TaskNode::order_in_chain #10102
Merged
TaskNode::order_in_chain #10102
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
chengtbf
commented
Apr 10, 2023
| file_stream << "i : " << std::to_string(i) << " , actor id : " << std::to_string(task_id) | ||
| << " thrd : " << std::to_string(thrd_id) << " name : " << task_id2name.at(task_id) | ||
| << "\n chain_id : " << std::to_string(task->chain_id()) | ||
| << " order_in_chain : " << std::to_string(task->order_in_chain()) |
There was a problem hiding this comment.
增加了 chain id 的信息，格式修改：
before：
order : 39 , actor id : 8796126576640 name : reduce_sum-12 thrd : 4194320 device_type : kCPU stream_index : 16 {
consume : in : <- [ reshape-11/__out_0 ] ( actor_id: 8796124479488, regst: {regust_num: 1, device: cpu, time_shape: (1,1,4), shape: (16,), dtype: kFloat} )
produce : tmp regst: {regust_num: 1, device: cpu, time_shape: (1,1,4), shape: (64,), dtype: kChar} {
}
produce : __output_tensor_0 regst: {regust_num: 1, device: cpu, time_shape: (1,1,4), shape: (), dtype: kFloat} {
-> [ pack-21 ] ( actor_id: 8796147548160 )
-> [ ones_like-13 ] ( actor_id: 8796128673792 )
}
}
order : 40 , actor id : 8796147548160 name : pack-21 thrd : 4194330 device_type : kCPU stream_index : 26 {
consume : in : <- [ reduce_sum-12/__output_tensor_0 ] ( actor_id: 8796126576640, regst: {regust_num: 1, device: cpu, time_shape: (1,1,4), shape: (), dtype: kFloat} )
produce : out regst: {regust_num: 1, device: cpu, time_shape: (1,1), shape: (4,), dtype: kFloat} {
-> [ _LinearTrainGraph_0_output.0.0.1_4 ] ( actor_id: 8796149645312 )
}
}
after:
i : 37 , actor id : 17592186044430 thrd : 8388608 name : add_n-10
chain_id : 0 order_in_chain : 4 device_type : kCUDA stream_index : 0 {
consume : in : <- [ broadcast_add-5/__z_0 ] ( actor_id: 17592186044426, regst: {regust_num: 1, device: cuda, time_shape: (1,1,4), shape: (2,8), dtype: kFloat} )
consume : in : <- [ constant-8/__out_0 ] ( actor_id: 17592186044429, regst: {regust_num: 1, device: cuda, time_shape: (1,1,4), shape: (2,8), dtype: kFloat} )
produce : __out_0 regst: {regust_num: 1, device: cuda, time_shape: (1,1,4), shape: (2,8), dtype: kFloat} {
-> [ reshape-11 ] ( actor_id: 17592186044431 )
}
}
i : 38 , actor id : 17592186044431 thrd : 8388608 name : reshape-11
chain_id : 0 order_in_chain : 5 device_type : kCUDA stream_index : 0 {
consume : in : <- [ add_n-10/__out_0 ] ( actor_id: 17592186044430, regst: {regust_num: 1, device: cuda, time_shape: (1,1,4), shape: (2,8), dtype: kFloat} )
produce : __out_0 regst: {regust_num: 1, device: cuda, time_shape: (1,1,4), shape: (16,), dtype: kFloat} {
-> [ pack-20 ] ( actor_id: 17592186044440 )
-> [ broadcast_like-14 ] ( actor_id: 17592186044434 )
-> [ reduce_sum-12 ] ( actor_id: 17592186044432 )
}
}
i : 39 , actor id : 17592186044432 thrd : 8388608 name : reduce_sum-12
chain_id : 0 order_in_chain : 7 device_type : kCUDA stream_index : 0 {
consume : in_ctrl : <- [ pack-20/out_ctrl_103 ] ( actor_id: 17592186044440, regst: {regust_num: 1, device: cuda, ctrl} )
consume : in : <- [ reshape-11/__out_0 ] ( actor_id: 17592186044431, regst: {regust_num: 1, device: cuda, time_shape: (1,1,4), shape: (16,), dtype: kFloat} )
produce : __output_tensor_0 regst: {regust_num: 1, device: cuda, time_shape: (1,1,4), shape: (), dtype: kFloat} {
-> [ pack-21 ] ( actor_id: 17592186044442 )
-> [ ones_like-13 ] ( actor_id: 17592186044433 )
}
produce : tmp regst: {regust_num: 1, device: cuda, time_shape: (1,1,4), shape: (512,), dtype: kChar} {
}
}
chengtbf
commented
Apr 10, 2023
| @@ -606,11 +603,7 @@ void StraightenNodes(TaskGraph* task_graph, std::vector<TaskNode*>* ordered_task | |||
|
|
|||
| std::vector<int32_t> remain_task_nums(num_classifier, 0); | |||
|
|
|||
| auto SetOrderInGraph = [&](TaskNode* task_node) { | |||
There was a problem hiding this comment.
对物理图上的拉直算法做了一点 refine，移除了 order in graph 概念。 仅提供 ordered task nodes 。 @Yipeng1994
Speed stats: |
|
View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/10102/ |
strint
reviewed
Apr 11, 2023
strint
reviewed
Apr 11, 2023
oneflow/core/job/task.proto
Outdated
| map<string, RegstDescProto> produced_regst_desc = 8; | ||
| map<string, RegstDescIdSet> consumed_regst_desc_id = 9; | ||
| optional bool all_register_num_eq_one_hint = 10 [default = false]; | ||
| required int64 chain_id = 10; |
There was a problem hiding this comment.
这些 id 按说还是可以复用的，比如这里还用 6 ？
proto 部分一直在更新，一直也没有保证兼容
There was a problem hiding this comment.
这些 id 按说还是可以复用的
可以用 6，但用了也不兼容，因为之前是：task_set_info
这里是希望如果后续再插入字段，比如 xx_id，可以不影响 chain id 之后的字段。 参考 op_conf 中对不同类型的 type 的字段分割。
There was a problem hiding this comment.
proto 部分一直在更新，一直也没有保证兼容
是的，后续还会大改，因为 plan/job 里有很多冗余字段
只能保证大版本内的可用性。 如果跨版本，job/plan 重新走一次编译就好了
cache plan 可以做一些检查，如果发现存储的 plan 是老版本的，就自动重新编译覆盖掉
strint
approved these changes
Apr 11, 2023
Co-authored-by: Xiaoyu Xu <xiaoyulink@gmail.com>
…dev_cc_order_in_chain
leaves-zwx
approved these changes
Apr 11, 2023
chengtbf
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
Speed stats: |
|
View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/10102/ |
|
CI failed when running job: cpu-misc. PR label automerge has been removed |
|
CI failed when running job: cuda-module. PR label automerge has been removed |
|
View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/10102/ |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
拆分离编译下的 ：
依赖：
移除 order_in_graph，使用 order_in_chain，在 LogicalChainPass 打开的情况下（分离编译强制 LogicalChain），logical chain 将 order_in_logical_chain 写入各个 op，从逻辑图上读取 order，跳过物理图的拓扑信息。
refine LightPlan 的输出信息