This is the final version of my patch for module-loopback. It is on top of the patch I sent about an hour ago and contains a lot more changes than the previous versions: - Honor specified latency if possible, if not adjust to the lowest possible value - Smooth switching from fixed latency to dynamic latency source or sink and vice versa - good rate and latency stability, no rate oscillation - adjusts latency as good as your setup allows - fast regulation of latency offsets, adjusts 100 ms offset within 22 seconds (adjust time=1) to 60 seconds (adjust_time=10) - usable latency range 4 - 30000 ms - Avoid rewinds and "cannot peek into queue" messages during startup and switching - works with rates between 200 and 190000 Hz - maximum latency offset after source/sink switch or at startup around is 200 ms I also introduced a new parameter, buffer_latency_msec which can be used together with latency_msec. If buffer_latency_msec is specified, the resulting latency will be latency_msec + buffer_latency_msec. Latency_msec then refers only to the source/sink latency while buffer_latency_msec specifies the buffer part. This can be used to save a lot of CPU at low latencies, running 10 ms latency with latency_msec=6 buffer_latency_msec=4 gives 8% CPU on my system compared to 12% when I only specify latency_msec=10. Additionally you can go beyond the safe-guard limits that are built in, you can access the range 1 - 3 ms or lower the buffer latency for fixed latency devices. Some of my USB devices run fine at a buffer latency of fragment size + 4 ms instead of the dfault fragment size + 20 ms. I tested it all with Intel HDA, USB and bluetooth sound devices. I would like to see some test results from other people. --- src/modules/module-loopback.c | 539 +++++++++++++++++++++++++++++++----------- 1 file changed, 397 insertions(+), 142 deletions(-) diff --git a/src/modules/module-loopback.c b/src/modules/module-loopback.c index 7e2b92a..8def03e 100644 --- a/src/modules/module-loopback.c +++ b/src/modules/module-loopback.c @@ -47,6 +47,7 @@ PA_MODULE_USAGE( "sink=<sink to connect to> " "adjust_time=<how often to readjust rates in s> " "latency_msec=<latency in ms> " + "buffer_latency_msec=<buffer latency in ms> " "format=<sample format> " "rate=<sample rate> " "channels=<number of channels> " @@ -59,7 +60,9 @@ PA_MODULE_USAGE( #define DEFAULT_LATENCY_MSEC 200 -#define MEMBLOCKQ_MAXLENGTH (1024*1024*16) +#define MEMBLOCKQ_MAXLENGTH (1024*1024*32) + +#define DEFAULT_BUFFER_MARGIN_MSEC 20 #define DEFAULT_ADJUST_TIME_USEC (10*PA_USEC_PER_SEC) @@ -80,12 +83,26 @@ struct userdata { int64_t recv_counter; int64_t send_counter; + uint32_t sink_adjust_counter; + uint32_t source_adjust_counter; + uint32_t underruns; + uint32_t no_peeks; - size_t skip; pa_usec_t latency; + pa_usec_t buffer_latency; + pa_usec_t initial_buffer_latency; + pa_usec_t configured_sink_latency; + pa_usec_t configured_source_latency; + + pa_usec_t source_latency_sum; + pa_usec_t sink_latency_sum; + pa_usec_t next_latency; + double latency_error; bool in_pop; - size_t min_memblockq_length; + bool pop_called; + bool source_sink_changed; + bool buffer_latency_set; struct { int64_t send_counter; @@ -95,9 +112,6 @@ struct userdata { int64_t recv_counter; size_t sink_input_buffer; pa_usec_t sink_latency; - - size_t min_memblockq_length; - size_t max_request; } latency_snapshot; }; @@ -106,6 +120,7 @@ static const char* const valid_modargs[] = { "sink", "adjust_time", "latency_msec", + "buffer_latency_msec", "format", "rate", "channels", @@ -121,8 +136,7 @@ static const char* const valid_modargs[] = { enum { SINK_INPUT_MESSAGE_POST = PA_SINK_INPUT_MESSAGE_MAX, SINK_INPUT_MESSAGE_REWIND, - SINK_INPUT_MESSAGE_LATENCY_SNAPSHOT, - SINK_INPUT_MESSAGE_MAX_REQUEST_CHANGED + SINK_INPUT_MESSAGE_LATENCY_SNAPSHOT }; enum { @@ -168,84 +182,134 @@ static void teardown(struct userdata *u) { } /* Called from main context */ -static void adjust_rates(struct userdata *u) { - size_t buffer, fs; - uint32_t old_rate, base_rate, new_rate; - pa_usec_t buffer_latency; +static void adjust_rates(struct userdata *u, pa_usec_t snapshot_delay) { + size_t buffer; + uint32_t old_rate, base_rate, new_rate, hours, cut_off_frequency; + pa_usec_t final_latency, source_sink_latency, current_buffer_latency, current_latency, corrected_latency; + double min_cycles; + int32_t latency_difference; pa_assert(u); pa_assert_ctl_context(); - pa_asyncmsgq_send(u->source_output->source->asyncmsgq, PA_MSGOBJECT(u->source_output), SOURCE_OUTPUT_MESSAGE_LATENCY_SNAPSHOT, NULL, 0, NULL); - pa_asyncmsgq_send(u->sink_input->sink->asyncmsgq, PA_MSGOBJECT(u->sink_input), SINK_INPUT_MESSAGE_LATENCY_SNAPSHOT, NULL, 0, NULL); + /* Runtime and counters since last change of source or sink */ + hours = PA_MIN(u->sink_adjust_counter, u->source_adjust_counter) * u->adjust_time / PA_USEC_PER_SEC / 3600; + u->sink_adjust_counter +=1; + u->source_adjust_counter +=1; + + /* Latency sums */ + u->source_latency_sum += u->latency_snapshot.source_latency; + u->sink_latency_sum += u->latency_snapshot.sink_latency; + + /* If we are seeing underruns or cannot peek into the queue during pop then the latency is too small */ + if (u->underruns > 2 || u->no_peeks > 30) { + u->buffer_latency += 5 * PA_USEC_PER_MSEC; + pa_log_warn("Too many underruns, increasing buffer latency to %0.2f ms", (double) u->buffer_latency / PA_USEC_PER_MSEC); + u->underruns = 0; + u->no_peeks = 0; + } - buffer = - u->latency_snapshot.sink_input_buffer + - u->latency_snapshot.source_output_buffer; + /* Allow one underrun and 15 no peek messages per hour */ + if (PA_MIN(u->sink_adjust_counter, u->source_adjust_counter) * u->adjust_time / PA_USEC_PER_SEC / 3600 > hours) { + pa_log_info("Underrun counters: %u, %u", u->underruns, u->no_peeks); + u->underruns = PA_CLIP_SUB(u->underruns, 1u); + u->no_peeks = PA_CLIP_SUB(u->no_peeks, 15u); + } + + /* Rates and latencies*/ + old_rate = u->sink_input->sample_spec.rate; + base_rate = u->source_output->sample_spec.rate; + buffer = u->latency_snapshot.sink_input_buffer + u->latency_snapshot.source_output_buffer; if (u->latency_snapshot.recv_counter <= u->latency_snapshot.send_counter) buffer += (size_t) (u->latency_snapshot.send_counter - u->latency_snapshot.recv_counter); else - buffer += PA_CLIP_SUB(buffer, (size_t) (u->latency_snapshot.recv_counter - u->latency_snapshot.send_counter)); + buffer = PA_CLIP_SUB(buffer, (size_t) (u->latency_snapshot.recv_counter - u->latency_snapshot.send_counter)); + current_buffer_latency = pa_bytes_to_usec(buffer, &u->sink_input->sample_spec); - buffer_latency = pa_bytes_to_usec(buffer, &u->sink_input->sample_spec); + current_latency = u->latency_snapshot.sink_latency + current_buffer_latency + u->latency_snapshot.source_latency - snapshot_delay; - pa_log_debug("Loopback overall latency is %0.2f ms + %0.2f ms + %0.2f ms = %0.2f ms", + source_sink_latency = u->sink_latency_sum / u->sink_adjust_counter + + u->source_latency_sum / u->source_adjust_counter; + final_latency = u->latency; + if (u->buffer_latency_set) + final_latency += u->initial_buffer_latency; + final_latency = PA_MAX(final_latency, source_sink_latency + u->buffer_latency); + + pa_log_debug("Loopback overall latency is %0.2f ms + %0.2f ms + %0.2f ms = %0.2f ms, latency difference: %0.2f ms, rate difference: %i Hz", (double) u->latency_snapshot.sink_latency / PA_USEC_PER_MSEC, - (double) buffer_latency / PA_USEC_PER_MSEC, + (double) current_buffer_latency / PA_USEC_PER_MSEC, (double) u->latency_snapshot.source_latency / PA_USEC_PER_MSEC, - ((double) u->latency_snapshot.sink_latency + buffer_latency + u->latency_snapshot.source_latency) / PA_USEC_PER_MSEC); - - pa_log_debug("Should buffer %zu bytes, buffered at minimum %zu bytes", - u->latency_snapshot.max_request*2, - u->latency_snapshot.min_memblockq_length); - - fs = pa_frame_size(&u->sink_input->sample_spec); - old_rate = u->sink_input->sample_spec.rate; - base_rate = u->source_output->sample_spec.rate; - - if (u->latency_snapshot.min_memblockq_length < u->latency_snapshot.max_request*2) - new_rate = base_rate - (((u->latency_snapshot.max_request*2 - u->latency_snapshot.min_memblockq_length) / fs) *PA_USEC_PER_SEC)/u->adjust_time; - else - new_rate = base_rate + (((u->latency_snapshot.min_memblockq_length - u->latency_snapshot.max_request*2) / fs) *PA_USEC_PER_SEC)/u->adjust_time; - - if (new_rate < (uint32_t) (base_rate*0.8) || new_rate > (uint32_t) (base_rate*1.25)) { - pa_log_warn("Sample rates too different, not adjusting (%u vs. %u).", base_rate, new_rate); - new_rate = base_rate; - } else { - if (base_rate < new_rate + 20 && new_rate < base_rate + 20) - new_rate = base_rate; - /* Do the adjustment in small steps; 2â?° can be considered inaudible */ - if (new_rate < (uint32_t) (old_rate*0.998) || new_rate > (uint32_t) (old_rate*1.002)) { - pa_log_info("New rate of %u Hz not within 2â?° of %u Hz, forcing smaller adjustment", new_rate, old_rate); - new_rate = PA_CLAMP(new_rate, (uint32_t) (old_rate*0.998), (uint32_t) (old_rate*1.002)); - } + (double) current_latency / PA_USEC_PER_MSEC, + (double) (int32_t)(current_latency - final_latency) / PA_USEC_PER_MSEC, + (int32_t)(old_rate - base_rate)); + + /* Low pass filtered difference between expectation value and observed latency */ + if (!u->source_sink_changed) + u->latency_error = (4 * u->latency_error + (double)abs((int32_t)(current_latency - u->next_latency)) / final_latency) / 5; + else + u->source_sink_changed = false; + + /* Latency and latency difference at base rate */ + corrected_latency = u->latency_snapshot.source_latency + (u->latency_snapshot.sink_latency + current_buffer_latency) * old_rate / base_rate - snapshot_delay; + latency_difference = (int32_t)(corrected_latency - final_latency); + + /* Minimum number of adjust times + 1 needed to adjust at 0.75% deviation from base rate */ + min_cycles = (double)abs(latency_difference) / u->adjust_time / 0.0075 + 1; + + /* Rate calculation, maximum deviation from base rate will be less than 0.75% due to min_cycles */ + new_rate = base_rate * (1.0 + latency_difference / min_cycles / u->adjust_time) + 0.5; + + /* Adjust as good as physics allows (with some safety margin) */ + if (abs(latency_difference) <= 2.5 * u->latency_error * final_latency + u->adjust_time / 2 / base_rate + 100) + new_rate = base_rate; + + /* Do the adjustment in small steps; 2â?° of base rate can be considered inaudible, use 1 Hz below 500 Hz base rate */ + cut_off_frequency = PA_MAX(1, 0.002 * base_rate); + if (new_rate < old_rate - cut_off_frequency || new_rate > old_rate + cut_off_frequency) { + pa_log_info("New rate of %u Hz not within 2â?° of %u Hz, forcing smaller adjustment", new_rate, old_rate); + new_rate = PA_CLAMP(new_rate, old_rate - cut_off_frequency, old_rate + cut_off_frequency); } + /* Predictor */ + u->next_latency = (corrected_latency * base_rate + (int32_t)(base_rate - new_rate) * (int64_t)u->adjust_time) / new_rate; + + /* Set rate */ pa_sink_input_set_rate(u->sink_input, new_rate); pa_log_debug("[%s] Updated sampling rate to %lu Hz.", u->sink_input->sink->name, (unsigned long) new_rate); - - pa_core_rttime_restart(u->core, u->time_event, pa_rtclock_now() + u->adjust_time); } /* Called from main context */ static void time_callback(pa_mainloop_api *a, pa_time_event *e, const struct timeval *t, void *userdata) { struct userdata *u = userdata; + pa_usec_t timestamp; pa_assert(u); pa_assert(a); pa_assert(u->time_event == e); - adjust_rates(u); + /* Restart timer right away */ + pa_core_rttime_restart(u->core, u->time_event, pa_rtclock_now() + u->adjust_time); + + /* Get sink and source latency snapshot. The resulting latency must be corrected by the delay between sink and + * source snapshot which can somtimes be quite large (>>10 ms) */ + pa_asyncmsgq_send(u->sink_input->sink->asyncmsgq, PA_MSGOBJECT(u->sink_input), SINK_INPUT_MESSAGE_LATENCY_SNAPSHOT, NULL, 0, NULL); + timestamp = pa_rtclock_now(); + pa_asyncmsgq_send(u->source_output->source->asyncmsgq, PA_MSGOBJECT(u->source_output), SOURCE_OUTPUT_MESSAGE_LATENCY_SNAPSHOT, NULL, 0, NULL); + + adjust_rates(u, pa_rtclock_now() - timestamp); } -/* Called from main context */ +/* Called from main context + * When source or sink changes, give it a third of a second to settle down, then call adjust_rates for the first time */ static void enable_adjust_timer(struct userdata *u, bool enable) { if (enable) { - if (u->time_event || u->adjust_time <= 0) + if (!u->adjust_time) return; + if (u->time_event) + u->core->mainloop->time_free(u->time_event); - u->time_event = pa_core_rttime_new(u->module->core, pa_rtclock_now() + u->adjust_time, time_callback, u); + u->time_event = pa_core_rttime_new(u->module->core, pa_rtclock_now() + 333 * PA_USEC_PER_MSEC, time_callback, u); } else { if (!u->time_event) return; @@ -263,29 +327,62 @@ static void update_adjust_timer(struct userdata *u) { enable_adjust_timer(u, true); } +static pa_usec_t get_requested_latency(struct userdata *u) { + pa_usec_t requested_latency; + + requested_latency = u->latency; + if(u->buffer_latency_set) + requested_latency += u->buffer_latency; + return PA_MAX(u->configured_sink_latency + u->buffer_latency, requested_latency); +} + +/* Called from all contexts */ +static void memblockq_adjust(struct userdata *u, int32_t offset, bool allow_push) { + size_t memblock_bytes, requested_buffer_bytes; + pa_usec_t requested_buffer_latency; + size_t buffer_offset; + pa_memchunk silence; + + requested_buffer_latency = get_requested_latency(u); + if (offset > 0) + requested_buffer_latency = PA_CLIP_SUB(requested_buffer_latency, (pa_usec_t)offset); + else + requested_buffer_latency = requested_buffer_latency - offset; + + requested_buffer_bytes = pa_usec_to_bytes(requested_buffer_latency, &u->sink_input->sample_spec); + memblock_bytes = pa_memblockq_get_length(u->memblockq); + + /* Drop audio from queue */ + if ((int32_t)(memblock_bytes - requested_buffer_bytes) > 0) { + buffer_offset = memblock_bytes - requested_buffer_bytes; + pa_log_info("Dropping %zd bytes from queue", buffer_offset); + pa_memblockq_drop(u->memblockq, buffer_offset); + } + /* Add silence to queue, will never happen from IO-thread */ + else if ((int32_t)(memblock_bytes - requested_buffer_bytes) < 0 && allow_push) { + requested_buffer_bytes = requested_buffer_bytes - memblock_bytes; + pa_log_info("Adding %zd bytes of silence to queue", requested_buffer_bytes); + pa_sink_input_get_silence(u->sink_input, &silence); + while (requested_buffer_bytes >= silence.length) { + pa_memblockq_push_align(u->memblockq, &silence); + requested_buffer_bytes -= silence.length; + } + if (requested_buffer_bytes > 0) { + silence.length = requested_buffer_bytes; + pa_memblockq_push_align(u->memblockq, &silence); + } + pa_memblock_unref(silence.memblock); + } +} + /* Called from input thread context */ static void source_output_push_cb(pa_source_output *o, const pa_memchunk *chunk) { struct userdata *u; - pa_memchunk copy; pa_source_output_assert_ref(o); pa_source_output_assert_io_context(o); pa_assert_se(u = o->userdata); - if (u->skip >= chunk->length) { - u->skip -= chunk->length; - return; - } - - if (u->skip > 0) { - copy = *chunk; - copy.index += u->skip; - copy.length -= u->skip; - u->skip = 0; - - chunk = © - } - pa_asyncmsgq_post(u->asyncmsgq, PA_MSGOBJECT(u->sink_input), SINK_INPUT_MESSAGE_POST, NULL, 0, chunk, NULL); u->send_counter += (int64_t) chunk->length; } @@ -324,6 +421,36 @@ static int source_output_process_msg_cb(pa_msgobject *obj, int code, void *data, return pa_source_output_process_msg(obj, code, data, offset, chunk); } +static void set_source_output_latency(struct userdata *u, pa_source *source) { + pa_usec_t min_latency, max_latency, buffer_msec, latency; + + /* Set lower limit of source latency to 2.333 ms, if buffer + * latency is specified, use latency_msec as source latency */ + latency = PA_MAX(u->latency / 3, 2.333 * PA_USEC_PER_MSEC); + if (u->buffer_latency_set) + latency = u->latency; + + if(source->flags & PA_SOURCE_DYNAMIC_LATENCY) { + pa_source_get_latency_range(source, &min_latency, &max_latency); + if (min_latency > latency && !u->buffer_latency_set) { + u->buffer_latency = PA_MAX(u->buffer_latency, (pa_usec_t)(min_latency * 0.75)); + pa_log_warn("Cannot set requested source latency, adjusting buffer to %0.2f ms", (double)u->buffer_latency / PA_USEC_PER_MSEC); + } + latency = PA_CLAMP(latency, min_latency, max_latency); + } + else { + latency = pa_source_get_latency(source); + if (latency == 0) + latency = pa_source_get_fixed_latency(source); + buffer_msec = u->core->default_fragment_size_msec + DEFAULT_BUFFER_MARGIN_MSEC; + if (!u->buffer_latency_set && u->buffer_latency < buffer_msec * PA_USEC_PER_MSEC) { + pa_log_warn("Fixed latency device, setting buffer latency to %zd.00 ms", buffer_msec); + u->buffer_latency = buffer_msec * PA_USEC_PER_MSEC; + } + } + u->configured_source_latency = pa_source_output_set_requested_latency(u->source_output, latency); +} + /* Called from output thread context */ static void source_output_attach_cb(pa_source_output *o) { struct userdata *u; @@ -350,24 +477,12 @@ static void source_output_detach_cb(pa_source_output *o) { pa_rtpoll_item_free(u->rtpoll_item_write); u->rtpoll_item_write = NULL; } -} - -/* Called from output thread context */ -static void source_output_state_change_cb(pa_source_output *o, pa_source_output_state_t state) { - struct userdata *u; - - pa_source_output_assert_ref(o); - pa_source_output_assert_io_context(o); - pa_assert_se(u = o->userdata); - - if (PA_SOURCE_OUTPUT_IS_LINKED(state) && o->thread_info.state == PA_SOURCE_OUTPUT_INIT) { - - u->skip = pa_usec_to_bytes(PA_CLIP_SUB(pa_source_get_latency_within_thread(o->source), - u->latency), - &o->sample_spec); - - pa_log_info("Skipping %lu bytes", (unsigned long) u->skip); - } + u->source_sink_changed = true; + u->source_latency_sum = 0; + u->source_adjust_counter = 0; + u->underruns = 0; + u->no_peeks = 0; + u->buffer_latency = u->initial_buffer_latency; } /* Called from main thread */ @@ -393,7 +508,12 @@ static bool source_output_may_move_to_cb(pa_source_output *o, pa_source *dest) { if (!u->sink_input || !u->sink_input->sink) return true; - return dest != u->sink_input->sink->monitor_source; + /* We may still be adjusting, so reset rate to default before moving the source */ + if (dest != u->sink_input->sink->monitor_source) { + pa_sink_input_set_rate(u->sink_input, u->source_output->sample_spec.rate); + return true; + } + return false; } /* Called from main thread */ @@ -401,6 +521,7 @@ static void source_output_moving_cb(pa_source_output *o, pa_source *dest) { pa_proplist *p; const char *n; struct userdata *u; + pa_usec_t sink_latency; if (!dest) return; @@ -418,6 +539,26 @@ static void source_output_moving_cb(pa_source_output *o, pa_source *dest) { pa_sink_input_update_proplist(u->sink_input, PA_UPDATE_REPLACE, p); pa_proplist_free(p); + /* Set latency and calculate necessary buffer length */ + set_source_output_latency(u, dest); + if (!u->buffer_latency_set) { + if (u->sink_input->sink->flags & PA_SINK_DYNAMIC_LATENCY) + u->buffer_latency = PA_MAX(u->buffer_latency, (pa_usec_t)(u->configured_sink_latency * 0.75)); + else + u->buffer_latency = PA_MAX(u->buffer_latency, (u->core->default_fragment_size_msec + DEFAULT_BUFFER_MARGIN_MSEC) * PA_USEC_PER_MSEC); + } + + pa_sink_input_get_latency(u->sink_input, &sink_latency); + if (u->send_counter > u->recv_counter) + sink_latency += pa_bytes_to_usec(u->send_counter - u->recv_counter, &u->sink_input->sample_spec); + if (dest->flags & PA_SOURCE_DYNAMIC_LATENCY) + sink_latency += pa_source_get_latency(dest); + else + sink_latency = PA_CLIP_SUB(sink_latency, pa_source_get_latency(dest)); + memblockq_adjust(u, sink_latency, true); + + u->latency_error = 400.0 / get_requested_latency(u); + if (pa_source_get_state(dest) == PA_SOURCE_SUSPENDED) pa_sink_input_cork(u->sink_input, true); else @@ -439,18 +580,29 @@ static void source_output_suspend_cb(pa_source_output *o, bool suspended) { update_adjust_timer(u); } -/* Called from output thread context */ -static void update_min_memblockq_length(struct userdata *u) { - size_t length; - - pa_assert(u); - pa_sink_input_assert_io_context(u->sink_input); +/* Called from input thread context */ +static void update_source_requested_latency_cb(pa_source_output *i) { + struct userdata *u; + pa_usec_t source_latency; - length = pa_memblockq_get_length(u->memblockq); + pa_source_output_assert_ref(i); + pa_source_output_assert_io_context(i); + pa_assert_se(u = i->userdata); - if (u->min_memblockq_length == (size_t) -1 || - length < u->min_memblockq_length) - u->min_memblockq_length = length; + /* Source latency may have changed */ + source_latency = pa_source_get_requested_latency_within_thread(u->source_output->source); + if (source_latency > u->configured_source_latency) { + pa_log_warn("Source latency increased to %0.2f ms", (double)source_latency / PA_USEC_PER_MSEC); + u->configured_source_latency = source_latency; + if (!u->buffer_latency_set && u->buffer_latency < source_latency * 0.75) + u->buffer_latency = source_latency * 0.75; + if (!u->source_sink_changed) { + u->source_adjust_counter = 0; + u->source_latency_sum = 0; + u->underruns = 0; + u->no_peeks = 0; + } + } } /* Called from output thread context */ @@ -462,21 +614,23 @@ static int sink_input_pop_cb(pa_sink_input *i, size_t nbytes, pa_memchunk *chunk pa_assert_se(u = i->userdata); pa_assert(chunk); + u->pop_called = true; u->in_pop = true; while (pa_asyncmsgq_process_one(u->asyncmsgq) > 0) ; u->in_pop = false; if (pa_memblockq_peek(u->memblockq, chunk) < 0) { - pa_log_info("Could not peek into queue"); + if (!u->source_sink_changed) { + u->no_peeks +=1; + pa_log_info("Could not peek into queue"); + } return -1; } chunk->length = PA_MIN(chunk->length, nbytes); pa_memblockq_drop(u->memblockq, chunk->length); - update_min_memblockq_length(u); - return 0; } @@ -513,12 +667,13 @@ static int sink_input_process_msg_cb(pa_msgobject *obj, int code, void *data, in pa_sink_input_assert_io_context(u->sink_input); - if (PA_SINK_IS_OPENED(u->sink_input->sink->thread_info.state)) - pa_memblockq_push_align(u->memblockq, chunk); - else - pa_memblockq_flush_write(u->memblockq, true); + pa_memblockq_push_align(u->memblockq, chunk); + u->recv_counter += (int64_t) chunk->length; - update_min_memblockq_length(u); + if (!PA_SINK_IS_OPENED(u->sink_input->sink->thread_info.state) || !u->pop_called) { + memblockq_adjust(u, (int32_t)(-u->configured_sink_latency / 4), false); + return 0; + } /* Is this the end of an underrun? Then let's start things * right-away */ @@ -527,67 +682,76 @@ static int sink_input_process_msg_cb(pa_msgobject *obj, int code, void *data, in pa_memblockq_is_readable(u->memblockq)) { pa_log_debug("Requesting rewind due to end of underrun."); + if (!u->source_sink_changed) + u->underruns +=1; pa_sink_input_request_rewind(u->sink_input, (size_t) (u->sink_input->thread_info.underrun_for == (size_t) -1 ? 0 : u->sink_input->thread_info.underrun_for), false, true, false); } - u->recv_counter += (int64_t) chunk->length; - return 0; case SINK_INPUT_MESSAGE_REWIND: pa_sink_input_assert_io_context(u->sink_input); - if (PA_SINK_IS_OPENED(u->sink_input->sink->thread_info.state)) + if (PA_SINK_IS_OPENED(u->sink_input->sink->thread_info.state) && u->pop_called) pa_memblockq_seek(u->memblockq, -offset, PA_SEEK_RELATIVE, true); else - pa_memblockq_flush_write(u->memblockq, true); + memblockq_adjust(u, (int32_t)(-u->configured_sink_latency / 4), false); u->recv_counter -= offset; - update_min_memblockq_length(u); - return 0; case SINK_INPUT_MESSAGE_LATENCY_SNAPSHOT: { size_t length; - update_min_memblockq_length(u); - length = pa_memblockq_get_length(u->sink_input->thread_info.render_memblockq); u->latency_snapshot.recv_counter = u->recv_counter; - u->latency_snapshot.sink_input_buffer = - pa_memblockq_get_length(u->memblockq) + - (u->sink_input->thread_info.resampler ? pa_resampler_request(u->sink_input->thread_info.resampler, length) : length); + u->latency_snapshot.sink_input_buffer = pa_memblockq_get_length(u->memblockq) + + (u->sink_input->thread_info.resampler ? pa_resampler_request(u->sink_input->thread_info.resampler, length) : length); u->latency_snapshot.sink_latency = pa_sink_get_latency_within_thread(u->sink_input->sink); - u->latency_snapshot.max_request = pa_sink_input_get_max_request(u->sink_input); - - u->latency_snapshot.min_memblockq_length = u->min_memblockq_length; - u->min_memblockq_length = (size_t) -1; - return 0; } - case SINK_INPUT_MESSAGE_MAX_REQUEST_CHANGED: { - /* This message is sent from the IO thread to the main - * thread! So don't be confused. All the user cases above - * are executed in thread context, but this one is not! */ - - pa_assert_ctl_context(); - - if (u->time_event) - adjust_rates(u); - return 0; - } } return pa_sink_input_process_msg(obj, code, data, offset, chunk); } +static void set_sink_input_latency(struct userdata *u, pa_sink *sink) { + pa_usec_t min_latency, max_latency, buffer_msec, latency; + + /* Set lower limit of sink latency to 2.333 ms, if buffer + * latency is specified, use latency_msec as sink latency */ + latency = PA_MAX(u->latency / 3, 2.333 * PA_USEC_PER_MSEC); + if (u->buffer_latency_set) + latency = u->latency; + + if(sink->flags & PA_SINK_DYNAMIC_LATENCY) { + pa_sink_get_latency_range(sink, &min_latency, &max_latency); + if (min_latency > latency && !u->buffer_latency_set) { + u->buffer_latency = PA_MAX(u->buffer_latency, (pa_usec_t)(min_latency * 0.75)); + pa_log_warn("Cannot set requested sink latency, adjusting buffer to %0.2f ms", (double)u->buffer_latency / PA_USEC_PER_MSEC); + } + latency = PA_CLAMP(latency, min_latency, max_latency); + } + else { + latency = pa_sink_get_latency(sink); + if (latency == 0) + latency = pa_sink_get_fixed_latency(sink); + buffer_msec = u->core->default_fragment_size_msec + DEFAULT_BUFFER_MARGIN_MSEC; + if (!u->buffer_latency_set && u->buffer_latency < buffer_msec * PA_USEC_PER_MSEC) { + pa_log_warn("Fixed latency device, setting buffer latency to %zd.00 ms", buffer_msec); + u->buffer_latency = buffer_msec * PA_USEC_PER_MSEC; + } + } + u->configured_sink_latency = pa_sink_input_set_requested_latency(u->sink_input, latency); +} + /* Called from output thread context */ static void sink_input_attach_cb(pa_sink_input *i) { struct userdata *u; @@ -603,8 +767,6 @@ static void sink_input_attach_cb(pa_sink_input *i) { pa_memblockq_set_prebuf(u->memblockq, pa_sink_input_get_max_request(i)*2); pa_memblockq_set_maxrewind(u->memblockq, pa_sink_input_get_max_rewind(i)); - - u->min_memblockq_length = (size_t) -1; } /* Called from output thread context */ @@ -619,6 +781,13 @@ static void sink_input_detach_cb(pa_sink_input *i) { pa_rtpoll_item_free(u->rtpoll_item_read); u->rtpoll_item_read = NULL; } + u->source_sink_changed = true; + u->pop_called = false; + u->sink_latency_sum = 0; + u->sink_adjust_counter = 0; + u->underruns = 0; + u->no_peeks = 0; + u->buffer_latency = u->initial_buffer_latency; } /* Called from output thread context */ @@ -642,7 +811,6 @@ static void sink_input_update_max_request_cb(pa_sink_input *i, size_t nbytes) { pa_memblockq_set_prebuf(u->memblockq, nbytes*2); pa_log_info("Max request changed"); - pa_asyncmsgq_post(pa_thread_mq_get()->outq, PA_MSGOBJECT(u->sink_input), SINK_INPUT_MESSAGE_MAX_REQUEST_CHANGED, NULL, 0, NULL, NULL); } /* Called from main thread */ @@ -673,6 +841,7 @@ static void sink_input_moving_cb(pa_sink_input *i, pa_sink *dest) { struct userdata *u; pa_proplist *p; const char *n; + pa_usec_t source_latency; if (!dest) return; @@ -690,6 +859,22 @@ static void sink_input_moving_cb(pa_sink_input *i, pa_sink *dest) { pa_source_output_update_proplist(u->source_output, PA_UPDATE_REPLACE, p); pa_proplist_free(p); + /* Set latency and calculate necessary buffer length */ + set_sink_input_latency(u, dest); + if (!u->buffer_latency_set) { + if (u->source_output->source->flags & PA_SOURCE_DYNAMIC_LATENCY) + u->buffer_latency = PA_MAX(u->buffer_latency, (pa_usec_t)(u->configured_source_latency * 0.75)); + else + u->buffer_latency = PA_MAX(u->buffer_latency, (u->core->default_fragment_size_msec + DEFAULT_BUFFER_MARGIN_MSEC) * PA_USEC_PER_MSEC); + } + + pa_source_output_get_latency(u->source_output, &source_latency); + if (u->send_counter > u->recv_counter) + source_latency += pa_bytes_to_usec(u->send_counter - u->recv_counter, &u->sink_input->sample_spec); + memblockq_adjust(u, source_latency, true); + + u->latency_error = 400.0 / get_requested_latency(u); + if (pa_sink_get_state(dest) == PA_SINK_SUSPENDED) pa_source_output_cork(u->source_output, true); else @@ -709,7 +894,37 @@ static bool sink_input_may_move_to_cb(pa_sink_input *i, pa_sink *dest) { if (!u->source_output || !u->source_output->source) return true; - return dest != u->source_output->source->monitor_of; + /* We may still be adjusting, so reset rate to default before moving the sink */ + if (dest != u->source_output->source->monitor_of) { + pa_sink_input_set_rate(u->sink_input, u->source_output->sample_spec.rate); + return true; + } + return false; +} + +/* Called from output thread context */ +static void update_sink_requested_latency_cb(pa_sink_input *i) { + struct userdata *u; + pa_usec_t sink_latency; + + pa_sink_input_assert_ref(i); + pa_sink_input_assert_io_context(i); + pa_assert_se(u = i->userdata); + + /* Sink latency may have changed */ + sink_latency = pa_sink_get_requested_latency_within_thread(u->sink_input->sink); + if (sink_latency > u->configured_sink_latency) { + pa_log_warn("Sink latency increased to %0.2f ms", (double)sink_latency / PA_USEC_PER_MSEC); + u->configured_sink_latency = sink_latency; + if (!u->buffer_latency_set && u->buffer_latency < sink_latency * 0.75) + u->buffer_latency = sink_latency * 0.75; + if (!u->source_sink_changed) { + u->sink_adjust_counter = 0; + u->sink_latency_sum = 0; + u->underruns = 0; + u->no_peeks = 0; + } + } } /* Called from main thread */ @@ -734,12 +949,13 @@ int pa__init(pa_module *m) { pa_source *source = NULL; pa_source_output_new_data source_output_data; bool source_dont_move; - uint32_t latency_msec; + uint32_t latency_msec, buffer_latency_msec; pa_sample_spec ss; pa_channel_map map; bool format_set = false; bool rate_set = false; bool channels_set = false; + bool buffer_latency_set = false; pa_memchunk silence; uint32_t adjust_time_sec; const char *n; @@ -799,6 +1015,11 @@ int pa__init(pa_module *m) { goto fail; } + if (ss.rate < 20 || ss.rate > 190500) { + pa_log("Invalid rate specification, valid range is 20 Hz to 190500 Hz"); + goto fail; + } + if (pa_modargs_get_value(ma, "format", NULL)) format_set = true; @@ -808,8 +1029,18 @@ int pa__init(pa_module *m) { if (pa_modargs_get_value(ma, "channels", NULL) || pa_modargs_get_value(ma, "channel_map", NULL)) channels_set = true; - latency_msec = DEFAULT_LATENCY_MSEC; - if (pa_modargs_get_value_u32(ma, "latency_msec", &latency_msec) < 0 || latency_msec < 1 || latency_msec > 30000) { + buffer_latency_msec = 0; + if (pa_modargs_get_value_u32(ma, "buffer_latency_msec", &buffer_latency_msec) < 0 || buffer_latency_msec > 30000) { + pa_log_info("Invalid buffer latency specification"); + goto fail; + } + else if (buffer_latency_msec > 0) + buffer_latency_set = true; + + latency_msec = 0; + if (!buffer_latency_set) + latency_msec = DEFAULT_LATENCY_MSEC; + if (pa_modargs_get_value_u32(ma, "latency_msec", &latency_msec) < 0 || latency_msec > 30000) { pa_log("Invalid latency specification"); goto fail; } @@ -818,6 +1049,18 @@ int pa__init(pa_module *m) { u->core = m->core; u->module = m; u->latency = (pa_usec_t) latency_msec * PA_USEC_PER_MSEC; + if (buffer_latency_set) + u->initial_buffer_latency = (pa_usec_t) buffer_latency_msec * PA_USEC_PER_MSEC; + else + u->initial_buffer_latency = PA_MAX(u->latency / 4, 1.667 * PA_USEC_PER_MSEC); + u->buffer_latency = u->initial_buffer_latency; + u->buffer_latency_set = buffer_latency_set; + u->sink_latency_sum = 0; + u->source_latency_sum = 0; + u->pop_called = false; + u->underruns = 0; + u->no_peeks = 0; + u->source_sink_changed = true; adjust_time_sec = DEFAULT_ADJUST_TIME_USEC / PA_USEC_PER_SEC; if (pa_modargs_get_value_u32(ma, "adjust_time", &adjust_time_sec) < 0) { @@ -894,9 +1137,13 @@ int pa__init(pa_module *m) { u->sink_input->may_move_to = sink_input_may_move_to_cb; u->sink_input->moving = sink_input_moving_cb; u->sink_input->suspend = sink_input_suspend_cb; + u->sink_input->update_sink_requested_latency = update_sink_requested_latency_cb; u->sink_input->userdata = u; - pa_sink_input_set_requested_latency(u->sink_input, u->latency/3); + if (u->latency < 4 * PA_USEC_PER_MSEC && !buffer_latency_set) + pa_log_warn("Latency limited to 4 ms, try buffer_latency_msec together with latency_msec if you know what you are doing"); + + set_sink_input_latency(u, u->sink_input->sink); pa_source_output_new_data_init(&source_output_data); source_output_data.driver = __FILE__; @@ -941,13 +1188,13 @@ int pa__init(pa_module *m) { u->source_output->kill = source_output_kill_cb; u->source_output->attach = source_output_attach_cb; u->source_output->detach = source_output_detach_cb; - u->source_output->state_change = source_output_state_change_cb; u->source_output->may_move_to = source_output_may_move_to_cb; u->source_output->moving = source_output_moving_cb; u->source_output->suspend = source_output_suspend_cb; + u->source_output->update_source_requested_latency = update_source_requested_latency_cb; u->source_output->userdata = u; - pa_source_output_set_requested_latency(u->source_output, u->latency/3); + set_source_output_latency(u, u->source_output->source); pa_sink_input_get_silence(u->sink_input, &silence); u->memblockq = pa_memblockq_new( @@ -962,6 +1209,14 @@ int pa__init(pa_module *m) { &silence); /* silence frame */ pa_memblock_unref(silence.memblock); + if(u->sink_input->sink->flags & PA_SINK_DYNAMIC_LATENCY) + memblockq_adjust(u, 0, true); + else + memblockq_adjust(u, (int32_t)(-u->configured_sink_latency / 4), true); + + /* Initialize latency error (allow at least 1 ms latency offset) */ + u->latency_error = 400.0 / get_requested_latency(u); + u->asyncmsgq = pa_asyncmsgq_new(0); if (!pa_proplist_contains(u->source_output->proplist, PA_PROP_MEDIA_NAME)) -- 2.1.4