- Remove unnecessary imports from test/AutoDiff/stdlib/simd.swift.
- Use platform-correct `Builtin` integer type in
`PullbackEmitter::getArrayAdjointElementBuffer`.
JVP functions are forward-mode derivative functions. They take original
arguments and return original results and a differential function. Differential
functions take derivatives wrt arguments and return derivatives wrt results.
`JVPEmitter` is a cloner that emits JVP and differential functions at the same
time. In JVP functions, function applications are replaced with JVP function
applications. In differential functions, function applications are replaced
with differential function applications.
In JVP functions, each basic block takes a differential struct containing callee
differentials. These structs are consumed by differential functions.
Disable `SILCombiner::visitPartialApplyInst` from rewriting `partial_apply` with
with `@convention(method)` callee to `thin_to_thick_function`.
This fixes SIL verification errors: `thin_to_thick_function` only supports
`@convention(thin)` operands.
Resolves SR-12548.
Test SR-12548: `SILCombiner::visitPartialApplyInst` rewrites `partial_apply`
with `@convention(method)` callee to `thin_to_thick_function`.
This produces a SIL verification error: `thin_to_thick_function` only supports
`@convention(thin)` operands.
Add implicit declarations generated by the differentiation transform to a
`SynthesizedFileUnit` instead of an ad-hoc pre-existing `SourceFile`.
Resolves TF-1232: type reconstruction for AutoDiff-generated declarations.
Previously, type reconstruction failed because retroactively adding declarations
to a `SourceFile` did not update name lookup caches.
Lift temporary cross-file derivative registration restriction.
`@derivative` attribute type-checking simplications coming soon: TF-1099.
Original function and derivative function must have same access level, with one
exception: public original functions may have internal `@usableFromInline`
derivatives.
Canonicalizes `differentiable_function` instructions by filling in missing
derivative function operands.
Derivative function emission rules, based on the original function value:
- `function_ref`: look up differentiability witness with the exact or a minimal
superset derivative configuration. Emit a `differentiability_witness_function`
for the derivative function.
- `witness_method`: emit a `witness_method` with the minimal superset derivative
configuration for the derivative function.
- `class_method`: emit a `class_method` with the minimal superset derivative
configuration for the derivative function.
If an *actual* emitted derivative function has a superset derivative
configuration versus the *desired* derivative configuration, create a "subset
parameters thunk" to thunk the actual derivative to the desired type.
For `differentiable_function` instructions formed from curry thunk applications:
clone the curry thunk (with type `(Self) -> (T, ...) -> U`) and create a new
version with type `(Self) -> @differentiable (T, ...) -> U`.
Progress towards TF-1211.
The differentiation transform does the following:
- Canonicalizes differentiability witnesses by filling in missing derivative
function entries.
- Canonicalizes `differentiable_function` instructions by filling in missing
derivative function operands.
- If necessary, performs automatic differentiation: generating derivative
functions for original functions.
- When encountering non-differentiability code, produces a diagnostic and
errors out.
Partially resolves TF-1211: add the main canonicalization loop.
To incrementally stage changes, derivative functions are currently created
with empty bodies that fatal error with a nice message.
Derivative emitters will be upstreamed separately.
Previously, two conditions were necessary to enable differentiable programming:
- Using the `-enable-experimental-differentiable-programming` frontend flag.
- Importing the `_Differentiation` module.
Importing the `_Differentiation` module is the true condition because it
contains the required compiler-known `Differentiable` protocol. The frontend
flag is redundant and cumbersome.
Now, the frontend flag is removed.
Importing `_Differentiation` is the only condition.
* Add all [differential operators](https://github.com/apple/swift/blob/master/docs/DifferentiableProgramming.md#list-of-differential-operators).
* Add `withoutDerivative(at:)`, used for efficiently stopping the derivative propagation at a value and causing the derivative at the value to be zero.
* Add utility `differentiableFunction(from:)`, used for creating a `@differentiable` function from an original function and a derivative function.
Mostly work done by @marcrasi and @dan-zheng.
Partially resolves TF-843.
TODO:
* Add `AnyDerivative`.
* Add `Array.differentiableMap(_:)` and `differentiableReduce(_:_:)`.
Add mangling scheme for `@differentiable` and `@differentiable(linear)` function
types. Mangling support is important for debug information, among other things.
Update docs and add tests.
Resolves TF-948.
The `@transpose` attribute registers a function as the transpose of another
function-like declaration: a `func`, `init`, `subscript`, or `var` computed
property declaration.
The `@transpose` attribute also has an optional `wrt:` clause specifying the
linearity parameters, i.e. the parameters that are transposed with respect to.
The linearity parameters must conform to the `Differentiable` protocol and
satisfy `Self == TangentVector`.
If the `wrt:` clause is unspecified, the linearity parameters are inferred to be
all parameters that conform to `Differentiable` and that satisfy
`Self == TangentVector`.
`@transpose` attribute type-checking verifies that the type of the transpose
function declaration is consistent with the type of the referenced original
declaration and the linearity parameters.
Resolves TF-830.
Add `AdditiveArithmetic` derived conformances for structs and classes, gated by
the `-enable-experimental-differentiable-programming` flag.
Structs and classes whose stored properties all conform to `Differentiable` can
derive `Differentiable`:
- `associatedtype TangentVector: Differentiable & AdditiveArithmetic`
- Member `TangentVector` structs are synthesized whose stored properties are
all `var` stored properties that conform to `Differentiable` and that are
not `@noDerivative`.
- `mutating func move(along: TangentVector)`
The `@noDerivative` attribute may be declared on stored properties to opt out of
inclusion in synthesized `TangentVector` structs.
Some stored properties cannot be used in `TangentVector` struct synthesis and
are implicitly marked as `@noDerivative`, with a warning:
- `let` stored properties.
- These cannot be updated by `mutating func move(along: TangentVector)`.
- Non-`Differentiable`-conforming stored properties.
`@noDerivative` also implies `@_semantics("autodiff.nonvarying")`, which is
relevant for differentiable activity analysis.
Add type-checking and SILGen tests.
Resolves TF-845.
Add the `@differentiable` function conversion pipeline:
- New expressions that convert between `@differentiable`,
`@differentiable(linear)`, and non-`@differentiable` functions:
- `DifferentiableFunction`
- `LinearFunction`
- `DifferentiableFunctionExtractOriginal`
- `LinearFunctionExtractOriginal`
- `LinearToDifferentiableFunction`
- All the AST handling (e.g. printing) necessary for those expressions.
- SILGen for those expressions.
- CSApply code that inserts these expressions to implicitly convert between
the various function types.
- Sema tests for the implicit conversions.
- SILGen tests for the SILGen of these expressions.
Resolves TF-833.
Add type checking for `@differentiable` function types:
- Check that parameters and results conform to `Differentiable`.
- Implicitly conform parameters and results whose types are generic parameters
to `Differentiable`.
- Upstream most of the differentiable_func_type_type_checking.swift test from
`tensorflow` branch. A few function conversion tests have not been added
because they depend on the `@differentiable` function conversion pipeline.
Diagnose gracefully when the `Differentiable` protocol is unavailable because
`_Differentiation` has not been imported.
Resolves TF-823 and TF-1219.
Add `linear_function` and `linear_function_extract` instructions.
`linear_function` creates a `@differentiable(linear)` function-typed value from
an original function operand and a transpose function operand (optional).
`linear_function_extract` extracts either the original or transpose function
value from a `@differentiable(linear)` function.
Resolves TF-1142 and TF-1143.
Generate `differentiable_function` and `differentiable_function_extract` in
derivative witness table/vtable thunks.
`differentiation_function` is later canonicalized by the differentiation
transform.
Add SIL FileCheck tests.
