Add normalized_total_op_time_ps (on default dvfs) to opstats.device_op_metrics_db, together with the total_op_time_ps, it will give us a weighted average dvfs multiplier, which can be used to calculate dvfs normalized flops utilization and so on.

- The aggregated dvfs multiplier will be shown in roofline page top as part of the device info, and used for device spec normalization.
- The agg dvfs multiplier will also be used a an identifier if dvfs normalization control knobs should be displayed on UI or not. (show the knob with a valid multiplier, which is >0 and <1)

PiperOrigin-RevId: 817431887

Author: zzzaries

plugin/xprof/protobuf/op_metrics.proto

@@ -120,8 +120,8 @@ message OpMetrics {
   uint32 occurrences = 3;
   // Total time (self + children) in picoseconds.
   uint64 time_ps = 7;
-  // Total time (self + children) in picoseconds normalized to the default
-  // device capability.
+  // Total time (self + children) in picoseconds needs to be spent if running on
+  // the default pstate.
   uint64 normalized_time_ps = 27;
   // Minimum time (self + children) among all occurrences.
   uint64 min_time_ps = 17;
@@ -179,7 +179,7 @@ message PrecisionStats {
 }
 
 // A database for OpMetrics.
-// Next ID: 16
+// Next ID: 17
 message OpMetricsDb {
   // A bunch of OpMetrics.
   repeated OpMetrics metrics_db = 10;
@@ -192,6 +192,9 @@ message OpMetricsDb {
   uint64 total_time_ps = 11;
   // The total time incurred by OPs in picoseconds.
   uint64 total_op_time_ps = 12;
+  // The weighted average total op time in picoseconds if running all ops on the
+  // default pstate.
+  uint64 normalized_total_op_time_ps = 16;
   // Precision-related stats.
   PrecisionStats precision_stats = 13;
   // The below two stats will be different from the total time ps and total op

xprof/convert/op_metrics_db_combiner.cc

@@ -136,6 +136,8 @@ void OpMetricsDbCombiner::Combine(const OpMetricsDb& src,
   dst->set_total_op_time_ps(src.total_op_time_ps() + dst->total_op_time_ps());
   dst->set_idle_time_ps(src.idle_time_ps() + dst->idle_time_ps());
   dst->set_busy_time_ps(src.busy_time_ps() + dst->busy_time_ps());
+  dst->set_normalized_total_op_time_ps(src.normalized_total_op_time_ps() +
+                                       dst->normalized_total_op_time_ps());
   CombinePrecisionStats(src.precision_stats(), dst->mutable_precision_stats());
 
   for (const auto& src_metrics : src.metrics_db()) {

xprof/convert/xplane_to_op_metrics_db_test.cc

@@ -278,6 +278,7 @@ TEST(ConvertXPlaneToOpMetricsDb, TpuDeviceOpMetricsDb) {
                                    metrics_db { name: "IDLE" category: "IDLE" }
                                    total_time_ps: 10000
                                    total_op_time_ps: 10000
+                                   normalized_total_op_time_ps: 20000
                               )pb")));
 #endif
 }

xprof/utils/op_metrics_db_utils.cc

@@ -321,14 +321,19 @@ OpMetricsDb XEventsOpMetricsDbBuilder::Finalize(uint64_t total_time_ps) {
 OpMetricsDb XEventsOpMetricsDbBuilder::Finalize() {
   OpMetricsDb db;
   uint64_t total_op_time_ps = 0;
+  uint64_t normalized_total_op_time_ps = 0;
   for (auto& [program_id, op_metric_by_symbol] : flat_op_metric_) {
     for (auto& [symbol_id, op_metrics] : op_metric_by_symbol) {
       AdjustFlopsAndBytesAccessed(op_metrics);
       total_op_time_ps += op_metrics.self_time_ps();
+      normalized_total_op_time_ps +=
+          op_metrics.self_time_ps() *
+          (op_metrics.normalized_time_ps() * 1.0 / op_metrics.time_ps());
       db.add_metrics_db()->Swap(&op_metrics);
     }
   }
   db.set_total_op_time_ps(total_op_time_ps);
+  db.set_normalized_total_op_time_ps(normalized_total_op_time_ps);
   return db;
 }
 

xprof/utils/op_metrics_db_utils_test.cc

@@ -237,6 +237,7 @@ TEST(OpMetricsDbTest, AddOpMetric) {
                   num_cores: 1
                 }
                 total_op_time_ps: 300
+                normalized_total_op_time_ps: 350
               )pb")));
 #endif
 }