[asterisk-commits] dvossel: branch 1.8 r282047 - in /branches/1.8: ./ include/asterisk/ main/
SVN commits to the Asterisk project
asterisk-commits at lists.digium.com
Thu Aug 12 15:15:46 CDT 2010
Author: dvossel
Date: Thu Aug 12 15:15:41 2010
New Revision: 282047
URL: http://svnview.digium.com/svn/asterisk?view=rev&rev=282047
Log:
improved translation paths for wideband codecs
The problem I'm addressing is that Asterisk's current
method of building the least cost translation paths
between codecs does not take into account sample rate.
For instance, it was possible for siren14 (a 32khz codec),
to contain the a translation path to siren7 (a 16khz
audio codec) that goes through slin at 8khz. In this
case Asterisk takes a 32khz codec, down samples it to
8khz and then up samples it to 16khz which is terrible
regardless if it is computationally less expensive. This
patch now builds translation paths that give priority to
maintaining the best possible sample rate before taking
into consideration computational cost. This patch also
adds cli commands to expose what translation paths are
actually being used.
Changes:
1. Translation paths will never contain a step that changes
the sample rate unless absolutely necessary.
2. When choosing the best codec to make two channels compatible.
Shared codecs with the highest sample rate are given priority.
3. A new cli command to show all translation paths available
for a specific codec 'core show translation paths [codec name]'
has been added.
4. 'core show translation' which displays the translation
matrix now includes the new higher bit audio codecs in the table.
5. 'core show channel [channel name]' now displays the
translation paths if translation is used.
(closes issue #16841)
Reported by: dvossel
Review: https://reviewboard.asterisk.org/r/842/
Modified:
branches/1.8/CHANGES
branches/1.8/include/asterisk/translate.h
branches/1.8/main/cli.c
branches/1.8/main/translate.c
Modified: branches/1.8/CHANGES
URL: http://svnview.digium.com/svn/asterisk/branches/1.8/CHANGES?view=diff&rev=282047&r1=282046&r2=282047
==============================================================================
--- branches/1.8/CHANGES (original)
+++ branches/1.8/CHANGES Thu Aug 12 15:15:41 2010
@@ -535,6 +535,8 @@
* The UNISTIM channel driver (chan_unistim) has been updated to support devices that
have less than 3 lines on the LCD.
* Realtime now supports database failover. See the sample extconfig.conf for details.
+ * The addition of improved translation path building for wideband codecs. Sample
+ rate changes during translation are now avoided unless absolutely necessary.
CLI Changes
-----------
Modified: branches/1.8/include/asterisk/translate.h
URL: http://svnview.digium.com/svn/asterisk/branches/1.8/include/asterisk/translate.h?view=diff&rev=282047&r1=282046&r2=282047
==============================================================================
--- branches/1.8/include/asterisk/translate.h (original)
+++ branches/1.8/include/asterisk/translate.h Thu Aug 12 15:15:41 2010
@@ -254,6 +254,14 @@
*/
format_t ast_translate_available_formats(format_t dest, format_t src);
+/*!
+ * \brief Puts a string representation of the translation path into outbuf
+ * \param translator structure containing the translation path
+ * \param ast_str output buffer
+ * \retval on success pointer to beginning of outbuf. on failure "".
+ */
+const char *ast_translate_path_to_str(struct ast_trans_pvt *t, struct ast_str **str);
+
#if defined(__cplusplus) || defined(c_plusplus)
}
#endif
Modified: branches/1.8/main/cli.c
URL: http://svnview.digium.com/svn/asterisk/branches/1.8/main/cli.c?view=diff&rev=282047&r1=282046&r2=282047
==============================================================================
--- branches/1.8/main/cli.c (original)
+++ branches/1.8/main/cli.c Thu Aug 12 15:15:41 2010
@@ -46,6 +46,7 @@
#include "asterisk/lock.h"
#include "editline/readline/readline.h"
#include "asterisk/threadstorage.h"
+#include "asterisk/translate.h"
/*!
* \brief List of restrictions per user.
@@ -1343,6 +1344,8 @@
struct ast_str *out = ast_str_thread_get(&ast_str_thread_global_buf, 16);
char cdrtime[256];
char nf[256], wf[256], rf[256];
+ struct ast_str *write_transpath = ast_str_alloca(256);
+ struct ast_str *read_transpath = ast_str_alloca(256);
long elapsed_seconds=0;
int hour=0, min=0, sec=0;
#ifdef CHANNEL_TRACE
@@ -1398,8 +1401,8 @@
" NativeFormats: %s\n"
" WriteFormat: %s\n"
" ReadFormat: %s\n"
- " WriteTranscode: %s\n"
- " ReadTranscode: %s\n"
+ " WriteTranscode: %s %s\n"
+ " ReadTranscode: %s %s\n"
"1st File Descriptor: %d\n"
" Frames in: %d%s\n"
" Frames out: %d%s\n"
@@ -1426,7 +1429,9 @@
ast_getformatname_multiple(wf, sizeof(wf), c->writeformat),
ast_getformatname_multiple(rf, sizeof(rf), c->readformat),
c->writetrans ? "Yes" : "No",
+ ast_translate_path_to_str(c->writetrans, &write_transpath),
c->readtrans ? "Yes" : "No",
+ ast_translate_path_to_str(c->readtrans, &read_transpath),
c->fds[0],
c->fin & ~DEBUGCHAN_FLAG, (c->fin & DEBUGCHAN_FLAG) ? " (DEBUGGED)" : "",
c->fout & ~DEBUGCHAN_FLAG, (c->fout & DEBUGCHAN_FLAG) ? " (DEBUGGED)" : "",
Modified: branches/1.8/main/translate.c
URL: http://svnview.digium.com/svn/asterisk/branches/1.8/main/translate.c?view=diff&rev=282047&r1=282046&r2=282047
==============================================================================
--- branches/1.8/main/translate.c (original)
+++ branches/1.8/main/translate.c Thu Aug 12 15:15:41 2010
@@ -45,10 +45,23 @@
/*! \brief the list of translators */
static AST_RWLIST_HEAD_STATIC(translators, ast_translator);
+
+/*! \brief these values indicate how a translation path will affect the sample rate
+ *
+ * \note These must stay in this order. They are ordered by most optimal selection first.
+ */
+enum path_samp_change {
+ RATE_CHANGE_NONE = 0, /*!< path uses the same sample rate consistently */
+ RATE_CHANGE_UPSAMP = 1, /*!< path will up the sample rate during a translation */
+ RATE_CHANGE_DOWNSAMP = 2, /*!< path will have to down the sample rate during a translation. */
+ RATE_CHANGE_UPSAMP_DOWNSAMP = 3, /*!< path will both up and down the sample rate during translation */
+};
+
struct translator_path {
struct ast_translator *step; /*!< Next step translator */
unsigned int cost; /*!< Complete cost to destination */
unsigned int multistep; /*!< Multiple conversions required for this translation */
+ enum path_samp_change rate_change; /*!< does this path require a sample rate change, if so what kind. */
};
/*! \brief a matrix that, for any pair of supported formats,
@@ -402,6 +415,24 @@
t->cost = 1;
}
+static enum path_samp_change get_rate_change_result(format_t src, format_t dst)
+{
+ int src_rate = ast_format_rate(src);
+ int dst_rate = ast_format_rate(dst);
+
+ /* if src rate is less than dst rate, a sample upgrade is required */
+ if (src_rate < dst_rate) {
+ return RATE_CHANGE_UPSAMP;
+ }
+
+ /* if src rate is larger than dst rate, a downgrade is required */
+ if (src_rate > dst_rate) {
+ return RATE_CHANGE_DOWNSAMP;
+ }
+
+ return RATE_CHANGE_NONE;
+}
+
/*!
* \brief rebuild a translation matrix.
* \note This function expects the list of translators to be locked
@@ -409,6 +440,8 @@
static void rebuild_matrix(int samples)
{
struct ast_translator *t;
+ int new_rate_change;
+ int newcost;
int x; /* source format index */
int y; /* intermediate format index */
int z; /* destination format index */
@@ -427,10 +460,21 @@
if (samples)
calc_cost(t, samples);
-
- if (!tr_matrix[x][z].step || t->cost < tr_matrix[x][z].cost) {
+
+ new_rate_change = get_rate_change_result(1LL << t->srcfmt, 1LL << t->dstfmt);
+
+ /* this translator is the best choice if any of the below are true.
+ * 1. no translation path is set between x and z yet.
+ * 2. the new translation costs less and sample rate is no worse than old one.
+ * 3. the new translation has a better sample rate conversion than the old one.
+ */
+ if (!tr_matrix[x][z].step ||
+ ((t->cost < tr_matrix[x][z].cost) && (new_rate_change <= tr_matrix[x][z].rate_change)) ||
+ (new_rate_change < tr_matrix[x][z].rate_change)) {
+
tr_matrix[x][z].step = t;
tr_matrix[x][z].cost = t->cost;
+ tr_matrix[x][z].rate_change = new_rate_change;
}
}
@@ -442,31 +486,73 @@
*/
for (;;) {
int changed = 0;
+ int better_choice = 0;
for (x = 0; x < MAX_FORMAT; x++) { /* source format */
for (y = 0; y < MAX_FORMAT; y++) { /* intermediate format */
if (x == y) /* skip ourselves */
continue;
-
- for (z = 0; z<MAX_FORMAT; z++) { /* dst format */
- int newcost;
-
+ for (z = 0; z < MAX_FORMAT; z++) { /* dst format */
if (z == x || z == y) /* skip null conversions */
continue;
if (!tr_matrix[x][y].step) /* no path from x to y */
continue;
if (!tr_matrix[y][z].step) /* no path from y to z */
continue;
+
+ /* Does x->y->z result in a less optimal sample rate change?
+ * Never downgrade the sample rate conversion quality regardless
+ * of any cost improvements */
+ if (tr_matrix[x][z].step &&
+ ((tr_matrix[x][z].rate_change < tr_matrix[x][y].rate_change) ||
+ (tr_matrix[x][z].rate_change < tr_matrix[y][z].rate_change))) {
+ continue;
+ }
+
+ /* is x->y->z a better sample rate confersion that the current x->z? */
+ new_rate_change = tr_matrix[x][y].rate_change + tr_matrix[y][z].rate_change;
+
+ /* calculate cost from x->y->z */
newcost = tr_matrix[x][y].cost + tr_matrix[y][z].cost;
- if (tr_matrix[x][z].step && newcost >= tr_matrix[x][z].cost)
- continue; /* x->y->z is more expensive than
- * the existing path */
+
+ /* Is x->y->z a better choice than x->z?
+ * There are three conditions for x->y->z to be a better choice than x->z
+ * 1. if there is no step directly between x->z then x->y->z is the best and only current option.
+ * 2. if x->y->z costs less and the sample rate conversion is no less optimal.
+ * 3. if x->y->z results in a more optimal sample rate conversion. */
+ if (!tr_matrix[x][z].step) {
+ better_choice = 1;
+ } else if ((newcost < tr_matrix[x][z].cost) && (new_rate_change <= tr_matrix[x][z].rate_change)) {
+ better_choice = 1;
+ } else if (new_rate_change < tr_matrix[x][z].rate_change) {
+ better_choice = 1;
+ } else {
+ better_choice = 0;
+ }
+
+ if (!better_choice) {
+ continue;
+ }
/* ok, we can get from x to z via y with a cost that
- is the sum of the transition from x to y and
- from y to z */
-
+ is the sum of the transition from x to y and from y to z */
tr_matrix[x][z].step = tr_matrix[x][y].step;
tr_matrix[x][z].cost = newcost;
tr_matrix[x][z].multistep = 1;
+
+ /* now calculate what kind of sample rate change is required for this multi-step path
+ *
+ * if both paths require a change in rate, and they are not in the same direction
+ * then this is a up sample down sample conversion scenario. */
+ if ((tr_matrix[x][y].rate_change > RATE_CHANGE_NONE) &&
+ (tr_matrix[y][z].rate_change > RATE_CHANGE_NONE) &&
+ (tr_matrix[x][y].rate_change != tr_matrix[y][z].rate_change)) {
+
+ tr_matrix[x][z].rate_change = RATE_CHANGE_UPSAMP_DOWNSAMP;
+ } else {
+ /* else just set the rate change to whichever is worse */
+ tr_matrix[x][z].rate_change = tr_matrix[x][y].rate_change > tr_matrix[y][z].rate_change
+ ? tr_matrix[x][y].rate_change : tr_matrix[y][z].rate_change;
+ }
+
ast_debug(3, "Discovered %d cost path from %s to %s, via %s\n", tr_matrix[x][z].cost,
ast_getformatname(1LL << x), ast_getformatname(1LL << z), ast_getformatname(1LL << y));
changed++;
@@ -478,30 +564,134 @@
}
}
+const char *ast_translate_path_to_str(struct ast_trans_pvt *p, struct ast_str **str)
+{
+ struct ast_trans_pvt *pn = p;
+
+ if (!p || !p->t) {
+ return "";
+ }
+
+ ast_str_set(str, 0, "%s", ast_getformatname(1LL << p->t->srcfmt));
+
+ while ( (p = pn) ) {
+ pn = p->next;
+ ast_str_append(str, 0, "->%s", ast_getformatname(1LL << p->t->dstfmt));
+ }
+
+ return ast_str_buffer(*str);
+}
+
+static char *complete_trans_path_choice(const char *line, const char *word, int pos, int state)
+{
+ int which = 0;
+ int wordlen = strlen(word);
+ int i;
+ char *ret = NULL;
+ size_t len = 0;
+ const struct ast_format_list *format_list = ast_get_format_list(&len);
+
+ for (i = 0; i < len; i++) {
+ if (!(format_list[i].bits & AST_FORMAT_AUDIO_MASK)) {
+ continue;
+ }
+ if (!strncasecmp(word, format_list[i].name, wordlen) && ++which > state) {
+ ret = ast_strdup(format_list[i].name);
+ break;
+ }
+ }
+ return ret;
+}
+
static char *handle_cli_core_show_translation(struct ast_cli_entry *e, int cmd, struct ast_cli_args *a)
{
-#define SHOW_TRANS 16
+#define SHOW_TRANS 64
+ static const char * const option1[] = { "recalc", "paths", NULL };
int x, y, z;
int curlen = 0, longest = 0, magnitude[SHOW_TRANS] = { 0, };
switch (cmd) {
case CLI_INIT:
- e->command = "core show translation [recalc]";
+ e->command = "core show translation";
e->usage =
- "Usage: core show translation [recalc [<recalc seconds>]]\n"
- " Displays known codec translators and the cost associated\n"
- " with each conversion. If the argument 'recalc' is supplied along\n"
- " with optional number of seconds to test a new test will be performed\n"
- " as the chart is being displayed.\n";
+ "Usage: 'core show translation' can be used in two ways.\n"
+ " 1. 'core show translation [recalc [<recalc seconds>]]\n"
+ " Displays known codec translators and the cost associated\n"
+ " with each conversion. If the argument 'recalc' is supplied along\n"
+ " with optional number of seconds to test a new test will be performed\n"
+ " as the chart is being displayed.\n"
+ " 2. 'core show translation paths [codec]'\n"
+ " This will display all the translation paths associated with a codec\n";
return NULL;
case CLI_GENERATE:
+ if (a->pos == 3) {
+ return ast_cli_complete(a->word, option1, a->n);
+ }
+ if (a->pos == 4 && !strcasecmp(a->argv[3], option1[1])) {
+ return complete_trans_path_choice(a->line, a->word, a->pos, a->n);
+ }
return NULL;
}
if (a->argc > 5)
return CLI_SHOWUSAGE;
- if (a->argv[3] && !strcasecmp(a->argv[3], "recalc")) {
+ if (a->argv[3] && !strcasecmp(a->argv[3], option1[1]) && a->argc == 5) {
+ format_t input_src = 0;
+ format_t src = 0;
+ size_t len = 0;
+ int dst;
+ int i;
+ const struct ast_format_list *format_list = ast_get_format_list(&len);
+ struct ast_str *str = ast_str_alloca(256);
+ struct ast_translator *step;
+
+ for (i = 0; i < len; i++) {
+ if (!(format_list[i].bits & AST_FORMAT_AUDIO_MASK)) {
+ continue;
+ }
+ if (!strncasecmp(format_list[i].name, a->argv[4], strlen(format_list[i].name))) {
+ input_src = format_list[i].bits;
+ }
+ }
+
+ if (!input_src) {
+ ast_cli(a->fd, "Source codec \"%s\" is not found.\n", a->argv[4]);
+ return CLI_FAILURE;
+ }
+
+ AST_RWLIST_RDLOCK(&translators);
+ ast_cli(a->fd, "--- Translation paths SRC Codec \"%s\" sample rate %d ---\n", a->argv[4], ast_format_rate(input_src));
+ for (i = 0; i < len; i++) {
+ if (!(format_list[i].bits & AST_FORMAT_AUDIO_MASK) || (format_list[i].bits == input_src)) {
+ continue;
+ }
+ dst = powerof(format_list[i].bits);
+ src = powerof(input_src);
+ ast_str_reset(str);
+ if (tr_matrix[src][dst].step) {
+ ast_str_append(&str, 0, "%s", ast_getformatname(1LL << tr_matrix[src][dst].step->srcfmt));
+ while (src != dst) {
+ step = tr_matrix[src][dst].step;
+ if (!step) {
+ ast_str_reset(str);
+ break;
+ }
+ ast_str_append(&str, 0, "->%s", ast_getformatname(1LL << step->dstfmt));
+ src = step->dstfmt;
+ }
+ }
+
+ if (ast_strlen_zero(ast_str_buffer(str))) {
+ ast_str_set(&str, 0, "No Translation Path");
+ }
+
+ ast_cli(a->fd, "\t%-10.10s To %-10.10s: %-60.60s\n", a->argv[4], format_list[i].name, ast_str_buffer(str));
+ }
+ AST_RWLIST_UNLOCK(&translators);
+
+ return CLI_SUCCESS;
+ } else if (a->argv[3] && !strcasecmp(a->argv[3], "recalc")) {
z = a->argv[4] ? atoi(a->argv[4]) : 1;
if (z <= 0) {
@@ -526,22 +716,33 @@
ast_cli(a->fd, " Source Format (Rows) Destination Format (Columns)\n\n");
/* Get the length of the longest (usable?) codec name, so we know how wide the left side should be */
for (x = 0; x < SHOW_TRANS; x++) {
+ /* translation only applies to audio right now. */
+ if (!(AST_FORMAT_AUDIO_MASK & (1LL << (x))))
+ continue;
curlen = strlen(ast_getformatname(1LL << (x)));
if (curlen > longest)
longest = curlen;
for (y = 0; y < SHOW_TRANS; y++) {
+ if (!(AST_FORMAT_AUDIO_MASK & (1LL << (y))))
+ continue;
if (tr_matrix[x][y].cost > pow(10, magnitude[x])) {
magnitude[y] = floor(log10(tr_matrix[x][y].cost));
}
}
}
for (x = -1; x < SHOW_TRANS; x++) {
- struct ast_str *out = ast_str_alloca(125);
+ struct ast_str *out = ast_str_alloca(256);
+ /* translation only applies to audio right now. */
+ if (x >= 0 && !(AST_FORMAT_AUDIO_MASK & (1LL << (x))))
+ continue;
/*Go ahead and move to next iteration if dealing with an unknown codec*/
if(x >= 0 && !strcmp(ast_getformatname(1LL << (x)), "unknown"))
continue;
ast_str_set(&out, -1, " ");
for (y = -1; y < SHOW_TRANS; y++) {
+ /* translation only applies to audio right now. */
+ if (y >= 0 && !(AST_FORMAT_AUDIO_MASK & (1LL << (y))))
+ continue;
/*Go ahead and move to next iteration if dealing with an unknown codec*/
if (y >= 0 && !strcmp(ast_getformatname(1LL << (y)), "unknown"))
continue;
@@ -715,37 +916,65 @@
format_t ast_translator_best_choice(format_t *dst, format_t *srcs)
{
int x,y;
+ int better = 0;
+ int besttime = INT_MAX;
+ int beststeps = INT_MAX;
+ unsigned int best_rate_change = INT_MAX;
format_t best = -1;
format_t bestdst = 0;
format_t cur, cursrc;
- int besttime = INT_MAX;
- int beststeps = INT_MAX;
format_t common = ((*dst) & (*srcs)) & AST_FORMAT_AUDIO_MASK; /* are there common formats ? */
if (common) { /* yes, pick one and return */
for (cur = 1, y = 0; y <= MAX_AUDIO_FORMAT; cur <<= 1, y++) {
- if (cur & common) /* guaranteed to find one */
- break;
+ if (!(cur & common)) {
+ continue;
+ }
+
+ /* We are guaranteed to find one common format. */
+ if (best == -1) {
+ best = cur;
+ continue;
+ }
+ /* If there are multiple common formats, pick the one with the highest sample rate */
+ if (ast_format_rate(best) < ast_format_rate(cur)) {
+ best = cur;
+ continue;
+ }
}
/* We are done, this is a common format to both. */
- *srcs = *dst = cur;
+ *srcs = *dst = best;
return 0;
- } else { /* No, we will need to translate */
+ } else { /* No, we will need to translate */
AST_RWLIST_RDLOCK(&translators);
for (cur = 1, y = 0; y <= MAX_AUDIO_FORMAT; cur <<= 1, y++) {
- if (! (cur & *dst))
+ if (! (cur & *dst)) {
continue;
+ }
for (cursrc = 1, x = 0; x <= MAX_AUDIO_FORMAT; cursrc <<= 1, x++) {
- if (!(*srcs & cursrc) || !tr_matrix[x][y].step ||
- tr_matrix[x][y].cost > besttime)
- continue; /* not existing or no better */
- if (tr_matrix[x][y].cost < besttime ||
- tr_matrix[x][y].multistep < beststeps) {
+ if (!(*srcs & cursrc) || !tr_matrix[x][y].step) {
+ continue;
+ }
+
+ /* This is a better choice if any of the following are true.
+ * 1. The sample rate conversion is better than the current pick.
+ * 2. the sample rate conversion is no worse than the current pick and the cost or multistep is better
+ */
+ better = 0;
+ if (tr_matrix[x][y].rate_change < best_rate_change) {
+ better = 1; /* this match has a better rate conversion */
+ }
+ if ((tr_matrix[x][y].rate_change <= best_rate_change) &&
+ (tr_matrix[x][y].cost < besttime || tr_matrix[x][y].multistep < beststeps)) {
+ better = 1; /* this match has no worse rate conversion and the conversion cost is less */
+ }
+ if (better) {
/* better than what we have so far */
best = cursrc;
bestdst = cur;
besttime = tr_matrix[x][y].cost;
beststeps = tr_matrix[x][y].multistep;
+ best_rate_change = tr_matrix[x][y].rate_change;
}
}
}
More information about the asterisk-commits
mailing list