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 `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.
* Move muliti-functionality SIL tests to test/AutoDiff/SIL.
* Add `with_derivative` clause to `differentiable_function` instructions.
Otherwise, IRGen for test/AutoDiff/SIL/differentiable_function_inst.sil fails on
tensorflow branch because the differentiation transform cannot differentiate
external functions.
Add test/AutoDiff/lit.local.cfg: run tests only when `differentiable_programming`
is enabled in lit. With this, individual tests no longer need
`REQUIRES: differentiable_programming`.
Move multi-functionality SIL tests from test/AutoDiff/SIL/Serialization to
test/AutoDiff/SIL.
Garden test filenames.
Add `differentiable_function` and `differentiable_function_extract`
instructions.
`differentiable_function` creates a `@differentiable` function-typed
value from an original function operand and derivative function operands
(optional).
`differentiable_function_extract` extracts either the original or
derivative function value from a `@differentiable` function.
The differentiation transform canonicalizes `differentiable_function`
instructions, filling in derivative function operands if missing.
Resolves TF-1139 and TF-1140.
`@differentiable` attribute on protocol requirements and non-final class
members now produces derivative function entries in witness tables and vtables.
This enables `witness_method` and `class_method` differentiation.
Existing type-checking rules:
- Witness declarations of `@differentiable` protocol requirements must have a
`@differentiable` attribute with the same configuration (or a configuration
with superset parameter indices).
- Witness table derivative function entries are SILGen'd for `@differentiable`
witness declarations.
- Class vtable derivative function entries are SILGen'd for non-final
`@differentiable` class members.
- These derivative entries can be overridden or inherited, just like other
vtable entries.
Resolves TF-1212.
`@differentiable` attribute on protocol requirements and non-final class members
will produce derivative function entries in witness tables and vtables.
This patch adds an optional derivative function configuration
(`AutoDiffDerivativeFunctionIdentifier`) to `SILDeclRef` to represent these
derivative function entries.
Derivative function configurations consist of:
- A derivative function kind (JVP or VJP).
- Differentiability parameter indices.
Resolves TF-1209.
Enables TF-1212: upstream derivative function entries in witness tables/vtables.
This is necessary because the `Differentiable` protocol exists in stdlib core
on `tensorflow` branch but in the `_Differentiation` module on `master` branch.
The robust solution is to add auto-import `_Differentiation` logic to `tensorflow`.
The `differentiability_witness_function` instruction looks up a
differentiability witness function (JVP, VJP, or transpose) for a referenced
function via SIL differentiability witnesses.
Add round-trip parsing/serialization and IRGen tests.
Notes:
- Differentiability witnesses for linear functions require more support.
`differentiability_witness_function [transpose]` instructions do not yet
have IRGen.
- Nothing currently generates `differentiability_witness_function` instructions.
The differentiation transform does, but it hasn't been upstreamed yet.
Resolves TF-1141.
There is currently a difference between the tensorflow branch and the
master branch. On tensorflow, the differentiability support is merged
into the standard library. This changes the decoration of the witness.
Loosen the test to accept either.
We should change the tensorflow branch to generate the
`_Differentiation` module with the support and then auto-import the
module in the longer term. This can be gated by the
`-enable-experimental-autodifferentiation` flag to the driver to gain
the same behaviour on both the branches.
SIL differentiability witnesses are a new top-level SIL construct mapping
an "original" SIL function and derivative configuration to derivative SIL
functions.
This patch adds `SILDifferentiabilityWitness` IRGen.
`SILDifferentiabilityWitness` has a fixed `{ i8*, i8* }` layout:
JVP and VJP derivative function pointers.
Resolves TF-1146.
SIL differentiability witnesses are a new top-level SIL construct mapping
an "original" SIL function and derivative configuration to derivative SIL
functions.
This patch adds `SILDifferentiabilityWitness` serialization/deserialization.
Resolves TF-1136.
SIL differentiability witnesses are a new top-level SIL construct mapping
"original" SIL functions to derivative SIL functions.
SIL differentiability witnesses have the following components:
- "Original" `SILFunction`.
- SIL linkage.
- Differentiability parameter indices (`IndexSubset`).
- Differentiability result indices (`IndexSubset`).
- Derivative `GenericSignature` representing differentiability generic
requirements (optional).
- JVP derivative `SILFunction` (optional).
- VJP derivative `SILFunction` (optional).
- "Is serialized?" bit.
This patch adds the `SILDifferentiabilityWitness` data structure, with
documentation, parsing, and printing.
Resolves TF-911.
Todos:
- TF-1136: upstream `SILDifferentiabilityWitness` serialization.
- TF-1137: upstream `SILDifferentiabilityWitness` verification.
- TF-1138: upstream `SILDifferentiabilityWitness` SILGen from
`@differentiable` and `@derivative` attributes.
- TF-20: robust mangling for `SILDifferentiabilityWitness` names.
The `@noDerivative` attribute marks the non-differentiability parameters of a
`@differentiable` function type. All parameters except those marked with
`@noDerivative` are differentiability parameters.
For example, `@differentiable (Float, @noDerivative Float) -> Float` is only
differentiable with respect to its first parameter.
The `@noDerivative` attribute is represented as a
`SILParameterDifferentiability` bit on `SILParameterInfo`.
Add round-trip serialization tests.
Resolves TF-872.
