v0.12.0
Enzyme v0.12.0
Merged pull requests:
- TA pass julia type info via param attr (#1300) (@wsmoses)
- Memoize typetree calls (#1302) (@vchuravy)
- add complex sqrt (#1324) (@wsmoses)
- Remove invariant loads on primal computation (#1325) (@wsmoses)
- Add Julia formatter (#1326) (@simsurace)
- Define
compiler_job_from_backend
(#1328) (@vchuravy) - Add FAQ section to docs (#1343) (@gdalle)
- Set
parallel
tofalse
ifparent_job
is notnothing
(#1348) (@giordano) - Bump Enzyme_jll to 0.0.103 (#1351) (@vchuravy)
- Skip intermittant builds (#1352) (@vchuravy)
- Use
make_zero
andRef
insidegradient
(#1356) (@mcabbott) - Add additional docs (#1357) (@wsmoses)
- Set type warning false by default (#1360) (@wsmoses)
- Fix assertion on nocapture llvm.enzymejl.sret (#1361) (@wsmoses)
- Fix CUDA tests (#1364) (@vchuravy)
- Support GPUCompiler 0.26 (#1365) (@vchuravy)
- Load Adapt unconditionally for Julia 1.8 (#1369) (@vchuravy)
- Backport EnzymeCore Adapt fix to 0.6 (#1374) (@vchuravy)
- Set development branch to 0.12 (#1375) (@vchuravy)
- Fix undefined memory in faq (#1376) (@wsmoses)
- Add differential use handler (#1377) (@wsmoses)
- Mark randn! as inactive (#1378) (@wsmoses)
- [EnzymeTestUtils] Mark 1.8 batch test as failing (#1379) (@wsmoses)
- Restore 1.9+ setfield (#1383) (@wsmoses)
- Update Project.toml (#1384) (@wsmoses)
Closed issues:
- Why does
Duplicated(x, dx)
assumex
anddx
have the same type? (#1329) - Forward over reverse for variadic function (#1336)
- gradient (and gradient!) fails on basic example (#1340)
- Warning with Flux.Bilinear (#1341)
- Matrix exponential fails (#1346)
- EnzymeTestUtils tests failing on main (#1349)
- Using Julia's LLVM and Enzyme.jl's artifacts to differentiate C code (#1367)
- Enzyme requests reverse rule on Const returns (#1380)
- Gradient returns wrong result for static arrays (#1382)