wehub-resource-sync
67 lines
2.3 KiB
C++
67 lines
2.3 KiB
C++
/* Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
you may not use this file except in compliance with the License.
|
|
You may obtain a copy of the License at
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
See the License for the specific language governing permissions and
|
|
limitations under the License. */
|
|
|
|
#pragma once
|
|
|
|
#include <vector>
|
|
|
|
#include "paddle/phi/core/distributed/auto_parallel/dist_meta_tensor.h"
|
|
#include "paddle/phi/core/distributed/type_defs.h"
|
|
#include "paddle/phi/infermeta/spmd_rules/utils.h"
|
|
|
|
namespace phi {
|
|
namespace distributed {
|
|
/**
|
|
* A **hack** rule with a strong assumption that the first dimension of
|
|
* all the input and output tensors is the batch dimension (broadcast
|
|
* dimension), therefore, if any tensor's first dimension is sharded, the
|
|
* sharding would be propagating to all the other tensors (for tensor first
|
|
* dimension). All the other axes of tensors would be set as unshard (-1).
|
|
*
|
|
*
|
|
* This rule is used to support emerging op for hybrid parallelism quickly, and
|
|
* once there is a specific rule for that op, we should remove that op from
|
|
* this rule.
|
|
*
|
|
* Vector of input tensors and output tensors used as arguments (for both
|
|
* inferfw & inferbw) to support any kind of op.
|
|
*
|
|
*/
|
|
SpmdInfo DefaultDataParallelInferSpmd(
|
|
const std::vector<const DistMetaTensor*>& ins,
|
|
const std::vector<const DistMetaTensor*>& outs);
|
|
|
|
SpmdInfo DefaultDataParallelInferSpmdReverse(
|
|
const std::vector<const DistMetaTensor*>& ins,
|
|
const std::vector<const DistMetaTensor*>& outs);
|
|
|
|
// For phi api
|
|
template <typename... Args>
|
|
SpmdInfo VariadicDefaultDataParallelInferSpmd(const Args&... args) {
|
|
return detail::VariadicSpmdRuleArgumentParser<DefaultDataParallelInferSpmd>()
|
|
.apply(args...)
|
|
.InferForward();
|
|
}
|
|
|
|
template <typename... Args>
|
|
SpmdInfo VariadicDefaultDataParallelInferSpmdReverse(const Args&... args) {
|
|
return detail::VariadicSpmdRuleArgumentParser<
|
|
DefaultDataParallelInferSpmdReverse>()
|
|
.apply(args...)
|
|
.InferBackward();
|
|
}
|
|
|
|
} // namespace distributed
|
|
} // namespace phi
|