Fixed submix buffering when input and output stream formats differ.
When channel conversion is enabled and the pipe is opened with less channels
(e.g 1) than the output stream's channels (e.g 2) the input stream's buffer
size was calculated incorrectly which resulted in the input stream buffer size
being larger than the output stream buffer size in terms of time. This
changes the pipe size to take into the account maximum frame size so the
output stream buffer size will be greater or equal to the input stream buffer
size.
In addition, if the pipe is created when the input stream is opened and the
output stream is opened at a different sample rate (e.g input = 8KHz,
output = 48Khz) the pipe will incorrectly rate limit writes to the pipe
resulting in gaps of silence in the data read from the pipe by the input
stream.
Bug: 15291446
Change-Id: I01480edff7ef3a199f509866c1e95ecd4c437a92
diff --git a/modules/audio_remote_submix/audio_hw.cpp b/modules/audio_remote_submix/audio_hw.cpp
index f11b207..51a5a29 100644
--- a/modules/audio_remote_submix/audio_hw.cpp
+++ b/modules/audio_remote_submix/audio_hw.cpp
@@ -388,8 +388,17 @@
// If a pipe isn't associated with the device, create one.
if (rsxadev->rsxSink == NULL || rsxadev->rsxSource == NULL) {
struct submix_config * const device_config = &rsxadev->config;
- const NBAIO_Format format = Format_from_SR_C(config->sample_rate,
- get_channel_count_from_mask(config->channel_mask), config->format);
+ const uint32_t channel_count = get_channel_count_from_mask(config->channel_mask);
+#if ENABLE_CHANNEL_CONVERSION
+ // If channel conversion is enabled, allocate enough space for the maximum number of
+ // possible channels stored in the pipe for the situation when the number of channels in
+ // the output stream don't match the number in the input stream.
+ const uint32_t pipe_channel_count = max(channel_count, 2);
+#else
+ const uint32_t pipe_channel_count = channel_count;
+#endif // ENABLE_CHANNEL_CONVERSION
+ const NBAIO_Format format = Format_from_SR_C(config->sample_rate, pipe_channel_count,
+ config->format);
const NBAIO_Format offers[1] = {format};
size_t numCounterOffers = 0;
// Create a MonoPipe with optional blocking set to true.
@@ -417,6 +426,11 @@
buffer_period_count;
if (in) device_config->pipe_frame_size = audio_stream_frame_size(&in->stream.common);
if (out) device_config->pipe_frame_size = audio_stream_frame_size(&out->stream.common);
+#if ENABLE_CHANNEL_CONVERSION
+ // Calculate the pipe frame size based upon the number of channels.
+ device_config->pipe_frame_size = (device_config->pipe_frame_size * pipe_channel_count) /
+ channel_count;
+#endif // ENABLE_CHANNEL_CONVERSION
SUBMIX_ALOGV("submix_audio_device_create_pipe(): pipe frame size %zd, pipe size %zd, "
"period size %zd", device_config->pipe_frame_size,
device_config->buffer_size_frames, device_config->buffer_period_size_frames);
@@ -669,15 +683,10 @@
const struct submix_config * const config = &out->dev->config;
const size_t buffer_size_frames = calculate_stream_pipe_size_in_frames(
&stream->common, config, config->buffer_size_frames);
-#if ENABLE_RESAMPLING
- // Sample rate conversion occurs when data is read from the input so data in the buffer is
- // at output_sample_rate Hz.
- const uint32_t latency_ms = (buffer_size_frames * 1000) / config->output_sample_rate;
-#else
- const uint32_t latency_ms = (buffer_size_frames * 1000) / config->common.sample_rate;
-#endif // ENABLE_RESAMPLING
+ const uint32_t sample_rate = out_get_sample_rate(&stream->common);
+ const uint32_t latency_ms = (buffer_size_frames * 1000) / sample_rate;
SUBMIX_ALOGV("out_get_latency() returns %u ms, size in frames %zu, sample rate %u",
- latency_ms, buffer_size_frames, config->common.sample_rate);
+ latency_ms, buffer_size_frames, sample_rate);
return latency_ms;
}
@@ -1114,7 +1123,7 @@
if (remaining_frames > 0) {
const size_t remaining_bytes = remaining_frames * frame_size;
- SUBMIX_ALOGV(" remaining_frames = %zu", remaining_frames);
+ SUBMIX_ALOGV(" clearing remaining_frames = %zu", remaining_frames);
memset(((char*)buffer)+ bytes - remaining_bytes, 0, remaining_bytes);
}
@@ -1186,6 +1195,7 @@
struct submix_audio_device * const rsxadev = audio_hw_device_get_submix_audio_device(dev);
ALOGV("adev_open_output_stream()");
struct submix_stream_out *out;
+ bool force_pipe_creation = false;
(void)handle;
(void)devices;
(void)flags;
@@ -1221,9 +1231,16 @@
out->stream.get_render_position = out_get_render_position;
out->stream.get_next_write_timestamp = out_get_next_write_timestamp;
- // If the sink has been shutdown, delete the pipe so that it's recreated.
+#if ENABLE_RESAMPLING
+ // Recreate the pipe with the correct sample rate so that MonoPipe.write() rate limits
+ // writes correctly.
+ force_pipe_creation = rsxadev->config.common.sample_rate != config->sample_rate;
+#endif // ENABLE_RESAMPLING
+
+ // If the sink has been shutdown or pipe recreation is forced (see above), delete the pipe so
+ // that it's recreated.
pthread_mutex_lock(&rsxadev->lock);
- if (rsxadev->rsxSink != NULL && rsxadev->rsxSink->isShutdown()) {
+ if ((rsxadev->rsxSink != NULL && rsxadev->rsxSink->isShutdown()) || force_pipe_creation) {
submix_audio_device_release_pipe(rsxadev);
}
pthread_mutex_unlock(&rsxadev->lock);
@@ -1347,7 +1364,7 @@
const size_t frame_size_in_bytes = get_channel_count_from_mask(config->channel_mask) *
audio_bytes_per_sample(config->format);
const size_t buffer_size = buffer_period_size_frames * frame_size_in_bytes;
- SUBMIX_ALOGV("out_get_buffer_size() returns %zu bytes, %zu frames",
+ SUBMIX_ALOGV("adev_get_input_buffer_size() returns %zu bytes, %zu frames",
buffer_size, buffer_period_size_frames);
return buffer_size;
}