Allow cost estimates to differ per backend and include the estimates into the HLO profile. Add a summary table for what categories have the most opportunity for optimization left in them.

PiperOrigin-RevId: 163780413

1 files changed, 5 insertions, 5 deletions

diff --git a/tensorflow/compiler/xla/service/hlo_execution_profile.cc b/tensorflow/compiler/xla/service/hlo_execution_profile.cc
index 7a83a92404..654565ab17 100644
--- a/tensorflow/compiler/xla/service/hlo_execution_profile.cc
+++ b/tensorflow/compiler/xla/service/hlo_execution_profile.cc
@@ -44,10 +44,9 @@ uint64 HloExecutionProfile::GetProfileResult(const HloInstruction& hlo) const {
 string HloExecutionProfile::ToString(
     const HloComputation& computation,
     const DeviceDescription& device_description,
-    const HloCostAnalysis::ShapeSizeFunction& shape_size) const {
-  HloCostAnalysis cost_analysis(shape_size);
+    HloCostAnalysis* cost_analysis) const {
   tensorflow::Status analysis_status =
-      computation.root_instruction()->Accept(&cost_analysis);
+      computation.root_instruction()->Accept(cost_analysis);
   if (!analysis_status.ok()) {
     return "";
   }
@@ -61,8 +60,9 @@ string HloExecutionProfile::ToString(
 
     builder.AddOp(/*op_name=*/hlo->ToString(),
                   /*short_name=*/hlo->ToString(/*compact_operands=*/true),
-                  hlo->ToCategory(), cycles, cost_analysis.flop_count(*hlo),
-                  cost_analysis.bytes_accessed(*hlo));
+                  hlo->ToCategory(), cycles, cost_analysis->flop_count(*hlo),
+                  cost_analysis->bytes_accessed(*hlo),
+                  cost_analysis->seconds(*hlo));
   }
   return builder.ToString();
 }