#include "st_context.h"
#include "st_cb_bitmap.h"
#include "st_cb_perfmon.h"
+#include "st_util.h"
#include "util/bitset.h"
struct st_context *st = st_context(ctx);
struct st_perf_monitor_object *stm = st_perf_monitor_object(m);
struct pipe_context *pipe = st->pipe;
+ unsigned *batch = NULL;
unsigned num_active_counters = 0;
+ unsigned max_batch_counters = 0;
+ unsigned num_batch_counters = 0;
int gid, cid;
st_flush_bitmap_cache(st);
/* Determine the number of active counters. */
for (gid = 0; gid < ctx->PerfMonitor.NumGroups; gid++) {
const struct gl_perf_monitor_group *g = &ctx->PerfMonitor.Groups[gid];
+ const struct st_perf_monitor_group *stg = &st->perfmon[gid];
if (m->ActiveGroups[gid] > g->MaxActiveCounters) {
/* Maximum number of counters reached. Cannot start the session. */
}
num_active_counters += m->ActiveGroups[gid];
+ if (stg->has_batch)
+ max_batch_counters += m->ActiveGroups[gid];
}
if (!num_active_counters)
if (!stm->active_counters)
return false;
+ if (max_batch_counters) {
+ batch = CALLOC(max_batch_counters, sizeof(*batch));
+ if (!batch)
+ return false;
+ }
+
/* Create a query for each active counter. */
for (gid = 0; gid < ctx->PerfMonitor.NumGroups; gid++) {
const struct gl_perf_monitor_group *g = &ctx->PerfMonitor.Groups[gid];
const struct st_perf_monitor_group *stg = &st->perfmon[gid];
- BITSET_WORD tmp;
- BITSET_FOREACH_SET(cid, tmp, m->ActiveCounters[gid], g->NumCounters) {
+ BITSET_FOREACH_SET(cid, m->ActiveCounters[gid], g->NumCounters) {
const struct st_perf_monitor_counter *stc = &stg->counters[cid];
struct st_perf_counter_object *cntr =
&stm->active_counters[stm->num_active_counters];
- cntr->query = pipe->create_query(pipe, stc->query_type, 0);
cntr->id = cid;
cntr->group_id = gid;
+ if (stc->flags & PIPE_DRIVER_QUERY_FLAG_BATCH) {
+ cntr->batch_index = num_batch_counters;
+ batch[num_batch_counters++] = stc->query_type;
+ } else {
+ cntr->query = pipe->create_query(pipe, stc->query_type, 0);
+ if (!cntr->query)
+ goto fail;
+ }
++stm->num_active_counters;
}
}
+
+ /* Create the batch query. */
+ if (num_batch_counters) {
+ stm->batch_query = pipe->create_batch_query(pipe, num_batch_counters,
+ batch);
+ stm->batch_result = CALLOC(num_batch_counters, sizeof(stm->batch_result->batch[0]));
+ if (!stm->batch_query || !stm->batch_result)
+ goto fail;
+ }
+
+ FREE(batch);
return true;
+
+fail:
+ FREE(batch);
+ return false;
}
static void
FREE(stm->active_counters);
stm->active_counters = NULL;
stm->num_active_counters = 0;
+
+ if (stm->batch_query) {
+ pipe->destroy_query(pipe, stm->batch_query);
+ stm->batch_query = NULL;
+ }
+ FREE(stm->batch_result);
+ stm->batch_result = NULL;
}
static struct gl_perf_monitor_object *
/* Start the query for each active counter. */
for (i = 0; i < stm->num_active_counters; ++i) {
struct pipe_query *query = stm->active_counters[i].query;
- if (!pipe->begin_query(pipe, query))
+ if (query && !pipe->begin_query(pipe, query))
goto fail;
}
+
+ if (stm->batch_query && !pipe->begin_query(pipe, stm->batch_query))
+ goto fail;
+
return true;
fail:
/* Stop the query for each active counter. */
for (i = 0; i < stm->num_active_counters; ++i) {
struct pipe_query *query = stm->active_counters[i].query;
- pipe->end_query(pipe, query);
+ if (query)
+ pipe->end_query(pipe, query);
}
+
+ if (stm->batch_query)
+ pipe->end_query(pipe, stm->batch_query);
}
static void
for (i = 0; i < stm->num_active_counters; ++i) {
struct pipe_query *query = stm->active_counters[i].query;
union pipe_query_result result;
- if (!pipe->get_query_result(pipe, query, FALSE, &result)) {
+ if (query && !pipe->get_query_result(pipe, query, FALSE, &result)) {
/* The query is busy. */
return false;
}
}
+
+ if (stm->batch_query &&
+ !pipe->get_query_result(pipe, stm->batch_query, FALSE, stm->batch_result))
+ return false;
+
return true;
}
* active counter. The API allows counters to appear in any order.
*/
GLsizei offset = 0;
+ bool have_batch_query = false;
+
+ if (stm->batch_query)
+ have_batch_query = pipe->get_query_result(pipe, stm->batch_query, TRUE,
+ stm->batch_result);
/* Read query results for each active counter. */
for (i = 0; i < stm->num_active_counters; ++i) {
gid = cntr->group_id;
type = ctx->PerfMonitor.Groups[gid].Counters[cid].Type;
- if (!pipe->get_query_result(pipe, cntr->query, TRUE, &result))
- continue;
+ if (cntr->query) {
+ if (!pipe->get_query_result(pipe, cntr->query, TRUE, &result))
+ continue;
+ } else {
+ if (!have_batch_query)
+ continue;
+ result.batch[0] = stm->batch_result->batch[cntr->batch_index];
+ }
data[offset++] = gid;
data[offset++] = cid;
switch (type) {
case GL_UNSIGNED_INT64_AMD:
- *(uint64_t *)&data[offset] = result.u64;
+ memcpy(&data[offset], &result.u64, sizeof(uint64_t));
offset += sizeof(uint64_t) / sizeof(GLuint);
break;
case GL_UNSIGNED_INT:
- *(uint32_t *)&data[offset] = result.u32;
+ memcpy(&data[offset], &result.u32, sizeof(uint32_t));
offset += sizeof(uint32_t) / sizeof(GLuint);
break;
case GL_FLOAT:
case GL_PERCENTAGE_AMD:
- *(GLfloat *)&data[offset] = result.f;
+ memcpy(&data[offset], &result.f, sizeof(GLfloat));
offset += sizeof(GLfloat) / sizeof(GLuint);
break;
}
bool
-st_init_perfmon(struct st_context *st)
+st_have_perfmon(struct st_context *st)
{
- struct gl_perf_monitor_state *perfmon = &st->ctx->PerfMonitor;
+ struct pipe_screen *screen = st->pipe->screen;
+
+ if (!screen->get_driver_query_info || !screen->get_driver_query_group_info)
+ return false;
+
+ return screen->get_driver_query_group_info(screen, 0, NULL) != 0;
+}
+
+static void
+st_InitPerfMonitorGroups(struct gl_context *ctx)
+{
+ struct st_context *st = st_context(ctx);
+ struct gl_perf_monitor_state *perfmon = &ctx->PerfMonitor;
struct pipe_screen *screen = st->pipe->screen;
struct gl_perf_monitor_group *groups = NULL;
struct st_perf_monitor_group *stgroups = NULL;
int num_counters, num_groups;
int gid, cid;
- if (!screen->get_driver_query_info || !screen->get_driver_query_group_info)
- return false;
-
/* Get the number of available queries. */
num_counters = screen->get_driver_query_info(screen, 0, NULL);
- if (!num_counters)
- return false;
/* Get the number of available groups. */
num_groups = screen->get_driver_query_group_info(screen, 0, NULL);
- if (num_groups)
- groups = CALLOC(num_groups, sizeof(*groups));
+ groups = CALLOC(num_groups, sizeof(*groups));
if (!groups)
- return false;
+ return;
stgroups = CALLOC(num_groups, sizeof(*stgroups));
if (!stgroups)
for (gid = 0; gid < num_groups; gid++) {
struct gl_perf_monitor_group *g = &groups[perfmon->NumGroups];
+ struct st_perf_monitor_group *stg = &stgroups[perfmon->NumGroups];
struct pipe_driver_query_group_info group_info;
struct gl_perf_monitor_counter *counters = NULL;
struct st_perf_monitor_counter *stcounters = NULL;
stcounters = CALLOC(group_info.num_queries, sizeof(*stcounters));
if (!stcounters)
goto fail;
- stgroups[perfmon->NumGroups].counters = stcounters;
+ stg->counters = stcounters;
for (cid = 0; cid < num_counters; cid++) {
struct gl_perf_monitor_counter *c = &counters[g->NumCounters];
case PIPE_DRIVER_QUERY_TYPE_MICROSECONDS:
case PIPE_DRIVER_QUERY_TYPE_HZ:
c->Minimum.u64 = 0;
- c->Maximum.u64 = info.max_value.u64 ? info.max_value.u64 : -1;
+ c->Maximum.u64 = info.max_value.u64 ? info.max_value.u64 : UINT64_MAX;
c->Type = GL_UNSIGNED_INT64_AMD;
break;
case PIPE_DRIVER_QUERY_TYPE_UINT:
c->Minimum.u32 = 0;
- c->Maximum.u32 = info.max_value.u32 ? info.max_value.u32 : -1;
+ c->Maximum.u32 = info.max_value.u32 ? info.max_value.u32 : UINT32_MAX;
c->Type = GL_UNSIGNED_INT;
break;
case PIPE_DRIVER_QUERY_TYPE_FLOAT:
c->Minimum.f = 0.0;
- c->Maximum.f = info.max_value.f ? info.max_value.f : -1;
+ c->Maximum.f = info.max_value.f ? info.max_value.f : FLT_MAX;
c->Type = GL_FLOAT;
break;
case PIPE_DRIVER_QUERY_TYPE_PERCENTAGE:
}
stc->query_type = info.query_type;
+ stc->flags = info.flags;
+ if (stc->flags & PIPE_DRIVER_QUERY_FLAG_BATCH)
+ stg->has_batch = true;
g->NumCounters++;
}
perfmon->Groups = groups;
st->perfmon = stgroups;
- return true;
+ return;
fail:
for (gid = 0; gid < num_groups; gid++) {
FREE(stgroups);
fail_only_groups:
FREE(groups);
- return false;
}
void
void st_init_perfmon_functions(struct dd_function_table *functions)
{
+ functions->InitPerfMonitorGroups = st_InitPerfMonitorGroups;
functions->NewPerfMonitor = st_NewPerfMonitor;
functions->DeletePerfMonitor = st_DeletePerfMonitor;
functions->BeginPerfMonitor = st_BeginPerfMonitor;