v0.0.14

What's Changed

• Update EnzymeAD/Enzyme to commit bec2714865d54b62bd8ce1b19d3eefaa1f3943d9 by @enzymead-bot[bot] in #2505

Full Changelog: v0.0.13...v0.0.14

v0.0.13

What's Changed

• [CI] Set container image for Tag workflow by @giordano in #2517
• [CI] Use release/v1 ref for pypa/gh-action-pypi-publish by @giordano in #2518
• [CI] Move PyPI publishing to a non-Docker job by @giordano in #2519
• Test macos build by @wsmoses in #2520
• [CI] Publish package to PyPI from a single job by @giordano in #2521

Full Changelog: v0.0.12...v0.0.13

Nightly Release

Latest nightly build of Enzyme-JAX.

v0.0.11

What's Changed

• Checkpointing take 2 by @Pangoraw in #1014
• Add linalg.qr op by @mofeing in #985
• chore(deps): bump protobuf from 5.28.2 to 5.29.5 in /builddeps by @dependabot[bot] in #1048
• fix(python): update tests by @avik-pal in #1046
• Raise with gpu id by @wsmoses in #1047
• chore(deps): bump requests from 2.32.3 to 2.32.4 in /builddeps by @dependabot[bot] in #1052
• Revert "feat: transpose of scatter (#1042)" by @avik-pal in #1051
• Fix while dataflow by @wsmoses in #1054
• Raising fixes by @pengmai in #1055
• Fix lower kernel to cpu by @wsmoses in #1056
• Explicit check for gpu.alloc async token by @pengmai in #1057
• Add ability to print to file by @pengmai in #1058
• Fix barrier by @wsmoses in #1059
• chore(deps): bump aiohttp from 3.10.10 to 3.10.11 in /builddeps by @dependabot[bot] in #1061
• While loop multi induction by @wsmoses in #1062
• [Enzyme][While] Fix multi ind usage by @wsmoses in #1063
• Fix stream type by @ivanradanov in #1065
• fix: mapping in TransposeAllUsersSlice by @avik-pal in #1064
• fix: add checks for gather_elementwise by @avik-pal in #1069
• feat: scatter setindex reverse mode by @avik-pal in #1071
• fix: scatter derivative for all constant inputs by @avik-pal in #1073
• Fix TPU kernel name on enzymexla.linalg.qr lowering by @mofeing in #1070
• feat: various passes to fuse mul chains by @avik-pal in #1079
• feat: more all inf simplifications by @avik-pal in #1085
• feat: reshape(bcast) simplify by @avik-pal in #1089
• feat: transpose fft optimizations by @avik-pal in #1091
• feat: transpose reshape optimization for dimension expansion by @avik-pal in #1093
• bump jax by @wsmoses in #1088
• Libdevice workaround lack of arith pointer support by @wsmoses in #1097
• fix if statement issue by @wsmoses in #1100
• While checkpointing fix by @wsmoses in #1092
• Fix wheel build by @wsmoses in #1102
• Don't discard value names by @wsmoses in #1101
• Fix release build of libRaise by @wsmoses in #1103
• Update workspace.bzl by @wsmoses in #1106
• Fix bounds error of gpu error lowering by @wsmoses in #1105
• Address space cast lowering by @wsmoses in #1107
• Remove bad pad simplify by @wsmoses in #1108
• [GPU1] don't serialize upgraded parallel loops from within a kernel by @wsmoses in #1109
• Add parallel serialization by @wsmoses in #1110
• Switch2If fix if no returned value by @wsmoses in #1111
• Fix libdevice lowering, and gpu/barrier lowering by @wsmoses in #1113
• ci: format bazel build files + format checker by @avik-pal in #1112
• Fix affinecfg forop raising by @wsmoses in #1114
• GPU: emit noop alternative, if available by @wsmoses in #1115
• Add linalg.svd op by @mofeing in #1049
• Move dus derivative rule to tablegen by @wsmoses in #1121
• Debug while issue by @wsmoses in #1122
• Bump jax by @wsmoses in #1123
• Don't use wheel on arm by @wsmoses in #1125
• Fix infinite compile time by @wsmoses in #1127
• fix remaining perf bug by @wsmoses in #1128
• chore(deps): bump actions/checkout from 3 to 4 by @dependabot[bot] in #1126
• feat: add activation functions by @avik-pal in #1104
• Fix memory error by @wsmoses in #1130
• Nondom lift by @wsmoses in #1132
• Fix typo by @giordano in #1133
• LLVM affine access fix by @wsmoses in #1134
• Canonicalize memref.load(pointer2memref(gep)) by @tyb0807 in #1119
• AffineCFG fix constant raise by @wsmoses in #1139
• getIntOrFloatBitwidth only if type is int or float by @tyb0807 in #1140
• While dead arg speedup by @wsmoses in #1146
• WhileDeadResult: intermediate deletion by @wsmoses in #1148
• Fast negation check by @wsmoses in #1149
• XLA exec by @wsmoses in #1153
• Add support for non-constant bounds in do-while to for pattern by @tyb0807 in #1152
• Code hygiene for CanonicalizeFor by @tyb0807 in #1154
• feat: register triton passes + dialects by @avik-pal in #1156
• Bump jax by @wsmoses in #1157
• fix: transpose batchnorm ops by @avik-pal in #1159
• Generalize while2for by @wsmoses in #1160
• Fix integer or float temporary shim by @wsmoses in #1162
• Use cast to element type by @wsmoses in #1163
• feat: register stablehlo_ext passes by @avik-pal in #1165
• fix isLegalConcatToOneDimDUS by @jumerckx in #1166
• Remove IV related iter args from scf.for by @tyb0807 in #1167
• Bump dependencies by @wsmoses in #1168
• Add a dialect-specific canonicalization pass by @ftynse in #1172
• Register string debug info by @wsmoses in #1175
• Fix concat to dus by @wsmoses in #1176
• Bump enzyme commit by @wsmoses in #1177
• Use new runners by @wsmoses in #1173
• More tests to RemoveIVRelated pattern by @tyb0807 in #1181
• feat: negate/add/subtract simplify by @avik-pal in #1182
• Attempt full python testing by @wsmoses in #1180
• Update StableHLOAutoDiffOpInterfaceImpl.cpp by @wsmoses in #1192
• TPU CI fixes by @wsmoses in #1184
• Don't raise zero arg functions by @wsmoses in #1195
• feat: diag matrix multiply optimize by @avik-pal in #1194
• Fix constants brought into kernel, and loads. multi dim store wip by @wsmoses in #1196
• [CI] Run format bazel workflow only when necessary by @giordano in #1198
• [CI] Add GHA job to test downstream GB-25 by @giordano in #1197
• [CI] Also run correctness tests on GB-25 code by @giordano in #1210
• feat: compare negate const simplify by @avik-pal in #1215
• Add widen wrap and extend optimizations by @wsmoses in #1212
• feat: elementwise pad optimizations by @avik-pal in #1219
• Support negate in sum to conv by @wsmoses in #1222
• Bump jax by @wsmoses in #1223
• feat: simplify concatenate of subtract by @avik-pal in #1221
• Fix assertion concat bcast by @wsmoses in #1226
• Don't recogn...

v0.0.10

update buildkite

v0.0.9

What's Changed

• optimize transpose(convolution) -> convolution by @Pangoraw in #117
• Force bump xla commit by @wsmoses in #119
• Simplify path infra by @wsmoses in #128
• Bump internals by @wsmoses in #130
• Attempt gpu ci fix by @wsmoses in #125
• Fast path slice contiguous constant by @wsmoses in #137
• Add jaxmd tests by @wsmoses in #136
• Transpose batch by @wsmoses in #138
• Maxtext by @wsmoses in #139
• implement derivative for real and imag by @Pangoraw in #145
• Update compile_with_xla.cc by @klucke in #147
• Clamp derivatives by @Pangoraw in #148
• IfOp by @Pangoraw in #149
• bump enzyme and jax version by @Pangoraw in #150
• Update test-requirements.txt by @wsmoses in #155
• Try a100 CI by @wsmoses in #156
• CUDA ci by @wsmoses in #154
• add neuralgcm by @wsmoses in #141
• Generic batch op interface by @Pangoraw in #151
• Use absl for neuralgcm by @wsmoses in #158
• DynamicUpdateSliceOp reverse derivative by @Pangoraw in #159
• Try caching again by @wsmoses in #163
• WhileOp reverse derivative by @Pangoraw in #160
• remove transform.with_named_sequence attribute by @Pangoraw in #153
• Keras3 by @wsmoses in #166
• Merge concatenate of contiguous slices in a single slice by @Pangoraw in #168
• Improve enzyme gradient ops removal in while op by @Pangoraw in #167
• Fix stablehlo ffi by @wsmoses in #171
• Nicer hlo ffi by @wsmoses in #172

New Contributors

• @klucke made their first contribution in #147

Full Changelog: v0.0.8...v0.0.9

v0.0.8

What's Changed

• ci(buildkite): add gpu_pipeline by @avik-pal in #114
• Update tag.yml by @wsmoses in #115
• Fix build deps by @wsmoses in #116

New Contributors

• @avik-pal made their first contribution in #114

Full Changelog: v0.0.7...v0.0.8

v0.0.7

What's Changed

• Benchmark llama by @wsmoses in #23
• Update README.md by @wsmoses in #26
• Rebase to later llvm/xla by @wsmoses in #25
• Add support for use of xla runtime by @wsmoses in #27
• [JAX] Remove uses of dialect="mhlo" from the JAX compiler_ir() function by @wsmoses in #28
• Bump XLA by @wsmoses in #29
• Custom MLIR lowering pipeline by @wsmoses in #30
• Add missing ')' to enzyme_call and add tests for old pipeline by @itf in #31
• Make enzyme_ref delete in_shapes instead of out_shapes by @itf in #32
• Pipelinemod by @wsmoses in #33
• reuse Common.td from Enzyme by @ftynse in #34
• Fix segfault by @wsmoses in #35
• Jax reverse more by @wsmoses in #36
• MLIR Reverse Mode by @wsmoses in #37
• Update xla by @wsmoses in #38
• Fix unused warning by @wsmoses in #39
• XLA: vendor the runtime mlir backend by @wsmoses in #40
• Handle inactive args from context by @wsmoses in #41
• Fix shape error by @wsmoses in #42
• generalize mlir zeroing by @wsmoses in #43
• Generalize mktup by @wsmoses in #44
• Cleanup warnings by @wsmoses in #45
• bugfix else by @wsmoses in #46
• Reduction optimization by @wsmoses in #47
• More optimization fixes by @wsmoses in #48
• Add pad folding optimization by @wsmoses in #49
• Add unrolling by @wsmoses in #50
• Cleanup bazel files by @wsmoses in #51
• transpose opts by @ftynse in #52
• Add missing dependencies by @ivanradanov in #53
• Fix 0 dim reshape concat case by @ivanradanov in #55
• Handle full reduce of reshape by @wsmoses in #56
• simply mul-of-pad by @ftynse in #57
• Generalize dot general pad by @wsmoses in #59
• Add gradient of pad operation by @wsmoses in #58
• slice(reshape) -> reshape(slice) by @ftynse in #60
• Fix likely index bug in reshape helper by @wsmoses in #61
• Add chlo dialect by @wsmoses in #63
• Add .clang-format by @ftynse in #64
• Fix and add test for slice of pad by @wsmoses in #65
• propagate location information by @ftynse in #66
• simplify pad(pad) by @ftynse in #67
• transform dialect for pattern combination by @ftynse in #62
• Add slice of dot general by @wsmoses in #70
• Update XLA/LLVM by @wsmoses in #72
• Bitflag mismatch size in transform[er] dialect by @wsmoses in #73
• Try to fix macos by @wsmoses in #69
• Transform dialect ops for all patterns by @ftynse in #71
• Actuall take control over llvm build flags by @wsmoses in #75
• Just use c api for pass pipeline by @wsmoses in #76
• Use capsule by @wsmoses in #77
• Bump jax commit by @wsmoses in #78
• Bump enzyme commit by @wsmoses in #79
• Bump enzyme commit by @wsmoses in #80
• Fix up rebase by @wsmoses in #81
• Negate of int by @wsmoses in #86
• Add all StableHLO ops from spec by @mofeing in #87
• Refactor diff rules to prepare for more rules by @mofeing in #89
• Add stablehlo.einsum, stablehlo.unary_einsum diffrules by @mofeing in #83
• Fix for lowering change by @wsmoses in #92
• Fix grad sum by @wsmoses in #93
• Add complex support by @mofeing in #94
• Fix complex dialect registration by @mofeing in #95
• Update Enzyme commit by @mofeing in #96
• Add a bunch of rules for scalar and non-differentiable functions by @mofeing in #90
• Fix depency import of "ChloOps.h" by @mofeing in #97
• Fix broadcast derivative by @wsmoses in #98
• bump xla by @wsmoses in #99
• Add compile to llvm backend by @wsmoses in #100
• Bump Enzyme commit by @wsmoses in #101
• Bump JaX commit by @wsmoses in #102
• Mark stablehlo.compare as inactive by @Pangoraw in #106
• sink transposes in einsum by @Pangoraw in #105
• optimize convolution of transpose by @Pangoraw in #103
• Add batching of constants by @wsmoses in #107
• Attempt linker fix by @wsmoses in #108
• More linker fix by @wsmoses in #109
• Llama 3 gap size by @wsmoses in #110
• Bump xla and fix ci by @wsmoses in #111
• Fix CI by @wsmoses in #112
• Fix signature by @wsmoses in #113

New Contributors

• @itf made their first contribution in #31
• @ivanradanov made their first contribution in #53
• @mofeing made their first contribution in #87
• @Pangoraw made their first contribution in #106

Full Changelog: v0.0.6...v0.0.7

v0.0.6

Repackage as enzyme_ad with a jax subpackage to avoid name clashes on PyPI.

v0.0.5

Bump version (#8)