swift

ba93f64e - [AutoDiff] Cache derivative function types for a 3.5x compile time speedup. (#29590)

[AutoDiff] Cache derivative function types for a 3.5x compile time speedup. (#29590) Compiling packages that use AutoDiff is currently extremely slow. For example, compiling tensorflow/swift-apis takes ~5 minutes on a slower machine. This is largely due to calls to `SILFunctionType::getAutoDiffDerivativeFunctionType` in SIL verifier, which spends a long time computing generic signatures with GSB. This PR improves this situation by caching computed derivative function types in `ASTContext`. This resulted in a 1.6x speedup in running the AutoDiff test suite, and a 3.5x speedup in compiling tensorflow/swift-apis. Future investigations: - Cache generic signatures computed from `getAutoDiffDerivativeFunctionGenericSignature`. - Cache transpose function types computed from `SILFunctionType::getAutoDiffTransposeFunctionType()`. -------------------------------------------------------------------------------- Here are the results building tensorflow/swift-apis with `-j56`. Before: `swift build 411.45s user 31.20s system 334% cpu 2:12.52 total` After: `swift build 154.03s user 28.26s system 485% cpu 37.519 total` Resolves TF-1133.