mirror of
https://gitlab.freedesktop.org/pipewire/pipewire.git
synced 2025-11-06 13:30:01 -05:00
WIP: work on per driver graphs
Make a graph per driver node and let nodes that link to this join this graph Work on an idea for an even simpler scheduling mechanism.
This commit is contained in:
Wim Taymans
parent
2d77c4dc34
commit
933635f63f
17 changed files with 325 additions and 239 deletions
|
|
@ -36,15 +36,12 @@ static inline void spa_graph_data_init(struct spa_graph_data *data,
|
|||
data->graph = graph;
|
||||
}
|
||||
|
||||
static inline int spa_graph_impl_process(void *data, struct spa_graph_node *node)
|
||||
static inline int spa_graph_trigger(struct spa_graph *g, struct spa_graph_node *node)
|
||||
{
|
||||
struct spa_graph_data *d = (struct spa_graph_data *) data;
|
||||
struct spa_graph *g = d->graph;
|
||||
int old = node->state->status, res = 0;
|
||||
uint32_t val;
|
||||
|
||||
spa_debug("node %p: pending %d required %d %d", node,
|
||||
node->state->pending, node->state->required, old);
|
||||
spa_debug("node %p: pending %d required %d", node,
|
||||
node->state->pending, node->state->required);
|
||||
|
||||
if (node->state->pending == 0) {
|
||||
spa_debug("node %p: nothing pending", node);
|
||||
|
|
@ -52,25 +49,10 @@ static inline int spa_graph_impl_process(void *data, struct spa_graph_node *node
|
|||
}
|
||||
|
||||
val = __atomic_sub_fetch(&node->state->pending, 1, __ATOMIC_SEQ_CST);
|
||||
if (val == 0) {
|
||||
if (old == SPA_STATUS_NEED_BUFFER &&
|
||||
node->implementation->process_input) {
|
||||
res = spa_node_process_input(node->implementation);
|
||||
}
|
||||
else {
|
||||
res = spa_node_process_output(node->implementation);
|
||||
}
|
||||
spa_debug("node %p: process %d", node, res);
|
||||
if (val == 0)
|
||||
spa_graph_node_process(node);
|
||||
|
||||
if (res == SPA_STATUS_HAVE_BUFFER)
|
||||
spa_graph_have_output(g, node);
|
||||
|
||||
node->state->status = res;
|
||||
|
||||
spa_debug("node %p: end %d", node, res);
|
||||
}
|
||||
return node->state->status;
|
||||
|
||||
}
|
||||
|
||||
static inline int spa_graph_impl_need_input(void *data, struct spa_graph_node *node)
|
||||
|
|
@ -118,8 +100,9 @@ static inline int spa_graph_impl_need_input(void *data, struct spa_graph_node *n
|
|||
if (pn->sched_link.next != NULL)
|
||||
continue;
|
||||
|
||||
if (pp->io->status == SPA_STATUS_NEED_BUFFER) {
|
||||
pn->state->status = spa_node_process_output(pn->implementation);
|
||||
if (pp->io->status == SPA_STATUS_NEED_BUFFER &&
|
||||
pn->state->status == SPA_STATUS_HAVE_BUFFER) {
|
||||
pn->state->status = spa_graph_node_process(pn);
|
||||
} else {
|
||||
n->state->pending--;
|
||||
}
|
||||
|
|
@ -132,7 +115,7 @@ static inline int spa_graph_impl_need_input(void *data, struct spa_graph_node *n
|
|||
n->sched_link.next = NULL;
|
||||
|
||||
spa_debug("schedule node %p: %d", n, n->state->status);
|
||||
spa_graph_process(d->graph, n);
|
||||
spa_graph_trigger(d->graph, n);
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
|
|
@ -145,19 +128,22 @@ static inline int spa_graph_impl_have_output(void *data, struct spa_graph_node *
|
|||
spa_debug("node %p start push", node);
|
||||
|
||||
spa_list_for_each(p, &node->ports[SPA_DIRECTION_OUTPUT], link)
|
||||
spa_graph_process(d->graph, p->peer->node);
|
||||
spa_graph_trigger(d->graph, p->peer->node);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static inline int spa_graph_impl_run(void *data)
|
||||
{
|
||||
return 0;
|
||||
}
|
||||
|
||||
static const struct spa_graph_callbacks spa_graph_impl_default = {
|
||||
SPA_VERSION_GRAPH_CALLBACKS,
|
||||
.need_input = spa_graph_impl_need_input,
|
||||
.have_output = spa_graph_impl_have_output,
|
||||
.process = spa_graph_impl_process,
|
||||
};
|
||||
|
||||
|
||||
#ifdef __cplusplus
|
||||
} /* extern "C" */
|
||||
#endif
|
||||
|
|
|
|||
|
|
@ -26,158 +26,143 @@ extern "C" {
|
|||
|
||||
#include <spa/graph/graph.h>
|
||||
|
||||
static inline int spa_graph_scheduler_default(struct spa_graph_node *node)
|
||||
struct spa_graph_data {
|
||||
struct spa_graph *graph;
|
||||
};
|
||||
|
||||
static inline void spa_graph_data_init(struct spa_graph_data *data,
|
||||
struct spa_graph *graph)
|
||||
{
|
||||
int res;
|
||||
struct spa_node *n = node->user_data;
|
||||
|
||||
if (node->action == SPA_GRAPH_ACTION_IN)
|
||||
res = spa_node_process_input(n);
|
||||
else if (node->action == SPA_GRAPH_ACTION_OUT)
|
||||
res = spa_node_process_output(n);
|
||||
else
|
||||
res = -EBADF;
|
||||
|
||||
return res;
|
||||
data->graph = graph;
|
||||
}
|
||||
|
||||
static inline void spa_graph_port_check(struct spa_graph *graph, struct spa_graph_port *port)
|
||||
static inline int spa_graph_trigger(struct spa_graph *g, struct spa_graph_node *node)
|
||||
{
|
||||
struct spa_graph_node *node = port->node;
|
||||
uint32_t val;
|
||||
|
||||
if (port->io->status == SPA_STATUS_HAVE_BUFFER)
|
||||
node->ready++;
|
||||
spa_debug("node %p: pending %d required %d", node,
|
||||
node->state->pending, node->state->required);
|
||||
|
||||
spa_debug("port %p node %p check %d %d %d", port, node, port->io->status, node->ready, node->required);
|
||||
|
||||
if (node->required > 0 && node->ready == node->required) {
|
||||
node->action = SPA_GRAPH_ACTION_IN;
|
||||
if (node->ready_link.next == NULL)
|
||||
spa_list_append(&graph->ready, &node->ready_link);
|
||||
} else if (node->ready_link.next) {
|
||||
spa_list_remove(&node->ready_link);
|
||||
node->ready_link.next = NULL;
|
||||
if (node->state->pending == 0) {
|
||||
spa_debug("node %p: nothing pending", node);
|
||||
return node->state->status;
|
||||
}
|
||||
|
||||
val = __atomic_sub_fetch(&node->state->pending, 1, __ATOMIC_SEQ_CST);
|
||||
if (val == 0)
|
||||
spa_graph_node_process(node);
|
||||
|
||||
return node->state->status;
|
||||
}
|
||||
|
||||
static inline void spa_graph_node_update(struct spa_graph *graph, struct spa_graph_node *node) {
|
||||
struct spa_graph_port *p;
|
||||
|
||||
node->ready = 0;
|
||||
spa_list_for_each(p, &node->ports[SPA_DIRECTION_INPUT], link) {
|
||||
if (p->io->status == SPA_STATUS_OK && !(node->flags & SPA_GRAPH_NODE_FLAG_ASYNC))
|
||||
node->ready++;
|
||||
}
|
||||
spa_debug("node %p update %d ready", node, node->ready);
|
||||
}
|
||||
|
||||
static inline bool spa_graph_scheduler_iterate(struct spa_graph *graph)
|
||||
static inline int spa_graph_impl_need_input(void *data, struct spa_graph_node *node)
|
||||
{
|
||||
bool empty;
|
||||
struct spa_graph_port *p;
|
||||
struct spa_graph_node *n;
|
||||
int iter = 1;
|
||||
uint32_t state;
|
||||
#if 0
|
||||
struct spa_graph_data *d = (struct spa_graph_data *) data;
|
||||
struct spa_list queue, pending;
|
||||
struct spa_graph_node *n, *pn;
|
||||
struct spa_graph_port *p, *pp;
|
||||
|
||||
next:
|
||||
empty = spa_list_is_empty(&graph->ready);
|
||||
if (empty && !spa_list_is_empty(&graph->pending)) {
|
||||
spa_debug("copy pending");
|
||||
spa_list_insert_list(&graph->ready, &graph->pending);
|
||||
spa_list_init(&graph->pending);
|
||||
empty = false;
|
||||
}
|
||||
if (iter-- == 0 || empty)
|
||||
return !empty;
|
||||
|
||||
n = spa_list_first(&graph->ready, struct spa_graph_node, ready_link);
|
||||
spa_list_remove(&n->ready_link);
|
||||
n->ready_link.next = NULL;
|
||||
|
||||
spa_debug("node %p state %d", n, n->state);
|
||||
|
||||
switch (n->state) {
|
||||
case SPA_GRAPH_STATE_IN:
|
||||
case SPA_GRAPH_STATE_OUT:
|
||||
case SPA_GRAPH_STATE_END:
|
||||
if (n->state == SPA_GRAPH_STATE_END)
|
||||
n->state = SPA_GRAPH_STATE_OUT;
|
||||
|
||||
state = n->schedule(n);
|
||||
spa_debug("node %p schedule %d res %d", n, action, state);
|
||||
|
||||
if (n->state == SPA_GRAPH_STATE_IN && n == graph->node)
|
||||
break;
|
||||
|
||||
if (n->state != SPA_GRAPH_STATE_END) {
|
||||
spa_debug("node %p add ready for CHECK", n);
|
||||
if (state == SPA_STATUS_NEED_BUFFER)
|
||||
n->state = SPA_GRAPH_STATE_CHECK_IN;
|
||||
else if (state == SPA_STATUS_HAVE_BUFFER)
|
||||
n->state = SPA_GRAPH_STATE_CHECK_OUT;
|
||||
else if (state == SPA_STATUS_OK)
|
||||
n->state = SPA_GRAPH_STATE_CHECK_OK;
|
||||
spa_list_append(&graph->ready, &n->ready_link);
|
||||
}
|
||||
else {
|
||||
spa_graph_node_update(graph, n);
|
||||
}
|
||||
break;
|
||||
|
||||
case SPA_GRAPH_STATE_CHECK_IN:
|
||||
n->ready = 0;
|
||||
spa_list_for_each(p, &n->ports[SPA_DIRECTION_INPUT], link) {
|
||||
struct spa_graph_node *pn = p->peer->node;
|
||||
if (p->io->status == SPA_STATUS_NEED_BUFFER) {
|
||||
if (pn != graph->node
|
||||
|| pn->flags & SPA_GRAPH_NODE_FLAG_ASYNC) {
|
||||
pn->state = SPA_GRAPH_STATE_OUT;
|
||||
spa_debug("node %p add ready OUT", n);
|
||||
spa_list_append(&graph->ready, &pn->ready_link);
|
||||
}
|
||||
} else if (p->io->status == SPA_STATUS_OK)
|
||||
n->ready++;
|
||||
}
|
||||
break;
|
||||
|
||||
case SPA_GRAPH_STATE_CHECK_OUT:
|
||||
spa_list_for_each(p, &n->ports[SPA_DIRECTION_OUTPUT], link)
|
||||
spa_graph_port_check(graph, p->peer);
|
||||
|
||||
spa_debug("node %p add pending", n);
|
||||
n->state = SPA_GRAPH_STATE_END;
|
||||
spa_list_insert(&graph->pending, &n->ready_link);
|
||||
break;
|
||||
|
||||
case SPA_GRAPH_STATE_CHECK_OK:
|
||||
spa_graph_node_update(graph, n);
|
||||
break;
|
||||
|
||||
default:
|
||||
break;
|
||||
}
|
||||
goto next;
|
||||
}
|
||||
|
||||
static inline void spa_graph_scheduler_pull(struct spa_graph *graph, struct spa_graph_node *node)
|
||||
{
|
||||
node->action = SPA_GRAPH_ACTION_CHECK;
|
||||
node->state = SPA_STATUS_NEED_BUFFER;
|
||||
graph->node = node;
|
||||
spa_debug("node %p start pull", node);
|
||||
if (node->ready_link.next == NULL)
|
||||
spa_list_append(&graph->ready, &node->ready_link);
|
||||
|
||||
spa_list_init(&queue);
|
||||
spa_list_init(&pending);
|
||||
|
||||
node->state->status = SPA_STATUS_NEED_BUFFER;
|
||||
|
||||
if (node->sched_link.next == NULL)
|
||||
spa_list_append(&queue, &node->sched_link);
|
||||
|
||||
while (!spa_list_is_empty(&queue)) {
|
||||
|
||||
n = spa_list_first(&queue, struct spa_graph_node, sched_link);
|
||||
spa_list_remove(&n->sched_link);
|
||||
n->sched_link.next = NULL;
|
||||
|
||||
n->state->pending = n->state->required + 1;
|
||||
spa_debug("node %p: add %d %d status %d", n,
|
||||
n->state->pending, n->state->required,
|
||||
n->state->status);
|
||||
|
||||
spa_list_prepend(&pending, &n->sched_link);
|
||||
|
||||
if (n->state->status == SPA_STATUS_HAVE_BUFFER)
|
||||
continue;
|
||||
|
||||
spa_list_for_each(p, &n->ports[SPA_DIRECTION_INPUT], link) {
|
||||
pp = p->peer;
|
||||
if (pp == NULL)
|
||||
continue;
|
||||
pn = pp->node;
|
||||
|
||||
spa_debug("node %p: %p in io:%d state:%d %p", n, pn, pp->io->status,
|
||||
pn->state->status, pn->sched_link.next);
|
||||
|
||||
if (pn->sched_link.next != NULL)
|
||||
continue;
|
||||
|
||||
if (pp->io->status == SPA_STATUS_NEED_BUFFER &&
|
||||
pn->state->status == SPA_STATUS_HAVE_BUFFER) {
|
||||
pn->state->status = spa_graph_node_process(pn);
|
||||
} else {
|
||||
n->state->pending--;
|
||||
}
|
||||
spa_list_append(&queue, &pn->sched_link);
|
||||
}
|
||||
}
|
||||
while (!spa_list_is_empty(&pending)) {
|
||||
n = spa_list_first(&pending, struct spa_graph_node, sched_link);
|
||||
spa_list_remove(&n->sched_link);
|
||||
n->sched_link.next = NULL;
|
||||
|
||||
spa_debug("schedule node %p: %d", n, n->state->status);
|
||||
spa_graph_trigger(d->graph, n);
|
||||
}
|
||||
#endif
|
||||
return 0;
|
||||
}
|
||||
|
||||
static inline void spa_graph_scheduler_push(struct spa_graph *graph, struct spa_graph_node *node)
|
||||
static inline int spa_graph_impl_have_output(void *data, struct spa_graph_node *node)
|
||||
{
|
||||
node->action = SPA_GRAPH_ACTION_OUT;
|
||||
graph->node = node;
|
||||
struct spa_graph_data *d = (struct spa_graph_data *) data;
|
||||
struct spa_graph_port *p;
|
||||
|
||||
spa_debug("node %p start push", node);
|
||||
if (node->ready_link.next == NULL)
|
||||
spa_list_append(&graph->ready, &node->ready_link);
|
||||
|
||||
spa_list_for_each(p, &node->ports[SPA_DIRECTION_OUTPUT], link)
|
||||
if (p->peer)
|
||||
spa_graph_trigger(d->graph, p->peer->node);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static inline int spa_graph_impl_run(void *data)
|
||||
{
|
||||
struct spa_graph_data *d = (struct spa_graph_data *) data;
|
||||
struct spa_graph *g = d->graph;
|
||||
struct spa_graph_node *n;
|
||||
|
||||
spa_debug("graph %p run", d->graph);
|
||||
|
||||
spa_list_for_each(n, &g->nodes, link) {
|
||||
n->state->pending = n->state->required + 1;
|
||||
spa_debug("node %p: add %d %d status %d", n,
|
||||
n->state->pending, n->state->required,
|
||||
n->state->status);
|
||||
}
|
||||
spa_list_for_each(n, &g->nodes, link)
|
||||
spa_graph_trigger(d->graph, n);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static const struct spa_graph_callbacks spa_graph_impl_default = {
|
||||
SPA_VERSION_GRAPH_CALLBACKS,
|
||||
.need_input = spa_graph_impl_need_input,
|
||||
.have_output = spa_graph_impl_have_output,
|
||||
.run = spa_graph_impl_run,
|
||||
};
|
||||
|
||||
#ifdef __cplusplus
|
||||
} /* extern "C" */
|
||||
#endif
|
||||
|
|
|
|||
|
|
@ -43,9 +43,7 @@ struct spa_graph_callbacks {
|
|||
|
||||
int (*need_input) (void *data, struct spa_graph_node *node);
|
||||
int (*have_output) (void *data, struct spa_graph_node *node);
|
||||
int (*process) (void *data, struct spa_graph_node *node);
|
||||
int (*reuse_buffer) (void *data, struct spa_graph_node *node,
|
||||
uint32_t port_id, uint32_t buffer_id);
|
||||
int (*run) (void *data);
|
||||
};
|
||||
|
||||
struct spa_graph {
|
||||
|
|
@ -56,8 +54,7 @@ struct spa_graph {
|
|||
|
||||
#define spa_graph_need_input(g,n) ((g)->callbacks->need_input((g)->callbacks_data, (n)))
|
||||
#define spa_graph_have_output(g,n) ((g)->callbacks->have_output((g)->callbacks_data, (n)))
|
||||
#define spa_graph_process(g,n) ((g)->callbacks->process((g)->callbacks_data, (n)))
|
||||
#define spa_graph_reuse_buffer(g,n,p,i) ((g)->callbacks->reuse_buffer((g)->callbacks_data, (n),(p),(i)))
|
||||
#define spa_graph_run(g) ((g)->callbacks->run((g)->callbacks_data))
|
||||
|
||||
struct spa_graph_state {
|
||||
int status; /**< status of the node */
|
||||
|
|
@ -65,18 +62,31 @@ struct spa_graph_state {
|
|||
uint32_t pending; /**< number of input nodes pending */
|
||||
};
|
||||
|
||||
struct spa_graph_node_callbacks {
|
||||
#define SPA_VERSION_GRAPH_NODE_CALLBACKS 0
|
||||
uint32_t version;
|
||||
|
||||
int (*process) (void *data, struct spa_graph_node *node);
|
||||
int (*reuse_buffer) (void *data, struct spa_graph_node *node,
|
||||
uint32_t port_id, uint32_t buffer_id);
|
||||
};
|
||||
|
||||
struct spa_graph_node {
|
||||
struct spa_list link; /**< link in graph nodes list */
|
||||
struct spa_graph *graph; /**< owner graph */
|
||||
struct spa_list ports[2]; /**< list of input and output ports */
|
||||
#define SPA_GRAPH_NODE_FLAG_ASYNC (1 << 0)
|
||||
uint32_t flags; /**< node flags */
|
||||
struct spa_node *implementation;/**< node implementation */
|
||||
struct spa_graph_state *state; /**< state of the node */
|
||||
const struct spa_graph_node_callbacks *callbacks;
|
||||
void *callbacks_data;
|
||||
struct spa_list sched_link; /**< link for scheduler */
|
||||
void *scheduler_data; /**< scheduler private data */
|
||||
};
|
||||
|
||||
#define spa_graph_node_process(n) ((n)->callbacks->process((n)->callbacks_data, (n)))
|
||||
#define spa_graph_node_reuse_buffer(n,p,i) ((n)->callbacks->reuse_buffer((n)->callbacks_data, (n),(p),(i)))
|
||||
|
||||
struct spa_graph_port {
|
||||
struct spa_list link; /**< link in node port list */
|
||||
struct spa_graph_node *node; /**< owner node */
|
||||
|
|
@ -115,10 +125,12 @@ spa_graph_node_init(struct spa_graph_node *node, struct spa_graph_state *state)
|
|||
}
|
||||
|
||||
static inline void
|
||||
spa_graph_node_set_implementation(struct spa_graph_node *node,
|
||||
struct spa_node *implementation)
|
||||
spa_graph_node_set_callbacks(struct spa_graph_node *node,
|
||||
const struct spa_graph_node_callbacks *callbacks,
|
||||
void *callbacks_data)
|
||||
{
|
||||
node->implementation = implementation;
|
||||
node->callbacks = callbacks;
|
||||
node->callbacks_data = callbacks_data;
|
||||
}
|
||||
|
||||
static inline void
|
||||
|
|
@ -202,6 +214,45 @@ spa_graph_port_unlink(struct spa_graph_port *port)
|
|||
}
|
||||
}
|
||||
|
||||
static inline int spa_graph_node_impl_process(void *data, struct spa_graph_node *node)
|
||||
{
|
||||
struct spa_graph *g = node->graph;
|
||||
struct spa_node *n = data;
|
||||
//int old = node->state->status, res = 0;
|
||||
int res = 0;
|
||||
|
||||
// if (old == SPA_STATUS_NEED_BUFFER && n->process_input &&
|
||||
if (n->process_input &&
|
||||
!spa_list_is_empty(&node->ports[SPA_DIRECTION_INPUT]))
|
||||
res = spa_node_process_input(n);
|
||||
else
|
||||
res = spa_node_process_output(n);
|
||||
|
||||
spa_debug("node %p: process %d", node, res);
|
||||
|
||||
node->state->status = res;
|
||||
|
||||
if (res == SPA_STATUS_HAVE_BUFFER)
|
||||
spa_graph_have_output(g, node);
|
||||
|
||||
spa_debug("node %p: end %d", node, res);
|
||||
|
||||
return res;
|
||||
}
|
||||
|
||||
static inline int spa_graph_node_impl_reuse_buffer(void *data, struct spa_graph_node *node,
|
||||
uint32_t port_id, uint32_t buffer_id)
|
||||
{
|
||||
struct spa_node *n = data;
|
||||
return spa_node_port_reuse_buffer(n, port_id, buffer_id);
|
||||
}
|
||||
|
||||
static const struct spa_graph_node_callbacks spa_graph_node_impl_default = {
|
||||
SPA_VERSION_GRAPH_NODE_CALLBACKS,
|
||||
.process = spa_graph_node_impl_process,
|
||||
.reuse_buffer = spa_graph_node_impl_reuse_buffer,
|
||||
};
|
||||
|
||||
#ifdef __cplusplus
|
||||
} /* extern "C" */
|
||||
#endif
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue