#include "fips-dispatch-gl.h"
#include "metrics.h"
+#include "context.h"
+#include "metrics-info.h"
#include "xmalloc.h"
+int frames;
+int verbose;
+
+#define MAX_MONITORS_IN_FLIGHT 1000
+
/* Timer query */
typedef struct timer_query
{
struct monitor *next;
} monitor_t;
-
typedef struct op_metrics
{
/* This happens to also be the index into the
- * ctx->op_metrics array currently
+ * metrics->op_metrics array currently
*/
metrics_op_t op;
double time_ns;
- double *counters;
- unsigned num_counters;
+ double **counters;
} op_metrics_t;
-typedef struct counter_group_info
-{
- GLuint id;
- GLint num_counters;
- GLint max_active_counters;
- GLuint *counters;
-} counter_group_info_t;
-
-typedef struct metrics_info
-{
- int num_groups;
- int max_counters_per_group;
- counter_group_info_t *groups;
-} metrics_info_t;
-
-typedef struct context
+struct metrics
{
- metrics_info_t metrics_info;
+ /* Description of all available peformance counters, counter
+ * groups, their names and IDs, etc. */
+ metrics_info_t *info;
+ /* The current operation being measured. */
metrics_op_t op;
+ /* GL_TIME_ELAPSED query for which glEndQuery has not yet
+ * been called. */
+ unsigned timer_begun_id;
+
+ /* GL_TIME_ELAPSED queries for which glEndQuery has been
+ * called, (but results have not yet been queried). */
timer_query_t *timer_head;
timer_query_t *timer_tail;
+ /* Performance monitor for which glEndPerfMonitorAMD has not
+ * yet been called. */
+ unsigned monitor_begun_id;
+
+ /* Performance monitors for which glEndPerfMonitorAMD has
+ * been called, (but results have not yet been queried). */
monitor_t *monitor_head;
monitor_t *monitor_tail;
+ int monitors_in_flight;
+
unsigned num_op_metrics;
op_metrics_t *op_metrics;
-} context_t;
+};
-/* FIXME: Need a map from integers to context objects and track the
- * current context with glXMakeContextCurrent, eglMakeCurrent, etc. */
+metrics_t *
+metrics_create (metrics_info_t *info)
+{
+ metrics_t *metrics;
-context_t current_context;
+ metrics = xmalloc (sizeof (metrics_t));
-int frames;
-int verbose;
+ metrics->info = info;
-void
-metrics_info_init (void)
-{
- int i;
- GLuint *group_ids;
- metrics_info_t *metrics_info = ¤t_context.metrics_info;
+ metrics->op = 0;
- glGetPerfMonitorGroupsAMD (&metrics_info->num_groups, 0, NULL);
+ metrics->timer_begun_id = 0;
- group_ids = xmalloc (metrics_info->num_groups * sizeof (GLuint));
+ metrics->timer_head = NULL;
+ metrics->timer_tail = NULL;
- glGetPerfMonitorGroupsAMD (NULL, metrics_info->num_groups, group_ids);
+ metrics->monitor_begun_id = 0;
- metrics_info->max_counters_per_group = 0;
+ metrics->monitor_head = NULL;
+ metrics->monitor_tail = NULL;
- metrics_info->groups = xmalloc (metrics_info->num_groups * sizeof (counter_group_info_t));
+ metrics->monitors_in_flight = 0;
- for (i = 0; i < metrics_info->num_groups; i++)
- {
- counter_group_info_t *group;
+ metrics->num_op_metrics = 0;
+ metrics->op_metrics = NULL;
- group = &metrics_info->groups[i];
+ return metrics;
+}
+
+void
+metrics_fini (metrics_t *metrics)
+{
+ timer_query_t *timer, *timer_next;
+ monitor_t *monitor, *monitor_next;
+
+ /* Discard and cleanup any outstanding queries. */
+ if (metrics->timer_begun_id) {
+ glEndQuery (GL_TIME_ELAPSED);
+ glDeleteQueries (1, &metrics->timer_begun_id);
+ metrics->timer_begun_id = 0;
+ }
- group->id = group_ids[i];
+ for (timer = metrics->timer_head;
+ timer;
+ timer = timer_next)
+ {
+ glDeleteQueries (1, &timer->id);
+ timer_next = timer->next;
+ free (timer);
+ }
+ metrics->timer_head = NULL;
+ metrics->timer_tail = NULL;
- glGetPerfMonitorCountersAMD (group->id, &group->num_counters,
- &group->max_active_counters, 0, NULL);
+ if (metrics->info->have_perfmon) {
- group->counters = xmalloc (group->num_counters * sizeof (GLuint));
+ if (metrics->monitor_begun_id) {
+ glEndPerfMonitorAMD (metrics->monitor_begun_id);
+ glDeletePerfMonitorsAMD (1, &metrics->monitor_begun_id);
+ metrics->monitor_begun_id = 0;
+ }
- glGetPerfMonitorCountersAMD (group->id, NULL, NULL,
- group->num_counters,
- group->counters);
+ for (monitor = metrics->monitor_head;
+ monitor;
+ monitor = monitor_next)
+ {
+ glDeletePerfMonitorsAMD (1, &monitor->id);
+ monitor_next = monitor->next;
+ free (monitor);
+ }
+ metrics->monitor_head = NULL;
+ metrics->monitor_tail = NULL;
- if (group->num_counters > metrics_info->max_counters_per_group)
- metrics_info->max_counters_per_group = group->num_counters;
}
- free (group_ids);
+ metrics->monitors_in_flight = 0;
+}
+
+void
+metrics_destroy (metrics_t *metrics)
+{
+ metrics_fini (metrics);
+
+ free (metrics);
}
static const char *
}
void
-metrics_counter_start (void)
+metrics_counter_start (metrics_t *metrics)
{
- context_t *ctx = ¤t_context;
- timer_query_t *timer;
- monitor_t *monitor;
- int i;
-
- /* Create new timer query, add to list */
- timer = xmalloc (sizeof (timer_query_t));
+ unsigned i;
- timer->op = ctx->op;
- timer->next = NULL;
+ /* Initialize the timer_query object. */
+ glGenQueries (1, &metrics->timer_begun_id);
- if (ctx->timer_tail) {
- ctx->timer_tail->next = timer;
- ctx->timer_tail = timer;
- } else {
- ctx->timer_tail = timer;
- ctx->timer_head = timer;
+ /* Most everything else in this function is
+ * performance-monitor related. If we don't have that
+ * extension, just start the timer query and be done. */
+ if (! metrics->info->have_perfmon) {
+ glBeginQuery (GL_TIME_ELAPSED, metrics->timer_begun_id);
+ return;
}
- /* Create a new performance-monitor query */
- monitor = xmalloc (sizeof (monitor_t));
-
- monitor->op = ctx->op;
- monitor->next = NULL;
-
- if (ctx->monitor_tail) {
- ctx->monitor_tail->next = monitor;
- ctx->monitor_tail = monitor;
- } else {
- ctx->monitor_tail = monitor;
- ctx->monitor_head = monitor;
- }
+ /* Initialize the performance-monitor object */
+ glGenPerfMonitorsAMD (1, &metrics->monitor_begun_id);
- /* Initialize the timer_query and monitor objects */
- glGenQueries (1, &timer->id);
-
- glGenPerfMonitorsAMD (1, &monitor->id);
-
- for (i = 0; i < ctx->metrics_info.num_groups; i++)
+ for (i = 0; i < metrics->info->num_groups; i++)
{
- counter_group_info_t *group;
+ metrics_group_info_t *group;
int num_counters;
- group = &ctx->metrics_info.groups[i];
+ group = &metrics->info->groups[i];
num_counters = group->num_counters;
if (group->max_active_counters < group->num_counters)
}
- glSelectPerfMonitorCountersAMD(monitor->id,
+ glSelectPerfMonitorCountersAMD(metrics->monitor_begun_id,
GL_TRUE, group->id,
num_counters,
- group->counters);
+ group->counter_ids);
}
/* Start the queries */
- glBeginQuery (GL_TIME_ELAPSED, timer->id);
+ glBeginQuery (GL_TIME_ELAPSED, metrics->timer_begun_id);
- glBeginPerfMonitorAMD (monitor->id);
+ glBeginPerfMonitorAMD (metrics->monitor_begun_id);
}
void
-metrics_counter_stop (void)
+metrics_counter_stop (metrics_t *metrics)
{
+ timer_query_t *timer;
+ monitor_t *monitor;
+
+ /* Stop the current timer and monitor. */
glEndQuery (GL_TIME_ELAPSED);
- glEndPerfMonitorAMD (current_context.monitor_tail->id);
+
+ if (metrics->info->have_perfmon)
+ glEndPerfMonitorAMD (metrics->monitor_begun_id);
+
+ /* Add these IDs to our lists of outstanding queries and
+ * monitors so the results can be collected later. */
+ timer = xmalloc (sizeof (timer_query_t));
+
+ timer->op = metrics->op;
+ timer->id = metrics->timer_begun_id;
+ timer->next = NULL;
+
+ if (metrics->timer_tail) {
+ metrics->timer_tail->next = timer;
+ metrics->timer_tail = timer;
+ } else {
+ metrics->timer_tail = timer;
+ metrics->timer_head = timer;
+ }
+
+ if (metrics->info->have_perfmon) {
+ /* Create a new performance-monitor query */
+ monitor = xmalloc (sizeof (monitor_t));
+
+ monitor->op = metrics->op;
+ monitor->id = metrics->monitor_begun_id;
+ monitor->next = NULL;
+
+ if (metrics->monitor_tail) {
+ metrics->monitor_tail->next = monitor;
+ metrics->monitor_tail = monitor;
+ } else {
+ metrics->monitor_tail = monitor;
+ metrics->monitor_head = monitor;
+ }
+ }
+
+ metrics->monitors_in_flight++;
+
+ /* Avoid being a resource hog and collect outstanding results
+ * once we have sent off a large number of
+ * queries. (Presumably, many of the outstanding queries are
+ * available by now.)
+ */
+ if (metrics->monitors_in_flight > MAX_MONITORS_IN_FLIGHT)
+ metrics_collect_available (metrics);
}
void
-metrics_set_current_op (metrics_op_t op)
+metrics_set_current_op (metrics_t *metrics, metrics_op_t op)
{
- current_context.op = op;
+ metrics->op = op;
}
metrics_op_t
-metrics_get_current_op (void)
+metrics_get_current_op (metrics_t *metrics)
{
- return current_context.op;
+ return metrics->op;
}
static void
-op_metrics_init (context_t *ctx, op_metrics_t *metrics, metrics_op_t op)
+op_metrics_init (metrics_info_t *info, op_metrics_t *metrics, metrics_op_t op)
{
- metrics_info_t *info = &ctx->metrics_info;
- unsigned i;
+ unsigned i, j;
metrics->op = op;
metrics->time_ns = 0.0;
- metrics->num_counters = info->num_groups * info->max_counters_per_group;
- metrics->counters = xmalloc (sizeof(double) * metrics->num_counters);
+ metrics->counters = xmalloc (sizeof(double *) * info->num_groups);
- for (i = 0; i < metrics->num_counters; i++)
- metrics->counters[i] = 0.0;
+ for (i = 0; i < info->num_groups; i++) {
+ metrics->counters[i] = xmalloc (sizeof (double) *
+ info->groups[i].num_counters);
+ for (j = 0; j < info->groups[i].num_counters; j++)
+ metrics->counters[i][j] = 0.0;
+ }
}
static op_metrics_t *
-ctx_get_op_metrics (context_t *ctx, metrics_op_t op)
+_get_op_metrics (metrics_t *metrics, metrics_op_t op)
{
unsigned i;
- if (op >= ctx->num_op_metrics)
+ if (op >= metrics->num_op_metrics)
{
- ctx->op_metrics = realloc (ctx->op_metrics,
- (op + 1) * sizeof (op_metrics_t));
- for (i = ctx->num_op_metrics; i < op + 1; i++)
- op_metrics_init (ctx, &ctx->op_metrics[i], i);
+ metrics->op_metrics = realloc (metrics->op_metrics,
+ (op + 1) * sizeof (op_metrics_t));
+ for (i = metrics->num_op_metrics; i < op + 1; i++)
+ op_metrics_init (metrics->info, &metrics->op_metrics[i], i);
- ctx->num_op_metrics = op + 1;
+ metrics->num_op_metrics = op + 1;
}
- return &ctx->op_metrics[op];
+ return &metrics->op_metrics[op];
}
static void
-accumulate_program_metrics (metrics_op_t op, GLuint *result, GLuint size)
+accumulate_program_metrics (metrics_t *metrics, metrics_op_t op,
+ GLuint *result, GLuint size)
{
#define CONSUME(var) \
if (p + sizeof(var) > ((unsigned char *) result) + size) \
(var) = *((typeof(var) *) p); \
p += sizeof(var);
- context_t *ctx = ¤t_context;
+ metrics_info_t *info = metrics->info;
+ op_metrics_t *op_metrics = _get_op_metrics (metrics, op);
unsigned char *p = (unsigned char *) result;
while (p < ((unsigned char *) result) + size)
{
- GLuint group_id, counter_id, counter_type;
- uint32_t value;
+ GLuint group_id, group_index;
+ GLuint counter_id, counter_index;
+ metrics_group_info_t *group;
+ double value = 0.0;
unsigned i;
CONSUME (group_id);
CONSUME (counter_id);
- glGetPerfMonitorCounterInfoAMD (group_id, counter_id,
- GL_COUNTER_TYPE_AMD,
- &counter_type);
-
- /* We assume that all peformance counters are made
- * available as uint32 values. This code can easily be
- * extended as needed. */
- if (counter_type != GL_UNSIGNED_INT) {
- fprintf (stderr, "Warning: Non-uint counter value. Ignoring remainder of results\n");
- break;
+ for (i = 0; i < info->num_groups; i++) {
+ if (info->groups[i].id == group_id)
+ break;
}
+ group_index = i;
+ assert (group_index < info->num_groups);
+ group = &info->groups[group_index];
- CONSUME (value);
-
- i = (group_id * ctx->metrics_info.max_counters_per_group +
- counter_id);
+ for (i = 0; i < group->num_counters; i++) {
+ if (group->counter_ids[i] == counter_id)
+ break;
+ }
+ counter_index = i;
+ assert (counter_index < group->num_counters);
- assert (i < ctx->op_metrics[op].num_counters);
+ switch (group->counter_types[counter_index])
+ {
+ uint uint_value;
+ uint64_t uint64_value;
+ float float_value;
+ case GL_UNSIGNED_INT:
+ CONSUME (uint_value);
+ value = uint_value;
+ break;
+ case GL_UNSIGNED_INT64_AMD:
+ CONSUME (uint64_value);
+ value = uint64_value;
+ break;
+ case GL_PERCENTAGE_AMD:
+ case GL_FLOAT:
+ CONSUME (float_value);
+ value = float_value;
+ break;
+ default:
+ fprintf (stderr, "fips: Warning: Unknown counter value type (%d)\n",
+ group->counter_types[counter_index]);
+ value = 0.0;
+ break;
+ }
- /* FIXME: While I'm still occasionally getting bogus
- * numbers from the performance counters, I'm simply
- * going to discard anything larger than half the
- * range, (something that looks like a negative signed
- * quantity).
- */
- if (((int32_t) value) < 0)
- fprintf (stderr, ".");
- else
- ctx->op_metrics[op].counters[i] += value;
+ op_metrics->counters[group_index][counter_index] += value;
}
}
static void
-accumulate_program_time (metrics_op_t op, unsigned time_ns)
+accumulate_program_time (metrics_t *metrics, metrics_op_t op, unsigned time_ns)
{
- op_metrics_t *metrics;
+ op_metrics_t *op_metrics;
- metrics = ctx_get_op_metrics (¤t_context, op);
+ op_metrics = _get_op_metrics (metrics, op);
- metrics->time_ns += time_ns;
+ op_metrics->time_ns += time_ns;
}
+typedef struct per_stage_metrics
+{
+ op_metrics_t *metrics;
+ shader_stage_info_t *stage;
+ double time_ns;
+ double active;
+} per_stage_metrics_t;
+
static int
-time_compare(const void *in_a, const void *in_b, void *arg)
+_is_shader_stage_counter (metrics_info_t *info,
+ unsigned group_index,
+ unsigned counter_index)
{
- int a = *(const int *)in_a;
- int b = *(const int *)in_b;
- struct op_metrics *metrics = arg;
+ shader_stage_info_t *stage;
+ unsigned i;
+
+ for (i = 0; i < info->num_shader_stages; i++) {
+ stage = &info->stages[i];
+
+ if (stage->active_group_index == group_index &&
+ stage->active_counter_index == counter_index)
+ {
+ return 1;
+ }
+
+ if (stage->stall_group_index == group_index &&
+ stage->stall_counter_index == counter_index)
+ {
+ return 1;
+ }
+ }
- if (metrics[a].time_ns < metrics[b].time_ns)
- return -1;
- if (metrics[a].time_ns > metrics[b].time_ns)
- return 1;
return 0;
}
static void
-print_program_metrics (void)
+print_per_stage_metrics (metrics_t *metrics,
+ per_stage_metrics_t *per_stage,
+ double total)
{
- context_t *ctx = ¤t_context;
- int *sorted; /* Sorted indices into the ctx->op_metrics */
- double total = 0;
- unsigned i, j;
+ metrics_info_t *info = metrics->info;
+ op_metrics_t *op_metrics = per_stage->metrics;
+ metrics_group_info_t *group;
+ const char *op_string;
+ unsigned group_index, counter;
+ double value;
+
+ /* Don't print anything for stages with no alloted time. */
+ if (per_stage->time_ns == 0.0)
+ return;
+
+ op_string = metrics_op_string (op_metrics->op);
+
+ printf ("%21s", op_string);
+
+ if (op_metrics->op >= METRICS_OP_SHADER) {
+ printf (" %3d", op_metrics->op - METRICS_OP_SHADER);
+ } else {
+ printf (" ");
- /* Make a sorted list of the operations by time used, and figure
- * out the total so we can print percentages.
- */
- sorted = calloc(ctx->num_op_metrics, sizeof(*sorted));
- for (i = 0; i < ctx->num_op_metrics; i++) {
- sorted[i] = i;
- total += ctx->op_metrics[i].time_ns;
}
- qsort_r(sorted, ctx->num_op_metrics, sizeof(*sorted),
- time_compare, ctx->op_metrics);
- for (i = 0; i < ctx->num_op_metrics; i++) {
- const char *op_string;
- op_metrics_t *metric =&ctx->op_metrics[sorted[i]];
+ if (per_stage->stage)
+ printf (" %cS:", per_stage->stage->name[0]);
+ else
+ printf (" :");
- /* Since we sparsely fill the array based on program
- * id, many "programs" have no time.
- */
- if (metric->time_ns == 0.0)
- continue;
+ printf ("\t%7.2f ms (%4.1f%%)",
+ per_stage->time_ns / 1e6,
+ per_stage->time_ns / total * 100);
- op_string = metrics_op_string (metric->op);
+ if (per_stage->active)
+ printf (", %4.1f%% active", per_stage->active * 100);
- printf ("%s", op_string);
- if (metric->op >= METRICS_OP_SHADER) {
- printf (" %d:", metric->op - METRICS_OP_SHADER);
- } else {
- printf (":");
- for (j = strlen (op_string); j < 20; j++)
- printf (" ");
- }
- printf ("\t%7.2f ms (% 2.1f%%)",
- metric->time_ns / 1e6,
- metric->time_ns / total * 100);
- printf ("[");
- for (j = 0; j < metric->num_counters; j++) {
- if (metric->counters[j] == 0.0)
+ printf ("\n");
+
+ /* I'm not seeing a lot of value printing the rest of these
+ * performance counters by default yet. Use --verbose to get
+ * them for now. */
+ if (! verbose)
+ return;
+
+ printf ("[");
+ for (group_index = 0; group_index < info->num_groups; group_index++) {
+ group = &info->groups[group_index];
+ for (counter = 0; counter < group->num_counters; counter++) {
+
+ /* Don't print this counter value if it's a
+ * per-stage cycle counter, (which we have
+ * already accounted for). */
+ if (_is_shader_stage_counter (info, group_index, counter))
continue;
- printf ("%d: %.2f ms ", j, metric->counters[j] / 1e6);
+
+ value = op_metrics->counters[group_index][counter];
+ if (value == 0.0)
+ continue;
+ printf ("%s: %.2f ", group->counter_names[counter],
+ value / 1e6);
}
- printf ("]\n");
}
+ printf ("]\n");
}
-/* Called at program exit */
-static void
-metrics_exit (void)
+static int
+time_compare(const void *in_a, const void *in_b, void *arg unused)
{
- if (verbose)
- printf ("fips: terminating\n");
-}
+ const per_stage_metrics_t *a = in_a;
+ const per_stage_metrics_t *b = in_b;
-void
-metrics_end_frame (void)
+ if (a->time_ns < b->time_ns)
+ return -1;
+ if (a->time_ns > b->time_ns)
+ return 1;
+ return 0;
+}
+
+static void
+print_program_metrics (metrics_t *metrics)
{
- static int initialized = 0;
- static struct timeval tv_start, tv_now;
+ metrics_info_t *info = metrics->info;
+ unsigned num_shader_stages = info->num_shader_stages;
+ per_stage_metrics_t *sorted, *per_stage;
+ double total_time, op_cycles;
+ op_metrics_t *op;
+ unsigned group_index, counter_index;
+ unsigned i, j, num_sorted;
+
+ /* Make a sorted list of the per-stage operations by time
+ * used, and figure out the total so we can print percentages.
+ */
+ if (num_shader_stages)
+ num_sorted = metrics->num_op_metrics * num_shader_stages;
+ else
+ num_sorted = metrics->num_op_metrics;
- if (! initialized) {
- gettimeofday (&tv_start, NULL);
- atexit (metrics_exit);
- if (getenv ("FIPS_VERBOSE"))
- verbose = 1;
- initialized = 1;
+ sorted = xmalloc (sizeof (*sorted) * num_sorted);
+
+ total_time = 0.0;
+
+ for (i = 0; i < metrics->num_op_metrics; i++) {
+
+ op = &metrics->op_metrics[i];
+
+ /* Accumulate total time across all ops. */
+ total_time += op->time_ns;
+
+ /* Also, find total cycles in all stages of this op. */
+ op_cycles = 0.0;
+
+ if (num_shader_stages == 0) {
+ per_stage = &sorted[i];
+ per_stage->metrics = op;
+ per_stage->stage = NULL;
+ per_stage->time_ns = op->time_ns;
+ per_stage->active = 0.0;
+ }
+
+ for (j = 0; j < num_shader_stages; j++) {
+ /* Active cycles */
+ group_index = info->stages[j].active_group_index;
+ counter_index = info->stages[j].active_counter_index;
+ op_cycles += op->counters[group_index][counter_index];
+
+ /* Stall cycles */
+ group_index = info->stages[j].stall_group_index;
+ counter_index = info->stages[j].stall_counter_index;
+ op_cycles += op->counters[group_index][counter_index];
+ }
+
+ for (j = 0; j < num_shader_stages; j++) {
+ double active_cycles, stall_cycles, stage_cycles;
+
+ /* Active cycles */
+ group_index = info->stages[j].active_group_index;
+ counter_index = info->stages[j].active_counter_index;
+ active_cycles = op->counters[group_index][counter_index];
+
+ /* Stall cycles */
+ group_index = info->stages[j].stall_group_index;
+ counter_index = info->stages[j].stall_counter_index;
+ stall_cycles = op->counters[group_index][counter_index];
+
+ stage_cycles = active_cycles + stall_cycles;
+
+ per_stage = &sorted[i * num_shader_stages + j];
+ per_stage->metrics = op;
+
+ if (op_cycles) {
+ per_stage->stage = &info->stages[j];
+ per_stage->time_ns = op->time_ns * (stage_cycles / op_cycles);
+ } else {
+ /* If we don't have any per-stage cycle counts
+ * for this operation, then use the first
+ * stage as a placeholder for all the time,
+ * but NULL-ify the stage info so that the
+ * report doesn't lie about this time being
+ * from any particular stage. */
+ per_stage->stage = NULL;
+ if (j == 0) {
+ per_stage->time_ns = op->time_ns;
+ } else {
+ per_stage->time_ns = 0.0;
+ }
+ }
+
+ if (stage_cycles) {
+ per_stage->active = active_cycles / stage_cycles;
+ } else {
+ per_stage->active = 0.0;
+ }
+ }
}
- if (verbose)
- printf ("fips: frame %d complete\n", frames);
+ qsort_r (sorted, num_sorted, sizeof (*sorted),
+ time_compare, metrics->op_metrics);
- frames++;
- gettimeofday (&tv_now, NULL);
+ for (i = 0; i < num_sorted; i++)
+ print_per_stage_metrics (metrics, &sorted[i], total_time);
+ free (sorted);
+}
+
+void
+metrics_collect_available (metrics_t *metrics)
+{
/* Consume all timer queries that are ready. */
- timer_query_t *timer = current_context.timer_head;
+ timer_query_t *timer = metrics->timer_head;
while (timer) {
GLuint available, elapsed;
glGetQueryObjectuiv (timer->id,
GL_QUERY_RESULT, &elapsed);
- accumulate_program_time (timer->op, elapsed);
+ accumulate_program_time (metrics, timer->op, elapsed);
- current_context.timer_head = timer->next;
- if (current_context.timer_head == NULL)
- current_context.timer_tail = NULL;
+ metrics->timer_head = timer->next;
+ if (metrics->timer_head == NULL)
+ metrics->timer_tail = NULL;
glDeleteQueries (1, &timer->id);
free (timer);
- timer = current_context.timer_head;
+ timer = metrics->timer_head;
}
+ if (! metrics->info->have_perfmon)
+ return;
+
/* And similarly for all performance monitors that are ready. */
- monitor_t *monitor = current_context.monitor_head;
+ monitor_t *monitor = metrics->monitor_head;
while (monitor) {
GLuint available, result_size, *result;
result_size, result,
&bytes_written);
- accumulate_program_metrics (monitor->op, result, result_size);
+ accumulate_program_metrics (metrics, monitor->op, result, result_size);
+
+ free (result);
- current_context.monitor_head = monitor->next;
- if (current_context.monitor_head == NULL)
- current_context.monitor_tail = NULL;
+ metrics->monitor_head = monitor->next;
+ if (metrics->monitor_head == NULL)
+ metrics->monitor_tail = NULL;
glDeletePerfMonitorsAMD (1, &monitor->id);
free (monitor);
- monitor = current_context.monitor_head;
+
+ metrics->monitors_in_flight--;
+
+ monitor = metrics->monitor_head;
}
+}
- if (frames % 60 == 0) {
+static void
+metrics_exit (void)
+{
+ if (verbose)
+ printf ("fips: terminating\n");
+}
+
+void
+metrics_end_frame (metrics_t *metrics)
+{
+ static int initialized = 0;
+ static struct timeval tv_start, tv_now;
+
+ if (! initialized) {
+ gettimeofday (&tv_start, NULL);
+ atexit (metrics_exit);
+ if (getenv ("FIPS_VERBOSE"))
+ verbose = 1;
+ initialized = 1;
+ }
+
+ frames++;
+
+ metrics_collect_available (metrics);
+
+ if (frames % 15 == 0) {
double fps;
+ gettimeofday (&tv_now, NULL);
+
fps = (double) frames / (tv_now.tv_sec - tv_start.tv_sec +
(tv_now.tv_usec - tv_start.tv_usec) / 1.0e6);
printf("FPS: %.3f\n", fps);
- print_program_metrics ();
+ print_program_metrics (metrics);
}
}
projects / fips / blobdiff