v0.12.33

Enzyme v0.12.33

Diff since v0.12.32

Merged pull requests:

• Fix cpu features (#1744) (@wsmoses)
• Try diferent name for gpu exception (#1746) (@wsmoses)
• Handle const addr casts (#1752) (@wsmoses)
• Clarify use cases for DuplicatedNoNeed (#1753) (@danielwe)
• Fix multidim solve (#1754) (@wsmoses)
• Type unstable custom rule tape (#1755) (@wsmoses)

Closed issues:

• EnzymeVJP failed when training Neural ODE with DiscreteCallback (#1743)
• CUDA.jl CI broken again because of Enzyme (#1745)
• 'sensealg=InterpolatingAdjoint(autojacvec=EnzymeVJP(), checkpointing=true) ' failed (#1749)
• left division with two matrices (#1750)
• Mixed activity error for 2nd order diff of fused broadcast into preallocated array (#1751)

v0.12.32

Enzyme v0.12.32

Diff since v0.12.31

Merged pull requests:

• small fix for static array onehot (#1732) (@ExpandingMan)
• Fix cuda tape type (#1737) (@wsmoses)
• Fix return diffe (#1740) (@wsmoses)
• Update Project.toml (#1742) (@wsmoses)

Closed issues:

• LLVM error (#1693)
• Enzyme doesn't work for AdvancedVI Part VIII: FP32 + Base.Fix1 ignores gradients (#1735)
• Update broke CUDA.jl downstream CI (#1736)

v0.12.31

Enzyme v0.12.31

Diff since v0.12.30

Merged pull requests:

• chore: bump jll to 0.0.144 (#1731) (@avik-pal)

Closed issues:

• Error in forward over reverse differentiation (KeyError: key not found) (#1713)

v0.12.30

Enzyme v0.12.30

Diff since v0.12.29

Merged pull requests:

• Precompilation simplification (#1715) (@wsmoses)
• Mark gc preserve as readonly (#1724) (@wsmoses)
• fix: BFloat16 extension and tests (#1725) (@avik-pal)
• boxfloat fixup (#1726) (@wsmoses)
• Unhash the jl_idtable_rehash call (#1727) (@wsmoses)
• More bfloat fix (#1728) (@wsmoses)
• Update compiler.jl (#1729) (@wsmoses)

Closed issues:

• LLVM error (#1693)
• Function argument passed to autodiff cannot be proven readonly (#1719)
• Error in BFloat16 extension (#1720)
• ERROR: LoadError: UndefVarError: BFloat16 not defined (#1723)

v0.12.29

Enzyme v0.12.29

Diff since v0.12.28

Merged pull requests:

• fix: typetree_inner is not in Compiler module (#1716) (@avik-pal)
• Fix doctests, remove Formatter CI, add caching to Documenter CI (#1721) (@vchuravy)

v0.12.28

Enzyme v0.12.28

Diff since v0.12.27

Merged pull requests:

• Add optional [but presently on] check if function contains written-to data (#1701) (@wsmoses)
• allow forward mode reflection (#1702) (@vchuravy)
• Interpreter: single return type (#1703) (@wsmoses)
• more 1.11 workarounds (#1704) (@wsmoses)
• Add bfloat16 (#1708) (@wsmoses)
• Add ABI setter (#1709) (@wsmoses)
• More diverse error types (#1710) (@wsmoses)
• Fixup more than simple jacobian (#1712) (@wsmoses)

Closed issues:

• doc: Got confused with EnzymeCore.Duplicated doc (#1624)
• Duplicating closures (#1669)
• Incorrect gradients with BFloat16 (#1695)
• Enzyme doesn't work for AdvancedVI Part VII: Type instability with FP32 (#1700)

v0.12.27

Enzyme v0.12.27

Diff since v0.12.26

Merged pull requests:

• Add dump mod post optimizations option (#1694) (@wsmoses)
• Fix interpreter caches (#1698) (@wsmoses)

Closed issues:

• Segfault with iteration (#1310)
• fail with simple sum gradient on CUDA (#1347)
• CuArray broadcasting (#1454)
• Cannot deduce type for QuadGK (#1599)
• Indeterministic illegal instruction: 4 (#1626)
• Differentiating FFTW fftshift (#1652)
• Incompatible with MKL.jl (#1683)
• Recent updates broke LLVM.jl downstream CI (#1688)
• Seemingly wrong derivative (#1691)
• Enzyme doesn't work for AdvancedVI Part VI: Bizzare type instability (#1697)

v0.12.26

Enzyme v0.12.26

Diff since v0.12.25

Merged pull requests:

• ci: add DynamicExpressions integration tests (#1675) (@MilesCranmer)
• Bump jll with extract tuple for reverse fix (#1678) (@wsmoses)
• Fix bitcast arg (#1679) (@wsmoses)
• First relocation of globals (#1684) (@wsmoses)
• Fix array of tuple any (#1685) (@wsmoses)
• Simplify libjulia usage (#1687) (@wsmoses)
• Update Project.toml (#1689) (@wsmoses)

Closed issues:

• Enzyme failing in Julia: ERROR: LoadError: Enzyme execution failed. (#1571)
• Does Enzyme's high-level API assume f is constant? (#1670)
• Missing 2-argument method for bitcast! (#1674)
• Compilation Error: Unsafe copyto! (#1681)

v0.12.25

Enzyme v0.12.25

Diff since v0.12.24

Merged pull requests:

• Add internal forward-mode rules for ranges (#1655) (@ChrisRackauckas)
• Fix tape union check (#1660) (@wsmoses)
• Pushforward dvals in the ranges forward rule (#1663) (@ChrisRackauckas)
• Fix make_zero box infinite recursion (#1665) (@wsmoses)
• Tape type of error (#1666) (@wsmoses)
• Inactive boxed cache (#1667) (@wsmoses)
• Update Project.toml (#1668) (@wsmoses)

Closed issues:

• Hitting new segfaults in DynamicExpressions.jl (#1662)
• StackoverFlow with KernelAbstractions (#1664)

v0.12.24

Enzyme v0.12.24

Diff since v0.12.23

Merged pull requests:

• Fix #1630 (#1631) (@danielwe)
• Reduce time for enzyme import (#1645) (@wsmoses)
• Update Project.toml (#1646) (@wsmoses)
• Fix typo in jl_array_ptr_copy_fwd (#1648) (@mhauru)
• Mark regex fns as nofree (#1654) (@wsmoses)
• Bump jll (#1657) (@wsmoses)
• Bigfloat constructor rules (#1658) (@wsmoses)
• GPU report exception: fix linkage (#1659) (@wsmoses)

Closed issues:

• Illegal type analysis - BigFloat (#1621)
• unwrapMode != UnwrapMode::LegalFullUnwrap with Julia <1.8 (#1629)
• "UndefVarError: N not defined" when writing custom rule (#1630)
• Raise Julia errors rather than crashing on Assertion failures (#1634)
• LoopVectorization leads to Assertion orig failed (#1635)
• Second order AD of broadcast function (#1637)
• Adding Jacobian computation support for AbstractArray{<:AbstractArray{T}} type function argument (#1644)
• AD on WaterLily using Enzyme (#1647)
• Cholesky solve segfault (#1649)
• Failure with label2index from ComponentArrays (#1650)