#include <pulsecore/log.h>
#include <pulsecore/play-memblockq.h>
#include <pulsecore/namereg.h>
+#include <pulsecore/core-util.h>
#include "sink-input.h"
+#define MEMBLOCKQ_MAXLENGTH (32*1024*1024)
#define CONVERT_BUFFER_LENGTH (PA_PAGE_SIZE)
-#define SILENCE_BUFFER_LENGTH (PA_PAGE_SIZE*12)
#define MOVE_BUFFER_LENGTH (PA_PAGE_SIZE*256)
static PA_DEFINE_CHECK_TYPE(pa_sink_input, pa_msgobject);
memset(data, 0, sizeof(*data));
data->resample_method = PA_RESAMPLER_INVALID;
+ data->proplist = pa_proplist_new();
return data;
}
+void pa_sink_input_new_data_set_sample_spec(pa_sink_input_new_data *data, const pa_sample_spec *spec) {
+ pa_assert(data);
+
+ if ((data->sample_spec_is_set = !!spec))
+ data->sample_spec = *spec;
+}
+
void pa_sink_input_new_data_set_channel_map(pa_sink_input_new_data *data, const pa_channel_map *map) {
pa_assert(data);
data->volume = *volume;
}
-void pa_sink_input_new_data_set_sample_spec(pa_sink_input_new_data *data, const pa_sample_spec *spec) {
+void pa_sink_input_new_data_set_muted(pa_sink_input_new_data *data, pa_bool_t mute) {
pa_assert(data);
- if ((data->sample_spec_is_set = !!spec))
- data->sample_spec = *spec;
+ data->muted_is_set = TRUE;
+ data->muted = !!mute;
}
-void pa_sink_input_new_data_set_muted(pa_sink_input_new_data *data, pa_bool_t mute) {
+void pa_sink_input_new_data_done(pa_sink_input_new_data *data) {
pa_assert(data);
- data->muted_is_set = TRUE;
- data->muted = !!mute;
+ pa_proplist_free(data->proplist);
+}
+
+static void reset_callbacks(pa_sink_input *i) {
+ pa_assert(i);
+
+ i->pop = NULL;
+ i->process_rewind = NULL;
+ i->update_max_rewind = NULL;
+ i->attach = NULL;
+ i->detach = NULL;
+ i->suspend = NULL;
+ i->moved = NULL;
+ i->kill = NULL;
+ i->get_latency = NULL;
+ i->state_change = NULL;
}
pa_sink_input* pa_sink_input_new(
return NULL;
pa_return_null_if_fail(!data->driver || pa_utf8_valid(data->driver));
- pa_return_null_if_fail(!data->name || pa_utf8_valid(data->name));
if (!data->sink)
data->sink = pa_namereg_get(core, NULL, PA_NAMEREG_SINK, 1);
pa_return_null_if_fail(pa_cvolume_valid(&data->volume));
pa_return_null_if_fail(data->volume.channels == data->sample_spec.channels);
+ if (!data->muted_is_set)
+ data->muted = FALSE;
+
if (flags & PA_SINK_INPUT_FIX_FORMAT)
data->sample_spec.format = data->sink->sample_spec.format;
if (data->volume.channels != data->sample_spec.channels)
pa_cvolume_set(&data->volume, data->sample_spec.channels, pa_cvolume_avg(&data->volume));
- if (!data->muted_is_set)
- data->muted = 0;
-
if (data->resample_method == PA_RESAMPLER_INVALID)
data->resample_method = core->resample_method;
i->core = core;
i->state = PA_SINK_INPUT_INIT;
i->flags = flags;
- i->name = pa_xstrdup(data->name);
+ i->proplist = pa_proplist_copy(data->proplist);
i->driver = pa_xstrdup(data->driver);
i->module = data->module;
i->sink = data->sink;
} else
i->sync_next = i->sync_prev = NULL;
- i->peek = NULL;
- i->drop = NULL;
- i->kill = NULL;
- i->get_latency = NULL;
- i->attach = NULL;
- i->detach = NULL;
- i->suspend = NULL;
- i->moved = NULL;
+ reset_callbacks(i);
i->userdata = NULL;
i->thread_info.state = i->state;
+ i->thread_info.attached = FALSE;
pa_atomic_store(&i->thread_info.drained, 1);
i->thread_info.sample_spec = i->sample_spec;
- i->thread_info.silence_memblock = NULL;
- i->thread_info.move_silence = 0;
- pa_memchunk_reset(&i->thread_info.resampled_chunk);
i->thread_info.resampler = resampler;
i->thread_info.volume = i->volume;
i->thread_info.muted = i->muted;
- i->thread_info.attached = FALSE;
+ i->thread_info.requested_sink_latency = (pa_usec_t) -1;
+ i->thread_info.rewrite_nbytes = 0;
+ i->thread_info.rewrite_flush = FALSE;
+ i->thread_info.underrun_for = (uint64_t) -1;
+ i->thread_info.playing_for = 0;
+
+ i->thread_info.render_memblockq = pa_memblockq_new(
+ 0,
+ MEMBLOCKQ_MAXLENGTH,
+ 0,
+ pa_frame_size(&i->sink->sample_spec),
+ 0,
+ 1,
+ 0,
+ &i->sink->silence);
pa_assert_se(pa_idxset_put(core->sink_inputs, pa_sink_input_ref(i), &i->index) == 0);
pa_assert_se(pa_idxset_put(i->sink->inputs, i, NULL) == 0);
pa_log_info("Created input %u \"%s\" on %s with sample spec %s and channel map %s",
i->index,
- i->name,
+ pa_strnull(pa_proplist_gets(i->proplist, PA_PROP_MEDIA_NAME)),
i->sink->name,
pa_sample_spec_snprint(st, sizeof(st), &i->sample_spec),
pa_channel_map_snprint(cm, sizeof(cm), &i->channel_map));
pa_sink_input_ref(i);
- linked = PA_SINK_INPUT_LINKED(i->state);
+ linked = PA_SINK_INPUT_IS_LINKED(i->state);
if (linked)
pa_hook_fire(&i->sink->core->hooks[PA_CORE_HOOK_SINK_INPUT_UNLINK], i);
if (pa_idxset_remove_by_data(i->sink->inputs, i, NULL))
pa_sink_input_unref(i);
- if (linked) {
- pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i->sink), PA_SINK_MESSAGE_REMOVE_INPUT, i, 0, NULL);
- sink_input_set_state(i, PA_SINK_INPUT_UNLINKED);
- pa_sink_update_status(i->sink);
- } else
- i->state = PA_SINK_INPUT_UNLINKED;
+ update_n_corked(i, PA_SINK_INPUT_UNLINKED);
+ i->state = PA_SINK_INPUT_UNLINKED;
- i->peek = NULL;
- i->drop = NULL;
- i->kill = NULL;
- i->get_latency = NULL;
- i->attach = NULL;
- i->detach = NULL;
- i->suspend = NULL;
- i->moved = NULL;
+ if (linked)
+ if (i->sink->asyncmsgq)
+ pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i->sink), PA_SINK_MESSAGE_REMOVE_INPUT, i, 0, NULL);
+
+ reset_callbacks(i);
if (linked) {
pa_subscription_post(i->sink->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_REMOVE, i->index);
pa_assert(i);
pa_assert(pa_sink_input_refcnt(i) == 0);
- if (PA_SINK_INPUT_LINKED(i->state))
+ if (PA_SINK_INPUT_IS_LINKED(i->state))
pa_sink_input_unlink(i);
- pa_log_info("Freeing output %u \"%s\"", i->index, i->name);
+ pa_log_info("Freeing input %u \"%s\"", i->index, pa_strnull(pa_proplist_gets(i->proplist, PA_PROP_MEDIA_NAME)));
pa_assert(!i->thread_info.attached);
- if (i->thread_info.resampled_chunk.memblock)
- pa_memblock_unref(i->thread_info.resampled_chunk.memblock);
+ if (i->thread_info.render_memblockq)
+ pa_memblockq_free(i->thread_info.render_memblockq);
if (i->thread_info.resampler)
pa_resampler_free(i->thread_info.resampler);
- if (i->thread_info.silence_memblock)
- pa_memblock_unref(i->thread_info.silence_memblock);
+ if (i->proplist)
+ pa_proplist_free(i->proplist);
- pa_xfree(i->name);
pa_xfree(i->driver);
pa_xfree(i);
}
void pa_sink_input_put(pa_sink_input *i) {
+ pa_sink_input_state_t state;
pa_sink_input_assert_ref(i);
pa_assert(i->state == PA_SINK_INPUT_INIT);
- pa_assert(i->peek);
- pa_assert(i->drop);
+ pa_assert(i->pop);
+ pa_assert(i->process_rewind);
- i->thread_info.state = i->state = i->flags & PA_SINK_INPUT_START_CORKED ? PA_SINK_INPUT_CORKED : PA_SINK_INPUT_RUNNING;
i->thread_info.volume = i->volume;
i->thread_info.muted = i->muted;
- if (i->state == PA_SINK_INPUT_CORKED)
- i->sink->n_corked++;
+ state = i->flags & PA_SINK_INPUT_START_CORKED ? PA_SINK_INPUT_CORKED : PA_SINK_INPUT_RUNNING;
+
+ update_n_corked(i, state);
+ i->state = state;
pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i->sink), PA_SINK_MESSAGE_ADD_INPUT, i, 0, NULL);
- pa_sink_update_status(i->sink);
pa_subscription_post(i->sink->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_NEW, i->index);
pa_hook_fire(&i->sink->core->hooks[PA_CORE_HOOK_SINK_INPUT_PUT], i);
-
- /* Please note that if you change something here, you have to
- change something in pa_sink_input_move() with the ghost stream
- registration too. */
}
void pa_sink_input_kill(pa_sink_input*i) {
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
if (i->kill)
i->kill(i);
pa_usec_t r = 0;
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
if (pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i), PA_SINK_INPUT_MESSAGE_GET_LATENCY, &r, 0, NULL) < 0)
r = 0;
}
/* Called from thread context */
-int pa_sink_input_peek(pa_sink_input *i, size_t length, pa_memchunk *chunk, pa_cvolume *volume) {
- int ret = -1;
- int do_volume_adj_here;
- int volume_is_norm;
- size_t block_size_max;
+int pa_sink_input_peek(pa_sink_input *i, size_t slength /* in sink frames */, pa_memchunk *chunk, pa_cvolume *volume) {
+ pa_bool_t do_volume_adj_here;
+ pa_bool_t volume_is_norm;
+ size_t block_size_max_sink, block_size_max_sink_input;
+ size_t ilength;
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->thread_info.state));
- pa_assert(pa_frame_aligned(length, &i->sink->sample_spec));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->thread_info.state));
+ pa_assert(pa_frame_aligned(slength, &i->sink->sample_spec));
pa_assert(chunk);
pa_assert(volume);
- if (!i->peek || !i->drop || i->thread_info.state == PA_SINK_INPUT_CORKED)
- goto finish;
+/* pa_log_debug("peek"); */
+
+ if (!i->pop)
+ return -1;
+
+ pa_assert(i->thread_info.state == PA_SINK_INPUT_RUNNING ||
+ i->thread_info.state == PA_SINK_INPUT_CORKED ||
+ i->thread_info.state == PA_SINK_INPUT_DRAINED);
+
+ /* If there's still some rewrite request the handle, but the sink
+ didn't do this for us, we do it here. However, since the sink
+ apparently doesn't support rewinding, we pass 0 here. This still
+ allows rewinding through the render buffer. */
+ pa_sink_input_process_rewind(i, 0);
+
+ block_size_max_sink_input = i->thread_info.resampler ?
+ pa_resampler_max_block_size(i->thread_info.resampler) :
+ pa_frame_align(pa_mempool_block_size_max(i->sink->core->mempool), &i->sample_spec);
- pa_assert(i->thread_info.state == PA_SINK_INPUT_RUNNING || i->thread_info.state == PA_SINK_INPUT_DRAINED);
+ block_size_max_sink = pa_frame_align(pa_mempool_block_size_max(i->sink->core->mempool), &i->sink->sample_spec);
/* Default buffer size */
- if (length <= 0)
- length = pa_frame_align(CONVERT_BUFFER_LENGTH, &i->sink->sample_spec);
-
- /* Make sure the buffer fits in the mempool tile */
- block_size_max = pa_mempool_block_size_max(i->sink->core->mempool);
- if (length > block_size_max)
- length = pa_frame_align(block_size_max, &i->sink->sample_spec);
-
- if (i->thread_info.move_silence > 0) {
- size_t l;
-
- /* We have just been moved and shall play some silence for a
- * while until the old sink has drained its playback buffer */
-
- if (!i->thread_info.silence_memblock)
- i->thread_info.silence_memblock = pa_silence_memblock_new(
- i->sink->core->mempool,
- &i->sink->sample_spec,
- pa_frame_align(SILENCE_BUFFER_LENGTH, &i->sink->sample_spec));
-
- chunk->memblock = pa_memblock_ref(i->thread_info.silence_memblock);
- chunk->index = 0;
- l = pa_memblock_get_length(chunk->memblock);
- chunk->length = i->thread_info.move_silence < l ? i->thread_info.move_silence : l;
-
- ret = 0;
- do_volume_adj_here = 1;
- goto finish;
- }
+ if (slength <= 0)
+ slength = pa_frame_align(CONVERT_BUFFER_LENGTH, &i->sink->sample_spec);
- if (!i->thread_info.resampler) {
- do_volume_adj_here = 0; /* FIXME??? */
- ret = i->peek(i, length, chunk);
- goto finish;
- }
+ if (slength > block_size_max_sink)
+ slength = block_size_max_sink;
+
+ if (i->thread_info.resampler) {
+ ilength = pa_resampler_request(i->thread_info.resampler, slength);
+
+ if (ilength <= 0)
+ ilength = pa_frame_align(CONVERT_BUFFER_LENGTH, &i->sample_spec);
+ } else
+ ilength = slength;
+
+ if (ilength > block_size_max_sink_input)
+ ilength = block_size_max_sink_input;
+
+ /* If the channel maps of the sink and this stream differ, we need
+ * to adjust the volume *before* we resample. Otherwise we can do
+ * it after and leave it for the sink code */
do_volume_adj_here = !pa_channel_map_equal(&i->channel_map, &i->sink->channel_map);
volume_is_norm = pa_cvolume_is_norm(&i->thread_info.volume) && !i->thread_info.muted;
- while (!i->thread_info.resampled_chunk.memblock) {
+ while (!pa_memblockq_is_readable(i->thread_info.render_memblockq)) {
pa_memchunk tchunk;
- size_t l, rmbs;
- l = pa_resampler_request(i->thread_info.resampler, length);
+ /* There's nothing in our render queue. We need to fill it up
+ * with data from the implementor. */
- if (l <= 0)
- l = pa_frame_align(CONVERT_BUFFER_LENGTH, &i->sample_spec);
+ if (i->thread_info.state == PA_SINK_INPUT_CORKED ||
+ i->pop(i, ilength, &tchunk) < 0) {
- rmbs = pa_resampler_max_block_size(i->thread_info.resampler);
- if (l > rmbs)
- l = rmbs;
+ /* OK, we're corked or the implementor didn't give us any
+ * data, so let's just hand out silence */
+ pa_atomic_store(&i->thread_info.drained, 1);
- if ((ret = i->peek(i, l, &tchunk)) < 0)
- goto finish;
+ pa_memblockq_seek(i->thread_info.render_memblockq, slength, PA_SEEK_RELATIVE);
+ i->thread_info.playing_for = 0;
+ if (i->thread_info.underrun_for != (uint64_t) -1)
+ i->thread_info.underrun_for += ilength;
+ break;
+ }
+
+ pa_atomic_store(&i->thread_info.drained, 0);
pa_assert(tchunk.length > 0);
+ pa_assert(tchunk.memblock);
- if (tchunk.length > l)
- tchunk.length = l;
+ i->thread_info.underrun_for = 0;
+ i->thread_info.playing_for += tchunk.length;
- i->drop(i, tchunk.length);
+ while (tchunk.length > 0) {
+ pa_memchunk wchunk;
- /* It might be necessary to adjust the volume here */
- if (do_volume_adj_here && !volume_is_norm) {
- pa_memchunk_make_writable(&tchunk, 0);
+ wchunk = tchunk;
+ pa_memblock_ref(wchunk.memblock);
- if (i->thread_info.muted)
- pa_silence_memchunk(&tchunk, &i->thread_info.sample_spec);
- else
- pa_volume_memchunk(&tchunk, &i->thread_info.sample_spec, &i->thread_info.volume);
+ if (wchunk.length > block_size_max_sink_input)
+ wchunk.length = block_size_max_sink_input;
+
+ /* It might be necessary to adjust the volume here */
+ if (do_volume_adj_here && !volume_is_norm) {
+ pa_memchunk_make_writable(&wchunk, 0);
+
+ if (i->thread_info.muted)
+ pa_silence_memchunk(&wchunk, &i->thread_info.sample_spec);
+ else
+ pa_volume_memchunk(&wchunk, &i->thread_info.sample_spec, &i->thread_info.volume);
+ }
+
+ if (!i->thread_info.resampler)
+ pa_memblockq_push_align(i->thread_info.render_memblockq, &wchunk);
+ else {
+ pa_memchunk rchunk;
+ pa_resampler_run(i->thread_info.resampler, &wchunk, &rchunk);
+
+/* pa_log_debug("pushing %lu", (unsigned long) rchunk.length); */
+
+ if (rchunk.memblock) {
+ pa_memblockq_push_align(i->thread_info.render_memblockq, &rchunk);
+ pa_memblock_unref(rchunk.memblock);
+ }
+ }
+
+ pa_memblock_unref(wchunk.memblock);
+
+ tchunk.index += wchunk.length;
+ tchunk.length -= wchunk.length;
}
- pa_resampler_run(i->thread_info.resampler, &tchunk, &i->thread_info.resampled_chunk);
pa_memblock_unref(tchunk.memblock);
}
- pa_assert(i->thread_info.resampled_chunk.memblock);
- pa_assert(i->thread_info.resampled_chunk.length > 0);
+ pa_assert_se(pa_memblockq_peek(i->thread_info.render_memblockq, chunk) >= 0);
- *chunk = i->thread_info.resampled_chunk;
- pa_memblock_ref(i->thread_info.resampled_chunk.memblock);
+ pa_assert(chunk->length > 0);
+ pa_assert(chunk->memblock);
- ret = 0;
+/* pa_log_debug("peeking %lu", (unsigned long) chunk->length); */
-finish:
+ if (chunk->length > block_size_max_sink)
+ chunk->length = block_size_max_sink;
- if (ret >= 0)
- pa_atomic_store(&i->thread_info.drained, 0);
- else if (ret < 0)
- pa_atomic_store(&i->thread_info.drained, 1);
-
- if (ret >= 0) {
- /* Let's see if we had to apply the volume adjustment
- * ourselves, or if this can be done by the sink for us */
+ /* Let's see if we had to apply the volume adjustment ourselves,
+ * or if this can be done by the sink for us */
- if (do_volume_adj_here)
- /* We had different channel maps, so we already did the adjustment */
- pa_cvolume_reset(volume, i->sink->sample_spec.channels);
- else if (i->thread_info.muted)
- /* We've both the same channel map, so let's have the sink do the adjustment for us*/
- pa_cvolume_mute(volume, i->sink->sample_spec.channels);
- else
- *volume = i->thread_info.volume;
- }
+ if (do_volume_adj_here)
+ /* We had different channel maps, so we already did the adjustment */
+ pa_cvolume_reset(volume, i->sink->sample_spec.channels);
+ else if (i->thread_info.muted)
+ /* We've both the same channel map, so let's have the sink do the adjustment for us*/
+ pa_cvolume_mute(volume, i->sink->sample_spec.channels);
+ else
+ *volume = i->thread_info.volume;
- return ret;
+ return 0;
}
/* Called from thread context */
-void pa_sink_input_drop(pa_sink_input *i, size_t length) {
+void pa_sink_input_drop(pa_sink_input *i, size_t nbytes /* in sink sample spec */) {
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->thread_info.state));
- pa_assert(pa_frame_aligned(length, &i->sink->sample_spec));
- pa_assert(length > 0);
- if (!i->peek || !i->drop || i->thread_info.state == PA_SINK_INPUT_CORKED)
- return;
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->thread_info.state));
+ pa_assert(pa_frame_aligned(nbytes, &i->sink->sample_spec));
+ pa_assert(nbytes > 0);
- if (i->thread_info.move_silence > 0) {
+/* pa_log_debug("dropping %lu", (unsigned long) nbytes); */
- if (i->thread_info.move_silence >= length) {
- i->thread_info.move_silence -= length;
- length = 0;
- } else {
- length -= i->thread_info.move_silence;
- i->thread_info.move_silence = 0;
- }
+ /* If there's still some rewrite request the handle, but the sink
+ didn't do this for us, we do it here. However, since the sink
+ apparently doesn't support rewinding, we pass 0 here. This still
+ allows rewinding through the render buffer. */
+ pa_sink_input_process_rewind(i, 0);
- if (i->thread_info.move_silence <= 0) {
- if (i->thread_info.silence_memblock) {
- pa_memblock_unref(i->thread_info.silence_memblock);
- i->thread_info.silence_memblock = NULL;
- }
- }
+ pa_memblockq_drop(i->thread_info.render_memblockq, nbytes);
+}
+
+/* Called from thread context */
+void pa_sink_input_process_rewind(pa_sink_input *i, size_t nbytes /* in sink sample spec */) {
+ size_t lbq;
+ pa_sink_input_assert_ref(i);
+
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->thread_info.state));
+ pa_assert(pa_frame_aligned(nbytes, &i->sink->sample_spec));
+
+/* pa_log_debug("rewind(%lu, %lu)", (unsigned long) nbytes, (unsigned long) i->thread_info.rewrite_nbytes); */
- if (length <= 0)
- return;
+ lbq = pa_memblockq_get_length(i->thread_info.render_memblockq);
+
+ if (nbytes > 0) {
+ pa_log_debug("Have to rewind %lu bytes on render memblockq.", (unsigned long) nbytes);
+ pa_memblockq_rewind(i->thread_info.render_memblockq, nbytes);
}
- if (i->thread_info.resampled_chunk.memblock) {
- size_t l = length;
+ if (i->thread_info.rewrite_nbytes == (size_t) -1) {
- if (l > i->thread_info.resampled_chunk.length)
- l = i->thread_info.resampled_chunk.length;
+ /* We were asked to drop all buffered data, and rerequest new
+ * data from implementor the next time push() is called */
- i->thread_info.resampled_chunk.index += l;
- i->thread_info.resampled_chunk.length -= l;
+ pa_memblockq_flush(i->thread_info.render_memblockq);
- if (i->thread_info.resampled_chunk.length <= 0) {
- pa_memblock_unref(i->thread_info.resampled_chunk.memblock);
- pa_memchunk_reset(&i->thread_info.resampled_chunk);
- }
+ } else if (i->thread_info.rewrite_nbytes > 0) {
+ size_t max_rewrite, amount;
+
+ /* Calculate how much make sense to rewrite at most */
+ max_rewrite = nbytes + lbq;
+
+ /* Transform into local domain */
+ if (i->thread_info.resampler)
+ max_rewrite = pa_resampler_request(i->thread_info.resampler, max_rewrite);
+
+ /* Calculate how much of the rewinded data should actually be rewritten */
+ amount = PA_MIN(i->thread_info.rewrite_nbytes, max_rewrite);
+
+ if (amount > 0) {
+ pa_log_debug("Have to rewind %lu bytes on implementor.", (unsigned long) amount);
- length -= l;
+ /* Tell the implementor */
+ if (i->process_rewind)
+ i->process_rewind(i, amount);
+
+ /* Convert back to to sink domain */
+ if (i->thread_info.resampler)
+ amount = pa_resampler_result(i->thread_info.resampler, amount);
+
+ if (amount > 0)
+ /* Ok, now update the write pointer */
+ pa_memblockq_seek(i->thread_info.render_memblockq, - ((int64_t) amount), PA_SEEK_RELATIVE);
+
+ if (i->thread_info.rewrite_flush)
+ pa_memblockq_silence(i->thread_info.render_memblockq);
+
+ /* And reset the resampler */
+ if (i->thread_info.resampler)
+ pa_resampler_reset(i->thread_info.resampler);
+ }
}
- if (length > 0) {
+ i->thread_info.rewrite_nbytes = 0;
+ i->thread_info.rewrite_flush = FALSE;
+}
+
+/* Called from thread context */
+void pa_sink_input_update_max_rewind(pa_sink_input *i, size_t nbytes /* in the sink's sample spec */) {
+ pa_sink_input_assert_ref(i);
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->thread_info.state));
+ pa_assert(pa_frame_aligned(nbytes, &i->sink->sample_spec));
- if (i->thread_info.resampler) {
- /* So, we have a resampler. To avoid discontinuities we
- * have to actually read all data that could be read and
- * pass it through the resampler. */
+ pa_memblockq_set_maxrewind(i->thread_info.render_memblockq, nbytes);
- while (length > 0) {
- pa_memchunk chunk;
- pa_cvolume volume;
+ if (i->update_max_rewind)
+ i->update_max_rewind(i, i->thread_info.resampler ? pa_resampler_request(i->thread_info.resampler, nbytes) : nbytes);
+}
- if (pa_sink_input_peek(i, length, &chunk, &volume) >= 0) {
- size_t l;
+static pa_usec_t fixup_latency(pa_sink *s, pa_usec_t usec) {
+ pa_sink_assert_ref(s);
- pa_memblock_unref(chunk.memblock);
+ if (usec == (pa_usec_t) -1)
+ return usec;
- l = chunk.length;
- if (l > length)
- l = length;
+ if (s->max_latency > 0 && usec > s->max_latency)
+ usec = s->max_latency;
- pa_sink_input_drop(i, l);
- length -= l;
+ if (s->min_latency > 0 && usec < s->min_latency)
+ usec = s->min_latency;
- } else {
- size_t l;
+ return usec;
+}
- l = pa_resampler_request(i->thread_info.resampler, length);
+pa_usec_t pa_sink_input_set_requested_latency_within_thread(pa_sink_input *i, pa_usec_t usec) {
+ pa_sink_input_assert_ref(i);
- /* Hmmm, peeking failed, so let's at least drop
- * the right amount of data */
+ usec = fixup_latency(i->sink, usec);
- if (l > 0)
- if (i->drop)
- i->drop(i, l);
+ i->thread_info.requested_sink_latency = usec;
+ pa_sink_invalidate_requested_latency(i->sink);
- break;
- }
- }
+ return usec;
+}
- } else {
+pa_usec_t pa_sink_input_set_requested_latency(pa_sink_input *i, pa_usec_t usec) {
+ pa_sink_input_assert_ref(i);
- /* We have no resampler, hence let's just drop the data */
+ usec = fixup_latency(i->sink, usec);
- if (i->drop)
- i->drop(i, length);
- }
+ if (PA_SINK_INPUT_IS_LINKED(i->state))
+ pa_asyncmsgq_post(i->sink->asyncmsgq, PA_MSGOBJECT(i), PA_SINK_INPUT_MESSAGE_SET_REQUESTED_LATENCY, NULL, (int64_t) usec, NULL, NULL);
+ else {
+ /* If this sink input is not realized yet, we have to touch
+ * the thread info data directly */
+ i->thread_info.requested_sink_latency = usec;
+ i->sink->thread_info.requested_latency_valid = FALSE;
}
+
+ return usec;
+}
+
+pa_usec_t pa_sink_input_get_requested_latency(pa_sink_input *i) {
+ pa_usec_t usec = 0;
+
+ pa_sink_input_assert_ref(i);
+
+ if (PA_SINK_INPUT_IS_LINKED(i->state))
+ pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i), PA_SINK_INPUT_MESSAGE_GET_REQUESTED_LATENCY, &usec, 0, NULL);
+ else
+ /* If this sink input is not realized yet, we have to touch
+ * the thread info data directly */
+ usec = i->thread_info.requested_sink_latency;
+
+ return usec;
}
void pa_sink_input_set_volume(pa_sink_input *i, const pa_cvolume *volume) {
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
if (pa_cvolume_equal(&i->volume, volume))
return;
const pa_cvolume *pa_sink_input_get_volume(pa_sink_input *i) {
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
return &i->volume;
}
void pa_sink_input_set_mute(pa_sink_input *i, pa_bool_t mute) {
pa_assert(i);
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
if (!i->muted == !mute)
return;
int pa_sink_input_get_mute(pa_sink_input *i) {
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
return !!i->muted;
}
void pa_sink_input_cork(pa_sink_input *i, pa_bool_t b) {
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
sink_input_set_state(i, b ? PA_SINK_INPUT_CORKED : PA_SINK_INPUT_RUNNING);
}
int pa_sink_input_set_rate(pa_sink_input *i, uint32_t rate) {
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
pa_return_val_if_fail(i->thread_info.resampler, -1);
if (i->sample_spec.rate == rate)
}
void pa_sink_input_set_name(pa_sink_input *i, const char *name) {
+ const char *old;
pa_sink_input_assert_ref(i);
- if (!i->name && !name)
+ if (!name && !pa_proplist_contains(i->proplist, PA_PROP_MEDIA_NAME))
return;
- if (i->name && name && !strcmp(i->name, name))
+ old = pa_proplist_gets(i->proplist, PA_PROP_MEDIA_NAME);
+
+ if (old && name && !strcmp(old, name))
return;
- pa_xfree(i->name);
- i->name = pa_xstrdup(name);
+ if (name)
+ pa_proplist_sets(i->proplist, PA_PROP_MEDIA_NAME, name);
+ else
+ pa_proplist_unset(i->proplist, PA_PROP_MEDIA_NAME);
- if (PA_SINK_INPUT_LINKED(i->state)) {
- pa_hook_fire(&i->sink->core->hooks[PA_CORE_HOOK_SINK_INPUT_NAME_CHANGED], i);
+ if (PA_SINK_INPUT_IS_LINKED(i->state)) {
+ pa_hook_fire(&i->sink->core->hooks[PA_CORE_HOOK_SINK_INPUT_PROPLIST_CHANGED], i);
pa_subscription_post(i->sink->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_CHANGE, i->index);
}
}
return i->resample_method;
}
-int pa_sink_input_move_to(pa_sink_input *i, pa_sink *dest, int immediately) {
+int pa_sink_input_move_to(pa_sink_input *i, pa_sink *dest) {
pa_resampler *new_resampler;
pa_sink *origin;
- pa_usec_t silence_usec = 0;
- pa_sink_input_move_info info;
pa_sink_input_move_hook_data hook_data;
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->state));
pa_sink_assert_ref(dest);
origin = i->sink;
hook_data.destination = dest;
pa_hook_fire(&i->sink->core->hooks[PA_CORE_HOOK_SINK_INPUT_MOVE], &hook_data);
- memset(&info, 0, sizeof(info));
- info.sink_input = i;
-
- if (!immediately) {
- pa_usec_t old_latency, new_latency;
-
- /* Let's do a little bit of Voodoo for compensating latency
- * differences. We assume that the accuracy for our
- * estimations is still good enough, even though we do these
- * operations non-atomic. */
-
- old_latency = pa_sink_get_latency(origin);
- new_latency = pa_sink_get_latency(dest);
-
- /* The already resampled data should go to the old sink */
-
- if (old_latency >= new_latency) {
-
- /* The latency of the old sink is larger than the latency
- * of the new sink. Therefore to compensate for the
- * difference we to play silence on the new one for a
- * while */
-
- silence_usec = old_latency - new_latency;
-
- } else {
-
- /* The latency of new sink is larger than the latency of
- * the old sink. Therefore we have to precompute a little
- * and make sure that this is still played on the old
- * sink, until we can play the first sample on the new
- * sink.*/
-
- info.buffer_bytes = pa_usec_to_bytes(new_latency - old_latency, &origin->sample_spec);
- }
-
- /* Okey, let's move it */
-
- if (info.buffer_bytes > 0) {
-
- info.ghost_sink_input = pa_memblockq_sink_input_new(
- origin,
- "Ghost Stream",
- &origin->sample_spec,
- &origin->channel_map,
- NULL,
- NULL);
-
- info.ghost_sink_input->thread_info.state = info.ghost_sink_input->state = PA_SINK_INPUT_RUNNING;
- info.ghost_sink_input->thread_info.volume = info.ghost_sink_input->volume;
- info.ghost_sink_input->thread_info.muted = info.ghost_sink_input->muted;
-
- info.buffer = pa_memblockq_new(0, MOVE_BUFFER_LENGTH, 0, pa_frame_size(&origin->sample_spec), 0, 0, NULL);
- }
- }
-
- pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i->sink), PA_SINK_MESSAGE_REMOVE_INPUT_AND_BUFFER, &info, 0, NULL);
-
- if (info.ghost_sink_input) {
- /* Basically, do what pa_sink_input_put() does ...*/
-
- pa_subscription_post(i->sink->core, PA_SUBSCRIPTION_EVENT_SINK_INPUT|PA_SUBSCRIPTION_EVENT_NEW, info.ghost_sink_input->index);
- pa_hook_fire(&i->sink->core->hooks[PA_CORE_HOOK_SINK_INPUT_PUT], info.ghost_sink_input);
- pa_sink_input_unref(info.ghost_sink_input);
- }
+ pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i->sink), PA_SINK_MESSAGE_START_MOVE, i, 0, NULL);
pa_idxset_remove_by_data(origin->inputs, i, NULL);
pa_idxset_put(dest->inputs, i, NULL);
dest->n_corked++;
}
- /* Replace resampler */
+ /* Replace resampler and render queue */
if (new_resampler != i->thread_info.resampler) {
+
if (i->thread_info.resampler)
pa_resampler_free(i->thread_info.resampler);
i->thread_info.resampler = new_resampler;
- /* if the resampler changed, the silence memblock is
- * probably invalid now, too */
- if (i->thread_info.silence_memblock) {
- pa_memblock_unref(i->thread_info.silence_memblock);
- i->thread_info.silence_memblock = NULL;
- }
+ pa_memblockq_free(i->thread_info.render_memblockq);
+
+ i->thread_info.render_memblockq = pa_memblockq_new(
+ 0,
+ MEMBLOCKQ_MAXLENGTH,
+ 0,
+ pa_frame_size(&i->sink->sample_spec),
+ 0,
+ 1,
+ 0,
+ &i->sink->silence);
}
- /* Dump already resampled data */
- if (i->thread_info.resampled_chunk.memblock) {
- /* Hmm, this data has already been added to the ghost queue, presumably, hence let's sleep a little bit longer */
- silence_usec += pa_bytes_to_usec(i->thread_info.resampled_chunk.length, &origin->sample_spec);
- pa_memblock_unref(i->thread_info.resampled_chunk.memblock);
- pa_memchunk_reset(&i->thread_info.resampled_chunk);
- }
-
- /* Calculate the new sleeping time */
- if (immediately)
- i->thread_info.move_silence = 0;
- else
- i->thread_info.move_silence = pa_usec_to_bytes(
- pa_bytes_to_usec(i->thread_info.move_silence, &origin->sample_spec) +
- silence_usec,
- &dest->sample_spec);
-
- pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i->sink), PA_SINK_MESSAGE_ADD_INPUT, i, 0, NULL);
-
pa_sink_update_status(origin);
pa_sink_update_status(dest);
+ pa_asyncmsgq_send(i->sink->asyncmsgq, PA_MSGOBJECT(i->sink), PA_SINK_MESSAGE_FINISH_MOVE, i, 0, NULL);
+
if (i->moved)
i->moved(i);
return 0;
}
+void pa_sink_input_set_state_within_thread(pa_sink_input *i, pa_sink_input_state_t state) {
+ pa_sink_input_assert_ref(i);
+
+ if (state == i->thread_info.state)
+ return;
+
+ if ((state == PA_SINK_INPUT_DRAINED || state == PA_SINK_INPUT_RUNNING) &&
+ !(i->thread_info.state == PA_SINK_INPUT_DRAINED || i->thread_info.state != PA_SINK_INPUT_RUNNING))
+ pa_atomic_store(&i->thread_info.drained, 1);
+
+ if (state == PA_SINK_INPUT_CORKED && i->thread_info.state != PA_SINK_INPUT_CORKED) {
+
+ /* This will tell the implementing sink input driver to rewind
+ * so that the unplayed already mixed data is not lost */
+ pa_sink_input_request_rewind(i, 0, TRUE, TRUE);
+
+ } else if (i->thread_info.state == PA_SINK_INPUT_CORKED && state != PA_SINK_INPUT_CORKED) {
+
+ /* OK, we're being uncorked. Make sure we're not rewound when
+ * the hw buffer is remixed and request a remix. */
+ pa_sink_input_request_rewind(i, 0, FALSE, TRUE);
+ }
+
+ if (i->state_change)
+ i->state_change(i, state);
+
+ i->thread_info.state = state;
+}
+
/* Called from thread context */
int pa_sink_input_process_msg(pa_msgobject *o, int code, void *userdata, int64_t offset, pa_memchunk *chunk) {
pa_sink_input *i = PA_SINK_INPUT(o);
pa_sink_input_assert_ref(i);
- pa_assert(PA_SINK_INPUT_LINKED(i->thread_info.state));
+ pa_assert(PA_SINK_INPUT_IS_LINKED(i->thread_info.state));
switch (code) {
case PA_SINK_INPUT_MESSAGE_SET_VOLUME:
i->thread_info.volume = *((pa_cvolume*) userdata);
+ pa_sink_input_request_rewind(i, 0, TRUE, FALSE);
return 0;
case PA_SINK_INPUT_MESSAGE_SET_MUTE:
i->thread_info.muted = PA_PTR_TO_UINT(userdata);
+ pa_sink_input_request_rewind(i, 0, TRUE, FALSE);
return 0;
case PA_SINK_INPUT_MESSAGE_GET_LATENCY: {
pa_usec_t *r = userdata;
- if (i->thread_info.resampled_chunk.memblock)
- *r += pa_bytes_to_usec(i->thread_info.resampled_chunk.length, &i->sink->sample_spec);
-
- if (i->thread_info.move_silence)
- *r += pa_bytes_to_usec(i->thread_info.move_silence, &i->sink->sample_spec);
-
+ *r += pa_bytes_to_usec(pa_memblockq_get_length(i->thread_info.render_memblockq), &i->sink->sample_spec);
return 0;
}
case PA_SINK_INPUT_MESSAGE_SET_STATE: {
pa_sink_input *ssync;
- if ((PA_PTR_TO_UINT(userdata) == PA_SINK_INPUT_DRAINED || PA_PTR_TO_UINT(userdata) == PA_SINK_INPUT_RUNNING) &&
- (i->thread_info.state != PA_SINK_INPUT_DRAINED) && (i->thread_info.state != PA_SINK_INPUT_RUNNING))
- pa_atomic_store(&i->thread_info.drained, 1);
+ pa_sink_input_set_state_within_thread(i, PA_PTR_TO_UINT(userdata));
- i->thread_info.state = PA_PTR_TO_UINT(userdata);
+ for (ssync = i->thread_info.sync_prev; ssync; ssync = ssync->thread_info.sync_prev)
+ pa_sink_input_set_state_within_thread(ssync, PA_PTR_TO_UINT(userdata));
- for (ssync = i->thread_info.sync_prev; ssync; ssync = ssync->thread_info.sync_prev) {
- if ((PA_PTR_TO_UINT(userdata) == PA_SINK_INPUT_DRAINED || PA_PTR_TO_UINT(userdata) == PA_SINK_INPUT_RUNNING) &&
- (ssync->thread_info.state != PA_SINK_INPUT_DRAINED) && (ssync->thread_info.state != PA_SINK_INPUT_RUNNING))
- pa_atomic_store(&ssync->thread_info.drained, 1);
- ssync->thread_info.state = PA_PTR_TO_UINT(userdata);
- }
+ for (ssync = i->thread_info.sync_next; ssync; ssync = ssync->thread_info.sync_next)
+ pa_sink_input_set_state_within_thread(ssync, PA_PTR_TO_UINT(userdata));
- for (ssync = i->thread_info.sync_next; ssync; ssync = ssync->thread_info.sync_next) {
- if ((PA_PTR_TO_UINT(userdata) == PA_SINK_INPUT_DRAINED || PA_PTR_TO_UINT(userdata) == PA_SINK_INPUT_RUNNING) &&
- (ssync->thread_info.state != PA_SINK_INPUT_DRAINED) && (ssync->thread_info.state != PA_SINK_INPUT_RUNNING))
- pa_atomic_store(&ssync->thread_info.drained, 1);
- ssync->thread_info.state = PA_PTR_TO_UINT(userdata);
- }
+ return 0;
+ }
+
+ case PA_SINK_INPUT_MESSAGE_SET_REQUESTED_LATENCY:
+
+ pa_sink_input_set_requested_latency_within_thread(i, (pa_usec_t) offset);
+ return 0;
+
+ case PA_SINK_INPUT_MESSAGE_GET_REQUESTED_LATENCY: {
+ pa_usec_t *r = userdata;
+ *r = i->thread_info.requested_sink_latency;
return 0;
}
}
return i->state;
}
+
+/* Called from IO context */
+pa_bool_t pa_sink_input_safe_to_remove(pa_sink_input *i) {
+ pa_sink_input_assert_ref(i);
+
+ if (PA_SINK_INPUT_IS_LINKED(i->thread_info.state))
+ return pa_memblockq_is_empty(i->thread_info.render_memblockq);
+
+ return TRUE;
+}
+
+void pa_sink_input_request_rewind(pa_sink_input *i, size_t nbytes /* in our sample spec */, pa_bool_t rewrite, pa_bool_t flush) {
+ size_t lbq;
+
+ /* If 'rewrite' is TRUE the sink is rewound as far as requested
+ * and possible and the exact value of this is passed back the
+ * implementor via process_rewind(). If 'flush' is also TRUE all
+ * already rendered data is also dropped.
+ *
+ * If 'rewrite' is FALSE the sink is rewound as far as requested
+ * and possible and the already rendered data is dropped so that
+ * in the next iteration we read new data from the
+ * implementor. This implies 'flush' is TRUE. */
+
+ pa_sink_input_assert_ref(i);
+ pa_assert(i->thread_info.rewrite_nbytes == 0);
+
+ /* We don't take rewind requests while we are corked */
+ if (i->thread_info.state == PA_SINK_INPUT_CORKED)
+ return;
+
+ pa_assert(rewrite || flush);
+
+ /* Calculate how much we can rewind locally without having to
+ * touch the sink */
+ if (rewrite)
+ lbq = pa_memblockq_get_length(i->thread_info.render_memblockq);
+ else
+ lbq = 0;
+
+ /* Check if rewinding for the maximum is requested, and if so, fix up */
+ if (nbytes <= 0) {
+
+ /* Calculate maximum number of bytes that could be rewound in theory */
+ nbytes = i->sink->thread_info.max_rewind + lbq;
+
+ /* Transform from sink domain */
+ if (i->thread_info.resampler)
+ nbytes = pa_resampler_request(i->thread_info.resampler, nbytes);
+ }
+
+ if (rewrite) {
+ /* Make sure to not overwrite over underruns */
+ if (nbytes > i->thread_info.playing_for)
+ nbytes = (size_t) i->thread_info.playing_for;
+
+ i->thread_info.rewrite_nbytes = nbytes;
+ } else
+ i->thread_info.rewrite_nbytes = (size_t) -1;
+
+ i->thread_info.rewrite_flush = flush && i->thread_info.rewrite_nbytes != 0;
+
+ /* Transform to sink domain */
+ if (i->thread_info.resampler)
+ nbytes = pa_resampler_result(i->thread_info.resampler, nbytes);
+
+ if (nbytes > lbq)
+ pa_sink_request_rewind(i->sink, nbytes - lbq);
+}
+
+pa_memchunk* pa_sink_input_get_silence(pa_sink_input *i, pa_memchunk *ret) {
+ pa_sink_input_assert_ref(i);
+ pa_assert(ret);
+
+ pa_silence_memchunk_get(
+ &i->sink->core->silence_cache,
+ i->sink->core->mempool,
+ ret,
+ &i->sample_spec,
+ i->thread_info.resampler ? pa_resampler_max_block_size(i->thread_info.resampler) : 0);
+
+ return ret;
+}