mpv/stream/cache.c

640 lines
20 KiB
C
Raw Normal View History

/*
* This file is part of MPlayer.
*
* MPlayer is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* MPlayer is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with MPlayer; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
// Time in seconds the main thread waits for the cache thread. On wakeups, the
// code checks for user requested aborts and also prints warnings that the
// cache is being slow.
#define CACHE_WAIT_TIME 0.5
// The time the cache sleeps in idle mode. This controls how often the cache
// retries reading from the stream after EOF has reached (in case the stream is
// actually readable again, for example if data has been appended to a file).
// Note that if this timeout is too low, the player will waste too much CPU
// when player is paused.
#define CACHE_IDLE_SLEEP_TIME 1.0
// Time in seconds the cache updates "cached" controls. Note that idle mode
// will block the cache from doing this, and this timeout is honored only if
// the cache is active.
#define CACHE_UPDATE_CONTROLS_TIME 2.0
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/types.h>
#include <unistd.h>
#include <errno.h>
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
#include <assert.h>
#include <pthread.h>
#include <time.h>
#include <libavutil/common.h>
#include "config.h"
#include "osdep/timer.h"
#include "core/mp_msg.h"
#include "stream.h"
#include "core/mp_common.h"
// Note: (struct priv*)(cache->priv)->cache == cache
struct priv {
pthread_t cache_thread;
bool cache_thread_running;
pthread_mutex_t mutex;
pthread_cond_t wakeup;
// Constants (as long as cache thread is running)
unsigned char *buffer; // base pointer of the allocated buffer memory
int64_t buffer_size; // size of the allocated buffer memory
int64_t back_size; // keep back_size amount of old bytes for backward seek
int64_t fill_limit; // we should fill buffer only if space>=fill_limit
int64_t seek_limit; // keep filling cache if distance is less that seek limit
struct byte_meta *bm; // additional per-byte metadata
// Owned by the main thread
stream_t *cache; // wrapper stream, used by demuxer etc.
// Owned by the cache thread
stream_t *stream; // "real" stream, used to read from the source media
// All the following members are shared between the threads.
// You must lock the mutex to access them.
// Ringbuffer
int64_t min_filepos; // range of file that is cached in the buffer
int64_t max_filepos; // ... max_filepos being the last read position
bool eof; // true if max_filepos = EOF
int64_t offset; // buffer[offset] correponds to max_filepos
bool idle; // cache thread has stopped reading
int64_t read_filepos; // client read position (mirrors cache->pos)
int control; // requested STREAM_CTRL_... or CACHE_CTRL_...
void *control_arg; // temporary for executing STREAM_CTRLs
int control_res;
bool control_flush;
// Cached STREAM_CTRLs
double stream_time_length;
double stream_start_time;
int64_t stream_size;
bool stream_manages_timeline;
unsigned int stream_num_chapters;
int stream_cache_idle;
int stream_cache_fill;
char **stream_metadata;
};
// Store additional per-byte metadata. Since per-byte would be way too
// inefficient, store it only for every BYTE_META_CHUNK_SIZE byte.
struct byte_meta {
float stream_pts;
};
enum {
BYTE_META_CHUNK_SIZE = 8 * 1024,
CACHE_INTERRUPTED = -1,
CACHE_CTRL_NONE = 0,
CACHE_CTRL_QUIT = -1,
CACHE_CTRL_PING = -2,
};
static int64_t mp_clipi64(int64_t val, int64_t min, int64_t max)
{
val = FFMIN(val, max);
val = FFMAX(val, min);
return val;
}
// pthread_cond_timedwait() with a relative timeout in seconds
static int cond_timed_wait(pthread_cond_t *cond, pthread_mutex_t *mutex,
double timeout)
{
struct timespec ts;
#if _POSIX_TIMERS > 0
clock_gettime(CLOCK_REALTIME, &ts);
#else
struct timeval tv;
gettimeofday(&tv, NULL);
ts.tv_sec = tv.tv_sec;
ts.tv_nsec = tv.tv_usec * 1000UL;
#endif
unsigned long seconds = (int)timeout;
unsigned long nsecs = (timeout - seconds) * 1000000000UL;
if (nsecs + ts.tv_nsec >= 1000000000UL) {
seconds += 1;
nsecs -= 1000000000UL;
}
ts.tv_sec += seconds;
ts.tv_nsec += nsecs;
return pthread_cond_timedwait(cond, mutex, &ts);
}
// Used by the main thread to wakeup the cache thread, and to wait for the
// cache thread. The cache mutex has to be locked when calling this function.
// *retry_time should be set to 0 on the first call.
// Returns CACHE_INTERRUPTED if the caller is supposed to abort.
static int cache_wakeup_and_wait(struct priv *s, double *retry_time)
{
if (stream_check_interrupt(0))
return CACHE_INTERRUPTED;
// Print a "more severe" warning after waiting 1 second and no new data
if ((*retry_time) * CACHE_WAIT_TIME >= 1.0) {
mp_msg(MSGT_CACHE, MSGL_ERR, "Cache keeps not responding.\n");
} else if (*retry_time > 0.1) {
mp_msg(MSGT_CACHE, MSGL_WARN,
"Cache is not responding - slow/stuck network connection?\n");
}
double start = mp_time_sec();
pthread_cond_signal(&s->wakeup);
cond_timed_wait(&s->wakeup, &s->mutex, CACHE_WAIT_TIME);
*retry_time += mp_time_sec() - start;
return 0;
}
// Runs in the cache thread
static void cache_drop_contents(struct priv *s)
{
s->offset = s->min_filepos = s->max_filepos = s->read_filepos;
s->eof = false;
}
// Runs in the main thread
// mutex must be held, but is sometimes temporarily dropped
static int cache_read(struct priv *s, unsigned char *buf, int size)
{
if (size <= 0)
return 0;
double retry = 0;
while (s->read_filepos >= s->max_filepos ||
s->read_filepos < s->min_filepos)
{
if (s->eof && s->read_filepos >= s->max_filepos)
return 0;
if (cache_wakeup_and_wait(s, &retry) == CACHE_INTERRUPTED)
return 0;
}
int64_t newb = s->max_filepos - s->read_filepos; // new bytes in the buffer
2013-06-04 23:58:36 +00:00
int64_t pos = s->read_filepos - s->offset; // file pos to buffer memory pos
if (pos < 0)
pos += s->buffer_size;
else if (pos >= s->buffer_size)
pos -= s->buffer_size;
2013-06-04 23:58:36 +00:00
if (newb > s->buffer_size - pos)
newb = s->buffer_size - pos; // handle wrap...
2013-06-04 23:58:36 +00:00
newb = FFMIN(newb, size);
2013-06-04 23:58:36 +00:00
memcpy(buf, &s->buffer[pos], newb);
s->read_filepos += newb;
return newb;
}
// Runs in the cache thread.
// Returns true if reading was attempted, and the mutex was shortly unlocked.
static bool cache_fill(struct priv *s)
{
2013-06-04 23:58:36 +00:00
int64_t read = s->read_filepos;
int len;
2013-06-04 23:58:36 +00:00
if (read < s->min_filepos || read > s->max_filepos) {
// seek...
mp_msg(MSGT_CACHE, MSGL_DBG2,
"Out of boundaries... seeking to 0x%" PRIX64 " \n", read);
// drop cache contents only if seeking backward or too much fwd.
// This is also done for on-disk files, since it loses the backseek cache.
// That in turn can cause major bandwidth increase and performance
// issues with e.g. mov or badly interleaved files
if (read < s->min_filepos || read >= s->max_filepos + s->seek_limit) {
mp_msg(MSGT_CACHE, MSGL_V, "Dropping cache at pos %"PRId64", "
"cached range: %"PRId64"-%"PRId64".\n", read,
s->min_filepos, s->max_filepos);
cache_drop_contents(s);
stream_seek(s->stream, read);
2013-06-04 23:58:36 +00:00
}
}
// number of buffer bytes which should be preserved in backwards direction
int64_t back = mp_clipi64(read - s->min_filepos, 0, s->back_size);
// number of buffer bytes that are valid and can be read
int64_t newb = FFMAX(s->max_filepos - read, 0);
// max. number of bytes that can be written (starting from max_filepos)
int64_t space = s->buffer_size - (newb + back);
// offset into the buffer that maps to max_filepos
int pos = s->max_filepos - s->offset;
2013-06-04 23:58:36 +00:00
if (pos >= s->buffer_size)
pos -= s->buffer_size; // wrap-around
2013-06-04 23:58:36 +00:00
if (space < s->fill_limit) {
s->idle = true;
return false;
2013-06-04 23:58:36 +00:00
}
// limit to end of buffer (without wrapping)
if (pos + space >= s->buffer_size)
space = s->buffer_size - pos;
// limit read size (or else would block and read the entire buffer in 1 call)
space = FFMIN(space, s->stream->read_chunk);
2013-06-04 23:58:36 +00:00
// back+newb+space <= buffer_size
int64_t back2 = s->buffer_size - (space + newb); // max back size
2013-06-04 23:58:36 +00:00
if (s->min_filepos < (read - back2))
s->min_filepos = read - back2;
// The read call might take a long time and block, so drop the lock.
pthread_mutex_unlock(&s->mutex);
len = stream_read_partial(s->stream, &s->buffer[pos], space);
pthread_mutex_lock(&s->mutex);
double pts;
if (stream_control(s->stream, STREAM_CTRL_GET_CURRENT_TIME, &pts) <= 0)
pts = MP_NOPTS_VALUE;
for (int64_t b_pos = pos; b_pos < pos + len + BYTE_META_CHUNK_SIZE;
b_pos += BYTE_META_CHUNK_SIZE)
{
s->bm[b_pos / BYTE_META_CHUNK_SIZE] = (struct byte_meta){.stream_pts = pts};
}
2013-06-04 23:58:36 +00:00
s->max_filepos += len;
if (pos + len == s->buffer_size)
2013-06-04 23:58:36 +00:00
s->offset += s->buffer_size; // wrap...
s->eof = len > 0 ? 0 : 1;
s->idle = s->eof;
pthread_cond_signal(&s->wakeup);
return true;
}
static void update_cached_controls(struct priv *s)
2013-06-04 23:58:36 +00:00
{
unsigned int ui;
double d;
char **m;
s->stream_time_length = 0;
if (stream_control(s->stream, STREAM_CTRL_GET_TIME_LENGTH, &d) == STREAM_OK)
s->stream_time_length = d;
s->stream_start_time = MP_NOPTS_VALUE;
if (stream_control(s->stream, STREAM_CTRL_GET_START_TIME, &d) == STREAM_OK)
s->stream_start_time = d;
s->stream_manages_timeline = false;
if (stream_control(s->stream, STREAM_CTRL_MANAGES_TIMELINE, NULL) == STREAM_OK)
s->stream_manages_timeline = true;
s->stream_num_chapters = 0;
if (stream_control(s->stream, STREAM_CTRL_GET_NUM_CHAPTERS, &ui) == STREAM_OK)
s->stream_num_chapters = ui;
if (stream_control(s->stream, STREAM_CTRL_GET_METADATA, &m) == STREAM_OK) {
talloc_free(s->stream_metadata);
s->stream_metadata = talloc_steal(s, m);
}
stream_update_size(s->stream);
s->stream_size = s->stream->end_pos;
}
// the core might call these every frame, so cache them...
static int cache_get_cached_control(stream_t *cache, int cmd, void *arg)
{
struct priv *s = cache->priv;
switch (cmd) {
case STREAM_CTRL_GET_CACHE_SIZE:
*(int64_t *)arg = s->buffer_size;
return STREAM_OK;
case STREAM_CTRL_GET_CACHE_FILL:
*(int64_t *)arg = s->max_filepos - s->read_filepos;
return STREAM_OK;
case STREAM_CTRL_GET_CACHE_IDLE:
*(int *)arg = s->idle;
return STREAM_OK;
case STREAM_CTRL_GET_TIME_LENGTH:
*(double *)arg = s->stream_time_length;
return s->stream_time_length ? STREAM_OK : STREAM_UNSUPPORTED;
case STREAM_CTRL_GET_START_TIME:
*(double *)arg = s->stream_start_time;
return s->stream_start_time !=
MP_NOPTS_VALUE ? STREAM_OK : STREAM_UNSUPPORTED;
case STREAM_CTRL_GET_SIZE:
*(int64_t *)arg = s->stream_size;
return STREAM_OK;
case STREAM_CTRL_MANAGES_TIMELINE:
return s->stream_manages_timeline ? STREAM_OK : STREAM_UNSUPPORTED;
case STREAM_CTRL_GET_NUM_CHAPTERS:
*(unsigned int *)arg = s->stream_num_chapters;
return STREAM_OK;
case STREAM_CTRL_GET_CURRENT_TIME: {
if (s->read_filepos >= s->min_filepos &&
s->read_filepos <= s->max_filepos &&
s->min_filepos < s->max_filepos)
{
int64_t fpos = FFMIN(s->read_filepos, s->max_filepos - 1);
int64_t pos = fpos - s->offset;
if (pos < 0)
pos += s->buffer_size;
else if (pos >= s->buffer_size)
pos -= s->buffer_size;
double pts = s->bm[pos / BYTE_META_CHUNK_SIZE].stream_pts;
*(double *)arg = pts;
return pts == MP_NOPTS_VALUE ? STREAM_UNSUPPORTED : STREAM_OK;
}
return STREAM_UNSUPPORTED;
}
case STREAM_CTRL_GET_METADATA: {
if (s->stream_metadata && s->stream_metadata[0]) {
char **m = talloc_new(NULL);
int num_m = 0;
for (int n = 0; s->stream_metadata[n]; n++) {
char *t = talloc_strdup(m, s->stream_metadata[n]);
MP_TARRAY_APPEND(NULL, m, num_m, t);
}
MP_TARRAY_APPEND(NULL, m, num_m, NULL);
MP_TARRAY_APPEND(NULL, m, num_m, NULL);
*(char ***)arg = m;
return STREAM_OK;
}
return STREAM_UNSUPPORTED;
}
2013-06-04 23:58:36 +00:00
}
return STREAM_ERROR;
}
static bool control_needs_flush(int stream_ctrl)
2013-06-04 23:58:36 +00:00
{
switch (stream_ctrl) {
case STREAM_CTRL_SEEK_TO_TIME:
case STREAM_CTRL_SEEK_TO_CHAPTER:
case STREAM_CTRL_SET_ANGLE:
return true;
2013-06-04 23:58:36 +00:00
}
return false;
}
// Runs in the cache thread
static void cache_execute_control(struct priv *s)
2013-06-04 23:58:36 +00:00
{
uint64_t old_pos = stream_tell(s->stream);
s->control_res = stream_control(s->stream, s->control, s->control_arg);
s->control_flush = false;
bool pos_changed = old_pos != stream_tell(s->stream);
bool ok = s->control_res == STREAM_OK;
if (pos_changed && !ok) {
mp_msg(MSGT_STREAM, MSGL_ERR, "STREAM_CTRL changed stream pos but "
"returned error, this is not allowed!\n");
} else if (pos_changed || (ok && control_needs_flush(s->control))) {
mp_msg(MSGT_CACHE, MSGL_V, "Dropping cache due to control()\n");
s->read_filepos = stream_tell(s->stream);
s->control_flush = true;
cache_drop_contents(s);
2013-06-04 23:58:36 +00:00
}
s->control = CACHE_CTRL_NONE;
pthread_cond_signal(&s->wakeup);
}
static void *cache_thread(void *arg)
2013-06-04 23:58:36 +00:00
{
struct priv *s = arg;
pthread_mutex_lock(&s->mutex);
update_cached_controls(s);
double last = mp_time_sec();
while (s->control != CACHE_CTRL_QUIT) {
if (mp_time_sec() - last > CACHE_UPDATE_CONTROLS_TIME) {
update_cached_controls(s);
last = mp_time_sec();
}
if (s->control > 0) {
cache_execute_control(s);
} else {
cache_fill(s);
}
if (s->control == CACHE_CTRL_PING) {
pthread_cond_signal(&s->wakeup);
s->control = CACHE_CTRL_NONE;
}
if (s->idle && s->control == CACHE_CTRL_NONE)
cond_timed_wait(&s->wakeup, &s->mutex, CACHE_IDLE_SLEEP_TIME);
}
pthread_cond_signal(&s->wakeup);
pthread_mutex_unlock(&s->mutex);
mp_msg(MSGT_CACHE, MSGL_V, "Cache exiting...\n");
return NULL;
}
static int cache_fill_buffer(struct stream *cache, char *buffer, int max_len)
2013-06-04 23:58:36 +00:00
{
struct priv *s = cache->priv;
assert(s->cache_thread_running);
pthread_mutex_lock(&s->mutex);
2013-06-04 23:58:36 +00:00
if (cache->pos != s->read_filepos)
2013-06-04 23:58:36 +00:00
mp_msg(MSGT_CACHE, MSGL_ERR,
"!!! read_filepos differs !!! report this bug...\n");
int t = cache_read(s, buffer, max_len);
// wakeup the cache thread, possibly make it read more data ahead
pthread_cond_signal(&s->wakeup);
pthread_mutex_unlock(&s->mutex);
return t;
}
static int cache_seek(stream_t *cache, int64_t pos)
2013-06-04 23:58:36 +00:00
{
struct priv *s = cache->priv;
assert(s->cache_thread_running);
pthread_mutex_lock(&s->mutex);
2013-06-04 23:58:36 +00:00
mp_msg(MSGT_CACHE, MSGL_DBG2, "CACHE2_SEEK: 0x%" PRIX64 " <= 0x%" PRIX64
" (0x%" PRIX64 ") <= 0x%" PRIX64 " \n",
s->min_filepos, pos, s->read_filepos, s->max_filepos);
cache->pos = s->read_filepos = pos;
s->eof = false; // so that cache_read() will actually wait for new data
pthread_cond_signal(&s->wakeup);
pthread_mutex_unlock(&s->mutex);
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
return 1;
}
static int cache_control(stream_t *cache, int cmd, void *arg)
2013-06-04 23:58:36 +00:00
{
struct priv *s = cache->priv;
int r = STREAM_ERROR;
assert(cmd > 0);
pthread_mutex_lock(&s->mutex);
r = cache_get_cached_control(cache, cmd, arg);
if (r != STREAM_ERROR)
goto done;
mp_msg(MSGT_CACHE, MSGL_V, "[cache] blocking for STREAM_CTRL %d\n", cmd);
s->control = cmd;
s->control_arg = arg;
double retry = 0;
while (s->control != CACHE_CTRL_NONE) {
if (cache_wakeup_and_wait(s, &retry) == CACHE_INTERRUPTED) {
2013-06-04 23:58:36 +00:00
s->eof = 1;
r = STREAM_UNSUPPORTED;
goto done;
2013-06-04 23:58:36 +00:00
}
}
r = s->control_res;
if (s->control_flush) {
cache->pos = s->read_filepos;
cache->eof = 0;
cache->buf_pos = cache->buf_len = 0;
2013-06-04 23:58:36 +00:00
}
done:
pthread_mutex_unlock(&s->mutex);
return r;
}
static void cache_uninit(stream_t *cache)
{
struct priv *s = cache->priv;
if (s->cache_thread_running) {
mp_msg(MSGT_CACHE, MSGL_V, "Terminating cache...\n");
pthread_mutex_lock(&s->mutex);
s->control = CACHE_CTRL_QUIT;
pthread_cond_signal(&s->wakeup);
pthread_mutex_unlock(&s->mutex);
pthread_join(s->cache_thread, NULL);
2013-06-04 23:58:36 +00:00
}
pthread_mutex_destroy(&s->mutex);
pthread_cond_destroy(&s->wakeup);
free(s->buffer);
free(s->bm);
talloc_free(s);
}
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
// return 1 on success, 0 if the function was interrupted and -1 on error, or
// if the cache is disabled
int stream_cache_init(stream_t *cache, stream_t *stream, int64_t size,
int64_t min, int64_t seek_limit)
{
if (size < 1)
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
return -1;
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
mp_tmsg(MSGT_NETWORK, MSGL_INFO, "Cache size set to %" PRId64 " KiB\n",
size / 1024);
if (size > SIZE_MAX) {
mp_msg(MSGT_CACHE, MSGL_FATAL,
"Cache size larger than max. allocation size\n");
return -1;
}
struct priv *s = talloc_zero(NULL, struct priv);
//64kb min_size
s->fill_limit = FFMAX(16 * 1024, BYTE_META_CHUNK_SIZE * 2);
s->buffer_size = FFMAX(size, s->fill_limit * 4);
s->back_size = s->buffer_size / 2;
s->buffer = malloc(s->buffer_size);
s->bm = malloc((s->buffer_size / BYTE_META_CHUNK_SIZE + 2) *
sizeof(struct byte_meta));
if (!s->buffer || !s->bm) {
mp_msg(MSGT_CACHE, MSGL_ERR, "Failed to allocate cache buffer.\n");
free(s->buffer);
free(s->bm);
talloc_free(s);
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
return -1;
}
pthread_mutex_init(&s->mutex, NULL);
pthread_cond_init(&s->wakeup, NULL);
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
cache->priv = s;
s->cache = cache;
s->stream = stream;
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
cache->seek = cache_seek;
cache->fill_buffer = cache_fill_buffer;
cache->control = cache_control;
cache->close = cache_uninit;
s->seek_limit = seek_limit;
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
//make sure that we won't wait from cache_fill
//more data than it is allowed to fill
if (s->seek_limit > s->buffer_size - s->fill_limit)
s->seek_limit = s->buffer_size - s->fill_limit;
if (min > s->buffer_size - s->fill_limit)
min = s->buffer_size - s->fill_limit;
if (pthread_create(&s->cache_thread, NULL, cache_thread, s) != 0) {
mp_msg(MSGT_CACHE, MSGL_ERR, "Starting cache process/thread failed: %s.\n",
strerror(errno));
return -1;
}
s->cache_thread_running = true;
// wait until cache is filled at least prefill_init %
for (;;) {
if (stream_check_interrupt(0))
return 0;
int64_t fill;
int idle;
if (stream_control(s->cache, STREAM_CTRL_GET_CACHE_FILL, &fill) < 0)
break;
if (stream_control(s->cache, STREAM_CTRL_GET_CACHE_IDLE, &idle) < 0)
break;
mp_tmsg(MSGT_CACHE, MSGL_STATUS, "\rCache fill: %5.2f%% "
"(%" PRId64 " bytes) ", 100.0 * fill / s->buffer_size, fill);
if (fill >= min)
break;
if (idle)
break; // file is smaller than prefill size
// Wake up if the cache is done reading some data (or on timeout/abort)
pthread_mutex_lock(&s->mutex);
s->control = CACHE_CTRL_PING;
pthread_cond_signal(&s->wakeup);
cache_wakeup_and_wait(s, &(double){0});
pthread_mutex_unlock(&s->mutex);
}
mp_msg(MSGT_CACHE, MSGL_STATUS, "\n");
return 1;
cache: make the stream cache a proper stream that wraps other streams Before this commit, the cache was franken-hacked on top of the stream API. You had to use special functions (like cache_stream_fill_buffer() instead of stream_fill_buffer()), which would access the stream in a cached manner. The whole idea about the previous design was that the cache runs in a thread or in a forked process, while the cache awa functions made sure the stream instance looked consistent to the user. If you used the normal functions instead of the special ones while the cache was running, you were out of luck. Make it a bit more reasonable by turning the cache into a stream on its own. This makes it behave exactly like a normal stream. The stream callbacks call into the original (uncached) stream to do work. No special cache functions or redirections are needed. The only different thing about cache streams is that they are created by special functions, instead of being part of the auto_open_streams[] array. To make things simpler, remove the threading implementation, which was messed into the code. The threading code could perhaps be kept, but I don't really want to have to worry about this special case. A proper threaded implementation will be added later. Remove the cache enabling code from stream_radio.c. Since enabling the cache involves replacing the old stream with a new one, the code as-is can't be kept. It would be easily possible to enable the cache by requesting a cache size (which is also much simpler). But nobody uses stream_radio.c and I can't even test this thing, and the cache is probably not really important for it either.
2013-05-24 16:49:09 +00:00
}