Use VJP in differentiation #21063
Merged
Use VJP in differentiation #21063
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
rxwei
approved these changes
Dec 6, 2018
rxwei
reviewed
Dec 6, 2018
| @@ -1,4 +1,4 @@ | |||
| // RUN: %target-swift-frontend -O -emit-sil %s | %FileCheck %s | |||
| // RUN: %target-swift-frontend -O -emit-sil -Xllvm -debug -Xllvm -debug-only -Xllvm differentiation %s | %FileCheck %s | |||
There was a problem hiding this comment.
Seems unnecessary. We are not CHECK'ing against any debug dump.
| // Get the pullback. | ||
| auto *field = getPrimalInfo().lookupPullbackDecl(ai); | ||
| assert(field); | ||
| SILValue pullback = builder.createStructExtract(remapLocation(ai->getLoc()), primalValueAggregateInAdj, field); |
| @@ -586,6 +612,14 @@ class PrimalInfo { | |||
| : lookup->getSecond(); | |||
| } | |||
|
|
|||
| /// Finds the pullback decl in the primal value struct for an `apply` in the | |||
| /// original function. | |||
| VarDecl *lookupPullbackDecl(ApplyInst *inst) { | |||
There was a problem hiding this comment.
We've been using "lookUp" instead of "lookup" in this file.
| // Get all results in type-defined order. | ||
| SmallVector<SILValue, 8> allResults; | ||
| collectAllActualResultsInTypeOrder( | ||
| pullbackCall, dirResults, pullbackCall->getIndirectSILResults(), allResults); |
| LLVM_DEBUG({ | ||
| auto &s = getADDebugStream(); | ||
| s << "All direct results of the nested pullback call: \n"; | ||
| interleave(dirResults, [&](SILValue v) { s << v; }, [&]{}); |
There was a problem hiding this comment.
If you aren't doing anything in between, then llvm::for_each is a better fit than interleave.
|
@swift-ci please test tensorflow |
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.
(This is on top of #21062. Ask GitHub to see the last commit to see the changes before I merge #21062).
This teaches the differentiation pass to use VJPs.
I put it under a flag because the TensorFlow autodiff tests are still failing. I will investigate those failures next.