Make Event::PollForStatus a virtual method, and override it where nec…

…essary.

This enables Event to be a completely virtual base class, and removes it from circular dependencies.

PiperOrigin-RevId: 636235458

tensorflow/c/experimental/stream_executor/stream_executor.cc

@@ -154,20 +154,6 @@ absl::Status ValidateSEPlatformRegistrationParams(
 }
 #undef TF_VALIDATE_NOT_NULL
 
-// Converts SE_EventStatus to Event::Status.
-Event::Status SEEventStatusToEventStatus(SE_EventStatus s) {
-  switch (s) {
-    case SE_EVENT_ERROR:
-      return Event::Status::kError;
-    case SE_EVENT_PENDING:
-      return Event::Status::kPending;
-    case SE_EVENT_COMPLETE:
-      return Event::Status::kComplete;
-    default:
-      return Event::Status::kUnknown;
-  }
-}
-
 // Converts DeviceMemoryBase to a C struct.
 SP_DeviceMemoryBase DeviceMemoryBaseToC(const DeviceMemoryBase* mem) {
   SP_DeviceMemoryBase device_memory_base{SP_DEVICE_MEMORY_BASE_STRUCT_SIZE};
@@ -422,12 +408,6 @@ class CStreamExecutor : public StreamExecutor {
     absl::Status s = StatusFromTF_Status(c_status.get());
     return s;
   }
-  Event::Status PollForEventStatus(Event* event) override {
-    SP_Event event_handle = static_cast<CEvent*>(event)->Handle();
-    SE_EventStatus event_status =
-        stream_executor_->get_event_status(&device_, event_handle);
-    return SEEventStatusToEventStatus(event_status);
-  }
   void DeallocateStream(Stream* stream) override {
     static_cast<CStream*>(stream->implementation())->Destroy();
   }
@@ -541,7 +521,7 @@ class CStreamExecutor : public StreamExecutor {
   }
 
   absl::StatusOr<std::unique_ptr<Event>> CreateEvent() override {
-    auto c_event = std::make_unique<CEvent>(&device_, stream_executor_, this);
+    auto c_event = std::make_unique<CEvent>(&device_, stream_executor_);
     TF_RETURN_IF_ERROR(c_event->Create());
     return std::move(c_event);
   }

tensorflow/c/experimental/stream_executor/stream_executor_internal.h

@@ -129,14 +129,28 @@ class CStream : public StreamInterface {
 
 class CEvent : public Event {
  public:
-  CEvent(SP_Device* device, SP_StreamExecutor* stream_executor,
-         StreamExecutorInterface* executor_interface)
-      : Event(executor_interface),
-        device_(device),
+  CEvent(SP_Device* device, SP_StreamExecutor* stream_executor)
+      : device_(device),
         stream_executor_(stream_executor),
         event_handle_(nullptr) {}
   ~CEvent() override { Destroy(); }
 
+  Event::Status PollForStatus() override {
+    SE_EventStatus event_status =
+        stream_executor_->get_event_status(device_, event_handle_);
+
+    switch (event_status) {
+      case SE_EVENT_ERROR:
+        return Event::Status::kError;
+      case SE_EVENT_PENDING:
+        return Event::Status::kPending;
+      case SE_EVENT_COMPLETE:
+        return Event::Status::kComplete;
+      default:
+        return Event::Status::kUnknown;
+    }
+  }
+
   absl::Status Create() {
     tensorflow::TF_StatusPtr c_status(TF_NewStatus());
     stream_executor_->create_event(device_, &event_handle_, c_status.get());

third_party/xla/xla/backends/interpreter/executor.h

@@ -121,10 +121,6 @@ class XlaInterpreterExecutor : public StreamExecutor {
     return absl::Status{absl::StatusCode::kUnimplemented, "WaitForEvent"};
   }
 
-  Event::Status PollForEventStatus(Event *event) override {
-    return Event::Status::kError;
-  }
-
   void DeallocateStream(Stream *stream) override {}
   bool CreateStreamDependency(Stream *dependent, Stream *other) override;
 
@@ -150,7 +146,7 @@ class XlaInterpreterExecutor : public StreamExecutor {
     return true;
   }
   absl::StatusOr<std::unique_ptr<Event>> CreateEvent() override {
-    return std::make_unique<Event>(this);
+    return std::make_unique<Event>();
   }
 
   absl::StatusOr<std::unique_ptr<Stream>> CreateStream(

third_party/xla/xla/stream_executor/BUILD

@@ -658,7 +658,6 @@ transitive_hdrs(
 cc_library(
     name = "stream_executor_pimpl",
     srcs = [
-        "event.cc",
         "stream.cc",
         "stream_executor_pimpl.cc",
     ],

third_party/xla/xla/stream_executor/cuda/cuda_event.cc

@@ -13,6 +13,8 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
+#include "xla/stream_executor/cuda/cuda_event.h"
+
 #include "absl/log/log.h"
 #include "absl/status/statusor.h"
 #include "third_party/gpus/cuda/include/cuda.h"
@@ -24,9 +26,9 @@ limitations under the License.
 namespace stream_executor {
 namespace gpu {
 
-Event::Status GpuEvent::PollForStatus() {
+Event::Status CudaEvent::PollForStatus() {
   absl::StatusOr<CUresult> status =
-      QueryEvent(parent_->gpu_context(), gpu_event_);
+      QueryEvent(parent()->gpu_context(), gpu_event());
   if (!status.ok()) {
     LOG(ERROR) << "Error polling for event status: "
                << status.status().message();

third_party/xla/xla/stream_executor/cuda/cuda_event.h

@@ -16,14 +16,20 @@ limitations under the License.
 #ifndef XLA_STREAM_EXECUTOR_CUDA_CUDA_EVENT_H_
 #define XLA_STREAM_EXECUTOR_CUDA_CUDA_EVENT_H_
 
+#include "xla/stream_executor/event.h"
 #include "xla/stream_executor/gpu/gpu_event.h"
+#include "xla/stream_executor/gpu/gpu_executor.h"
 
-namespace stream_executor {
-namespace cuda {
+namespace stream_executor::gpu {
 
-using CUDAEvent = gpu::GpuEvent;
+// This class implements Event::PollForStatus for CUDA devices.
+class CudaEvent : public GpuEvent {
+ public:
+  explicit CudaEvent(GpuExecutor *executor) : GpuEvent(executor) {}
 
-}  // namespace cuda
-}  // namespace stream_executor
+  Event::Status PollForStatus() override;
+};
+
+}  // namespace stream_executor::gpu
 
 #endif  // XLA_STREAM_EXECUTOR_CUDA_CUDA_EVENT_H_

third_party/xla/xla/stream_executor/cuda/cuda_executor.cc

@@ -60,6 +60,7 @@ limitations under the License.
 #include "xla/stream_executor/command_buffer.h"
 #include "xla/stream_executor/cuda/cuda_diagnostics.h"
 #include "xla/stream_executor/cuda/cuda_driver.h"
+#include "xla/stream_executor/cuda/cuda_event.h"
 #include "xla/stream_executor/cuda/cuda_platform_id.h"
 #include "xla/stream_executor/gpu/gpu_collectives.h"
 #include "xla/stream_executor/gpu/gpu_command_buffer.h"
@@ -772,10 +773,6 @@ absl::Status GpuExecutor::WaitForEvent(Stream* stream, Event* event) {
   }
 }
 
-Event::Status GpuExecutor::PollForEventStatus(Event* event) {
-  return AsGpuEvent(event)->PollForStatus();
-}
-
 void GpuExecutor::DeallocateStream(Stream* stream) {
   {
     absl::MutexLock lock(&mu_);
@@ -931,7 +928,7 @@ absl::Status FillBlockDimLimit(GpuDeviceHandle device,
 }
 
 absl::StatusOr<std::unique_ptr<Event>> GpuExecutor::CreateEvent() {
-  auto gpu_event = std::make_unique<GpuEvent>(this);
+  auto gpu_event = std::make_unique<CudaEvent>(this);
   TF_RETURN_IF_ERROR(gpu_event->Init());
   return std::move(gpu_event);
 }

third_party/xla/xla/stream_executor/event.h

@@ -22,8 +22,6 @@ limitations under the License.
 
 namespace stream_executor {
 
-class StreamExecutorInterface;
-
 // The Event class, when supported by a platform, enables low-overhead status
 // reporting for a Stream. An Event is inserted at a location in a stream via
 // the Stream::RecordEvent() API. From then on, the Event's status can be
@@ -41,30 +39,17 @@ class Event {
     kComplete,
   };
 
-  explicit Event(StreamExecutorInterface* stream_exec);
-
   // Releases any resources held by the Event object.
   virtual ~Event() = default;
 
   // Returns the current Status for the event.
-  Status PollForStatus();
+  virtual Status PollForStatus() { return Status::kError; }
 
   // Blocks `stream` on this event. `stream` is a raw platform-specific
   // stream (e.g. GpuStreamHandle).
   virtual absl::Status WaitForEventOnExternalStream(std::intptr_t stream) {
     return absl::UnimplementedError("Not supported for this Event.");
   }
-
-  Event(Event&&) = default;
-  Event& operator=(Event&&) = default;
-
- private:
-  // Pointer to the StreamExecutorInterface interface used to create this
-  // object. Not owned.
-  StreamExecutorInterface* stream_exec_;
-
-  Event(const Event&) = delete;
-  void operator=(const Event&) = delete;
 };
 
 }  // namespace stream_executor

third_party/xla/xla/stream_executor/gpu/gpu_event.cc

@@ -29,7 +29,7 @@ namespace stream_executor {
 namespace gpu {
 
 GpuEvent::GpuEvent(GpuExecutor* parent)
-    : Event(parent), parent_(parent), gpu_event_(nullptr) {}
+    : parent_(parent), gpu_event_(nullptr) {}
 
 GpuEvent::~GpuEvent() { Destroy().IgnoreError(); }
 

third_party/xla/xla/stream_executor/gpu/gpu_event.h

@@ -43,14 +43,14 @@ class GpuEvent : public Event {
   // Inserts the event at the current position into the specified stream.
   absl::Status Record(GpuStream* stream);
 
-  // Polls the CUDA platform for the event's current status.
-  Event::Status PollForStatus();
-
   // The underlying CUDA event element.
   GpuEventHandle gpu_event();
 
   absl::Status WaitForEventOnExternalStream(std::intptr_t stream) override;
 
+ protected:
+  GpuExecutor* parent() const { return parent_; }
+
  private:
   // The Executor used to which this object and GpuEventHandle are bound.
   GpuExecutor* parent_;

third_party/xla/xla/stream_executor/gpu/gpu_executor.h

@@ -237,8 +237,6 @@ class GpuExecutor : public StreamExecutor {
 
   absl::Status WaitForEvent(Stream* stream, Event* event) override;
 
-  Event::Status PollForEventStatus(Event* event) override;
-
   absl::Status BlockHostUntilDone(Stream* stream) override;
 
   absl::Status EnablePeerAccessTo(StreamExecutorInterface* other) override;

third_party/xla/xla/stream_executor/host/host_executor.cc

@@ -223,12 +223,15 @@ bool HostExecutor::CreateStreamDependency(Stream* dependent, Stream* other) {
 
 class HostEvent : public Event {
  public:
-  HostEvent(StreamExecutorInterface* executor)
-      : Event(executor),
-        notification_(std::make_shared<absl::Notification>()) {}
+  HostEvent() : notification_(std::make_shared<absl::Notification>()) {}
 
   std::shared_ptr<absl::Notification>& notification() { return notification_; }
 
+  Status PollForStatus() override {
+    return notification_->HasBeenNotified() ? Event::Status::kComplete
+                                            : Event::Status::kPending;
+  }
+
  private:
   // We use a std::shared_ptr here because the client may delete the HostEvent
   // object while there are still RecordEvent and WaitForEvent callbacks pending
@@ -237,7 +240,7 @@ class HostEvent : public Event {
 };
 
 absl::StatusOr<std::unique_ptr<Event>> HostExecutor::CreateEvent() {
-  return std::make_unique<HostEvent>(this);
+  return std::make_unique<HostEvent>();
 }
 
 static HostEvent* AsHostEvent(Event* event) {
@@ -263,12 +266,6 @@ absl::Status HostExecutor::WaitForEvent(Stream* stream, Event* event) {
   return absl::OkStatus();
 }
 
-Event::Status HostExecutor::PollForEventStatus(Event* event) {
-  absl::Notification& notification = *AsHostEvent(event)->notification();
-  return notification.HasBeenNotified() ? Event::Status::kComplete
-                                        : Event::Status::kPending;
-}
-
 absl::Status HostExecutor::BlockHostUntilDone(Stream* stream) {
   return AsHostStream(stream)->BlockUntilDone();
 }

third_party/xla/xla/stream_executor/host/host_executor.h

@@ -121,7 +121,6 @@ class HostExecutor : public StreamExecutor {
 
   absl::Status RecordEvent(Stream* stream, Event* event) override;
   absl::Status WaitForEvent(Stream* stream, Event* event) override;
-  Event::Status PollForEventStatus(Event* event) override;
 
   void DeallocateStream(Stream* stream) override;
   bool CreateStreamDependency(Stream* dependent, Stream* other) override;

third_party/xla/xla/stream_executor/mock_stream_executor.h

@@ -136,7 +136,6 @@ class MockStreamExecutor : public StreamExecutorInterface {
               (override));
   MOCK_METHOD(absl::Status, WaitForEvent, (Stream * stream, Event* event),
               (override));
-  MOCK_METHOD(Event::Status, PollForEventStatus, (Event * event), (override));
   MOCK_METHOD(void, DeallocateStream, (Stream * stream), (override));
   MOCK_METHOD(bool, CreateStreamDependency, (Stream * dependent, Stream* other),
               (override));

third_party/xla/xla/stream_executor/rocm/BUILD

@@ -142,6 +142,7 @@ cc_library(
 cc_library(
     name = "rocm_event",
     srcs = if_rocm_is_configured(["rocm_event.cc"]),
+    hdrs = if_rocm_is_configured(["rocm_event.h"]),
     deps = if_rocm_is_configured([
         # keep sorted
         ":rocm_driver",

third_party/xla/xla/stream_executor/rocm/rocm_event.cc

@@ -13,6 +13,8 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
+#include "xla/stream_executor/rocm/rocm_event.h"
+
 #include "xla/stream_executor/gpu/gpu_event.h"
 #include "xla/stream_executor/gpu/gpu_executor.h"
 #include "xla/stream_executor/gpu/gpu_stream.h"
@@ -21,9 +23,9 @@ limitations under the License.
 namespace stream_executor {
 namespace gpu {
 
-Event::Status GpuEvent::PollForStatus() {
+Event::Status RocmEvent::PollForStatus() {
   absl::StatusOr<hipError_t> status =
-      QueryEvent(parent_->gpu_context(), gpu_event_);
+      QueryEvent(parent()->gpu_context(), gpu_event());
   if (!status.ok()) {
     LOG(ERROR) << "Error polling for event status: "
                << status.status().message();

third_party/xla/xla/stream_executor/event.cc → third_party/xla/xla/stream_executor/rocm/rocm_event.h

@@ -1,4 +1,4 @@
-/* Copyright 2015 The OpenXLA Authors.
+/* Copyright 2018 The OpenXLA Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
@@ -13,23 +13,21 @@ See the License for the specific language governing permissions and
 limitations under the License.
 ==============================================================================*/
 
-#include "xla/stream_executor/event.h"
+#ifndef XLA_STREAM_EXECUTOR_ROCM_ROCM_EVENT_H_
+#define XLA_STREAM_EXECUTOR_ROCM_ROCM_EVENT_H_
 
-#include <cstdint>
-#include <memory>
-#include <utility>
+#include "xla/stream_executor/gpu/gpu_event.h"
+#include "xla/stream_executor/gpu/gpu_executor.h"
 
-#include "absl/log/log.h"
-#include "absl/status/status.h"
-#include "xla/stream_executor/stream_executor_interface.h"
+namespace stream_executor::gpu {
 
-namespace stream_executor {
+// This class implements Event::PollForStatus for ROCm devices.
+class RocmEvent : public GpuEvent {
+ public:
+  explicit RocmEvent(GpuExecutor *executor) : GpuEvent(executor) {}
 
-Event::Event(StreamExecutorInterface* stream_exec)
-    : stream_exec_(stream_exec) {}
+  Event::Status PollForStatus() override;
+};
+}  // namespace stream_executor::gpu
 
-Event::Status Event::PollForStatus() {
-  return stream_exec_->PollForEventStatus(this);
-}
-
-}  // namespace stream_executor
+#endif  // XLA_STREAM_EXECUTOR_ROCM_ROCM_EVENT_H_