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

2 files changed, 8 insertions, 1 deletions

diff --git a/tensorflow/compiler/plugin/executor/executable.cc b/tensorflow/compiler/plugin/executor/executable.cc
index 484465210b..e866dfef05 100644
--- a/tensorflow/compiler/plugin/executor/executable.cc
+++ b/tensorflow/compiler/plugin/executor/executable.cc
@@ -26,7 +26,7 @@ namespace se = ::perftools::gputools;
 namespace sep = ::perftools::gputools::executorplugin;
 
 ExecutorExecutable::ExecutorExecutable(std::unique_ptr<HloModule> hlo_module)
-    : Executable(std::move(hlo_module), ShapeSizeBytes) {}
+    : Executable(std::move(hlo_module)) {}
 
 ExecutorExecutable::~ExecutorExecutable() {}
 
@@ -140,5 +140,10 @@ StatusOr<se::DeviceMemoryBase> ExecutorExecutable::ExecuteAsyncOnStream(
   return ShapeUtil::ByteSizeOf(shape, sizeof(void*));
 }
 
+std::unique_ptr<HloCostAnalysis> ExecutorExecutable::CreateCostAnalysis()
+    const {
+  return MakeUnique<HloCostAnalysis>(ShapeSizeBytes);
+}
+
 }  // namespace executorplugin
 }  // namespace xla
diff --git a/tensorflow/compiler/plugin/executor/executable.h b/tensorflow/compiler/plugin/executor/executable.h
index ba3d4da21d..8572aba43d 100644
--- a/tensorflow/compiler/plugin/executor/executable.h
+++ b/tensorflow/compiler/plugin/executor/executable.h
@@ -55,6 +55,8 @@ class ExecutorExecutable : public Executable {
 
   static int64 ShapeSizeBytes(const Shape& shape);
 
+  std::unique_ptr<HloCostAnalysis> CreateCostAnalysis() const override;
+
  private:
   TF_DISALLOW_COPY_AND_ASSIGN(ExecutorExecutable);
 };