xref: /aosp_15_r20/external/tensorflow/tensorflow/core/common_runtime/collective_rma_local.h (revision b6fb3261f9314811a0f4371741dbb8839866f948)
1 /* Copyright 2018 The TensorFlow Authors. All Rights Reserved.
2 
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6 
7     http://www.apache.org/licenses/LICENSE-2.0
8 
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15 #ifndef TENSORFLOW_CORE_COMMON_RUNTIME_COLLECTIVE_RMA_LOCAL_H_
16 #define TENSORFLOW_CORE_COMMON_RUNTIME_COLLECTIVE_RMA_LOCAL_H_
17 
18 #include "tensorflow/core/common_runtime/buf_rendezvous.h"
19 #include "tensorflow/core/common_runtime/device_mgr.h"
20 #include "tensorflow/core/framework/collective.h"
21 #include "tensorflow/core/framework/rendezvous.h"
22 
23 namespace tensorflow {
24 
25 // Basic implementation of PerStepCollectiveRemoteAccess.
26 class CollectiveRemoteAccessLocal : public CollectiveRemoteAccess {
27  public:
CollectiveRemoteAccessLocal(const DeviceMgr * dev_mgr,DeviceResolverInterface * dev_resolver,int64_t step_id)28   CollectiveRemoteAccessLocal(const DeviceMgr* dev_mgr,
29                               DeviceResolverInterface* dev_resolver,
30                               int64_t step_id)
31       : dev_mgr_(dev_mgr),
32         dev_resolver_(dev_resolver),
33         buf_rendezvous_(step_id, dev_mgr),
34         step_id_(step_id) {}
35 
36   ~CollectiveRemoteAccessLocal() override = default;
37 
38   void StartAbort(const Status& s) override;
39 
40   void RecvFromPeer(const string& peer_device, const string& peer_task,
41                     bool peer_is_local, const string& key, Device* to_device,
42                     DeviceContext* to_device_ctx,
43                     const AllocatorAttributes& to_alloc_attr, Tensor* to_tensor,
44                     const DeviceLocality& client_locality,
45                     int dev_to_dev_stream_index,
46                     CancellationManager* cancellation_manager,
47                     const StatusCallback& done) override;
48 
49   void PostToPeer(const string& peer_device, const string& peer_task,
50                   const string& key, Device* from_device,
51                   DeviceContext* from_device_ctx,
52                   const AllocatorAttributes& from_alloc_attr,
53                   const Tensor* from_tensor,
54                   const DeviceLocality& client_locality,
55                   CancellationManager* cancellation_manager,
56                   const StatusCallback& done) override;
57 
58   void CheckPeerHealth(const string& peer_task, int64_t timeout_in_ms,
59                        const StatusCallback& done) override;
60 
buf_rendezvous()61   BufRendezvous* buf_rendezvous() override { return &buf_rendezvous_; }
62 
63   // Copy utility that always copies bytes from src to dst even if
64   // they are on the same device, unlike CopyTensor::ViaDMA which will
65   // just change the dst buffer pointer in that case.
66   static void MemCpyAsync(DeviceContext* src_dev_ctx,
67                           DeviceContext* dst_dev_ctx, Device* src_dev,
68                           Device* dst_dev, const AllocatorAttributes& src_attr,
69                           const AllocatorAttributes& dst_attr,
70                           const Tensor* src, Tensor* dst,
71                           int dev_to_dev_stream_index,
72                           const StatusCallback& done);
73 
74  protected:
75   const DeviceMgr* dev_mgr_;               // not owned
76   DeviceResolverInterface* dev_resolver_;  // not owned
77   BufRendezvous buf_rendezvous_;
78   int64_t step_id_;
79 };
80 
81 }  // namespace tensorflow
82 #endif  // TENSORFLOW_CORE_COMMON_RUNTIME_COLLECTIVE_RMA_LOCAL_H_
83