[refactor] Add SNode::GradInfoProvider to isolate SNode from Expr #2207
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
k-ye
commented
Mar 10, 2021
yuanming-hu
approved these changes
Mar 10, 2021
xumingkuan
approved these changes
Mar 10, 2021
| ExprGroup indices; | ||
| for (int i = 0; i < snode->num_active_indices; i++) { | ||
| indices.push_back(Expr::make<ArgLoadExpression>(i, PrimitiveType::i32)); | ||
| } | ||
| auto ret = Stmt::make<FrontendKernelReturnStmt>( | ||
| load_if_ptr((snode->expr)[indices])); | ||
| load_if_ptr(Expr(snode_to_glb_var_exprs_.at(snode))[indices])); |
There was a problem hiding this comment.
I wonder why we need this Expr()? (same question at line 663)
There was a problem hiding this comment.
Good point. Note that SNode::expr used to be of type Expr, which contains a GlobalVariableExpression. Here, snode_to_glb_var_exprs_ maps directly from SNode* to GlobalVariableExpression. As a result, when we do need to use an Expr, we have to wrap that up.
The reason for mapping to GlobalVariableExpression is that, If you look at the previous implementation of is_primal() (has_grad(), get_grad()), they do a lot of dynamic casts from Expr to GlobalVariableExpression.
Merged
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Outdated suggestions cannot be applied.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.
Related issue = #2196
This PR is still about decoupling
SNodefromtaichi/program. The major issue here is around its member variableexpr. BecauseExpris part of the frontend and relies ontaichi/program, we have to move it out fromSNode:place()andlazy_grad()fromSNodeitself to a separate filetaichi/program/snode_expr_utils.h/cpp(i cannot think of a better file name...). Other functions inSNodedo not depend onexpr.SNode::GradInfoProviderabstract class. This is essentially a wrapper aroundGlobalVariableExpression. From what I can tell,SNode::expris mostly used to provide gradient information, e.g.is_primal(),has_grad(), etc. The actual implementation,GradInfoImpl, is intaichi/program/snode_expr_utils.cpp.SNode*toGlobalVariableExpression. The map is stored insideProgram. While it is possible to retrieveGlobalVariableExpressionfromGradInfoImpl, that involves castingGradInfoProvidertoGradInfoImpland kind of exposes the implementation detail. I'd like to makeGradInfoProviderhave a single responsibility.[Click here for the format server]