What's Changed

• Fix Fwddiff canonicalization [MLIR] by @vimarsh6739 in #2333
• Fix ocml f64 by @wsmoses in #2336
• Add custom derivative use by @wsmoses in #2315
• MLIR: fix constant tablegen by @wsmoses in #2337
• Return activity optimization for reverse mode AD [MLIR] by @vimarsh6739 in #2332
• fix: remove ignore_derivatives pass by @avik-pal in #2339
• Try force marking hip math as used by @wsmoses in #2344
• Improve openmp error message by @wsmoses in #2343
• Bazel: generate compile_commands.json by @vimarsh6739 in #2338
• [MLIR][ActivityAnalysis] fix getFunctionIfArgument by @wsmoses in #2350
• [mlir] Fix failing lit test, remove warning by @pengmai in #2347
• [mlir] Summary-based activity analysis by @pengmai in #2114
• Integrate llvm by @wsmoses in #2351
• More llvm integration by @wsmoses in #2352
• mlir: Move cache mincut to enzyme proper by @Pangoraw in #2355
• WIP: fix gc_loaded shadow by @wsmoses in #2353

Full Changelog: v0.0.182...v0.0.183

What's Changed

• feat: op to detach tensor [MLIR] by @avik-pal in #2328
• TypeAnalysis: don't analyze tbaa of not analyzable by @wsmoses in #2335

Full Changelog: v0.0.181...v0.0.182

What's Changed

• Add ocml support by @wsmoses in #2324
• MLIR: Implement strong zero by @wsmoses in #2326
• MLIR: only value if broadcast by @wsmoses in #2327
• Move strong zero to not be a global & fix vector RT activity by @wsmoses in #2329

Full Changelog: v0.0.180...v0.0.181

What's Changed

• Only throw fwd mode runtime activity blas error if something is runti… by @wsmoses in #2318

Full Changelog: v0.0.179...v0.0.180

What's Changed

• Push write fix by @wsmoses in #2314

Full Changelog: v0.0.178...v0.0.179

What's Changed

• Julia: fix fwd mode batched write barrier by @wsmoses in #2313

Full Changelog: v0.0.177...v0.0.178

What's Changed

• Speed up may execute between by @wsmoses in #2299
• Bump DoozyX/clang-format-lint-action from 0.18.2 to 0.20 by @dependabot in #2296
• Update enzyme-mlir.yml by @wsmoses in #2300
• tblgen: Auto-generate ActivityOpInterface by @Pangoraw in #2303
• Add better error message for custom rule activity mismatch by @wsmoses in #2304
• PreserveNVVM: use early inc range for attribution by @wsmoses in #2307
• replace deprecated function in tblgen by @ZuseZ4 in #2306
• Julia: fix gc_loaded activity analysis by @wsmoses in #2312

Full Changelog: v0.0.176...v0.0.177

What's Changed

• Improve error message for incorrect type by @wsmoses in #2279
• Fwddiff ret_activity canonicalizations by @vimarsh6739 in #2271
• fix: remove uses of .cast/.dyn_cast/.isa by @avik-pal in #2280
• Report diagnostic as errors in ops removal interface by @Pangoraw in #2282
• MLIR: Add verifyPostPasses option by @Pangoraw in #2284
• don't access LLVMEnzyme on apple builds if LLVM_PLUGINS isn't set by @ZuseZ4 in #2285
• Try making c++ ci green by @wsmoses in #2286
• Fix and test LLVM 19/20 by @wsmoses in #2287
• fix apple builds against a no-plugin LLVM by @ZuseZ4 in #2288
• Disable truncate test on LLVM19 per LLVM issue by @wsmoses in #2291
• Fwd canonicalizer, fix const noneed by @wsmoses in #2292
• Move math simplification to interface by @wsmoses in #2293
• Update enzyme-mlir.yml by @wsmoses in #2295
• Rematerialization: handle unreachable default switch dest by @wsmoses in #2294
• Memcmp attribute: it is readonly, not write-only by @wsmoses in #2297

New Contributors

• @vimarsh6739 made their first contribution in #2271
• @avik-pal made their first contribution in #2280

Full Changelog: v0.0.175...v0.0.176

What's Changed

• fix-crash-by-coalese by @minansys in #2246
• update for llvm::Triple becoming a proper class by @ftynse in #2269
• Allow suppression of incompatible compiler warning by @dime10 in #2272
• Fix llvm rebase by @wsmoses in #2273
• support for LLVM 19 by @ved1beta in #2274
• [LLVM Integrate] adapt to new fabs by @wsmoses in #2276
• Fix phi address space casting by @wsmoses in #2277

New Contributors

• @dime10 made their first contribution in #2272
• @ved1beta made their first contribution in #2274

Full Changelog: v0.0.173...v0.0.174