DifferentiationInterface-v0.7.1

DifferentiationInterface DifferentiationInterface-v0.7.1

Diff since DifferentiationInterface-v0.7.0

Merged pull requests:

• docs: warn that preparation is not thread-safe (#801) (@FerreolS)
• feat: support DuplicatedNoNeed as function annotation for Enzyme (#805) (@gdalle)
• Revert "feat: support DuplicatedNoNeed as function annotation for Enzyme" (#807) (@gdalle)
• docs: make Enzyme warning more visible (#808) (@gdalle)
• feat: use Mooncake's copy utilities (#809) (@gdalle)
• fix: make (multi)basis work on CuArray (#810) (@gdalle)
• fix: take absstep into account for FiniteDiff (#812) (@gdalle)

Closed issues:

• Specify behavior of test_differentiation when users don't give scenarios (#797)
• is preparation thread safe? (#800)
• Absstep is not respected in AutoFiniteDiff (#811)

DifferentiationInterfaceTest-v0.10.0

DifferentiationInterfaceTest DifferentiationInterfaceTest-v0.10.0

Diff since DifferentiationInterfaceTest-v0.9.6

Breaking changes

• The Scenario type and its constructors have been modified
• Default sets of scenarios have been removed from the public API

See the changelog for details

Merged pull requests:

• fix: erase cache contexts before building function in Symbolics (#760) (@gdalle)
• fix: store sparsity pattern for symbolic backends (#764) (@gdalle)
• clarify doc for differentiate_with (#767) (@yebai)
• perf: remove Mooncake warmup during preparation (#772) (@gdalle)
• Fix typo in Scenario docstring (#773) (@penelopeysm)
• perf: optimize multibasis for sparse differentiation (#777) (@gdalle)
• CompatHelper: add new compat entry for GPUArraysCore in [weakdeps] at version 0.2 for package DifferentiationInterface, (keep existing compat) (#778) (@github-actions[bot])
• compintell --> chalk-lab (#781) (@yebai)
• perf: allocate Enzyme shadow memory during preparation (#782) (@gdalle)
• chore: apply JuliaFormatter v2 (#784) (@gdalle)
• feat!: specify preparation arguments in DIT Scenario (#786) (@gdalle)
• chore: add changelogs (#787) (@gdalle)
• feat: error hints for Enzyme (#788) (@gdalle)
• refactor: remove automatic code generation in DI (#789) (@gdalle)
• chore: add extra deps compats (#790) (@gdalle)
• chore: update citation with new Arxiv preprint (#795) (@gdalle)
• refactor!: remove scenario lists from public API (#796) (@gdalle)
• docs!: remove mention of default_scenarios in test_differentiation (#798) (@gdalle)

Closed issues:

• Better error messages for wrong preparation (#350)
• Better preparation for Enzyme (#481)
• Better Mooncake preparation (#518)
• Make it possible to overload basis and indices to iterate over non-arrays (#566)
• Dagger support (#567)
• Setting ForwardDiff chunk size can error for sparse AD (#593)
• Fix performance for ForwardDiff-over-ReverseDiff (#603)
• Add allocation benchmarks on StaticArrays (#616)
• Use native Enzyme gradient in reverse mode (#711)
• sparsity_pattern method is not defined for AutoSymbolics (#763)
• Assertion error with FastDifferentiation and ModelPredictiveControl (#765)
• Enzyme slower than FiniteDiff with Cache (#766)
• Specific preparation values for DifferentiationInterfaceTest.Scenarios (#771)
• Performance asymptotics for sparsity detection (#774)
• Avoid OneElement for basis creation by adding a GPUArrays extension (#776)
• Segfault with OrdinaryDiffEq and Enzyme (#779)
• Performance Regression from SparseDiffTools: Specialized sparse matrix decompression for structured matrices (#780)
• GC corruption with Enzyme cache modification (#785)
• Remove scenario lists from DIT interface (#793)
• Better document prep_args in DIT.Scenario (#794)

DifferentiationInterface-v0.7.0

DifferentiationInterface DifferentiationInterface-v0.7.0

Diff since DifferentiationInterface-v0.6.54

Breaking changes

• Preparation is now strict by default: if the argument types at execution differ from those at preparation, an error will be thrown. You can recover the old behavior by setting strict=Val(false) in the preparation functions.

Merged pull requests:

• feat!: specify preparation arguments in DIT Scenario (#786) (@gdalle)
• chore: update citation with new Arxiv preprint (#795) (@gdalle)
• refactor!: remove scenario lists from public API (#796) (@gdalle)
• docs!: remove mention of default_scenarios in test_differentiation (#798) (@gdalle)
• fix!: make strict preparation the default (#799) (@gdalle)

Closed issues:

• Make it possible to overload basis and indices to iterate over non-arrays (#566)
• Dagger support (#567)
• Setting ForwardDiff chunk size can error for sparse AD (#593)
• Add allocation benchmarks on StaticArrays (#616)
• Specific preparation values for DifferentiationInterfaceTest.Scenarios (#771)
• Remove scenario lists from DIT interface (#793)
• Better document prep_args in DIT.Scenario (#794)

DifferentiationInterface-v0.6.54

DifferentiationInterface DifferentiationInterface-v0.6.54

Diff since DifferentiationInterface-v0.6.53

Merged pull requests:

• chore: add changelogs (#787) (@gdalle)
• feat: error hints for Enzyme (#788) (@gdalle)
• refactor: remove automatic code generation in DI (#789) (@gdalle)
• chore: add extra deps compats (#790) (@gdalle)

Closed issues:

• Better error messages for wrong preparation (#350)
• Better Mooncake preparation (#518)
• Fix performance for ForwardDiff-over-ReverseDiff (#603)
• Use native Enzyme gradient in reverse mode (#711)

DifferentiationInterface-v0.6.53

DifferentiationInterface DifferentiationInterface-v0.6.53

Diff since DifferentiationInterface-v0.6.52

Merged pull requests:

• CompatHelper: add new compat entry for GPUArraysCore in [weakdeps] at version 0.2 for package DifferentiationInterface, (keep existing compat) (#778) (@github-actions[bot])
• compintell --> chalk-lab (#781) (@yebai)
• perf: allocate Enzyme shadow memory during preparation (#782) (@gdalle)
• chore: apply JuliaFormatter v2 (#784) (@gdalle)

Closed issues:

• Enzyme slower than FiniteDiff with Cache (#766)
• Performance asymptotics for sparsity detection (#774)
• Segfault with OrdinaryDiffEq and Enzyme (#779)
• Performance Regression from SparseDiffTools: Specialized sparse matrix decompression for structured matrices (#780)

DifferentiationInterface-v0.6.52

DifferentiationInterface DifferentiationInterface-v0.6.52

Diff since DifferentiationInterface-v0.6.51

Merged pull requests:

• perf: optimize multibasis for sparse differentiation (#777) (@gdalle)

Closed issues:

• Avoid OneElement for basis creation by adding a GPUArrays extension (#776)

DifferentiationInterface-v0.6.51

DifferentiationInterface DifferentiationInterface-v0.6.51

Diff since DifferentiationInterface-v0.6.50

Merged pull requests:

• clarify doc for differentiate_with (#767) (@yebai)
• perf: remove Mooncake warmup during preparation (#772) (@gdalle)
• Fix typo in Scenario docstring (#773) (@penelopeysm)

Closed issues:

• Better preparation for Enzyme (#481)
• Assertion error with FastDifferentiation and ModelPredictiveControl (#765)

DifferentiationInterface-v0.6.50

DifferentiationInterface DifferentiationInterface-v0.6.50

Diff since DifferentiationInterface-v0.6.49

Merged pull requests:

• fix: erase cache contexts before building function in Symbolics (#760) (@gdalle)
• fix: store sparsity pattern for symbolic backends (#764) (@gdalle)

Closed issues:

• sparsity_pattern method is not defined for AutoSymbolics (#763)

DifferentiationInterfaceTest-v0.9.6

DifferentiationInterfaceTest DifferentiationInterfaceTest-v0.9.6

Diff since DifferentiationInterfaceTest-v0.9.5

Merged pull requests:

• feat: fast prep modification for in-place FiniteDiff (#734) (@gdalle)
• fix: support Cache in sparsity detection with SparseConnectivityTracer (#739) (@gdalle)
• fix: check ForwardDiff tag manually to exclude contexts from it (#740) (@gdalle)
• perf: preallocate Cache during preparation for ForwardDiff (#741) (@gdalle)
• docs: improve and automate docstrings (#742) (@gdalle)
• perf: HVP with in-place gradient + inner preparation (#743) (@gdalle)
• chore: correct version number (#744) (@gdalle)
• feat: test type consistency between preparation and execution (#745) (@gdalle)
• fix: clarify dispatch for preparation (#746) (@gdalle)
• feat: support nested tuples of arrays as Caches (#748) (@gdalle)
• feat: add new ConstantOrCache context (#749) (@gdalle)
• Remove 1 backtick in excess (#752) (@franckgaga)
• fix: handle constant ConstantOrCache with Enzyme and SCT (#753) (@gdalle)
• compat: allow ForwardDiff v1.0 (#755) (@gdalle)
• compat: allow ForwardDiff v1.0 in DIT (#756) (@gdalle)
• perf: use common subexpression elimination in AutoSymbolics (#759) (@baggepinnen)

Closed issues:

• Preparation for second order (#86)
• Resizing preparation results (#519)
• Add Cache as another type of context (#551)
• Function annotation doesn't work for second-order Enzyme (#592)
• Fast preparation modification for FiniteDiff backend (#731)
• Mooncake Jacobian preperation fails when everything is a static array. (#736)
• TracerSparsityDetector + in-place RHS cache fails (#737)
• using prep for out of place method? (#738)
• Support more complex caches (#747)
• value_and_gradient with Enzyme forward-mode returns a different type (#754)
• Symbolics.build_funciton now supports common-subexpression elimination (#758)

DifferentiationInterface-v0.6.49

DifferentiationInterface DifferentiationInterface-v0.6.49

Diff since DifferentiationInterface-v0.6.48

Merged pull requests:

• compat: allow ForwardDiff v1.0 (#755) (@gdalle)
• compat: allow ForwardDiff v1.0 in DIT (#756) (@gdalle)
• perf: use common subexpression elimination in AutoSymbolics (#759) (@baggepinnen)

Closed issues:

• value_and_gradient with Enzyme forward-mode returns a different type (#754)
• Symbolics.build_funciton now supports common-subexpression elimination (#758)