Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
翠花的腿毛
Paddle
提交
06c3cce9
P
Paddle
项目概览
翠花的腿毛
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
1
Fork
0
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
未验证
提交
06c3cce9
编写于
12月 01, 2021
作者:
Z
Zhanlue Yang
提交者:
GitHub
12月 01, 2021
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Handled dispensable tensors in AutoCodeGen for Eager Dygraph (#37723)
上级
f91e2331
变更
3
展开全部
隐藏空白更改
内联
并排
Showing
3 changed file
with
314 addition
and
175 deletion
+314
-175
paddle/fluid/eager/auto_code_generator/eager_generator.cc
paddle/fluid/eager/auto_code_generator/eager_generator.cc
+191
-73
paddle/fluid/pybind/op_function_generator.cc
paddle/fluid/pybind/op_function_generator.cc
+2
-102
paddle/fluid/pybind/op_function_generator.h
paddle/fluid/pybind/op_function_generator.h
+121
-0
未找到文件。
paddle/fluid/eager/auto_code_generator/eager_generator.cc
浏览文件 @
06c3cce9
此差异已折叠。
点击以展开。
paddle/fluid/pybind/op_function_generator.cc
浏览文件 @
06c3cce9
...
...
@@ -12,6 +12,8 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#include "paddle/fluid/pybind/op_function_generator.h"
#include <algorithm>
#include <fstream>
#include <iostream>
...
...
@@ -30,108 +32,6 @@
#include "paddle/fluid/framework/fleet/ascend_wrapper.h"
#endif
// NOTE(zhiqiu): Commonly, the inputs in auto-generated OP function are
// determined by the OP`s proto automatically, i.e., all the inputs registered
// in OpMaker.
// However, some OPs have dispensable inputs, which means the input can
// be none for some conditions. It is discovered that most dispensable inputs
// is not used in imperative mode, so we drop those inputs when generating OP
// functions. While, for very few OPs, the dispensable inputs are used, we
// need to manually specify them in this map.
std
::
map
<
std
::
string
,
std
::
set
<
std
::
string
>>
op_ins_map
=
{
{
"layer_norm"
,
{
"X"
,
"Scale"
,
"Bias"
}},
{
"bincount"
,
{
"X"
,
"Weights"
}},
{
"fused_attention"
,
{
"X"
,
"LnScale"
,
"LnBias"
,
"QKVW"
,
"QKVBias"
,
"SrcMask"
,
"OutLinearW"
,
"OutLinearBias"
,
"Ln2Scale"
,
"Ln2Bias"
}},
{
"instance_norm"
,
{
"X"
,
"Scale"
,
"Bias"
}},
{
"gru_unit"
,
{
"Input"
,
"HiddenPrev"
,
"Weight"
,
"Bias"
}},
{
"label_smooth"
,
{
"X"
,
"PriorDist"
}},
{
"assign"
,
{
"X"
}},
{
"reshape2"
,
{
"X"
,
"Shape"
}},
{
"expand"
,
{
"X"
,
"ExpandTimes"
}},
{
"slice"
,
{
"Input"
,
"StartsTensor"
,
"EndsTensor"
}},
{
"fake_quantize_dequantize_moving_average_abs_max"
,
{
"X"
,
"InScale"
,
"InAccum"
,
"InState"
}},
{
"nll_loss"
,
{
"X"
,
"Label"
,
"Weight"
}},
{
"bilinear_tensor_product"
,
{
"X"
,
"Y"
,
"Weight"
,
"Bias"
}},
{
"gather"
,
{
"X"
,
"Index"
,
"Axis"
}},
{
"roi_pool"
,
{
"X"
,
"ROIs"
,
"RoisNum"
}},
{
"roi_align"
,
{
"X"
,
"ROIs"
,
"RoisNum"
}},
{
"psroi_pool"
,
{
"X"
,
"ROIs"
,
"RoisNum"
}},
{
"collect_fpn_proposals"
,
{
"MultiLevelRois"
,
"MultiLevelScores"
,
"MultiLevelRoIsNum"
}},
{
"distribute_fpn_proposals"
,
{
"FpnRois"
,
"RoisNum"
}},
{
"warpctc"
,
{
"Logits"
,
"Label"
,
"LogitsLength"
,
"LabelLength"
}},
{
"hierarchical_sigmoid"
,
{
"X"
,
"W"
,
"Label"
,
"PathTable"
,
"PathCode"
,
"Bias"
}},
{
"moving_average_abs_max_scale"
,
{
"X"
,
"InAccum"
,
"InState"
}},
{
"multiclass_nms3"
,
{
"BBoxes"
,
"Scores"
,
"RoisNum"
}},
{
"box_coder"
,
{
"PriorBox"
,
"PriorBoxVar"
,
"TargetBox"
}},
{
"momentum"
,
{
"Param"
,
"Grad"
,
"Velocity"
,
"LearningRate"
,
"MasterParam"
}},
{
"sparse_momentum"
,
{
"Param"
,
"Grad"
,
"Velocity"
,
"Index"
,
"LearningRate"
}},
{
"rnn"
,
{
"Input"
,
"PreState"
,
"WeightList"
,
"SequenceLength"
}},
{
"run_program"
,
{
"X"
,
"Params"
}},
{
"fused_feedforward"
,
{
"Dropout1Seed"
,
"Dropout2Seed"
,
"Linear1Bias"
,
"Linear2Bias"
,
"Ln1Scale"
,
"Ln1Bias"
,
"Ln2Scale"
,
"Ln2Bias"
}},
{
"faster_tokenizer"
,
{
"Text"
,
"Vocab"
,
"TextPair"
}},
{
"matrix_rank"
,
{
"X"
,
"TolTensor"
}},
{
"adam"
,
{
"Param"
,
"Grad"
,
"LearningRate"
,
"Moment1"
,
"Moment2"
,
"Beta1Pow"
,
"Beta2Pow"
,
"MasterParam"
}},
{
"adamw"
,
{
"Param"
,
"Grad"
,
"LearningRate"
,
"Moment1"
,
"Moment2"
,
"Beta1Pow"
,
"Beta2Pow"
,
"MasterParam"
}},
};
// NOTE(zhiqiu): Like op_ins_map.
// Commonly, the outputs in auto-generated OP function are determined by the
// OP`s proto automatically, i.e., all the outputs registered in OpMaker.
// However, some OPs have dispensable outputs, which means the output can
// be none for some conditions. It is discovered that most dispensable outputs
// is not used in imperative mode, so we drop those outputs when generating OP
// functions. While, for very few OPs, the dispensable outputs are used, we
// need to manually specify them in this map.
std
::
map
<
std
::
string
,
std
::
set
<
std
::
string
>>
op_outs_map
=
{
{
"fake_quantize_dequantize_moving_average_abs_max"
,
{
"Out"
,
"OutScale"
,
"OutAccum"
,
"OutState"
}},
{
"batch_norm"
,
{
"Y"
,
"MeanOut"
,
"VarianceOut"
,
"SavedMean"
,
"SavedVariance"
,
"ReserveSpace"
}},
{
"fused_attention"
,
{
"LnMean"
,
"LnVariance"
,
"LnOut"
,
"QKVOut"
,
"QKVBiasOut"
,
"TransposeOut2"
,
"QKOut"
,
"QKTVOut"
,
"SoftmaxOut"
,
"AttnDropoutMaskOut"
,
"AttnDropoutOut"
,
"SrcMaskOut"
,
"FMHAOut"
,
"OutLinearOut"
,
"DropoutMaskOut"
,
"Ln2Mean"
,
"Ln2Variance"
,
"BiasDropoutResidualOut"
,
"Y"
}},
{
"sync_batch_norm"
,
{
"Y"
,
"MeanOut"
,
"VarianceOut"
,
"SavedMean"
,
"SavedVariance"
,
"ReserveSpace"
}},
{
"unique"
,
{
"Out"
,
"Index"
,
"Indices"
,
"Counts"
}},
{
"unique_consecutive"
,
{
"Out"
,
"Index"
,
"Counts"
}},
{
"generate_proposals"
,
{
"RpnRois"
,
"RpnRoiProbs"
,
"RpnRoisNum"
}},
{
"collect_fpn_proposals"
,
{
"FpnRois"
,
"RoisNum"
}},
{
"matrix_nms"
,
{
"Out"
,
"Index"
,
"RoisNum"
}},
{
"distribute_fpn_proposals"
,
{
"MultiFpnRois"
,
"RestoreIndex"
,
"MultiLevelRoIsNum"
}},
{
"moving_average_abs_max_scale"
,
{
"Out"
,
"OutScale"
,
"OutAccum"
,
"OutState"
}},
{
"multiclass_nms3"
,
{
"Out"
,
"NmsRoisNum"
}},
{
"generate_proposals_v2"
,
{
"RpnRois"
,
"RpnRoiProbs"
,
"RpnRoisNum"
}},
{
"momentum"
,
{
"ParamOut"
,
"VelocityOut"
,
"MasterParamOut"
}},
{
"sparse_momentum"
,
{
"ParamOut"
,
"VelocityOut"
}},
{
"rnn"
,
{
"DropoutState"
,
"Reserve"
,
"Out"
,
"State"
}},
{
"lamb"
,
{
"ParamOut"
,
"Moment1Out"
,
"Moment2Out"
,
"Beta1PowOut"
,
"Beta2PowOut"
}},
{
"run_program"
,
{
"DOut"
}},
{
"adam"
,
{
"ParamOut"
,
"Moment1Out"
,
"Moment2Out"
,
"Beta1PowOut"
,
"Beta2PowOut"
,
"MasterParamOut"
}},
{
"adamw"
,
{
"ParamOut"
,
"Moment1Out"
,
"Moment2Out"
,
"Beta1PowOut"
,
"Beta2PowOut"
,
"MasterParamOut"
}},
};
// NOTE(zhiqiu): Commonly, the outputs in auto-generated OP function are
// generated in C++ automatically.
// However, some OPs need to pass the outputs from Python instead of generating
...
...
paddle/fluid/pybind/op_function_generator.h
0 → 100644
浏览文件 @
06c3cce9
// Copyright (c) 2021 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 <map>
#include <set>
#include <string>
// NOTE(zhiqiu): Commonly, the inputs in auto-generated OP function are
// determined by the OP`s proto automatically, i.e., all the inputs registered
// in OpMaker.
// However, some OPs have dispensable inputs, which means the input can
// be none for some conditions. It is discovered that most dispensable inputs
// is not used in imperative mode, so we drop those inputs when generating OP
// functions. While, for very few OPs, the dispensable inputs are used, we
// need to manually specify them in this map.
std
::
map
<
std
::
string
,
std
::
set
<
std
::
string
>>
op_ins_map
=
{
{
"layer_norm"
,
{
"X"
,
"Scale"
,
"Bias"
}},
{
"bincount"
,
{
"X"
,
"Weights"
}},
{
"fused_attention"
,
{
"X"
,
"LnScale"
,
"LnBias"
,
"QKVW"
,
"QKVBias"
,
"SrcMask"
,
"OutLinearW"
,
"OutLinearBias"
,
"Ln2Scale"
,
"Ln2Bias"
}},
{
"instance_norm"
,
{
"X"
,
"Scale"
,
"Bias"
}},
{
"gru_unit"
,
{
"Input"
,
"HiddenPrev"
,
"Weight"
,
"Bias"
}},
{
"label_smooth"
,
{
"X"
,
"PriorDist"
}},
{
"assign"
,
{
"X"
}},
{
"reshape2"
,
{
"X"
,
"Shape"
}},
{
"expand"
,
{
"X"
,
"ExpandTimes"
}},
{
"slice"
,
{
"Input"
,
"StartsTensor"
,
"EndsTensor"
}},
{
"fake_quantize_dequantize_moving_average_abs_max"
,
{
"X"
,
"InScale"
,
"InAccum"
,
"InState"
}},
{
"nll_loss"
,
{
"X"
,
"Label"
,
"Weight"
}},
{
"bilinear_tensor_product"
,
{
"X"
,
"Y"
,
"Weight"
,
"Bias"
}},
{
"gather"
,
{
"X"
,
"Index"
,
"Axis"
}},
{
"roi_pool"
,
{
"X"
,
"ROIs"
,
"RoisNum"
}},
{
"roi_align"
,
{
"X"
,
"ROIs"
,
"RoisNum"
}},
{
"psroi_pool"
,
{
"X"
,
"ROIs"
,
"RoisNum"
}},
{
"collect_fpn_proposals"
,
{
"MultiLevelRois"
,
"MultiLevelScores"
,
"MultiLevelRoIsNum"
}},
{
"distribute_fpn_proposals"
,
{
"FpnRois"
,
"RoisNum"
}},
{
"warpctc"
,
{
"Logits"
,
"Label"
,
"LogitsLength"
,
"LabelLength"
}},
{
"hierarchical_sigmoid"
,
{
"X"
,
"W"
,
"Label"
,
"PathTable"
,
"PathCode"
,
"Bias"
}},
{
"moving_average_abs_max_scale"
,
{
"X"
,
"InAccum"
,
"InState"
}},
{
"multiclass_nms3"
,
{
"BBoxes"
,
"Scores"
,
"RoisNum"
}},
{
"box_coder"
,
{
"PriorBox"
,
"PriorBoxVar"
,
"TargetBox"
}},
{
"momentum"
,
{
"Param"
,
"Grad"
,
"Velocity"
,
"LearningRate"
,
"MasterParam"
}},
{
"sparse_momentum"
,
{
"Param"
,
"Grad"
,
"Velocity"
,
"Index"
,
"LearningRate"
}},
{
"rnn"
,
{
"Input"
,
"PreState"
,
"WeightList"
,
"SequenceLength"
}},
{
"run_program"
,
{
"X"
,
"Params"
}},
{
"fused_feedforward"
,
{
"Dropout1Seed"
,
"Dropout2Seed"
,
"Linear1Bias"
,
"Linear2Bias"
,
"Ln1Scale"
,
"Ln1Bias"
,
"Ln2Scale"
,
"Ln2Bias"
}},
{
"faster_tokenizer"
,
{
"Text"
,
"Vocab"
,
"TextPair"
}},
{
"matrix_rank"
,
{
"X"
,
"TolTensor"
}},
{
"adam"
,
{
"Param"
,
"Grad"
,
"LearningRate"
,
"Moment1"
,
"Moment2"
,
"Beta1Pow"
,
"Beta2Pow"
,
"MasterParam"
}},
{
"adamw"
,
{
"Param"
,
"Grad"
,
"LearningRate"
,
"Moment1"
,
"Moment2"
,
"Beta1Pow"
,
"Beta2Pow"
,
"MasterParam"
}},
};
// NOTE(zhiqiu): Like op_ins_map.
// Commonly, the outputs in auto-generated OP function are determined by the
// OP`s proto automatically, i.e., all the outputs registered in OpMaker.
// However, some OPs have dispensable outputs, which means the output can
// be none for some conditions. It is discovered that most dispensable outputs
// is not used in imperative mode, so we drop those outputs when generating OP
// functions. While, for very few OPs, the dispensable outputs are used, we
// need to manually specify them in this map.
std
::
map
<
std
::
string
,
std
::
set
<
std
::
string
>>
op_outs_map
=
{
{
"fake_quantize_dequantize_moving_average_abs_max"
,
{
"Out"
,
"OutScale"
,
"OutAccum"
,
"OutState"
}},
{
"batch_norm"
,
{
"Y"
,
"MeanOut"
,
"VarianceOut"
,
"SavedMean"
,
"SavedVariance"
,
"ReserveSpace"
}},
{
"fused_attention"
,
{
"LnMean"
,
"LnVariance"
,
"LnOut"
,
"QKVOut"
,
"QKVBiasOut"
,
"TransposeOut2"
,
"QKOut"
,
"QKTVOut"
,
"SoftmaxOut"
,
"AttnDropoutMaskOut"
,
"AttnDropoutOut"
,
"SrcMaskOut"
,
"FMHAOut"
,
"OutLinearOut"
,
"DropoutMaskOut"
,
"Ln2Mean"
,
"Ln2Variance"
,
"BiasDropoutResidualOut"
,
"Y"
}},
{
"sync_batch_norm"
,
{
"Y"
,
"MeanOut"
,
"VarianceOut"
,
"SavedMean"
,
"SavedVariance"
,
"ReserveSpace"
}},
{
"unique"
,
{
"Out"
,
"Index"
,
"Indices"
,
"Counts"
}},
{
"unique_consecutive"
,
{
"Out"
,
"Index"
,
"Counts"
}},
{
"generate_proposals"
,
{
"RpnRois"
,
"RpnRoiProbs"
,
"RpnRoisNum"
}},
{
"collect_fpn_proposals"
,
{
"FpnRois"
,
"RoisNum"
}},
{
"matrix_nms"
,
{
"Out"
,
"Index"
,
"RoisNum"
}},
{
"distribute_fpn_proposals"
,
{
"MultiFpnRois"
,
"RestoreIndex"
,
"MultiLevelRoIsNum"
}},
{
"moving_average_abs_max_scale"
,
{
"Out"
,
"OutScale"
,
"OutAccum"
,
"OutState"
}},
{
"multiclass_nms3"
,
{
"Out"
,
"NmsRoisNum"
}},
{
"generate_proposals_v2"
,
{
"RpnRois"
,
"RpnRoiProbs"
,
"RpnRoisNum"
}},
{
"momentum"
,
{
"ParamOut"
,
"VelocityOut"
,
"MasterParamOut"
}},
{
"sparse_momentum"
,
{
"ParamOut"
,
"VelocityOut"
}},
{
"rnn"
,
{
"DropoutState"
,
"Reserve"
,
"Out"
,
"State"
}},
{
"lamb"
,
{
"ParamOut"
,
"Moment1Out"
,
"Moment2Out"
,
"Beta1PowOut"
,
"Beta2PowOut"
}},
{
"run_program"
,
{
"DOut"
}},
{
"adam"
,
{
"ParamOut"
,
"Moment1Out"
,
"Moment2Out"
,
"Beta1PowOut"
,
"Beta2PowOut"
,
"MasterParamOut"
}},
{
"adamw"
,
{
"ParamOut"
,
"Moment1Out"
,
"Moment2Out"
,
"Beta1PowOut"
,
"Beta2PowOut"
,
"MasterParamOut"
}},
};
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录