Replace StreamExecutorInterface::BlockHostUntilDone with BlockHostUntilDoneWithStatus

All known overrides of StreamExecutorInterface::BlockHostUntilDone are changed
by this CL.

PiperOrigin-RevId: 178492517

10 files changed, 17 insertions, 36 deletions

diff --git a/tensorflow/compiler/xla/service/interpreter/executor.cc b/tensorflow/compiler/xla/service/interpreter/executor.cc
index 0bb3259ef4..511de87b1b 100644
--- a/tensorflow/compiler/xla/service/interpreter/executor.cc
+++ b/tensorflow/compiler/xla/service/interpreter/executor.cc
@@ -100,9 +100,9 @@ bool InterpreterExecutor::StopTimer(Stream *stream, Timer *timer) {
   return true;
 }
 
-bool InterpreterExecutor::BlockHostUntilDone(Stream *stream) {
+port::Status InterpreterExecutor::BlockHostUntilDoneWithStatus(Stream *stream) {
   AsExecutorStream(stream)->BlockUntilDone();
-  return true;
+  return port::Status::OK();
 }
 
 DeviceDescription *InterpreterExecutor::PopulateDeviceDescription() const {
diff --git a/tensorflow/compiler/xla/service/interpreter/executor.h b/tensorflow/compiler/xla/service/interpreter/executor.h
index c59b2ccb15..d3753a6a65 100644
--- a/tensorflow/compiler/xla/service/interpreter/executor.h
+++ b/tensorflow/compiler/xla/service/interpreter/executor.h
@@ -157,7 +157,7 @@ class InterpreterExecutor : public internal::StreamExecutorInterface {
   bool StartTimer(Stream *stream, Timer *timer) override;
   bool StopTimer(Stream *stream, Timer *timer) override;
 
-  bool BlockHostUntilDone(Stream *stream) override;
+  port::Status BlockHostUntilDoneWithStatus(Stream *stream) override;
 
   int PlatformDeviceCount() override { return 1; }
 
diff --git a/tensorflow/stream_executor/cuda/cuda_driver.cc b/tensorflow/stream_executor/cuda/cuda_driver.cc
index b6a96ed3e5..a017ff64d4 100644
--- a/tensorflow/stream_executor/cuda/cuda_driver.cc
+++ b/tensorflow/stream_executor/cuda/cuda_driver.cc
@@ -1115,19 +1115,20 @@ CUDADriver::ContextGetSharedMemConfig(CudaContext* context) {
   return true;
 }
 
-/* static */ bool CUDADriver::SynchronizeStream(CudaContext* context,
-                                                CUstream stream) {
+/* static */ port::Status CUDADriver::SynchronizeStream(CudaContext *context,
+                                                        CUstream stream) {
   ScopedActivateContext activated{context};
   CHECK(stream != nullptr);
   CUresult res = cuStreamSynchronize(stream);
   if (res != CUDA_SUCCESS) {
-    LOG(ERROR) << "could not synchronize on CUDA stream: " << ToString(res)
-               << " :: " << port::CurrentStackTrace();
-    return false;
+    port::Status status = port::InternalError(
+        port::StrCat("could not synchronize on CUDA stream: ", ToString(res)));
+    LOG(ERROR) << status << " :: " << port::CurrentStackTrace();
+    return status;
   }
   VLOG(2) << "successfully synchronized stream " << stream << " on context "
           << context;
-  return true;
+  return port::Status::OK();
 }
 
 /* static */ bool CUDADriver::IsStreamIdle(CudaContext *context,
diff --git a/tensorflow/stream_executor/cuda/cuda_driver.h b/tensorflow/stream_executor/cuda/cuda_driver.h
index 68494aba65..4002ba2021 100644
--- a/tensorflow/stream_executor/cuda/cuda_driver.h
+++ b/tensorflow/stream_executor/cuda/cuda_driver.h
@@ -304,7 +304,7 @@ class CUDADriver {
   // amount of time?
   //
   // http://docs.nvidia.com/cuda/cuda-driver-api/group__CUDA__STREAM.html#group__CUDA__STREAM_1g15e49dd91ec15991eb7c0a741beb7dad
-  static bool SynchronizeStream(CudaContext* context, CUstream stream);
+  static port::Status SynchronizeStream(CudaContext* context, CUstream stream);
 
   // Blocks the calling thread until the operations associated with the context
   // have been completed, via cuCtxSynchronize.
diff --git a/tensorflow/stream_executor/cuda/cuda_gpu_executor.cc b/tensorflow/stream_executor/cuda/cuda_gpu_executor.cc
index 60eaaba21c..7f8a7ca7c7 100644
--- a/tensorflow/stream_executor/cuda/cuda_gpu_executor.cc
+++ b/tensorflow/stream_executor/cuda/cuda_gpu_executor.cc
@@ -664,7 +664,7 @@ bool CUDAExecutor::StopTimer(Stream *stream, Timer *timer) {
   return AsCUDATimer(timer)->Stop(AsCUDAStream(stream));
 }
 
-bool CUDAExecutor::BlockHostUntilDone(Stream *stream) {
+port::Status CUDAExecutor::BlockHostUntilDoneWithStatus(Stream *stream) {
   return CUDADriver::SynchronizeStream(context_, AsCUDAStreamValue(stream));
 }
 
diff --git a/tensorflow/stream_executor/cuda/cuda_gpu_executor.h b/tensorflow/stream_executor/cuda/cuda_gpu_executor.h
index 8ff4a30d62..5adbb59856 100644
--- a/tensorflow/stream_executor/cuda/cuda_gpu_executor.h
+++ b/tensorflow/stream_executor/cuda/cuda_gpu_executor.h
@@ -152,7 +152,7 @@ class CUDAExecutor : public internal::StreamExecutorInterface {
 
   Event::Status PollForEventStatus(Event *event) override;
 
-  bool BlockHostUntilDone(Stream *stream) override;
+  port::Status BlockHostUntilDoneWithStatus(Stream *stream) override;
 
   int PlatformDeviceCount() override { return CUDADriver::GetDeviceCount(); }
 
diff --git a/tensorflow/stream_executor/host/host_gpu_executor.cc b/tensorflow/stream_executor/host/host_gpu_executor.cc
index 0af2c8cc3d..1fd8eeb881 100644
--- a/tensorflow/stream_executor/host/host_gpu_executor.cc
+++ b/tensorflow/stream_executor/host/host_gpu_executor.cc
@@ -177,9 +177,9 @@ bool HostExecutor::StopTimer(Stream *stream, Timer *timer) {
   return true;
 }
 
-bool HostExecutor::BlockHostUntilDone(Stream *stream) {
+port::Status HostExecutor::BlockHostUntilDoneWithStatus(Stream *stream) {
   AsHostStream(stream)->BlockUntilDone();
-  return true;
+  return port::Status::OK();
 }
 
 DeviceDescription *HostExecutor::PopulateDeviceDescription() const {
diff --git a/tensorflow/stream_executor/host/host_gpu_executor.h b/tensorflow/stream_executor/host/host_gpu_executor.h
index 77b07e4a57..e884554a15 100644
--- a/tensorflow/stream_executor/host/host_gpu_executor.h
+++ b/tensorflow/stream_executor/host/host_gpu_executor.h
@@ -139,7 +139,7 @@ class HostExecutor : public internal::StreamExecutorInterface {
 
   bool StopTimer(Stream *stream, Timer *timer) override;
 
-  bool BlockHostUntilDone(Stream *stream) override;
+  port::Status BlockHostUntilDoneWithStatus(Stream *stream) override;
 
   int PlatformDeviceCount() override { return 1; }
 
diff --git a/tensorflow/stream_executor/stream_executor_internal.cc b/tensorflow/stream_executor/stream_executor_internal.cc
index 25b579fc16..273d970b6f 100644
--- a/tensorflow/stream_executor/stream_executor_internal.cc
+++ b/tensorflow/stream_executor/stream_executor_internal.cc
@@ -15,10 +15,6 @@ limitations under the License.
 
 #include "tensorflow/stream_executor/stream_executor_internal.h"
 
-#include "tensorflow/stream_executor/lib/error.h"
-#include "tensorflow/stream_executor/lib/statusor.h"
-#include "tensorflow/stream_executor/lib/stringprintf.h"
-
 namespace perftools {
 namespace gputools {
 namespace internal {
@@ -41,21 +37,6 @@ StreamExecutorFactory* MakeOpenCLExecutorImplementation() {
 
 StreamExecutorFactory MakeHostExecutorImplementation;
 
-// TODO(b/70298427) There are two similar methods:
-//   bool BlockHostUntilDone(Stream*);
-//   Status BlockHostUntilDoneWithStatus(Stream*);
-//
-// The intention is to replace all implementations of the bool version with the
-// Status version.  In the meantime, just implement one in terms of the other.
-port::Status StreamExecutorInterface::BlockHostUntilDoneWithStatus(
-    Stream* stream) {
-  if (!BlockHostUntilDone(stream)) {
-    return port::Status(port::error::INTERNAL,
-                        "Failed to block host until done.");
-  }
-  return port::Status::OK();
-}
-
 }  // namespace internal
 }  // namespace gputools
 }  // namespace perftools
diff --git a/tensorflow/stream_executor/stream_executor_internal.h b/tensorflow/stream_executor/stream_executor_internal.h
index d2426f46e2..0a9bef71d0 100644
--- a/tensorflow/stream_executor/stream_executor_internal.h
+++ b/tensorflow/stream_executor/stream_executor_internal.h
@@ -219,8 +219,7 @@ class StreamExecutorInterface {
   virtual void DeallocateTimer(Timer *timer) = 0;
   virtual bool StartTimer(Stream *stream, Timer *timer) = 0;
   virtual bool StopTimer(Stream *stream, Timer *timer) = 0;
-  virtual bool BlockHostUntilDone(Stream *stream) = 0;
-  virtual port::Status BlockHostUntilDoneWithStatus(Stream *stream);
+  virtual port::Status BlockHostUntilDoneWithStatus(Stream *stream) = 0;
   virtual int PlatformDeviceCount() = 0;
   virtual port::Status EnablePeerAccessTo(StreamExecutorInterface *other) = 0;
   virtual bool CanEnablePeerAccessTo(StreamExecutorInterface *other) = 0;