[AutoDiff] [stdlib] Conform 'Array.DifferentiableView' to collection protocols. #26062
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
- Remove `@trivial` type annotations. - Update SIL FileCheck based on changes in generated code.
Set type for implicit `NamedPattern`.
* Fix AutoDiff tests. - Fix `createAutoDiffThunk` to pass ownership verification. - The `@differentiable` function operand must be copied if non-trivial. - Fix `AutoDiff/builtin_differential_operators.swift` so that it actually runs. * Revert changes to `emitBuiltinAutoDiffApplyAssociatedFunction`. Re-add `createDestroyAddr`. Fix indentation.
Mark `var allKeyPaths` as `@inlinable` only when all stored properties are effectively public. Otherwise, SIL verification fails: "Key path in serialized function cannot reference non-public property".
- Fix `stdlib/public/TensorFlow/CMakeLists.txt`.
- List `ExecuteOp.swift.gyb` under GYB_SOURCES.
- Replace `count(where:)` with `lazy.filter { ... }.count`.
- `count(where:)` reverted in swiftlang#22289.
- Fix `ProtocolDecl::computeKnownProtocolKind()`.
- Update `test/TensorFlow/sema_errors.swift`.
- Unfortunately, diagnostics have regressed.
* Revamp TensorFlowRuntime tests after GPE removal. - Remove `%target-run-eager-swift` and `%target-run-gpe-swift`. - Use `%target-run-simple-swift` in RUN lines. - Remove `REQUIRES: swift_test_mode_optimize`. * Move `Tensor.concatenated` differentiation tests to `tensor_autodiff_runtime.swift`.
The type set by cs.setType() in the dynamic callable application did not have type-variables, so it was considering this string literal already checked. This made it to SILGen with a nullptr keyword argument.
Remove problematic call to `TC.conformsToProtocol` on synthesized `AllDifferentiableVariables` struct, which produced an `ErrorType` and caused an IRGen assertion failure.
- lib/AST/Type.cpp -> Brittle Vector type lookup leads to Self type not getting erased. - lib/SILGen/SILGenBuiltin.cpp -> builtin apply macro switched from Expr to [ManagedValue]. Need to remove all custom freeing logic. - utils/build-presets.ini -> importing tensorflow means we cannot use the new build-script impl.
`verify_all_overlays.swift` and `verify_all_overlays_O.swift` have been deleted upstream.
test/SILGen/builtins.swift -> Extra assert change hanging around. utils/PathSanitizingFileCheck -> injection bug. validation-test/ParseableInterface/verify_all_overlays.py -> Just a fix.
On Linux, adding an extra rpath fixes the following error: "error while loading shared libraries: lib_InternalSwiftSyntaxParser.so". Revisit, this change should not be necessary.
swiftlang#24630) Fix differentiation of `@inlinable public func differentiationFunction(from:)`.
These tests were previously disabled on tensorflow branch, but now pass.
…4640) This test was originally XFAIL on macOS, but now passes since it's been disabled. Re-enable XFAIL as part of TF-491.
…tlang#25914) `Differentiable` derived conformances now supports class types. Synthesis works just like for struct types, except `TangentVector = Self` is never synthesized even if `Self` conforms to `AdditiveArithmetic`. Class differentiation support requires further differentiation transform changes. Resolves TF-630.
Update to tensorflow/swift-apis@9a3f393, getting us function-builder-based `Sequential` layers.
Do the type checking for top level functions and methods that are marked as the transpose of another function.
…ding. (swiftlang#25958) Associated functions should be released when an `autodiff_function` instruction has been folded away.
…to `PullbackEmitter`. (swiftlang#25951)
…wiftlang#25967) In VJPEmitter, if the original function call has substitutions, we `partial_apply` it with no arguments to specialize it. This `partial_apply` is not being released. JVP and VJP are being specialized the same way, but they are not being released either. To fix this, we release the `@differentiable` function returned by `autodiff_function`, which will release the original and the associated functions tht are to be filled in later altogether. If the original function does not have substitutions, we retain the original function to balance out the release of the `@differentiable` function that comes later. As a result, `ADContext::promoteToDifferentiableFunction` no longer needs to retain the associated functions. Example where the original function has substitutions: ``` f' = partial_apply f<...>() f_diff = autodiff_function f' release_value f_diff ``` Example where the original function does not have substitutions: ``` retain_value f f_diff = autodiff_function f release_value f_diff ``` Note: This makes the `autodiff_function` folding optimization no longer able to detect the pattern, but it is necessary. We can rewrite the optimization later. This should fix [TF-621](https://bugs.swift.org/browse/TF-621).
…ay'. (swiftlang#26023) Add variants of map and reduce that take a `@differentiable` closure and are themselves differentiable. ```swift extension Array { @differentiable(wrt: self) func differentiableMap<Result: Differentiable>( _ body: @differentiable (Element) -> Result ) -> [Result] @differentiable(wrt: (self, initialResult)) func differentiableReduce<Result: Differentiable>( _ initialResult: Result, _ nextPartialResult: @differentiable (Result, Element) -> Result ) -> Result } ``` Also make `Array.DifferentiableView` conform to `ExpressibleByArrayLiteral` so that tests and user code are easier to write.
Fix formatting, fix unused variable warnings.
…ents. (swiftlang#25974) This PR enables the functionality shown in this example: ```swift public protocol Distribution { associatedtype Value func logProbability(of value: Value) -> Float } public protocol DifferentiableDistribution: Differentiable, Distribution { @differentiable(wrt: self) func logProbability(of value: Value) -> Float } struct Foo: DifferentiableDistribution { @differentiable(wrt: self) func logProbability(of value: Float) -> Float { .zero } } @differentiable func blah<T: DifferentiableDistribution>(_ x: T) -> Float where T.Value: AdditiveArithmetic { x.logProbability(of: .zero) } ``` The fix is based on the fact that the Swift compiler does not add entries to the witness tables of protocols for overridden functions, to avoid redundancy. However, the `@differentiable` attribute being added should not be interpreted as an override as it adds new functionality, and should result in entries being added to the witness table. This PR adds this check to the override checking code, thus enabling support for the aforementioned feature.
rxwei
changed the title
|
cc @eaplatanios |
|
@swift-ci please test tensorflow |
|
An IRGen crasher is expected. |
dan-zheng
force-pushed
the
tensorflow
branch
from
6dcf239 to
04dca63
Compare
|
I accidentally deleted |
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.
Uh oh!
There was an error while loading. Please reload this page.