v0.0.89

What's Changed

• Mpi allreduce no free by @ludgerpaehler in #1481
• Add flag to disable AD. by @ludgerpaehler in #1482
• Tblgen diffuse by @wsmoses in #1485
• cublas support by @ZuseZ4 in #1453
• Fix memory errors by @wsmoses in #1488
• Add nice helper loose type and print type lld cmake flags by @wsmoses in #1492
• Fix blas shadow differential use analysis by @wsmoses in #1493
• Add loose type fallback for adding type of cast by @wsmoses in #1494
• Fix legalOr bug in type tree by @wsmoses in #1497
• Add Enzyme print and cleanup TA by @wsmoses in #1496
• [NFC] Refactor adjoint generator by @wsmoses in #1498
• Fix Type index reuse by @wsmoses in #1499
• Add sparse to select by @wsmoses in #1495
• Fix additional loop var issue and speed up enzyme-print-type when deb… by @wsmoses in #1500
• Cleanup bazel and related warnings by @wsmoses in #1502
• attribute blas fnc in PreserveNVVM by @ZuseZ4 in #1505
• Fix transposed lda on cache by @wsmoses in #1501
• Zero alloca in Julia Calling convention fix by @wsmoses in #1506
• Fix unnecessary transpose usage by @wsmoses in #1507

Full Changelog: v0.0.88...v0.0.89

v0.0.88

What's Changed

• [SimplifyMPI] handle invoke by @wsmoses in #1467
• controlflow fixes by @tthsqe12 in #1425
• Fix must exit switch assertion error by @wsmoses in #1468
• Fix blas attributor by @wsmoses in #1470
• Fix enzyme_inactive removal on return by @wsmoses in #1469
• Fix LLVM16 build by @wsmoses in #1475
• Create c++ nofree attribute by @wsmoses in #1474
• Fix Llvm main by @wsmoses in #1476
• Add Metadata string and typetree API's by @wsmoses in #1479
• Support return attribute of fn being active/inactive by @wsmoses in #1478

Full Changelog: v0.0.87...v0.0.88

v0.0.87

What's Changed

• Don't store inverted global into map [fixes module linking state] by @wsmoses in #1456
• Add write barrier binding support by @wsmoses in #1457
• Fix forward mode nice c++ error by @wsmoses in #1458
• Simplify type analysis dump by @wsmoses in #1461
• Mark mpi functions of booleans as inactive by @wsmoses in #1464
• [Type Analysis] handle extract vector of i1 by @wsmoses in #1463
• Add collect offset c api function by @wsmoses in #1465

Full Changelog: v0.0.86...v0.0.87

v0.0.86

C Abi embarrassing bugfix

v0.0.85

What's Changed

• Do simple mem2reg for autodiff fn detection by @wsmoses in #1431
• Special case gep of small_typeof by @wsmoses in #1433
• Speed up and fix type analysis merges by @wsmoses in #1432
• Fix memset no type handling by @wsmoses in #1435
• Also memset undef as sret by @wsmoses in #1436
• Fix extract invertpointer ordering by @wsmoses in #1437
• Adding instructions for expm1f / expm1l by @jlk9 in #1438
• Fix mem2reg on input fn arg by @wsmoses in #1442
• Reuse intermediate loop variables on multi exit by @wsmoses in #1443
• [BLAS] fix row transpose arg by @wsmoses in #1440
• fix gemv rule for A by @ZuseZ4 in #1441
• Add blas integration infra tests by @wsmoses in #1439
• add blas axpy support by @ZuseZ4 in #1445
• add blas copy support by @ZuseZ4 in #1447
• Add dot blas test by @wsmoses in #1446
• Implement concat by @wsmoses in #1450
• Nice error message for undifferentiable functions by @wsmoses in #1451
• start fixing gemm by @ZuseZ4 in #1448

New Contributors

• @jlk9 made their first contribution in #1438

Full Changelog: v0.0.84...v0.0.85

v0.0.84

What's Changed

• Handle memset of undef memory, even if not type analyzable by @wsmoses in #1426
• Do not perform runtime activity check for known non-aliasing pointers by @wsmoses in #1427
• Print debug info on assertion by @wsmoses in #1428
• Handle null unwrap return in available map by @wsmoses in #1429
• Fix use of exit block shared by multiple loops by @wsmoses in #1430

Full Changelog: v0.0.83...v0.0.84

v0.0.83

What's Changed

• fix invsqrt example by @tthsqe12 in #1396
• Add const global test by @gaurav-arya in #1412
• Fix julia memset runtime err by @wsmoses in #1415
• Better c++ errors by @wsmoses in #1417
• Generalize binop inverse of active float by @wsmoses in #1418
• Fix adding type on mix of int/ptr/float by @wsmoses in #1419
• Fix memset to handle non-0 address spaces by @wsmoses in #1420
• Fix int/float memset by @wsmoses in #1421
• Simplify insert/extract, as required for address prop by @wsmoses in #1422
• Fix realloc for vector mode by @wsmoses in #1423
• Ensure return roots are remateralizable by @wsmoses in #1424

New Contributors

• @gaurav-arya made their first contribution in #1412

Full Changelog: v0.0.82...v0.0.83

v0.0.82

What's Changed

• Cleanup CI by @tgymnich in #1379
• fix c using 32bit chars instead of 8bit chars by @ZuseZ4 in #1383
• fix cblas layout arg by @ZuseZ4 in #1384
• fix transpose handling for 32 bit characters by @ZuseZ4 in #1385
• Handle batch ffi abi by @wsmoses in #1390
• support more names from MS by @tthsqe12 in #1338
• move test to right location and adjust lit syntax by @ZuseZ4 in #1398
• use errorhandler for more TA stuff by @ZuseZ4 in #1397
• Tblgen bugfix by @ZuseZ4 in #1378
• Fix constant global by @wsmoses in #1401
• Fix accidental one iter store if inactive val by @wsmoses in #1400
• Fix llvm main by @wsmoses in #1399
• Create nicer error for nofree issue by @wsmoses in #1407
• BCLoad pass update for CI by @wsmoses in #1409
• Fix indirect call activity by @wsmoses in #1408
• Use new from original in nofree by @wsmoses in #1410
• Fix lookup normalization by @wsmoses in #1405
• Actual fix nofree getnewfromorig by @wsmoses in #1411

Full Changelog: v0.0.81...v0.0.82

v0.0.81

What's Changed

• Add option for building LLVMEnzyme as static library. by @erick-xanadu in #1377
• Fix runtime memtransfer assertion into a julia rule by @wsmoses in #1380
• Conservatively mark active args as needed for reverse pass in BLAS ru… by @wsmoses in #1382
• Fix incorrect replacement of switch default branch if all other branc… by @wsmoses in #1381

New Contributors

• @erick-xanadu made their first contribution in #1377

Full Changelog: v0.0.80...v0.0.81

v0.0.80

What's Changed

• Handle the small_typeof global now in julia main by @wsmoses in #1369
• fix minor capitalization issue in installation directory by @samuelpmishLLNL in #1367
• Drop llvm 9 and 10 by @wsmoses in #1360
• Add merge_group by @tgymnich in #1370
• Handle unnecessary intermediate by @wsmoses in #1373
• Ficticious phi map fix by @wsmoses in #1372

New Contributors

• @samuelpmishLLNL made their first contribution in #1367

Full Changelog: v0.0.79...v0.0.80