xref: /aosp_15_r20/external/tensorflow/tensorflow/compiler/xla/service/reduce_scatter_combiner.h (revision b6fb3261f9314811a0f4371741dbb8839866f948)
1 /* Copyright 2021 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 
16 #ifndef TENSORFLOW_COMPILER_XLA_SERVICE_REDUCE_SCATTER_COMBINER_H_
17 #define TENSORFLOW_COMPILER_XLA_SERVICE_REDUCE_SCATTER_COMBINER_H_
18 
19 #include "tensorflow/compiler/xla/service/hlo_module.h"
20 #include "tensorflow/compiler/xla/service/hlo_pass_interface.h"
21 #include "tensorflow/compiler/xla/statusor.h"
22 
23 namespace xla {
24 
25 // Combines small non-dependent ReduceScatter ops into larger combined
26 // ReduceScatter ops. A typical ReduceScatter implementation has a minimum
27 // latency-induced time for a ReduceScatter op so a single combined op can be
28 // more efficient than many small ones.
29 class ReduceScatterCombiner : public HloModulePass {
30  public:
31   ReduceScatterCombiner(int64_t combine_threshold_in_bytes,
32                         int64_t combine_threshold_count);
33 
name()34   absl::string_view name() const override { return "reduce-scatter-combiner"; }
35 
36   using HloPassInterface::Run;
37   StatusOr<bool> Run(
38       HloModule* module,
39       const absl::flat_hash_set<absl::string_view>& execution_threads) override;
40 
41  private:
42   // Combine reduce-scatter ops up to this threshold.
43   int64_t combine_threshold_in_bytes_;
44 
45   // Combine reduce-scatter ops up to this threshold (number of operands).
46   int64_t combine_threshold_count_;
47 };
48 
49 }  // namespace xla
50 
51 #endif  // TENSORFLOW_COMPILER_XLA_SERVICE_REDUCE_SCATTER_COMBINER_H_
52