marcrasi

Incorrect reconstructed type for $sxq_Iegnr_D
Original type:
(sil_function_type type=@differentiable @callee_guaranteed (@in_guaranteed τ_0_0) -> @out τ_0_1)
Reconstructed type:
(sil_function_type type=@callee_guaranteed (@in_guaranteed τ_0_0) -> @out τ_0_1)
Stack dump:
0.	Program arguments: /usr/local/google/home/marcrasi/swift-base-merge/build/Ninja-RelWithDebInfoAssert/swift-linux-x86_64/bin/swift -frontend -c -filelist /tmp/sources-716834 -supplementary-output-file-map /tmp/supplementaryOutputs-c055d6 -disable-objc-attr-requires-foundation-module -target x86_64-unknown-linux-gnu -disable-objc-interop -sdk / -I /usr/local/google/home/marcrasi/swift-base-merge/build/Ninja-RelWithDebInfoAssert/swift-linux-x86_64/./lib/swift/linux/x86_64 -warn-swift3-objc-inference-complete -warn-implicit-overrides -enable-library-evolution -g -module-cache-path /usr/local/google/home/marcrasi/swift-base-merge/build/Ninja-RelWithDebInfoAssert/swift-linux-x86_64/./module-cache -module-link-name swiftCore -nostdimport -parse-st

marcrasi

struct MyTensor<T: Equatable & AdditiveArithmetic>: Equatable & AdditiveArithmetic {
  var value: T
}

extension MyTensor : Differentiable where T : AdditiveArithmetic & Differentiable {
  typealias TangentVector = MyTensor
  typealias CotangentVector = MyTensor
  typealias AllDifferentiableVariables = MyTensor
  func tangentVector(from cotangentVector: CotangentVector) -> TangentVector {
    return cotangentVector

marcrasi

protocol A {}

struct Wrapper<T> {
  var t: T
}

extension Wrapper: A where T: A {}

func inner<T: A>(_ t: T) {}

marcrasi

import TensorFlow

/// A Simple RNN Cell.
public struct SimpleRNNCell<Scalar: TensorFlowFloatingPoint>: RNNCell {
    public var weight: Tensor<Scalar>
    public var bias: Tensor<Scalar>

    @noDerivative public var stateShape: TensorShape {
        return TensorShape([1, weight.shape[1]])
    }

marcrasi

Differentiating Class Methods

marcrasi@

Last updated: 5/21/19

Problem

When c has class type with a method named f, Swift dispatches c.f() at runtime by looking for the concrete implementation of method inside a "vtable" referenced by c.

marcrasi

// MARK: - Some differentiable array manipulation functions used in the algorithms.

extension Array where Element: Differentiable {
  @differentiable(vjp: _vjpSwappedAt)
  func swappedAt(_ i: Int, _ j: Int) -> Array {
    var tmp = self
    tmp.swapAt(i, j)
    return tmp
  }

marcrasi

import TensorFlow

protocol MyLayer: Differentiable {
  associatedtype Input: Differentiable

  @differentiable
  func forward(_ x: Input) -> Tensor<Float>
}

extension MyLayer {

marcrasi

import TensorFlow
let g = gradient(at: Dense()) { m in m.callForward(Tensor<Float>(0)) }
print(g)

marcrasi

marcrasi