This repository was archived by the owner on Jan 24, 2024. It is now read-only.
upgrade op fusion lowering #1216
Open
Conversation
This file contains hidden or 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
thisjiang
reviewed
Mar 8, 2023
Collaborator
thisjiang
left a comment
thisjiang
left a comment
There was a problem hiding this comment.
大哥，你这PR好像做了很多非op lowering重构的工作啊，可以拆成多个PR么？
cinn/backends/compiler.cc
Outdated
|
|
||
| backends::nvrtc::Compiler compiler; | ||
|
|
||
| VLOG(3) << "[CUDA] device code:\n" << source_code; |
Collaborator
There was a problem hiding this comment.
SourceCodePrint就会打印这个源码，而且可以保证一个程序会将所有子图打印出来
cinn/common/cas.cc
Outdated
| }; | ||
|
|
||
| { | ||
| /* |
Collaborator
There was a problem hiding this comment.
冗余代码应该删除而非注释，而且注释应该统一用//而非/* */
cinn/common/cas_test.cc
Outdated
| auto p2 = CasSimplify(p0); | ||
| LOG(INFO) << "simplified " << p2; | ||
| EXPECT_EQ(GetStreamCnt(p2), "cinn_min(7, ((x) / (2)))"); | ||
| // EXPECT_EQ(GetStreamCnt(p2), "cinn_min(7, ((x) / (2)))"); |
Collaborator
There was a problem hiding this comment.
这是单测过不了了么？还是改之后结果是对的，但不是这样了？如果是前者那不能简单的注释掉啊，如果是后者那把这改为正确的值不就行了么
| } | ||
|
|
||
| std::unordered_set<Node*> NodeSet() { | ||
| std::unordered_set<Node*> node_set; |
Collaborator
There was a problem hiding this comment.
这函数功能完全和CollectNodes重复了而且也不常用啊。。。在用的地方直接定义
const auto& nodes = group->CollectNodes();
std::unordered_set<Node*> node_set(nodes.begin(), nodes.end());
不好么。。。
Collaborator
Author
There was a problem hiding this comment.
用的地方比较多 这里实现比较方便
cinn/hlir/framework/op_lowering.cc
Outdated
| } | ||
| } else { | ||
| LOG(FATAL) << "Previous IR Schedule Is Not Implemented!"; | ||
| LOG(FATAL) << "Previous IR Schedule Is Unsupport Now!"; |
Collaborator
There was a problem hiding this comment.
Suggested change
| LOG(FATAL) << "Previous IR Schedule Is Unsupport Now!"; | |
| LOG(FATAL) << "Previous IR Schedule Unsupported Now, Please set FLAGS_cinn_ir_schedule=1 to use new IR Schedule"; |
cinn/hlir/framework/op_lowering.cc
Outdated
| auto loops = ir_sch.GetLoops(GetNodeData(node)->id()); | ||
| if (op_pattern_dict[node->op()] == framework::kElementWise) { | ||
| ir_sch.FlattenLoops(loops, true); | ||
| } else if (op_pattern_dict[node->op()] == framework::kReduction) { |
Collaborator
There was a problem hiding this comment.
这if里啥都没有啊？那为啥不直接
else if (op_pattern_dict[node->op()] != framework::kReduction) {
ir_sch.FlattenLoops(loops, false);
}
| return tensors; | ||
| } | ||
|
|
||
| NodeData* GetNodeData(const Node* node) { |
Collaborator
There was a problem hiding this comment.
这函数名名不副实啊。。。
Suggested change
| NodeData* GetNodeData(const Node* node) { | |
| NodeData* GetFirstOutputNodeData(const Node* node) { |
Collaborator
Author
There was a problem hiding this comment.
我感觉这个名字没啥问题，而且用了很久了。
| return node_data; | ||
| } | ||
|
|
||
| std::vector<NodeData*> GetAllNodeData(const Node* node) { |
Collaborator
There was a problem hiding this comment.
Suggested change
| std::vector<NodeData*> GetAllNodeData(const Node* node) { | |
| std::vector<NodeData*> GetAllOutputNodeDatas(const Node* node) { |
Collaborator
Author
主要是op_loweing的，其他的工作只有一点点。 |
…pgrad_op_lowering
thisjiang
reviewed
Mar 21, 2023
cinn/hlir/framework/op_lowering.cc
Outdated
| stage->SimpleComputeAt(master_stage, master_stage->n_out_dims() - 1); | ||
| // do schedule | ||
| for (auto node : nodes_in_order) { | ||
| LOG(INFO) << node->id(); |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
2 participants
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.
这个PR主要是对融合算子lowering到ast进行升级，增加扩展性和兼容性，能够适配更加复杂的融合算子生成。
将elemenwise/kinjective/kbroadcast/reduce的循环融合放在一起。
此外，删除了旧的调度原语上的循环融合。