v0.11.14

Enzyme v0.11.14

Diff since v0.11.13

Merged pull requests:

• [WIP] First stab at Cholesky (#1220) (@michel2323)
• Handle batched inactive (#1247) (@wsmoses)
• Mark computesparms as inactivenoinl (#1248) (@wsmoses)
• [BLAS] Enable tablegen blas by default for reverse mode gemm/gemv/axp… (#1249) (@wsmoses)
• Cleanup and simplify cholesky handler (#1251) (@wsmoses)
• Mark gensym as inactive (#1252) (@wsmoses)
• Fix typo "appliable" (#1253) (@devmotion)
• Fix cholesky part 2 (#1256) (@wsmoses)
• Restrict ldiv! rules to Cholesky (#1257) (@devmotion)
• Custom rules overwritten fix (#1258) (@wsmoses)

Closed issues:

• Gradient correctness issues in expression trees (#1241)
• Trouble writing custom rule in Enzyme: AssertionError: !(overwritten[end]) (#1242)
• Error with Duplicated{<:Cholesky} (#1250)
• Wrong gradient for cholesky (#1254)
• Found unhandled active variable in tuple splat (#1255)

v0.11.13

Enzyme v0.11.13

Diff since v0.11.12

Merged pull requests:

• Update CI for 1.10 release (#1229) (@wsmoses)
• dont addrfix offsets of arg11 (#1230) (@wsmoses)
• Fix 1.10 for x86 (#1231) (@wsmoses)
• [EnzymeCore] Move adapt to extension (#1232) (@wsmoses)
• Add more info to gc select (#1237) (@wsmoses)
• Handle fact that addrspacecast can also change elemenent type (#1238) (@wsmoses)
• Fix GitHub action configuration – continue-on-error (#1240) (@MilesCranmer)
• Generalize to recursive argument-based decayed phi's (#1244) (@wsmoses)
• Second phi embarassing bugfix (#1245) (@wsmoses)
• Add additional support for svecref (#1246) (@wsmoses)

Closed issues:

• "Could not analyze garbage collection behavior" error with non-allocating code. (#1228)
• Package load time is quite long, and longer on Julia 1.10 (#1234)
• Enzyme v0.11.11+ much slower to load (#1243)

v0.11.12

Enzyme v0.11.12

Diff since v0.11.11

Merged pull requests:

• sort! rules (#1000) (@jgreener64)
• Docs: add links to more examples of custom rules. (#1183) (@lassepe)
• correct some formatting, variable names, and math in box example (#1184) (@just-walk)
• Docs: Mention gradient et al. in 'getting started' (#1185) (@lassepe)
• Turn code samples into doctests on first page of docs (#1187) (@gdalle)
• WIP mutability inspection fix (#1190) (@swilliamson7)
• Changes to track active variables with inactive instruction (#1191) (@jlk9)
• Array copy fix (#1194) (@wsmoses)
• Fix gc error (#1195) (@wsmoses)
• Mark arraycopy as nofree (#1197) (@wsmoses)
• Fix kindattr check (#1198) (@wsmoses)
• Misc ABI fixups (#1200) (@wsmoses)
• Mark 1.10 whocallsmorethan30args test as known failure (#1201) (@wsmoses)
• Disable failing blas tests on windows 1.6 (#1203) (@wsmoses)
• Arraycopy unused shadow fix (#1204) (@wsmoses)
• Absint typeof ijl_new_structt (#1206) (@wsmoses)
• Disable sort on integers (#1207) (@wsmoses)
• Add support for newstructt (#1209) (@wsmoses)
• Improve debug output for Not yet implemented constant table in jl_eqt… (#1210) (@wsmoses)
• Enable gradient wrapper to work with non-type inferrable code (#1211) (@wsmoses)
• add recursive add to accumulate (#1213) (@wsmoses)
• Handle apply iterate in fwd mode (#1214) (@wsmoses)
• Handle gep-like cast in GC fixup (#1215) (@wsmoses)
• Handle make_zero of box (#1216) (@wsmoses)
• [embarassing bugfix] actually call ntuple (#1217) (@wsmoses)
• Use recursive_add for getindex (#1218) (@wsmoses)
• Fix splat (#1221) (@wsmoses)
• Remove unused type var (#1225) (@wsmoses)
• Setindex fixup (#1226) (@wsmoses)
• Remove unnecessary extra shadow for generic call (#1227) (@wsmoses)

Closed issues:

• Sparse Array Structure Initializer (#373)
• Enzyme compilation error when using a sparse cholesky (#997)
• Mention gradient in the Getting started section (#1035)
• illegal insertion UNREACHABLE when calling gradient (#1101)
• GPU kernel error (#1166)
• Custom rules ignored on Julia 1.9.4 that work fine with 1.10.0-rc1 (#1177)
• ERROR: LoadError: Enzyme execution failed (#1181)
• Minor bug with Enzyme.Compiler.make_zero (#1186)
• Code sample on first page of docs is broken (#1188)
• How to build documentation locally? (#1189)
• Crash during GC (#1192)
• Error when range step is not 1 (#1196)
• GPU normalize error with CUDA.jl v5 (#1199)
• When I use structure as an input for a function to be differentiated to and supply a temporary variable for calculations I get 0 as derivative to the structure (#1223)
• Correctness issue in gradient between Julia 1.6 and 1.10 (#1224)

v0.11.11

Enzyme v0.11.11

Diff since v0.11.10

Merged pull requests:

• [EnzymeTestUtils] Mark tests as now passing (#1098) (@wsmoses)
• Import EnzymeCore into Enzyme (#1120) (@glwagner)
• Allow linsolv to have const args (#1121) (@wsmoses)
• Remove irrelevant setup code from, documentation (#1122) (@LilithHafner)
• Export annotation (#1124) (@wsmoses)
• Fix calling convention print and mismatch (#1125) (@jlk9)
• Bump JLL (#1126) (@wsmoses)
• [Nodecayed phis] handle repeated phi entry (#1127) (@wsmoses)
• Fix hcat_fill and custom rule type propagation (#1128) (@wsmoses)
• Add atan2 to cuda replacements (#1131) (@wsmoses)
• Fix return mismatch on partial information (#1133) (@wsmoses)
• Update JLL (#1135) (@wsmoses)
• [NFC] Move internal rules into separate files (#1136) (@wsmoses)
• JLL Bump (#1137) (@wsmoses)
• Add some support for tuple splat (#1141) (@wsmoses)
• Attempt to fix version testing mishap (#1142) (@wsmoses)
• Don't use Pkg in runtests.jl (#1143) (@vchuravy)
• Disable nodecay transformation for guaranteed inactive functions (#1144) (@wsmoses)
• Apply iterate with multiple iter args (#1146) (@wsmoses)
• Fix tuple number of fields warning (#1147) (@wsmoses)
• stablize active_reg_inner (#1148) (@wsmoses)
• Update Enzyme_jll (#1149) (@wsmoses)
• Fix union abi (#1150) (@wsmoses)
• fix is_arrayorvararg_ty (#1151) (@wsmoses)
• fix coupling of ptreltype and is_arrayorvararg_ty (#1153) (@wsmoses)
• Iddict (#1155) (@wsmoses)
• Abs int for n-ary arrays (#1157) (@wsmoses)
• additional partial activity (#1158) (@wsmoses)
• Expose make_zero (#1160) (@wsmoses)
• Improve custom rule sret detection (#1162) (@wsmoses)
• Update Project.toml (#1164) (@wsmoses)
• Absiint pointer_from_objre (#1165) (@wsmoses)
• Don't emit fallback error on known gpu paths (#1167) (@wsmoses)
• Fix forward mode type unstable more than 30 args (#1168) (@wsmoses)
• Update Project.toml (#1170) (@wsmoses)
• Update scripts_deploy.yml (#1171) (@wsmoses)
• Docs: discuss EnzymeRules.reverse return value in custom rules tutorial (#1174) (@lassepe)
• Better GC internal error message (#1175) (@wsmoses)
• [EnzymeTestUtils] mark test as now working (#1176) (@wsmoses)
• Handle load from global constant in gc fixup (#1179) (@wsmoses)

Closed issues:

• Bad TypeTree generation (#225)
• Union leads to broken function (#235)
• Enzyme + KA Stalls on Error instead of reporting it (#365)
• Enzyme generates malloc + free for KernelGradients matmul example (#577)
• augmented forward for jl_f__apply_iterate (#599)
• Enzyme compilation failed for cmpxchg (#655)
• [Forward-mode] Implement multi-typed store (#857)
• Enzyme cannot deduce type (#874)
• sum of array errors with recent Enzyme (#885)
• get for Dict errors with certain key types (#892)
• get error on earlier Julia versions when assertions are enabled (#993)
• Enzyme compilation failure (#1027)
• Enzyme execution failed in zygote tests (#1036)
• LLVM failed verification in custom batch rule (#1075)
• Garbage collection error (#1103)
• Decayattr error (#1108)
• get wrong gradient on Julia 1.9 (#1118)
• sum wrong gradient on Julia 1.10-nightly (#1119)
• Not yet implemented augmented forward for jl_f__apply_iterate when differentiating an ODE in reverse mode (#1123)
• GPU kernel error (#1129)
• KernelAbstractions and NTuple index (#1140)
• Tuple-modifying differentiates when imported from existing package, but not when written locally. (#1145)
• Enzyme.jl Error MethodError: no method matching ptreltype(::Type{Ptr}) (#1152)
• Suspected Garbage Collected Value Issue When Differentiating with Enzyme.jl (#1154)
• Crash on basic LoopVectorization operation (#1161)

v0.11.10

Enzyme v0.11.10

Diff since v0.11.9

Merged pull requests:

• Nofree array del (#1100) (@wsmoses)
• Add batch size for batchduplicatedtype (#1105) (@wsmoses)
• Fix abstract tuple (#1106) (@wsmoses)
• Create nicer decaayaddr error message (#1107) (@enigne)
• Fix decayaddr to handle stores into the decayed addr (#1109) (@enigne)
• Enable calling convention fixup for custom rules (#1110) (@enigne)
• Add linear solve derivative (#1111) (@enigne)
• Fix make_zero on undef arrays (#1112) (@enigne)
• Fix guaranteed_const instability and error message (#1113) (@wsmoses)
• Remove 1.7 from CUDA testing (#1114) (@wsmoses)
• Bump JLL (#1115) (@wsmoses)
• Bump Enzyme version (#1116) (@wsmoses)
• [EnzymeTestUtils] mark more tests as working (#1117) (@wsmoses)

v0.11.9

Enzyme v0.11.9

Diff since v0.11.8

Merged pull requests:

• Simplify make zero (#1063) (@wsmoses)
• Fix staticarrays active bug (#1071) (@wsmoses)
• Primitive deepcopy support (#1073) (@wsmoses)
• Consider jl_new_array (#1074) (@wsmoses)
• [EnzymeTestUtils] Fix batch duplicated return and aliasing return/argument (#1079) (@wsmoses)
• Fix constant, but aliasing, return (#1080) (@wsmoses)
• [EnzymeTestUtils] Prevent non-sroa'able tuple (#1082) (@wsmoses)
• Fix custom rule with batch fwd pointer (#1083) (@wsmoses)
• Add enzyme activity rule (#1084) (@wsmoses)
• CompatHelper: bump compat for GPUCompiler to 0.25, (keep existing compat) (#1085) (@github-actions[bot])
• Fix calling more than 30 arguments (#1087) (@swilliamson7)
• Generalize addr compuation (#1088) (@wsmoses)
• Delete range md when deleting return in julia (#1089) (@wsmoses)
• CompatHelper: bump compat for CEnum to 0.5, (keep existing compat) (#1090) (@github-actions[bot])
• Add missing type to ptreltype (#1092) (@jgreener64)
• Fix embarrassing array eltype size error (#1093) (@wsmoses)
• Fix offset calculation for i32 (#1094) (@wsmoses)
• Fix ptreltype (#1095) (@wsmoses)
• Simplify pmap rule [use EnzymeRules] (#1096) (@wsmoses)
• Add alloca type info metadata (#1097) (@wsmoses)
• Update Project.toml (#1099) (@wsmoses)

Closed issues:

• Support for jl_eqtable_get (#416)
• Enzyme using fallback BLAS replacements in deep neural network (#692)
• Enzyme compilation failed on IPU (#751)
• Assertion error with inlined function on recent Enzyme versions (#1025)
• error in nodecayed_phis (#1031)
• Simplified example of UDE dynamics with Lux neural networks (#1057)
• Handling of getrf! and getrs! (#1065)
• Constant variable as temporary storage for active memory (#1068)
• StaticMatrix in Struct - ERROR: Unsupported Active... (#1069)
• Forward mode of an ODE solver fails with TypeAnalysisDepthLimit (#1070)
• Fix Enzyme incompatibilities with Krylov methods (#1077)
• Issues with Wishart from Distributions.jl (#1081)
• Bounds error (#1086)

v0.11.8

Enzyme v0.11.8

Diff since v0.11.7

Merged pull requests:

• Add EnzymeTestUtils package for testing Enzyme rules (#782) (@sethaxen)
• Add tests for BLAS.dot, BLAS.dotc, and BLAS.dotu (#842) (@sethaxen)
• CompatHelper: bump compat for GPUCompiler to 0.22, (keep existing compat) (#1001) (@github-actions[bot])
• support wide ints in tape (#1002) (@motabbara)
• add 1.10 cpu features string (#1003) (@wsmoses)
• Fix poison value handling (#1006) (@wsmoses)
• Abstract state is default duplicated (#1007) (@wsmoses)
• Reduce amount of printing during 'cannot find shadow' error of an arg… (#1012) (@wsmoses)
• Attempt fix of nicer error handler for jl_f__apply_iterate (#1013) (@wsmoses)
• Fix 0 arg wait (#1014) (@wsmoses)
• Fix display of formula in documentation (#1015) (@metab0t)
• Fix fallback apply iterate in vector mode (#1016) (@wsmoses)
• Fix order of args to occursin to reduce printing on store error (#1017) (@wsmoses)
• Fix UndefVarError: ST not defined (#1019) (@devmotion)
• Mark custom rule tape loads as needing caching (#1024) (@wsmoses)
• Fix batched arraycopy (reverse mode) (#1033) (@wsmoses)
• CompatHelper: bump compat for GPUCompiler to 0.23, (keep existing compat) (#1034) (@github-actions[bot])
• Add getri nice backtrace (#1040) (@wsmoses)
• Add missing julia compat to EnzymeTestUtils (#1041) (@sethaxen)
• Unmark reverse mutating test as broken (#1042) (@sethaxen)
• Add nicer nc_sync error (#1044) (@wsmoses)
• Fix print for abi wrapper error (#1049) (@wsmoses)
• Fix array type cast (#1050) (@wsmoses)
• CompatHelper: bump compat for GPUCompiler to 0.24, (keep existing compat) (#1051) (@github-actions[bot])
• Set active_reg for mutable structs to be duplicated rather than active (#1052) (@wsmoses)
• Fix error when returning undef from function (#1054) (@wsmoses)
• Fix shadow segfault on custom rule (#1056) (@wsmoses)
• Add guess activity of ref of int (#1060) (@wsmoses)
• Fix ordering for vector fallback of apply iterate (#1062) (@wsmoses)
• Adapt to jll bump and use nice default error for unknown c functions (#1064) (@wsmoses)
• Better Store Activity Error Messages (#1066) (@wsmoses)

Closed issues:

• Adding testing utilities (#780)
• JLJit duplicate symbol (#991)
• tape width issue using InlineStrings (#998)
• Incorrect Jacobian when indexing an Array using begin & end (#1008)
• Catching assertions / debugging mode? (#1009)
• Assertion failed error (#1011)
• Illegal type analysis error with gemv call (#1020)
• Incorrect tape provided to custom reverse rule in for loop (#1022)
• Cannot deduce type error (#1023)
• mixed activity for jl_new_struct (#1026)
• dval of mutated vector not zeroed when other activities are Const (#1028)
• LLVM failed verification error with BatchDuplicated return type on reverse autodiff_thunk (#1032)
• No augmented forward pass found for dgemqrt_64_ (#1037)
• Precompilation failed; has_orc_v1 not defined (#1043)
• BLAS.scal! error with autodiff_thunk in batch reverse-mode (#1048)
• Crash with BLAS.scal! as argument in calling method with complex inputs (#1053)
• AssertionError: Tuple{Vector{Float64}, Float64} has mixed internal activity types (#1055)
• Invalid unwrap (#1058)

v0.11.7

Enzyme v0.11.7

Diff since v0.11.6

Closed issues:

• Better error for Active{Vector{Float64}} (#318)
• Differentiating Gridap code (#447)
• getting a function call location from a stacktrace (#854)
• improve debugging (#864)
• CUDA sin is broken (#924)
• runtime activity changes AD result (#947)
• quantile gives wrong gradient on Julia 1.6 (#973)
• hypot error on Julia 1.7 (#974)
• Set error with x86 (#975)
• Incorrect gradients when using views (#979)

Merged pull requests:

• Make bitcode replacement a preference (#960) (@gaurav-arya)
• Reduce warnings (and have as runtime errors) (#961) (@wsmoses)
• Fix phi of addrspace 13 (#962) (@wsmoses)
• Additional julia runtime function from int (#964) (@wsmoses)
• Fix hit c symbol map (#965) (@wsmoses)
• Extend make zero to handle generic structs (#966) (@wsmoses)
• Fix cuda implementedby (#967) (@wsmoses)
• Mark memhash as inactive (#968) (@wsmoses)
• Unreachable dce workaround (#972) (@wsmoses)
• Update GPU CI (#976) (@vchuravy)
• Better backtrace for some lapack functions (#980) (@wsmoses)
• Bump jll to 79 (#981) (@wsmoses)
• Fix unhandled handler (#982) (@wsmoses)
• nicer error fallback for getrf (#983) (@wsmoses)
• Don't mixed activity error if storing constant int (#984) (@wsmoses)
• Add structural check for view (#985) (@wsmoses)
• Disable attributor on known failing llvm versions (#986) (@wsmoses)
• Update compiler.jl (#987) (@wsmoses)
• Fix internal printing segfault on 1.6 where mustprogress attr not def… (#988) (@wsmoses)
• Update Project.toml (#989) (@wsmoses)
• Test hypot error with CI (#990) (@jgreener64)
• Applytup (#992) (@wsmoses)
• Fix active reg inner with vector (#994) (@wsmoses)
• Adapt to abi bump (#995) (@wsmoses)

v0.11.6

Enzyme v0.11.6

Diff since v0.11.5

Closed issues:

• autodiff_deferred failing on Metal.jl kernel (#925)
• 1.10 Forward mode segfault (#948)
• Compilation error for custom reverse rule (#952)

Merged pull requests:

• Demonstrate backtraces using the juliaojit (#863) (@gbaraldi)
• More stdlib and type tests (#872) (@jgreener64)
• Add memmove to nofreefns (#949) (@devmotion)
• Fix union return issue (#950) (@wsmoses)
• [CI] Update TagBot setup (#951) (@giordano)
• Fix runtime activity in generic (#953) (@wsmoses)
• Jll 76 (#955) (@wsmoses)
• fix newstruct (#957) (@wsmoses)
• Erase with placeholder (#958) (@wsmoses)
• Bump jll to 78 (#959) (@wsmoses)

v0.11.5

Enzyme v0.11.5

Diff since v0.11.4

Closed issues:

• Gradient computation "corrupted" on ARM-based M2 CPU (#611)
• Error with missing value (#801)
• Derivative leaks into Const input view (#804)
• Segfaults in LLVM.jl on Julia 1.9.0 (#860)
• Enzyme + KernelAbstractions: KA syntax changes and differentiating multiple kernels (#896)
• Handle the new calling convention that julia emits for cheaper ptls (#909)
• Enzyme compilation failed when using logpdf (#910)
• Enzyme troubles with ElectrochemicalKinetics (#931)
• Asserts on upcoming Julia 1.10 (#933)
• Regression on differentiating Bessels.jl (#941)

Merged pull requests:

• enable LLVM6 (#917) (@motabbara)
• Fix argbox issue (#932) (@wsmoses)
• Wrap gradient utils in julia_error (#938) (@wsmoses)
• Improve hessian docs (#939) (@wsmoses)
• Add noinl fix (#940) (@wsmoses)
• Add line break in hessian docs (#942) (@Vaibhavdixit02)
• Make more errors occur at runtime (#943) (@wsmoses)
• Use world in precomputing activity (#944) (@wsmoses)
• Missing closing backticks (#945) (@Vaibhavdixit02)
• Jll bump 75 (#946) (@wsmoses)