSTR str);
static std::unique_ptr<wchar_t const[]>
utf8_to_wstr(char const * str);
@ -314,6 +288,15 @@ utf8_to_wstr(char const * str);
class wasapi_collection_notification_client;
class monitor_device_notifications;
typedef enum {
/* Clear options */
CUBEB_AUDIO_CLIENT2_NONE,
/* Use AUDCLNT_STREAMOPTIONS_RAW */
CUBEB_AUDIO_CLIENT2_RAW,
/* Use CUBEB_STREAM_PREF_COMMUNICATIONS */
CUBEB_AUDIO_CLIENT2_VOICE
} AudioClient2Option;
struct cubeb {
cubeb_ops const * ops = &wasapi_ops;
owned_critical_section lock;
@ -331,13 +314,6 @@ struct cubeb {
nullptr;
void * output_collection_changed_user_ptr = nullptr;
UINT64 performance_counter_frequency;
/* Library dynamically opened to increase the render thread priority, and
the two function pointers we need. */
HMODULE mmcss_module = nullptr;
set_mm_thread_characteristics_function set_mm_thread_characteristics =
nullptr;
revert_mm_thread_characteristics_function revert_mm_thread_characteristics =
nullptr;
};
class wasapi_endpoint_notification_client;
@ -360,20 +336,33 @@ struct cubeb_stream {
/* Mixer pameters. We need to convert the input stream to this
samplerate/channel layout, as WASAPI does not resample nor upmix
itself. */
cubeb_stream_params input_mix_params = {CUBEB_SAMPLE_FLOAT32NE, 0, 0,
cubeb_stream_params input_mix_params = {CUBEB_SAMPLE_FLOAT32NE,
0,
0,
CUBEB_LAYOUT_UNDEFINED,
CUBEB_STREAM_PREF_NONE};
cubeb_stream_params output_mix_params = {CUBEB_SAMPLE_FLOAT32NE, 0, 0,
CUBEB_STREAM_PREF_NONE,
CUBEB_INPUT_PROCESSING_PARAM_NONE};
cubeb_stream_params output_mix_params = {CUBEB_SAMPLE_FLOAT32NE,
0,
0,
CUBEB_LAYOUT_UNDEFINED,
CUBEB_STREAM_PREF_NONE};
CUBEB_STREAM_PREF_NONE,
CUBEB_INPUT_PROCESSING_PARAM_NONE};
/* Stream parameters. This is what the client requested,
* and what will be presented in the callback. */
cubeb_stream_params input_stream_params = {CUBEB_SAMPLE_FLOAT32NE, 0, 0,
cubeb_stream_params input_stream_params = {CUBEB_SAMPLE_FLOAT32NE,
0,
0,
CUBEB_LAYOUT_UNDEFINED,
CUBEB_STREAM_PREF_NONE};
cubeb_stream_params output_stream_params = {CUBEB_SAMPLE_FLOAT32NE, 0, 0,
CUBEB_LAYOUT_UNDEFINED,
CUBEB_STREAM_PREF_NONE};
CUBEB_STREAM_PREF_NONE,
CUBEB_INPUT_PROCESSING_PARAM_NONE};
cubeb_stream_params output_stream_params = {
CUBEB_SAMPLE_FLOAT32NE,
0,
0,
CUBEB_LAYOUT_UNDEFINED,
CUBEB_STREAM_PREF_NONE,
CUBEB_INPUT_PROCESSING_PARAM_NONE};
/* A MMDevice role for this stream: either communication or console here. */
ERole role;
/* True if this stream will transport voice-data. */
@ -662,6 +651,10 @@ public:
LPCWSTR device_id)
{
LOG("collection: Audio device default changed, id = %S.", device_id);
/* Default device changes count as device collection changes */
monitor_notifications.notify(flow);
return S_OK;
}
@ -772,7 +765,7 @@ public:
LPCWSTR device_id)
{
LOG("endpoint: Audio device default changed flow=%d role=%d "
"new_device_id=%ws.",
"new_device_id=%S.",
flow, role, device_id);
/* we only support a single stream type for now. */
@ -783,11 +776,13 @@ public:
DWORD last_change_ms = timeGetTime() - last_device_change;
bool same_device = default_device_id && device_id &&
wcscmp(default_device_id.get(), device_id) == 0;
LOG("endpoint: Audio device default changed last_change=%u same_device=%d",
LOG("endpoint: Audio device default changed last_change=%lu same_device=%d",
last_change_ms, same_device);
if (last_change_ms > DEVICE_CHANGE_DEBOUNCE_MS || !same_device) {
if (device_id) {
default_device_id.reset(_wcsdup(device_id));
wchar_t * new_device_id = new wchar_t[wcslen(device_id) + 1];
wcscpy(new_device_id, device_id);
default_device_id.reset(new_device_id);
} else {
default_device_id.reset();
}
@ -863,16 +858,12 @@ intern_device_id(cubeb * ctx, wchar_t const * id)
auto_lock lock(ctx->lock);
char const * tmp = wstr_to_utf8(id);
std::unique_ptr<char const[]> tmp = wstr_to_utf8(id);
if (!tmp) {
return nullptr;
}
char const * interned = cubeb_strings_intern(ctx->device_ids, tmp);
free((void *)tmp);
return interned;
return cubeb_strings_intern(ctx->device_ids, tmp.get());
}
bool
@ -977,7 +968,7 @@ refill(cubeb_stream * stm, void * input_buffer, long input_frames_count,
cubeb_resampler_fill(stm->resampler.get(), input_buffer,
&input_frames_count, dest, output_frames_needed);
if (out_frames < 0) {
ALOGV("Callback refill error: %d", out_frames);
ALOGV("Callback refill error: %ld", out_frames);
wasapi_state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
return out_frames;
}
@ -1263,8 +1254,8 @@ refill_callback_duplex(cubeb_stream * stm)
XASSERT(has_input(stm) && has_output(stm));
if (stm->input_stream_params.prefs & CUBEB_STREAM_PREF_LOOPBACK) {
HRESULT rv = get_input_buffer(stm);
if (FAILED(rv)) {
rv = get_input_buffer(stm);
if (!rv) {
return rv;
}
}
@ -1274,7 +1265,6 @@ refill_callback_duplex(cubeb_stream * stm)
rv = get_output_buffer(stm, output_buffer, output_frames);
if (!rv) {
hr = stm->render_client->ReleaseBuffer(output_frames, 0);
return rv;
}
@ -1291,9 +1281,11 @@ refill_callback_duplex(cubeb_stream * stm)
stm->total_output_frames += output_frames;
ALOGV("in: %zu, out: %zu, missing: %ld, ratio: %f", stm->total_input_frames,
stm->total_output_frames,
static_cast<long>(stm->total_output_frames) - stm->total_input_frames,
ALOGV("in: %llu, out: %llu, missing: %ld, ratio: %f",
(unsigned long long)stm->total_input_frames,
(unsigned long long)stm->total_output_frames,
static_cast<long long>(stm->total_output_frames) -
static_cast<long long>(stm->total_input_frames),
static_cast<float>(stm->total_output_frames) / stm->total_input_frames);
long got;
@ -1438,8 +1430,7 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
/* We could consider using "Pro Audio" here for WebAudio and
maybe WebRTC. */
mmcss_handle =
stm->context->set_mm_thread_characteristics(L"Audio", &mmcss_task_index);
mmcss_handle = AvSetMmThreadCharacteristicsA("Audio", &mmcss_task_index);
if (!mmcss_handle) {
/* This is not fatal, but we might glitch under heavy load. */
LOG("Unable to use mmcss to bump the render thread priority: %lx",
@ -1519,8 +1510,8 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
is_playing = stm->refill_callback(stm);
break;
case WAIT_OBJECT_0 + 3: { /* input available */
HRESULT rv = get_input_buffer(stm);
if (FAILED(rv)) {
bool rv = get_input_buffer(stm);
if (!rv) {
is_playing = false;
continue;
}
@ -1532,8 +1523,11 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
break;
}
default:
LOG("case %lu not handled in render loop.", waitResult);
XASSERT(false);
LOG("render_loop: waitResult=%lu (lastError=%lu) unhandled, exiting",
waitResult, GetLastError());
is_playing = false;
hr = E_FAIL;
continue;
}
}
@ -1547,7 +1541,7 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
}
if (mmcss_handle) {
stm->context->revert_mm_thread_characteristics(mmcss_handle);
AvRevertMmThreadCharacteristics(mmcss_handle);
}
if (FAILED(hr)) {
@ -1560,18 +1554,6 @@ static unsigned int __stdcall wasapi_stream_render_loop(LPVOID stream)
void
wasapi_destroy(cubeb * context);
HANDLE WINAPI
set_mm_thread_characteristics_noop(LPCWSTR, LPDWORD mmcss_task_index)
{
return (HANDLE)1;
}
BOOL WINAPI
revert_mm_thread_characteristics_noop(HANDLE mmcss_handle)
{
return true;
}
HRESULT
register_notification_client(cubeb_stream * stm)
{
@ -1807,31 +1789,6 @@ wasapi_init(cubeb ** context, char const * context_name)
ctx->performance_counter_frequency = 0;
}
ctx->mmcss_module = LoadLibraryW(L"Avrt.dll");
bool success = false;
if (ctx->mmcss_module) {
ctx->set_mm_thread_characteristics =
reinterpret_cast<set_mm_thread_characteristics_function>(
GetProcAddress(ctx->mmcss_module, "AvSetMmThreadCharacteristicsW"));
ctx->revert_mm_thread_characteristics =
reinterpret_cast<revert_mm_thread_characteristics_function>(
GetProcAddress(ctx->mmcss_module,
"AvRevertMmThreadCharacteristics"));
success = ctx->set_mm_thread_characteristics &&
ctx->revert_mm_thread_characteristics;
}
if (!success) {
// This is not a fatal error, but we might end up glitching when
// the system is under high load.
LOG("Could not load avrt.dll or fetch AvSetMmThreadCharacteristicsW "
"AvRevertMmThreadCharacteristics: %lx",
GetLastError());
ctx->set_mm_thread_characteristics = &set_mm_thread_characteristics_noop;
ctx->revert_mm_thread_characteristics =
&revert_mm_thread_characteristics_noop;
}
*context = ctx;
return CUBEB_OK;
@ -1839,7 +1796,6 @@ wasapi_init(cubeb ** context, char const * context_name)
}
namespace {
enum ShutdownPhase { OnStop, OnDestroy };
bool
stop_and_join_render_thread(cubeb_stream * stm)
@ -1855,16 +1811,7 @@ stop_and_join_render_thread(cubeb_stream * stm)
return false;
}
/* Wait five seconds for the rendering thread to return. It's supposed to
* check its event loop very often, five seconds is rather conservative.
* Note: 5*1s loop to work around timer sleep issues on pre-Windows 8. */
DWORD r;
for (int i = 0; i < 5; ++i) {
r = WaitForSingleObject(stm->thread, 1000);
if (r == WAIT_OBJECT_0) {
break;
}
}
DWORD r = WaitForSingleObject(stm->thread, INFINITE);
if (r != WAIT_OBJECT_0) {
LOG("stop_and_join_render_thread: WaitForSingleObject on thread failed: "
"%lx, %lx",
@ -1888,10 +1835,6 @@ wasapi_destroy(cubeb * context)
}
}
if (context->mmcss_module) {
FreeLibrary(context->mmcss_module);
}
delete context;
}
@ -1949,44 +1892,6 @@ wasapi_get_min_latency(cubeb * ctx, cubeb_stream_params params,
return CUBEB_ERROR;
}
#if USE_AUDIO_CLIENT_3_MIN_PERIOD
// This is unreliable as we can't know the actual mixer format cubeb will
// ask for later on (nor we can branch on ALLOW_AUDIO_CLIENT_3_FOR_INPUT),
// and the min latency can change based on that.
com_ptr<IAudioClient3> client3;
hr = device->Activate(__uuidof(IAudioClient3), CLSCTX_INPROC_SERVER, NULL,
client3.receive_vpp());
if (SUCCEEDED(hr)) {
WAVEFORMATEX * mix_format = nullptr;
hr = client3->GetMixFormat(&mix_format);
if (SUCCEEDED(hr)) {
uint32_t default_period = 0, fundamental_period = 0, min_period = 0,
max_period = 0;
hr = client3->GetSharedModeEnginePeriod(mix_format, &default_period,
&fundamental_period, &min_period,
&max_period);
auto sample_rate = mix_format->nSamplesPerSec;
CoTaskMemFree(mix_format);
if (SUCCEEDED(hr)) {
// Print values in the same format as IAudioDevice::GetDevicePeriod()
REFERENCE_TIME min_period_rt(frames_to_hns(sample_rate, min_period));
REFERENCE_TIME default_period_rt(
frames_to_hns(sample_rate, default_period));
LOG("default device period: %I64d, minimum device period: %I64d",
default_period_rt, min_period_rt);
*latency_frames = hns_to_frames(params.rate, min_period_rt);
LOG("Minimum latency in frames: %u", *latency_frames);
return CUBEB_OK;
}
}
}
#endif
com_ptr<IAudioClient> client;
hr = device->Activate(__uuidof(IAudioClient), CLSCTX_INPROC_SERVER, NULL,
client.receive_vpp());
@ -2006,8 +1911,18 @@ wasapi_get_min_latency(cubeb * ctx, cubeb_stream_params params,
LOG("default device period: %I64d, minimum device period: %I64d",
default_period, minimum_period);
// The minimum_period is only relevant in exclusive streams.
/* If we're on Windows 10, we can use IAudioClient3 to get minimal latency.
Otherwise, according to the docs, the best latency we can achieve is by
synchronizing the stream and the engine.
http://msdn.microsoft.com/en-us/library/windows/desktop/dd370871%28v=vs.85%29.aspx
*/
// #ifdef _WIN32_WINNT_WIN10
#if 0
*latency_frames = hns_to_frames(params.rate, minimum_period);
#else
*latency_frames = hns_to_frames(params.rate, default_period);
#endif
LOG("Minimum latency in frames: %u", *latency_frames);
@ -2044,6 +1959,21 @@ wasapi_get_preferred_sample_rate(cubeb * ctx, uint32_t * rate)
return CUBEB_OK;
}
int
wasapi_get_supported_input_processing_params(
cubeb * ctx, cubeb_input_processing_params * params)
{
// This is not entirely accurate -- windows doesn't document precisely what
// AudioCategory_Communications does -- but assume that we can set all or none
// of them.
*params = static_cast<cubeb_input_processing_params>(
CUBEB_INPUT_PROCESSING_PARAM_ECHO_CANCELLATION |
CUBEB_INPUT_PROCESSING_PARAM_NOISE_SUPPRESSION |
CUBEB_INPUT_PROCESSING_PARAM_AUTOMATIC_GAIN_CONTROL |
CUBEB_INPUT_PROCESSING_PARAM_VOICE_ISOLATION);
return CUBEB_OK;
}
static void
waveformatex_update_derived_properties(WAVEFORMATEX * format)
{
@ -2097,10 +2027,7 @@ handle_channel_layout(cubeb_stream * stm, EDataFlow direction,
if (hr == S_FALSE) {
/* Channel layout not supported, but WASAPI gives us a suggestion. Use it,
and handle the eventual upmix/downmix ourselves. Ignore the subformat of
the suggestion, since it seems to always be IEEE_FLOAT.
This fallback doesn't update the bit depth, so if a device
only supported bit depths cubeb doesn't support, so IAudioClient3
streams might fail */
the suggestion, since it seems to always be IEEE_FLOAT. */
LOG("Using WASAPI suggested format: channels: %d", closest->nChannels);
XASSERT(closest->wFormatTag == WAVE_FORMAT_EXTENSIBLE);
WAVEFORMATEXTENSIBLE * closest_pcm =
@ -2122,7 +2049,8 @@ handle_channel_layout(cubeb_stream * stm, EDataFlow direction,
}
static int
initialize_iaudioclient2(com_ptr<IAudioClient> & audio_client)
initialize_iaudioclient2(com_ptr<IAudioClient> & audio_client,
AudioClient2Option option)
{
com_ptr<IAudioClient2> audio_client2;
audio_client->QueryInterface<IAudioClient2>(audio_client2.receive());
@ -2131,10 +2059,14 @@ initialize_iaudioclient2(com_ptr<IAudioClient> & audio_client)
"AUDCLNT_STREAMOPTIONS_RAW.");
return CUBEB_OK;
}
AudioClientProperties properties = {0};
AudioClientProperties properties = {};
properties.cbSize = sizeof(AudioClientProperties);
#ifndef __MINGW32__
properties.Options |= AUDCLNT_STREAMOPTIONS_RAW;
if (option == CUBEB_AUDIO_CLIENT2_RAW) {
properties.Options |= AUDCLNT_STREAMOPTIONS_RAW;
} else if (option == CUBEB_AUDIO_CLIENT2_VOICE) {
properties.eCategory = AudioCategory_Communications;
}
#endif
HRESULT hr = audio_client2->SetClientProperties(&properties);
if (FAILED(hr)) {
@ -2144,12 +2076,12 @@ initialize_iaudioclient2(com_ptr<IAudioClient> & audio_client)
return CUBEB_OK;
}
#if 0
bool
initialize_iaudioclient3(com_ptr<IAudioClient> & audio_client,
cubeb_stream * stm,
const com_heap_ptr<WAVEFORMATEX> & mix_format,
DWORD flags, EDataFlow direction,
REFERENCE_TIME latency_hns)
DWORD flags, EDataFlow direction)
{
com_ptr<IAudioClient3> audio_client3;
audio_client->QueryInterface<IAudioClient3>(audio_client3.receive());
@ -2165,22 +2097,24 @@ initialize_iaudioclient3(com_ptr<IAudioClient> & audio_client,
return false;
}
// Some people have reported glitches with capture streams:
// http://blog.nirbheek.in/2018/03/low-latency-audio-on-windows-with.html
if (direction == eCapture) {
LOG("Audio stream is capture, not using IAudioClient3");
return false;
}
// Possibly initialize a shared-mode stream using IAudioClient3. Initializing
// a stream this way lets you request lower latencies, but also locks the
// global WASAPI engine at that latency.
// - If we request a shared-mode stream, streams created with IAudioClient
// might have their latency adjusted to match. When the shared-mode stream
// is closed, they'll go back to normal.
// - If there's already a shared-mode stream running, if it created with the
// AUDCLNT_STREAMOPTIONS_MATCH_FORMAT option, the audio engine would be
// locked to that format, so we have to match it (a custom one would fail).
// - We don't lock the WASAPI engine to a format, as it's antisocial towards
// other apps, especially if we locked to a latency >= than its default.
// - If the user requested latency is >= the default one, we might still
// accept it (without locking the format) depending on
// ALLOW_AUDIO_CLIENT_3_LATENCY_OVER_DEFAULT, as we might want to prioritize
// to lower our latency over other apps
// (there might still be latency advantages compared to IAudioDevice(1)).
// will
// have their latency adjusted to match. When the shared-mode stream is
// closed, they'll go back to normal.
// - If there's already a shared-mode stream running, then we cannot request
// the engine change to a different latency - we have to match it.
// - It's antisocial to lock the WASAPI engine at its default latency. If we
// would do this, then stop and use IAudioClient instead.
HRESULT hr;
uint32_t default_period = 0, fundamental_period = 0, min_period = 0,
@ -2192,59 +2126,28 @@ initialize_iaudioclient3(com_ptr<IAudioClient> & audio_client,
LOG("Could not get shared mode engine period: error: %lx", hr);
return false;
}
uint32_t requested_latency =
hns_to_frames(mix_format->nSamplesPerSec, latency_hns);
#if !ALLOW_AUDIO_CLIENT_3_LATENCY_OVER_DEFAULT
uint32_t requested_latency = stm->latency;
if (requested_latency >= default_period) {
LOG("Requested latency %i equal or greater than default latency %i,"
" not using IAudioClient3",
LOG("Requested latency %i greater than default latency %i, not using "
"IAudioClient3",
requested_latency, default_period);
return false;
}
#elif REJECT_AUDIO_CLIENT_3_LATENCY_OVER_MAX
if (requested_latency > max_period) {
// Fallback to IAudioClient(1) as it's more accepting of large latencies
LOG("Requested latency %i greater than max latency %i,"
" not using IAudioClient3",
requested_latency, max_period);
return false;
}
#endif
LOG("Got shared mode engine period: default=%i fundamental=%i min=%i max=%i",
default_period, fundamental_period, min_period, max_period);
// Snap requested latency to a valid value
uint32_t old_requested_latency = requested_latency;
// The period is required to be a multiple of the fundamental period
// (and >= min and <= max, which should still be true)
requested_latency -= requested_latency % fundamental_period;
if (requested_latency < min_period) {
requested_latency = min_period;
}
// Likely unnecessary, but won't hurt
if (requested_latency > max_period) {
requested_latency = max_period;
}
requested_latency -= (requested_latency - min_period) % fundamental_period;
if (requested_latency != old_requested_latency) {
LOG("Requested latency %i was adjusted to %i", old_requested_latency,
requested_latency);
}
DWORD new_flags = flags;
// Always add these flags to IAudioClient3, they might help
// if the stream doesn't have the same format as the audio engine.
new_flags |= AUDCLNT_STREAMFLAGS_AUTOCONVERTPCM;
new_flags |= AUDCLNT_STREAMFLAGS_SRC_DEFAULT_QUALITY;
hr = audio_client3->InitializeSharedAudioStream(new_flags, requested_latency,
hr = audio_client3->InitializeSharedAudioStream(flags, requested_latency,
mix_format.get(), NULL);
// This error should be returned first even if
// the period was locked (AUDCLNT_E_ENGINE_PERIODICITY_LOCKED)
if (hr == AUDCLNT_E_INVALID_STREAM_FLAG) {
LOG("Got AUDCLNT_E_INVALID_STREAM_FLAG, removing some flags");
hr = audio_client3->InitializeSharedAudioStream(flags, requested_latency,
mix_format.get(), NULL);
}
if (SUCCEEDED(hr)) {
return true;
} else if (hr == AUDCLNT_E_ENGINE_PERIODICITY_LOCKED) {
@ -2256,37 +2159,22 @@ initialize_iaudioclient3(com_ptr<IAudioClient> & audio_client,
}
uint32_t current_period = 0;
WAVEFORMATEX * current_format_ptr = nullptr;
WAVEFORMATEX * current_format = nullptr;
// We have to pass a valid WAVEFORMATEX** and not nullptr, otherwise
// GetCurrentSharedModeEnginePeriod will return E_POINTER
hr = audio_client3->GetCurrentSharedModeEnginePeriod(&current_format_ptr,
hr = audio_client3->GetCurrentSharedModeEnginePeriod(&current_format,
&current_period);
CoTaskMemFree(current_format);
if (FAILED(hr)) {
LOG("Could not get current shared mode engine period: error: %lx", hr);
return false;
}
com_heap_ptr<WAVEFORMATEX> current_format(current_format_ptr);
if (current_format->nSamplesPerSec != mix_format->nSamplesPerSec) {
// Unless some other external app locked the shared mode engine period
// within our audio initialization, this is unlikely to happen, though we
// can't respect the user selected latency, so we fallback on IAudioClient
LOG("IAudioClient3::GetCurrentSharedModeEnginePeriod() returned a "
"different mixer format (nSamplesPerSec) from "
"IAudioClient::GetMixFormat(); not using IAudioClient3");
return false;
}
#if REJECT_AUDIO_CLIENT_3_LATENCY_OVER_MAX
// Reject IAudioClient3 if we can't respect the user target latency.
// We don't need to check against default_latency anymore,
// as the current_period is already the best one we could get.
if (old_requested_latency > current_period) {
LOG("Requested latency %i greater than currently locked shared mode "
"latency %i, not using IAudioClient3",
old_requested_latency, current_period);
if (current_period >= default_period) {
LOG("Current shared mode engine period %i too high, not using IAudioClient",
current_period);
return false;
}
#endif
hr = audio_client3->InitializeSharedAudioStream(flags, current_period,
mix_format.get(), NULL);
@ -2299,6 +2187,7 @@ initialize_iaudioclient3(com_ptr<IAudioClient> & audio_client,
LOG("Could not initialize shared stream with IAudioClient3: error: %lx", hr);
return false;
}
#endif
#define DIRECTION_NAME (direction == eCapture ? "capture" : "render")
@ -2322,12 +2211,6 @@ setup_wasapi_stream_one_side(cubeb_stream * stm,
return CUBEB_ERROR;
}
#if ALLOW_AUDIO_CLIENT_3_FOR_INPUT
constexpr bool allow_audio_client_3 = true;
#else
const bool allow_audio_client_3 = direction == eRender;
#endif
stm->stream_reset_lock.assert_current_thread_owns();
// If user doesn't specify a particular device, we can choose another one when
// the given devid is unavailable.
@ -2364,14 +2247,17 @@ setup_wasapi_stream_one_side(cubeb_stream * stm,
/* Get a client. We will get all other interfaces we need from
* this pointer. */
if (allow_audio_client_3) {
hr = device->Activate(__uuidof(IAudioClient3), CLSCTX_INPROC_SERVER, NULL,
audio_client.receive_vpp());
}
if (!allow_audio_client_3 || hr == E_NOINTERFACE) {
hr = device->Activate(__uuidof(IAudioClient), CLSCTX_INPROC_SERVER, NULL,
audio_client.receive_vpp());
#if 0 // See https://bugzilla.mozilla.org/show_bug.cgi?id=1590902
hr = device->Activate(__uuidof(IAudioClient3),
CLSCTX_INPROC_SERVER,
NULL, audio_client.receive_vpp());
if (hr == E_NOINTERFACE) {
#endif
hr = device->Activate(__uuidof(IAudioClient), CLSCTX_INPROC_SERVER, NULL,
audio_client.receive_vpp());
#if 0
}
#endif
if (FAILED(hr)) {
LOG("Could not activate the device to get an audio"
@ -2494,21 +2380,41 @@ setup_wasapi_stream_one_side(cubeb_stream * stm,
}
if (stream_params->prefs & CUBEB_STREAM_PREF_RAW) {
if (initialize_iaudioclient2(audio_client) != CUBEB_OK) {
if (initialize_iaudioclient2(audio_client, CUBEB_AUDIO_CLIENT2_RAW) !=
CUBEB_OK) {
LOG("Can't initialize an IAudioClient2, error: %lx", GetLastError());
// This is not fatal.
}
} else if (direction == eCapture &&
(stream_params->prefs & CUBEB_STREAM_PREF_VOICE) &&
stream_params->input_params != CUBEB_INPUT_PROCESSING_PARAM_NONE) {
if (stream_params->input_params ==
(CUBEB_INPUT_PROCESSING_PARAM_ECHO_CANCELLATION |
CUBEB_INPUT_PROCESSING_PARAM_NOISE_SUPPRESSION |
CUBEB_INPUT_PROCESSING_PARAM_AUTOMATIC_GAIN_CONTROL |
CUBEB_INPUT_PROCESSING_PARAM_VOICE_ISOLATION)) {
if (initialize_iaudioclient2(audio_client, CUBEB_AUDIO_CLIENT2_VOICE) !=
CUBEB_OK) {
LOG("Can't initialize an IAudioClient2, error: %lx", GetLastError());
// This is not fatal.
}
} else {
LOG("Invalid combination of input processing params %#x",
stream_params->input_params);
return CUBEB_ERROR;
}
}
if (allow_audio_client_3 &&
initialize_iaudioclient3(audio_client, stm, mix_format, flags, direction,
latency_hns)) {
#if 0 // See https://bugzilla.mozilla.org/show_bug.cgi?id=1590902
if (initialize_iaudioclient3(audio_client, stm, mix_format, flags, direction)) {
LOG("Initialized with IAudioClient3");
} else {
hr = audio_client->Initialize(AUDCLNT_SHAREMODE_SHARED, flags, latency_hns,
0, mix_format.get(), NULL);
#endif
hr = audio_client->Initialize(AUDCLNT_SHAREMODE_SHARED, flags, latency_hns, 0,
mix_format.get(), NULL);
#if 0
}
#endif
if (FAILED(hr)) {
LOG("Unable to initialize audio client for %s: %lx.", DIRECTION_NAME, hr);
return CUBEB_ERROR;
@ -2970,6 +2876,7 @@ wasapi_stream_init(cubeb * context, cubeb_stream ** stream,
}
}
cubeb_async_log_reset_threads();
stm->thread =
(HANDLE)_beginthreadex(NULL, 512 * 1024, wasapi_stream_render_loop, stm,
STACK_SIZE_PARAM_IS_A_RESERVATION, NULL);
@ -3031,7 +2938,7 @@ wasapi_stream_add_ref(cubeb_stream * stm)
{
XASSERT(stm);
LONG result = InterlockedIncrement(&stm->ref_count);
LOGV("Stream ref count incremented = %i (%p)", result, stm);
LOGV("Stream ref count incremented = %ld (%p)", result, stm);
return result;
}
@ -3041,7 +2948,7 @@ wasapi_stream_release(cubeb_stream * stm)
XASSERT(stm);
LONG result = InterlockedDecrement(&stm->ref_count);
LOGV("Stream ref count decremented = %i (%p)", result, stm);
LOGV("Stream ref count decremented = %ld (%p)", result, stm);
if (result == 0) {
LOG("Stream ref count hit zero, destroying (%p)", stm);
@ -3303,7 +3210,7 @@ wasapi_stream_set_volume(cubeb_stream * stm, float volume)
return CUBEB_OK;
}
static char const *
static std::unique_ptr<char const[]>
wstr_to_utf8(LPCWSTR str)
{
int size = ::WideCharToMultiByte(CP_UTF8, 0, str, -1, nullptr, 0, NULL, NULL);
@ -3311,8 +3218,8 @@ wstr_to_utf8(LPCWSTR str)
return nullptr;
}
char * ret = static_cast<char *>(malloc(size));
::WideCharToMultiByte(CP_UTF8, 0, str, -1, ret, size, NULL, NULL);
std::unique_ptr<char[]> ret(new char[size]);
::WideCharToMultiByte(CP_UTF8, 0, str, -1, ret.get(), size, NULL, NULL);
return ret;
}
@ -3440,7 +3347,7 @@ wasapi_create_device(cubeb * ctx, cubeb_device_info & ret,
prop_variant namevar;
hr = propstore->GetValue(PKEY_Device_FriendlyName, &namevar);
if (SUCCEEDED(hr) && namevar.vt == VT_LPWSTR) {
ret.friendly_name = wstr_to_utf8(namevar.pwszVal);
ret.friendly_name = wstr_to_utf8(namevar.pwszVal).release();
}
if (!ret.friendly_name) {
// This is not fatal, but a valid string is expected in all cases.
@ -3461,7 +3368,7 @@ wasapi_create_device(cubeb * ctx, cubeb_device_info & ret,
prop_variant instancevar;
hr = ps->GetValue(PKEY_Device_InstanceId, &instancevar);
if (SUCCEEDED(hr) && instancevar.vt == VT_LPWSTR) {
ret.group_id = wstr_to_utf8(instancevar.pwszVal);
ret.group_id = wstr_to_utf8(instancevar.pwszVal).release();
}
}
@ -3477,7 +3384,8 @@ wasapi_create_device(cubeb * ctx, cubeb_device_info & ret,
ret.preferred =
(cubeb_device_pref)(ret.preferred | CUBEB_DEVICE_PREF_MULTIMEDIA |
CUBEB_DEVICE_PREF_NOTIFICATION);
} else if (defaults->is_default(flow, eCommunications, device_id.get())) {
}
if (defaults->is_default(flow, eCommunications, device_id.get())) {
ret.preferred =
(cubeb_device_pref)(ret.preferred | CUBEB_DEVICE_PREF_VOICE);
}
@ -3504,7 +3412,6 @@ wasapi_create_device(cubeb * ctx, cubeb_device_info & ret,
CUBEB_DEVICE_FMT_S16NE);
ret.default_format = CUBEB_DEVICE_FMT_F32NE;
prop_variant fmtvar;
WAVEFORMATEX * wfx = NULL;
hr = propstore->GetValue(PKEY_AudioEngine_DeviceFormat, &fmtvar);
if (SUCCEEDED(hr) && fmtvar.vt == VT_BLOB) {
if (fmtvar.blob.cbSize == sizeof(PCMWAVEFORMAT)) {
@ -3514,7 +3421,8 @@ wasapi_create_device(cubeb * ctx, cubeb_device_info & ret,
ret.max_rate = ret.min_rate = ret.default_rate = pcm->wf.nSamplesPerSec;
ret.max_channels = pcm->wf.nChannels;
} else if (fmtvar.blob.cbSize >= sizeof(WAVEFORMATEX)) {
wfx = reinterpret_cast<WAVEFORMATEX *>(fmtvar.blob.pBlobData);
WAVEFORMATEX * wfx =
reinterpret_cast<WAVEFORMATEX *>(fmtvar.blob.pBlobData);
if (fmtvar.blob.cbSize >= sizeof(WAVEFORMATEX) + wfx->cbSize ||
wfx->wFormatTag == WAVE_FORMAT_PCM) {
@ -3524,30 +3432,9 @@ wasapi_create_device(cubeb * ctx, cubeb_device_info & ret,
}
}
#if USE_AUDIO_CLIENT_3_MIN_PERIOD
// Here we assume an IAudioClient3 stream will successfully
// be initialized later (it might fail)
#if ALLOW_AUDIO_CLIENT_3_FOR_INPUT
constexpr bool allow_audio_client_3 = true;
#else
const bool allow_audio_client_3 = flow == eRender;
#endif
com_ptr<IAudioClient3> client3;
uint32_t def, fun, min, max;
if (allow_audio_client_3 && wfx &&
SUCCEEDED(dev->Activate(__uuidof(IAudioClient3), CLSCTX_INPROC_SERVER,
NULL, client3.receive_vpp())) &&
SUCCEEDED(
client3->GetSharedModeEnginePeriod(wfx, &def, &fun, &min, &max))) {
ret.latency_lo = min;
// This latency might actually be used as "default" and not "max" later on,
// so we return the default (we never really want to use the max anyway)
ret.latency_hi = def;
} else
#endif
if (SUCCEEDED(dev->Activate(__uuidof(IAudioClient), CLSCTX_INPROC_SERVER,
NULL, client.receive_vpp())) &&
SUCCEEDED(client->GetDevicePeriod(&def_period, &min_period))) {
if (SUCCEEDED(dev->Activate(__uuidof(IAudioClient), CLSCTX_INPROC_SERVER,
NULL, client.receive_vpp())) &&
SUCCEEDED(client->GetDevicePeriod(&def_period, &min_period))) {
ret.latency_lo = hns_to_frames(ret.default_rate, min_period);
ret.latency_hi = hns_to_frames(ret.default_rate, def_period);
} else {
@ -3638,7 +3525,7 @@ wasapi_enumerate_devices(cubeb * context, cubeb_device_type type,
{
return wasapi_enumerate_devices_internal(
context, type, out,
DEVICE_STATE_ACTIVE /*| DEVICE_STATE_DISABLED | DEVICE_STATE_UNPLUGGED*/);
DEVICE_STATE_ACTIVE | DEVICE_STATE_DISABLED | DEVICE_STATE_UNPLUGGED);
}
static int
@ -3656,6 +3543,14 @@ wasapi_device_collection_destroy(cubeb * /*ctx*/,
return CUBEB_OK;
}
int
wasapi_set_input_processing_params(cubeb_stream * stream,
cubeb_input_processing_params params)
{
LOG("Cannot set voice processing params after init. Use cubeb_stream_init.");
return CUBEB_ERROR_NOT_SUPPORTED;
}
static int
wasapi_register_device_collection_changed(
cubeb * context, cubeb_device_type devtype,
@ -3736,7 +3631,8 @@ cubeb_ops const wasapi_ops = {
/*.get_max_channel_count =*/wasapi_get_max_channel_count,
/*.get_min_latency =*/wasapi_get_min_latency,
/*.get_preferred_sample_rate =*/wasapi_get_preferred_sample_rate,
/*.get_supported_input_processing_params =*/NULL,
/*.get_supported_input_processing_params =*/
wasapi_get_supported_input_processing_params,
/*.enumerate_devices =*/wasapi_enumerate_devices,
/*.device_collection_destroy =*/wasapi_device_collection_destroy,
/*.destroy =*/wasapi_destroy,
@ -3751,7 +3647,7 @@ cubeb_ops const wasapi_ops = {
/*.stream_set_name =*/NULL,
/*.stream_get_current_device =*/NULL,
/*.stream_set_input_mute =*/NULL,
/*.stream_set_input_processing_params =*/NULL,
/*.stream_set_input_processing_params =*/wasapi_set_input_processing_params,
/*.stream_device_destroy =*/NULL,
/*.stream_register_device_changed_callback =*/NULL,
/*.register_device_collection_changed =*/

View File

@ -41,10 +41,10 @@
#ifdef FLOATING_POINT
#error You cannot compile as floating point and fixed point at the same time
#endif
#ifdef _USE_SSE
#ifdef USE_SSE
#error SSE is only for floating-point
#endif
#if ((defined (ARM4_ASM)||defined (ARM4_ASM)) && defined(BFIN_ASM)) || (defined (ARM4_ASM)&&defined(ARM5E_ASM))
#if defined(ARM4_ASM) + defined(ARM5E_ASM) + defined(BFIN_ASM) > 1
#error Make up your mind. What CPU do you have?
#endif
#ifdef VORBIS_PSYCHO
@ -56,10 +56,10 @@
#ifndef FLOATING_POINT
#error You now need to define either FIXED_POINT or FLOATING_POINT
#endif
#if defined (ARM4_ASM) || defined(ARM5E_ASM) || defined(BFIN_ASM)
#if defined(ARM4_ASM) || defined(ARM5E_ASM) || defined(BFIN_ASM)
#error I suppose you can have a [ARM4/ARM5E/Blackfin] that has float instructions?
#endif
#ifdef FIXED_POINT_DEBUG
#ifdef FIXED_DEBUG
#error "Don't you think enabling fixed-point is a good thing to do if you want to debug that?"
#endif
@ -117,9 +117,9 @@ typedef spx_word32_t spx_sig_t;
#ifdef ARM5E_ASM
#include "fixed_arm5e.h"
#elif defined (ARM4_ASM)
#elif defined(ARM4_ASM)
#include "fixed_arm4.h"
#elif defined (BFIN_ASM)
#elif defined(BFIN_ASM)
#include "fixed_bfin.h"
#endif
@ -177,16 +177,13 @@ typedef float spx_word32_t;
#define ADD32(a,b) ((a)+(b))
#define SUB32(a,b) ((a)-(b))
#define MULT16_16_16(a,b) ((a)*(b))
#define MULT16_32_32(a,b) ((a)*(b))
#define MULT16_16(a,b) ((spx_word32_t)(a)*(spx_word32_t)(b))
#define MAC16_16(c,a,b) ((c)+(spx_word32_t)(a)*(spx_word32_t)(b))
#define MULT16_32_Q11(a,b) ((a)*(b))
#define MULT16_32_Q13(a,b) ((a)*(b))
#define MULT16_32_Q14(a,b) ((a)*(b))
#define MULT16_32_Q15(a,b) ((a)*(b))
#define MULT16_32_P15(a,b) ((a)*(b))
#define MAC16_32_Q11(c,a,b) ((c)+(a)*(b))
#define MAC16_32_Q15(c,a,b) ((c)+(a)*(b))
#define MAC16_16_Q11(c,a,b) ((c)+(a)*(b))
@ -210,7 +207,7 @@ typedef float spx_word32_t;
#endif
#if defined (CONFIG_TI_C54X) || defined (CONFIG_TI_C55X)
#if defined(CONFIG_TI_C54X) || defined(CONFIG_TI_C55X)
/* 2 on TI C5x DSP */
#define BYTES_PER_CHAR 2

View File

@ -69,22 +69,18 @@
/* result fits in 16 bits */
#define MULT16_16_16(a,b) ((((spx_word16_t)(a))*((spx_word16_t)(b))))
#define MULT16_16_16(a,b) (((spx_word16_t)(a))*((spx_word16_t)(b)))
/* result fits in 32 bits */
#define MULT16_32_32(a,b) (((spx_word16_t)(a))*((spx_word32_t)(b)))
/* (spx_word32_t)(spx_word16_t) gives TI compiler a hint that it's 16x16->32 multiply */
#define MULT16_16(a,b) (((spx_word32_t)(spx_word16_t)(a))*((spx_word32_t)(spx_word16_t)(b)))
#define MAC16_16(c,a,b) (ADD32((c),MULT16_16((a),(b))))
#define MULT16_32_Q12(a,b) ADD32(MULT16_16((a),SHR((b),12)), SHR(MULT16_16((a),((b)&0x00000fff)),12))
#define MULT16_32_Q13(a,b) ADD32(MULT16_16((a),SHR((b),13)), SHR(MULT16_16((a),((b)&0x00001fff)),13))
#define MULT16_32_Q14(a,b) ADD32(MULT16_16((a),SHR((b),14)), SHR(MULT16_16((a),((b)&0x00003fff)),14))
#define MULT16_32_Q11(a,b) ADD32(MULT16_16((a),SHR((b),11)), SHR(MULT16_16((a),((b)&0x000007ff)),11))
#define MAC16_32_Q11(c,a,b) ADD32(c,ADD32(MULT16_16((a),SHR((b),11)), SHR(MULT16_16((a),((b)&0x000007ff)),11)))
#define MULT16_32_P15(a,b) ADD32(MULT16_16((a),SHR((b),15)), PSHR(MULT16_16((a),((b)&0x00007fff)),15))
#define MULT16_32_Q15(a,b) ADD32(MULT16_16((a),SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15))
#define MAC16_32_Q15(c,a,b) ADD32(c,ADD32(MULT16_16((a),SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15)))
#define MULT16_32_P15(a,b) ADD32(MULT16_32_32(a,SHR((b),15)), PSHR(MULT16_16((a),((b)&0x00007fff)),15))
#define MULT16_32_Q15(a,b) ADD32(MULT16_32_32(a,SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15))
#define MAC16_32_Q15(c,a,b) ADD32(c,MULT16_32_Q15(a,b))
#define MAC16_16_Q11(c,a,b) (ADD32((c),SHR(MULT16_16((a),(b)),11)))

View File

@ -46,7 +46,7 @@
Smith, Julius O. Digital Audio Resampling Home Page
Center for Computer Research in Music and Acoustics (CCRMA),
Stanford University, 2007.
Web published at http://ccrma.stanford.edu/~jos/resample/.
Web published at https://ccrma.stanford.edu/~jos/resample/.
There is one main difference, though. This resampler uses cubic
interpolation instead of linear interpolation in the above paper. This
@ -63,9 +63,12 @@
#ifdef OUTSIDE_SPEEX
#include <stdlib.h>
static void *speex_alloc (int size) {return calloc(size,1);}
static void *speex_realloc (void *ptr, int size) {return realloc(ptr, size);}
static void speex_free (void *ptr) {free(ptr);}
static void *speex_alloc(int size) {return calloc(size,1);}
static void *speex_realloc(void *ptr, int size) {return realloc(ptr, size);}
static void speex_free(void *ptr) {free(ptr);}
#ifndef EXPORT
#define EXPORT
#endif
#include "speex_resampler.h"
#include "arch.h"
#else /* OUTSIDE_SPEEX */
@ -75,7 +78,6 @@ static void speex_free (void *ptr) {free(ptr);}
#include "os_support.h"
#endif /* OUTSIDE_SPEEX */
#include "stack_alloc.h"
#include <math.h>
#include <limits.h>
@ -91,18 +93,18 @@ static void speex_free (void *ptr) {free(ptr);}
#endif
#ifndef UINT32_MAX
#define UINT32_MAX 4294967296U
#define UINT32_MAX 4294967295U
#endif
#ifdef _USE_SSE
#ifdef USE_SSE
#include "resample_sse.h"
#endif
#ifdef _USE_NEON
#ifdef USE_NEON
#include "resample_neon.h"
#endif
/* Numer of elements to allocate on the stack */
/* Number of elements to allocate on the stack */
#ifdef VAR_ARRAYS
#define FIXED_STACK_ALLOC 8192
#else
@ -194,16 +196,14 @@ struct FuncDef {
int oversample;
};
static const struct FuncDef _KAISER12 = {kaiser12_table, 64};
#define KAISER12 (&_KAISER12)
/*static struct FuncDef _KAISER12 = {kaiser12_table, 32};
#define KAISER12 (&_KAISER12)*/
static const struct FuncDef _KAISER10 = {kaiser10_table, 32};
#define KAISER10 (&_KAISER10)
static const struct FuncDef _KAISER8 = {kaiser8_table, 32};
#define KAISER8 (&_KAISER8)
static const struct FuncDef _KAISER6 = {kaiser6_table, 32};
#define KAISER6 (&_KAISER6)
static const struct FuncDef kaiser12_funcdef = {kaiser12_table, 64};
#define KAISER12 (&kaiser12_funcdef)
static const struct FuncDef kaiser10_funcdef = {kaiser10_table, 32};
#define KAISER10 (&kaiser10_funcdef)
static const struct FuncDef kaiser8_funcdef = {kaiser8_table, 32};
#define KAISER8 (&kaiser8_funcdef)
static const struct FuncDef kaiser6_funcdef = {kaiser6_table, 32};
#define KAISER6 (&kaiser6_funcdef)
struct QualityMapping {
int base_length;
@ -473,7 +473,7 @@ static int resampler_basic_interpolate_single(SpeexResamplerState *st, spx_uint3
}
cubic_coef(frac, interp);
sum = MULT16_32_Q15(interp[0],SHR32(accum[0], 1)) + MULT16_32_Q15(interp[1],SHR32(accum[1], 1)) + MULT16_32_Q15(interp[2],SHR32(accum[2], 1)) + MULT16_32_Q15(interp[3],SHR32(accum[3], 1));
sum = MULT16_32_Q15(interp[0],accum[0]) + MULT16_32_Q15(interp[1],accum[1]) + MULT16_32_Q15(interp[2],accum[2]) + MULT16_32_Q15(interp[3],accum[3]);
sum = SATURATE32PSHR(sum, 15, 32767);
#else
cubic_coef(frac, interp);
@ -572,6 +572,7 @@ static int resampler_basic_zero(SpeexResamplerState *st, spx_uint32_t channel_in
const int frac_advance = st->frac_advance;
const spx_uint32_t den_rate = st->den_rate;
(void)in;
while (!(last_sample >= (spx_int32_t)*in_len || out_sample >= (spx_int32_t)*out_len))
{
out[out_stride * out_sample++] = 0;
@ -589,16 +590,15 @@ static int resampler_basic_zero(SpeexResamplerState *st, spx_uint32_t channel_in
return out_sample;
}
static int _muldiv(spx_uint32_t *result, spx_uint32_t value, spx_uint32_t mul, spx_uint32_t div)
static int multiply_frac(spx_uint32_t *result, spx_uint32_t value, spx_uint32_t num, spx_uint32_t den)
{
speex_assert(result);
spx_uint32_t major = value / div;
spx_uint32_t remainder = value % div;
spx_uint32_t major = value / den;
spx_uint32_t remain = value % den;
/* TODO: Could use 64 bits operation to check for overflow. But only guaranteed in C99+ */
if (remainder > UINT32_MAX / mul || major > UINT32_MAX / mul
|| major * mul > UINT32_MAX - remainder * mul / div)
if (remain > UINT32_MAX / num || major > UINT32_MAX / num
|| major * num > UINT32_MAX - remain * num / den)
return RESAMPLER_ERR_OVERFLOW;
*result = remainder * mul / div + major * mul;
*result = remain * num / den + major * num;
return RESAMPLER_ERR_SUCCESS;
}
@ -619,7 +619,7 @@ static int update_filter(SpeexResamplerState *st)
{
/* down-sampling */
st->cutoff = quality_map[st->quality].downsample_bandwidth * st->den_rate / st->num_rate;
if (_muldiv(&st->filt_len,st->filt_len,st->num_rate,st->den_rate) != RESAMPLER_ERR_SUCCESS)
if (multiply_frac(&st->filt_len,st->filt_len,st->num_rate,st->den_rate) != RESAMPLER_ERR_SUCCESS)
goto fail;
/* Round up to make sure we have a multiple of 8 for SSE */
st->filt_len = ((st->filt_len-1)&(~0x7))+8;
@ -638,12 +638,12 @@ static int update_filter(SpeexResamplerState *st)
st->cutoff = quality_map[st->quality].upsample_bandwidth;
}
/* Choose the resampling type that requires the least amount of memory */
#ifdef RESAMPLE_FULL_SINC_TABLE
use_direct = 1;
if (INT_MAX/sizeof(spx_word16_t)/st->den_rate < st->filt_len)
goto fail;
#else
/* Choose the resampling type that requires the least amount of memory */
use_direct = st->filt_len*st->den_rate <= st->filt_len*st->oversample+8
&& INT_MAX/sizeof(spx_word16_t)/st->den_rate >= st->filt_len;
#endif
@ -733,16 +733,18 @@ static int update_filter(SpeexResamplerState *st)
{
spx_uint32_t j;
spx_uint32_t olen = old_length;
spx_uint32_t start = i*st->mem_alloc_size;
spx_uint32_t magic_samples = st->magic_samples[i];
/*if (st->magic_samples[i])*/
{
/* Try and remove the magic samples as if nothing had happened */
/* FIXME: This is wrong but for now we need it to avoid going over the array bounds */
olen = old_length + 2*st->magic_samples[i];
for (j=old_length-1+st->magic_samples[i];j--;)
st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]] = st->mem[i*old_alloc_size+j];
for (j=0;j<st->magic_samples[i];j++)
st->mem[i*st->mem_alloc_size+j] = 0;
olen = old_length + 2*magic_samples;
for (j=old_length-1+magic_samples;j--;)
st->mem[start+j+magic_samples] = st->mem[i*old_alloc_size+j];
for (j=0;j<magic_samples;j++)
st->mem[start+j] = 0;
st->magic_samples[i] = 0;
}
if (st->filt_len > olen)
@ -750,17 +752,18 @@ static int update_filter(SpeexResamplerState *st)
/* If the new filter length is still bigger than the "augmented" length */
/* Copy data going backward */
for (j=0;j<olen-1;j++)
st->mem[i*st->mem_alloc_size+(st->filt_len-2-j)] = st->mem[i*st->mem_alloc_size+(olen-2-j)];
st->mem[start+(st->filt_len-2-j)] = st->mem[start+(olen-2-j)];
/* Then put zeros for lack of anything better */
for (;j<st->filt_len-1;j++)
st->mem[i*st->mem_alloc_size+(st->filt_len-2-j)] = 0;
st->mem[start+(st->filt_len-2-j)] = 0;
/* Adjust last_sample */
st->last_sample[i] += (st->filt_len - olen)/2;
} else {
/* Put back some of the magic! */
st->magic_samples[i] = (olen - st->filt_len)/2;
for (j=0;j<st->filt_len-1+st->magic_samples[i];j++)
st->mem[i*st->mem_alloc_size+j] = st->mem[i*st->mem_alloc_size+j+st->magic_samples[i]];
magic_samples = (olen - st->filt_len)/2;
for (j=0;j<st->filt_len-1+magic_samples;j++)
st->mem[start+j] = st->mem[start+j+magic_samples];
st->magic_samples[i] = magic_samples;
}
}
} else if (st->filt_len < old_length)
@ -977,8 +980,7 @@ EXPORT int speex_resampler_process_int(SpeexResamplerState *st, spx_uint32_t cha
const spx_uint32_t xlen = st->mem_alloc_size - (st->filt_len - 1);
#ifdef VAR_ARRAYS
const unsigned int ylen = (olen < FIXED_STACK_ALLOC) ? olen : FIXED_STACK_ALLOC;
VARDECL(spx_word16_t *ystack);
ALLOC(ystack, ylen, spx_word16_t);
spx_word16_t ystack[ylen];
#else
const unsigned int ylen = FIXED_STACK_ALLOC;
spx_word16_t ystack[FIXED_STACK_ALLOC];
@ -1093,7 +1095,7 @@ EXPORT void speex_resampler_get_rate(SpeexResamplerState *st, spx_uint32_t *in_r
*out_rate = st->out_rate;
}
static inline spx_uint32_t _gcd(spx_uint32_t a, spx_uint32_t b)
static inline spx_uint32_t compute_gcd(spx_uint32_t a, spx_uint32_t b)
{
while (b != 0)
{
@ -1123,7 +1125,7 @@ EXPORT int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t r
st->num_rate = ratio_num;
st->den_rate = ratio_den;
fact = _gcd (st->num_rate, st->den_rate);
fact = compute_gcd(st->num_rate, st->den_rate);
st->num_rate /= fact;
st->den_rate /= fact;
@ -1132,7 +1134,7 @@ EXPORT int speex_resampler_set_rate_frac(SpeexResamplerState *st, spx_uint32_t r
{
for (i=0;i<st->nb_channels;i++)
{
if (_muldiv(&st->samp_frac_num[i],st->samp_frac_num[i],st->den_rate,old_den) != RESAMPLER_ERR_SUCCESS)
if (multiply_frac(&st->samp_frac_num[i],st->samp_frac_num[i],st->den_rate,old_den) != RESAMPLER_ERR_SUCCESS)
return RESAMPLER_ERR_OVERFLOW;
/* Safety net */
if (st->samp_frac_num[i] >= st->den_rate)

View File

@ -36,14 +36,26 @@
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#include <arm_neon.h>
#include <stdint.h>
#ifdef FIXED_POINT
#ifdef __thumb2__
#if defined(__aarch64__)
static inline int32_t saturate_32bit_to_16bit(int32_t a) {
int32_t ret;
asm ("fmov s0, %w[a]\n"
"sqxtn h0, s0\n"
"sxtl v0.4s, v0.4h\n"
"fmov %w[ret], s0\n"
: [ret] "=r" (ret)
: [a] "r" (a)
: "v0" );
return ret;
}
#elif defined(__thumb2__)
static inline int32_t saturate_32bit_to_16bit(int32_t a) {
int32_t ret;
asm ("ssat %[ret], #16, %[a]"
: [ret] "=&r" (ret)
: [ret] "=r" (ret)
: [a] "r" (a)
: );
return ret;
@ -54,7 +66,7 @@ static inline int32_t saturate_32bit_to_16bit(int32_t a) {
asm ("vmov.s32 d0[0], %[a]\n"
"vqmovn.s32 d0, q0\n"
"vmov.s16 %[ret], d0[0]\n"
: [ret] "=&r" (ret)
: [ret] "=r" (ret)
: [a] "r" (a)
: "q0");
return ret;
@ -64,7 +76,63 @@ static inline int32_t saturate_32bit_to_16bit(int32_t a) {
#define WORD2INT(x) (saturate_32bit_to_16bit(x))
#define OVERRIDE_INNER_PRODUCT_SINGLE
/* Only works when len % 4 == 0 */
/* Only works when len % 4 == 0 and len >= 4 */
#if defined(__aarch64__)
static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
{
int32_t ret;
uint32_t remainder = len % 16;
len = len - remainder;
asm volatile (" cmp %w[len], #0\n"
" b.ne 1f\n"
" ld1 {v16.4h}, [%[b]], #8\n"
" ld1 {v20.4h}, [%[a]], #8\n"
" subs %w[remainder], %w[remainder], #4\n"
" smull v0.4s, v16.4h, v20.4h\n"
" b.ne 4f\n"
" b 5f\n"
"1:"
" ld1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%[b]], #32\n"
" ld1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%[a]], #32\n"
" subs %w[len], %w[len], #16\n"
" smull v0.4s, v16.4h, v20.4h\n"
" smlal v0.4s, v17.4h, v21.4h\n"
" smlal v0.4s, v18.4h, v22.4h\n"
" smlal v0.4s, v19.4h, v23.4h\n"
" b.eq 3f\n"
"2:"
" ld1 {v16.4h, v17.4h, v18.4h, v19.4h}, [%[b]], #32\n"
" ld1 {v20.4h, v21.4h, v22.4h, v23.4h}, [%[a]], #32\n"
" subs %w[len], %w[len], #16\n"
" smlal v0.4s, v16.4h, v20.4h\n"
" smlal v0.4s, v17.4h, v21.4h\n"
" smlal v0.4s, v18.4h, v22.4h\n"
" smlal v0.4s, v19.4h, v23.4h\n"
" b.ne 2b\n"
"3:"
" cmp %w[remainder], #0\n"
" b.eq 5f\n"
"4:"
" ld1 {v18.4h}, [%[b]], #8\n"
" ld1 {v22.4h}, [%[a]], #8\n"
" subs %w[remainder], %w[remainder], #4\n"
" smlal v0.4s, v18.4h, v22.4h\n"
" b.ne 4b\n"
"5:"
" saddlv d0, v0.4s\n"
" sqxtn s0, d0\n"
" sqrshrn h0, s0, #15\n"
" sxtl v0.4s, v0.4h\n"
" fmov %w[ret], s0\n"
: [ret] "=r" (ret), [a] "+r" (a), [b] "+r" (b),
[len] "+r" (len), [remainder] "+r" (remainder)
:
: "cc", "v0",
"v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23");
return ret;
}
#else
static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, unsigned int len)
{
int32_t ret;
@ -112,33 +180,104 @@ static inline int32_t inner_product_single(const int16_t *a, const int16_t *b, u
" vqmovn.s64 d0, q0\n"
" vqrshrn.s32 d0, q0, #15\n"
" vmov.s16 %[ret], d0[0]\n"
: [ret] "=&r" (ret), [a] "+r" (a), [b] "+r" (b),
: [ret] "=r" (ret), [a] "+r" (a), [b] "+r" (b),
[len] "+r" (len), [remainder] "+r" (remainder)
:
: "cc", "q0",
"d16", "d17", "d18", "d19",
"d20", "d21", "d22", "d23");
"d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23");
return ret;
}
#elif defined(FLOATING_POINT)
#endif // !defined(__aarch64__)
#elif defined(FLOATING_POINT)
#if defined(__aarch64__)
static inline int32_t saturate_float_to_16bit(float a) {
int32_t ret;
asm ("fcvtas s1, %s[a]\n"
"sqxtn h1, s1\n"
"sxtl v1.4s, v1.4h\n"
"fmov %w[ret], s1\n"
: [ret] "=r" (ret)
: [a] "w" (a)
: "v1");
return ret;
}
#else
static inline int32_t saturate_float_to_16bit(float a) {
int32_t ret;
asm ("vmov.f32 d0[0], %[a]\n"
"vcvt.s32.f32 d0, d0, #15\n"
"vqrshrn.s32 d0, q0, #15\n"
"vmov.s16 %[ret], d0[0]\n"
: [ret] "=&r" (ret)
: [ret] "=r" (ret)
: [a] "r" (a)
: "q0");
return ret;
}
#endif
#undef WORD2INT
#define WORD2INT(x) (saturate_float_to_16bit(x))
#define OVERRIDE_INNER_PRODUCT_SINGLE
/* Only works when len % 4 == 0 */
/* Only works when len % 4 == 0 and len >= 4 */
#if defined(__aarch64__)
static inline float inner_product_single(const float *a, const float *b, unsigned int len)
{
float ret;
uint32_t remainder = len % 16;
len = len - remainder;
asm volatile (" cmp %w[len], #0\n"
" b.ne 1f\n"
" ld1 {v16.4s}, [%[b]], #16\n"
" ld1 {v20.4s}, [%[a]], #16\n"
" subs %w[remainder], %w[remainder], #4\n"
" fmul v1.4s, v16.4s, v20.4s\n"
" b.ne 4f\n"
" b 5f\n"
"1:"
" ld1 {v16.4s, v17.4s, v18.4s, v19.4s}, [%[b]], #64\n"
" ld1 {v20.4s, v21.4s, v22.4s, v23.4s}, [%[a]], #64\n"
" subs %w[len], %w[len], #16\n"
" fmul v1.4s, v16.4s, v20.4s\n"
" fmul v2.4s, v17.4s, v21.4s\n"
" fmul v3.4s, v18.4s, v22.4s\n"
" fmul v4.4s, v19.4s, v23.4s\n"
" b.eq 3f\n"
"2:"
" ld1 {v16.4s, v17.4s, v18.4s, v19.4s}, [%[b]], #64\n"
" ld1 {v20.4s, v21.4s, v22.4s, v23.4s}, [%[a]], #64\n"
" subs %w[len], %w[len], #16\n"
" fmla v1.4s, v16.4s, v20.4s\n"
" fmla v2.4s, v17.4s, v21.4s\n"
" fmla v3.4s, v18.4s, v22.4s\n"
" fmla v4.4s, v19.4s, v23.4s\n"
" b.ne 2b\n"
"3:"
" fadd v16.4s, v1.4s, v2.4s\n"
" fadd v17.4s, v3.4s, v4.4s\n"
" cmp %w[remainder], #0\n"
" fadd v1.4s, v16.4s, v17.4s\n"
" b.eq 5f\n"
"4:"
" ld1 {v18.4s}, [%[b]], #16\n"
" ld1 {v22.4s}, [%[a]], #16\n"
" subs %w[remainder], %w[remainder], #4\n"
" fmla v1.4s, v18.4s, v22.4s\n"
" b.ne 4b\n"
"5:"
" faddp v1.4s, v1.4s, v1.4s\n"
" faddp %[ret].4s, v1.4s, v1.4s\n"
: [ret] "=w" (ret), [a] "+r" (a), [b] "+r" (b),
[len] "+r" (len), [remainder] "+r" (remainder)
:
: "cc", "v1", "v2", "v3", "v4",
"v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23");
return ret;
}
#else
static inline float inner_product_single(const float *a, const float *b, unsigned int len)
{
float ret;
@ -191,11 +330,12 @@ static inline float inner_product_single(const float *a, const float *b, unsigne
" vadd.f32 d0, d0, d1\n"
" vpadd.f32 d0, d0, d0\n"
" vmov.f32 %[ret], d0[0]\n"
: [ret] "=&r" (ret), [a] "+r" (a), [b] "+r" (b),
: [ret] "=r" (ret), [a] "+r" (a), [b] "+r" (b),
[len] "+l" (len), [remainder] "+l" (remainder)
:
: "cc", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7", "q8",
"q9", "q10", "q11");
: "cc", "q0", "q1", "q2", "q3",
"q4", "q5", "q6", "q7", "q8", "q9", "q10", "q11");
return ret;
}
#endif // defined(__aarch64__)
#endif

View File

@ -71,7 +71,7 @@ static inline float interpolate_product_single(const float *a, const float *b, u
return ret;
}
#ifdef _USE_SSE2
#ifdef USE_SSE2
#include <emmintrin.h>
#define OVERRIDE_INNER_PRODUCT_DOUBLE

View File

@ -288,9 +288,9 @@ std::vector<std::pair<std::string, std::string>> AudioStream::GetCubebDriverName
std::vector<std::pair<std::string, std::string>> names;
names.emplace_back(std::string(), TRANSLATE_STR("AudioStream", "Default"));
const char** cubeb_names = cubeb_get_backend_names();
for (u32 i = 0; cubeb_names[i] != nullptr; i++)
names.emplace_back(cubeb_names[i], cubeb_names[i]);
auto cubeb_names = cubeb_get_backend_names();
for (int i = 0; i < cubeb_names.count; i++)
names.emplace_back(cubeb_names.names[i], cubeb_names.names[i]);
return names;
}