metrics_op_t op;
double time_ns;
- double *counters;
- unsigned num_counters;
+ double **counters;
} op_metrics_t;
-typedef struct counter_group_info
+typedef struct metrics_group_info
{
GLuint id;
- GLint num_counters;
- GLint max_active_counters;
- GLuint *counters;
-} counter_group_info_t;
+ char *name;
+
+ GLuint num_counters;
+ GLuint max_active_counters;
+
+ GLuint *counter_ids;
+ char **counter_names;
+ GLuint *counter_types;
+
+} metrics_group_info_t;
+
+typedef struct shader_stage_info
+{
+ char *name;
+
+ GLuint active_group_index;
+ GLuint active_counter_index;
+
+ GLuint stall_group_index;
+ GLuint stall_counter_index;
+
+} shader_stage_info_t;
typedef struct metrics_info
{
- int num_groups;
- int max_counters_per_group;
- counter_group_info_t *groups;
+ int initialized;
+
+ unsigned num_groups;
+ metrics_group_info_t *groups;
+
+ unsigned num_shader_stages;
+ shader_stage_info_t *stages;
+
} metrics_info_t;
typedef struct context
int frames;
int verbose;
+static void
+metrics_group_info_init (metrics_group_info_t *group, GLuint id)
+{
+ GLsizei length;
+ unsigned i;
+
+ group->id = id;
+
+ /* Get name */
+ glGetPerfMonitorGroupStringAMD (id, 0, &length, NULL);
+
+ group->name = xmalloc (length + 1);
+
+ glGetPerfMonitorGroupStringAMD (id, length + 1, NULL, group->name);
+
+ /* Get number of counters */
+ group->num_counters = 0;
+ group->max_active_counters = 0;
+ glGetPerfMonitorCountersAMD (group->id,
+ (int *) &group->num_counters,
+ (int *) &group->max_active_counters,
+ 0, NULL);
+
+ /* Get counter numbers */
+ group->counter_ids = xmalloc (group->num_counters * sizeof (GLuint));
+
+ glGetPerfMonitorCountersAMD (group->id, NULL, NULL,
+ group->num_counters,
+ group->counter_ids);
+
+ /* Get counter names */
+ group->counter_names = xmalloc (group->num_counters * sizeof (char *));
+ group->counter_types = xmalloc (group->num_counters * sizeof (GLuint));
+
+ for (i = 0; i < group->num_counters; i++) {
+ glGetPerfMonitorCounterInfoAMD (group->id,
+ group->counter_ids[i],
+ GL_COUNTER_TYPE_AMD,
+ &group->counter_types[i]);
+
+ glGetPerfMonitorCounterStringAMD (group->id,
+ group->counter_ids[i],
+ 0, &length, NULL);
+
+ group->counter_names[i] = xmalloc (length + 1);
+
+ glGetPerfMonitorCounterStringAMD (group->id,
+ group->counter_ids[i],
+ length + 1, NULL,
+ group->counter_names[i]);
+ }
+}
+
+static void
+metrics_group_info_fini (metrics_group_info_t *group)
+{
+ unsigned i;
+
+ for (i = 0; i < group->num_counters; i++)
+ free (group->counter_names[i]);
+
+ free (group->counter_types);
+ free (group->counter_names);
+ free (group->counter_ids);
+
+ free (group->name);
+}
+
+static void
+metrics_info_fini (metrics_info_t *info);
+
+/* A helper function, part of metrics_info_init below. */
+
+typedef enum {
+ SHADER_ACTIVE,
+ SHADER_STALL
+} shader_phase_t;
+
+static void
+_add_shader_stage (metrics_info_t *info, const char *name,
+ GLuint group_index, GLuint counter_index,
+ shader_phase_t phase)
+{
+ shader_stage_info_t *stage;
+ char *stage_name, *space;
+ unsigned i;
+
+ stage_name = xstrdup (name);
+
+ /* Terminate the stage name at the first space.
+ *
+ * This is valid for counter names such as:
+ *
+ * "Vertex Shader Active Time"
+ * or
+ * "Vertex Shader Stall Time - Core Stall"
+ */
+ space = strchr (stage_name, ' ');
+ if (space)
+ *space = '\0';
+
+ /* Look for an existing stage of the given name. */
+ stage = NULL;
+
+ for (i = 0; i < info->num_shader_stages; i++) {
+ if (strcmp (info->stages[i].name, stage_name) == 0) {
+ stage = &info->stages[i];
+ break;
+ }
+ }
+
+ if (stage == NULL) {
+ info->num_shader_stages++;
+ info->stages = xrealloc (info->stages,
+ info->num_shader_stages *
+ sizeof (shader_stage_info_t));
+ stage = &info->stages[info->num_shader_stages - 1];
+ stage->name = xstrdup (stage_name);
+ stage->active_group_index = 0;
+ stage->active_counter_index = 0;
+ stage->stall_group_index = 0;
+ stage->stall_counter_index = 0;
+ }
+
+ if (phase == SHADER_ACTIVE) {
+ stage->active_group_index = group_index;
+ stage->active_counter_index = counter_index;
+ } else {
+ stage->stall_group_index = group_index;
+ stage->stall_counter_index = counter_index;
+ }
+
+ free (stage_name);
+}
+
void
metrics_info_init (void)
{
- int i;
+ unsigned i, j;
GLuint *group_ids;
- metrics_info_t *metrics_info = ¤t_context.metrics_info;
+ metrics_info_t *info = ¤t_context.metrics_info;
- glGetPerfMonitorGroupsAMD (&metrics_info->num_groups, 0, NULL);
+ if (info->initialized)
+ metrics_info_fini (info);
- group_ids = xmalloc (metrics_info->num_groups * sizeof (GLuint));
+ glGetPerfMonitorGroupsAMD ((int *) &info->num_groups, 0, NULL);
- glGetPerfMonitorGroupsAMD (NULL, metrics_info->num_groups, group_ids);
+ group_ids = xmalloc (info->num_groups * sizeof (GLuint));
- metrics_info->max_counters_per_group = 0;
+ glGetPerfMonitorGroupsAMD (NULL, info->num_groups, group_ids);
- metrics_info->groups = xmalloc (metrics_info->num_groups * sizeof (counter_group_info_t));
+ info->groups = xmalloc (info->num_groups * sizeof (metrics_group_info_t));
- for (i = 0; i < metrics_info->num_groups; i++)
- {
- counter_group_info_t *group;
+ for (i = 0; i < info->num_groups; i++)
+ metrics_group_info_init (&info->groups[i], group_ids[i]);
- group = &metrics_info->groups[i];
+ free (group_ids);
- group->id = group_ids[i];
+ /* Identify each shader stage (by looking at
+ * performance-counter names for specific patterns) and
+ * initialize structures referring to the corresponding
+ * counter numbers for each stage. */
+ info->num_shader_stages = 0;
+ info->stages = NULL;
+
+ for (i = 0; i < info->num_groups; i++) {
+ metrics_group_info_t *group = &info->groups[i];
+ for (j = 0; j < group->num_counters; j++) {
+ char *name = group->counter_names[j];
+ if (strstr (name, "Shader Active Time")) {
+ _add_shader_stage (info, name, i, j,
+ SHADER_ACTIVE);
+ }
+ if (strstr (name, "Shader Stall Time")) {
+ _add_shader_stage (info, name, i, j,
+ SHADER_STALL);
+ }
+ }
+ }
- glGetPerfMonitorCountersAMD (group->id, &group->num_counters,
- &group->max_active_counters, 0, NULL);
+ info->initialized = 1;
+}
- group->counters = xmalloc (group->num_counters * sizeof (GLuint));
+static void
+metrics_info_fini (metrics_info_t *info)
+{
+ unsigned i;
- glGetPerfMonitorCountersAMD (group->id, NULL, NULL,
- group->num_counters,
- group->counters);
+ for (i = 0; i < info->num_groups; i++)
+ metrics_group_info_fini (&info->groups[i]);
- if (group->num_counters > metrics_info->max_counters_per_group)
- metrics_info->max_counters_per_group = group->num_counters;
- }
+ free (info->groups);
- free (group_ids);
+ for (i = 0; i < info->num_shader_stages; i++)
+ free (info->stages[i].name);
+
+ free (info->stages);
}
static const char *
context_t *ctx = ¤t_context;
timer_query_t *timer;
monitor_t *monitor;
- int i;
+ unsigned i;
/* Create new timer query, add to list */
timer = xmalloc (sizeof (timer_query_t));
for (i = 0; i < ctx->metrics_info.num_groups; i++)
{
- counter_group_info_t *group;
+ metrics_group_info_t *group;
int num_counters;
group = &ctx->metrics_info.groups[i];
glSelectPerfMonitorCountersAMD(monitor->id,
GL_TRUE, group->id,
num_counters,
- group->counters);
+ group->counter_ids);
}
/* Start the queries */
op_metrics_init (context_t *ctx, 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 *
p += sizeof(var);
context_t *ctx = ¤t_context;
+ metrics_info_t *info = &ctx->metrics_info;
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;
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;
+ ctx->op_metrics[op].counters[group_index][counter_index] += value;
}
}
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_op_metrics (op_metrics_t *metric, double total)
+print_per_stage_metrics (context_t *ctx,
+ per_stage_metrics_t *per_stage,
+ double total)
{
+ metrics_info_t *info = &ctx->metrics_info;
+ op_metrics_t *metric = per_stage->metrics;
+ metrics_group_info_t *group;
const char *op_string;
- unsigned i;
+ unsigned group_index, counter;
+ double value;
- /* Since we sparsely fill the array based on program
- * id, many "programs" have no time.
- */
- if (metric->time_ns == 0.0)
+ /* Don't print anything for stages with no alloted time. */
+ if (per_stage->time_ns == 0.0)
return;
op_string = metrics_op_string (metric->op);
- printf ("%s", op_string);
+ printf ("%21s", op_string);
+
if (metric->op >= METRICS_OP_SHADER) {
- printf (" %d:", metric->op - METRICS_OP_SHADER);
+ printf (" %3d", metric->op - METRICS_OP_SHADER);
} else {
- printf (":");
- for (i = strlen (op_string); i < 20; i++)
- printf (" ");
+ printf (" ");
+
}
+ printf (" %cS:", per_stage->stage->name[0]);
+
+ printf ("\t%7.2f ms (%4.1f%%)",
+ per_stage->time_ns / 1e6,
+ per_stage->time_ns / total * 100);
+
+ if (per_stage->active)
+ printf (", %4.1f%% active", per_stage->active * 100);
- printf ("\t%7.2f ms (% 2.1f%%)",
- metric->time_ns / 1e6,
- metric->time_ns / total * 100);
+ 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 (i = 0; i < metric->num_counters; i++) {
- if (metric->counters[i] == 0.0)
- continue;
- printf ("%d: %.2f ms ", i, metric->counters[i] / 1e6);
+ 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;
+
+ value = metric->counters[group_index][counter];
+ if (value == 0.0)
+ continue;
+ printf ("%s: %.2f ", group->counter_names[counter],
+ value / 1e6);
+ }
}
printf ("]\n");
}
+static int
+time_compare(const void *in_a, const void *in_b, void *arg unused)
+{
+ const per_stage_metrics_t *a = in_a;
+ const per_stage_metrics_t *b = in_b;
+
+
+ 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 (void)
{
context_t *ctx = ¤t_context;
- int *sorted; /* Sorted indices into the ctx->op_metrics */
- double total = 0;
- unsigned i;
-
- /* Make a sorted list of the operations by time used, and figure
- * out the total so we can print percentages.
+ metrics_info_t *info = &ctx->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.
*/
- sorted = calloc(ctx->num_op_metrics, sizeof(*sorted));
+ num_sorted = ctx->num_op_metrics * num_shader_stages;
+
+ sorted = xmalloc (sizeof (*sorted) * num_sorted);
+
+ total_time = 0.0;
+
for (i = 0; i < ctx->num_op_metrics; i++) {
- sorted[i] = i;
- total += ctx->op_metrics[i].time_ns;
+
+ op = &ctx->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;
+
+ 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;
+ per_stage->stage = &info->stages[j];
+ if (op_cycles)
+ per_stage->time_ns = op->time_ns * (stage_cycles / op_cycles);
+ else
+ per_stage->time_ns = 0.0;
+ if (stage_cycles)
+ per_stage->active = active_cycles / stage_cycles;
+ else
+ per_stage->active = 0.0;
+ }
}
- qsort_r(sorted, ctx->num_op_metrics, sizeof(*sorted),
- time_compare, ctx->op_metrics);
- for (i = 0; i < ctx->num_op_metrics; i++)
- print_op_metrics (&ctx->op_metrics[sorted[i]], total);
+ qsort_r (sorted, num_sorted, sizeof (*sorted),
+ time_compare, ctx->op_metrics);
+
+ for (i = 0; i < num_sorted; i++)
+ print_per_stage_metrics (ctx, &sorted[i], total_time);
+
+ free (sorted);
}
/* Called at program exit */
{
if (verbose)
printf ("fips: terminating\n");
+
+ metrics_info_fini (¤t_context.metrics_info);
}
initialized = 1;
}
- if (verbose)
- printf ("fips: frame %d complete\n", frames);
-
frames++;
gettimeofday (&tv_now, NULL);
accumulate_program_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;