v1.1.0

Major Changes

• Platforms
  • Improved support for utilization of NVIDIA GPUs via CUDA
  • New support for utilization of AMD GPUs via HIP
  • Use https://github.com/victor-anisimov/Librett for tensor permutations on GPUs
• API:
  • einsum for general binary tensor products
  • Reference support for DistArrays with nested (Tensors-of-Tensors, aka ToT) tiles
  • Efficient re-ranging (retiling, subarrays, etc.)
• External dependencies
  • Use blaspp/lapackpp (aka linalgpp) for 1-node C++ linear algebra
  • Use blacspp/scalapackpp for multi-node C++ linear algebra
  • Use https://github.com/wavefunction91/linalg-cmake-modules for linear algebra discovery
  • Can build modularized Boost from source

Detailed List of Changes

• Andrey's revamp of generic algebra by @evaleev in #226
• Generic solver interface by @evaleev in #222
• Fix eigen MD5 hash 05b1f7511c93980c385ebe11bd3c93fa --> b9e98a200d245… by @powellsr in #228
• support for arrays of tensors (aka nested tensors, or tensors-of-tensors) by @evaleev in #223
• Feature/gitlab ci by @asadchev in #231
• Asadchev/refactor/math by @asadchev in #230
• Asadchev/refactor/lapack by @asadchev in #232
• small Ranges on stack by @evaleev in #233
• default TA_ERROR to throw if building unit tests by @evaleev in #234
• Asadchev/refactor/unit tests by @asadchev in #239
• converted to C++ BLAS/LAPACK interface by @evaleev in #237
• scalapack usable with distarrays of btas (and other) Tiles. by @evaleev in #241
• blaspp_headers to btas by @evaleev in #242
• APPLE does not imply x86_64 on Apple ARM hardware by @evaleev in #244
• kmp5VT [Feature] Round Robin pmap by @kmp5VT in #235
• Fix CMake Boost discovery by @asadchev in #245
• send notification from travis builds to VG slack by @evaleev in #246
• build from-source-dependencies before building TA by @evaleev in #249
• Asadchev/feature/gitlab cuda build by @asadchev in #250
• Asadchev/feature/GitHub actions ci by @asadchev in #253
• Update README.md for TA::TiledRange1 by @bimalgaudel in #257
• Evaleev/update/btas by @evaleev in #260
• Refactor TA_ASSERT by @asadchev in #259
• numeric_type trait for Eigen matrices not needed since Eigen 3.3 by @evaleev in #264
• Fix inconsistencies in rank-local SVD wrapper by @wavefunction91 in #263
• Fixes #265 by @ryanmrichard in #266
• Evaleev/fix/nonintrusive solver adaptors by @evaleev in #270
• Evaleev/fix/make ta range by @evaleev in #273
• update travis clang to 11 by @evaleev in #274
• SparseShape ctor taking scaled norms zeroes out values below threshold by @evaleev in #275
• Kmp5/feature/btas update by @kmp5VT in #272
• [cmake] use wfn91's linear algebra discovery modules by @evaleev in #254
• Range avoids serializes rank only once by @evaleev in #276
• cuda callback is prebuf-aware by @evaleev in #277
• Updates for revised madness serialization by @evaleev in #279
• Evaleev/fix/tensorimpl distributed ctor by @evaleev in #278
• DistArray::lazy_deleter waits for delayed sets by @evaleev in #280
• set CUDA vars before calling enable_language(CUDA) + misc cleanup by @evaleev in #283
• DistArray::set can properly avoid copies (unless setting remote data)… by @evaleev in #284
• Bump ScaLAPACK++ by @wavefunction91 in #287
• removed residual uses of TA_DEFAULT_ERROR by @evaleev in #289
• Fix the path for MADNESS config.h by @keceli in #293
• Bug fix fill_random() method by @bimalgaudel in #295
• Functions to change taskq wait policy by @asadchev in #294
• Tensor fwddecl "moved" to fwd.h by @evaleev in #296
• introduced umpire host allocator by @evaleev in #297
• clang does not like vector_il/matrix_il/etc. ... by @evaleev in #298
• Bumped BTAS tag to use most recent linalgpp by @evaleev in #299
• Build deps, if not found, via FetchContent, NOT ExternalProject by @evaleev in #300
• follow-up to failed ExternalProject elimination by @evaleev in #302
• run unit tests with raised log_level by @evaleev in #303
• Asadchev/feature/einsum by @evaleev in #285
• [cmake] bump MADNESS tag by @evaleev in #304
• umpire: skip std::filesystem if using old gcc by @evaleev in #306
• installation fixes by @evaleev in #307
• [cmake] BTAS fixes by @evaleev in #308
• foreach works with ShareReductionMethod::Union by @evaleev in #310
• bump umpire to v6+ by @evaleev in #311
• Add QR Implementations by @wavefunction91 in #316
• PaRSEC in MADNESS CI by @therault in #301
• Change ExternalProject prefixes to always match the FetchContent location by @awild82 in #315
• eigen {matrix,vector}_to_array UT needs to replicate the data... by @evaleev in #319
• block tsr expression assignments fix by @evaleev in #318
• DistArray conversion to/fro Eigen::Tensor by @evaleev in #320
• TiledArray/tensor.h: must #include <TiledArray/tile_op/tile_interface.h> by @evaleev in #321
• SparseShape maintains its own sparse threshold by @evaleev in #322
• Bump VG cmake kit tag by @evaleev in #323
• moved FindOrFetchScaLAPACK to vg cmake kit + bump vg cmake kit and BT… by @evaleev in #326
• moar small_vector by @evaleev in #328
• wrong operator[] used in Index::indexof() by @evaleev in #329
• fix einsum permutes by @evaleev in #331
• bump VG's cmake kit tag to allow the use of LAPACK_CXX_COMPILE_OPTIONS by @evaleev in #334
• Evaleev/fix/dox by @evaleev in #336
• [ci] dox fixes + introduced VALEEVGROUP_UBUNTU_TAG envvar to control … by @evaleev in #337
• Fix type signature bug(?) by @bimalgaudel in #338
• bump BTAS tag + make imported Boost targets IMPORTED_GLOBAL ... regim… by @evaleev in #339
• Fix parallel compilation on Umpire by @wavefunction91 in #340
• std::result_of -> std::invoke_result by @evaleev in #341
• TiledArray_{UMPIRE,CUTT} targets usable from the build tree at configure time by @evaleev in #342
• allow extended character set in annotations by @evaleev in #343
• Asadchev/feature/eigen einsum by @asadchev in #344
• Einsum hadamard reduction, eg c('h') = a('hi')*b('hi') by @asadchev in #346
• [followup] allow extended character set in annotations by @evaleev in #347
• const-correct serialize methods by @evaleev in #349
• fix: linalgpp compile noise by @evaleev in #351
• LibreTT integration into TiledArray by @victor-anisimov in #352
• Distributed einsum by @asadchev in #348
• Asadchev/feature/einsum ta sparse by @asadchev in #356
• Asadchev/bug/ta einsum permute by @asadchev in #361
• fix distarray lifetime in init_tiles and assignment by @evaleev in #360
• discover/fetch TTG + use for cholesky_linv by @evaleev in #355
• #include linalg/basic.h from top-level solver-specific headers also by @evaleev in #363
• Asadchev/feature/einsum replicated by @asadchev in #362
• suppress warning re std::complex return by C-linkage LAPACK functions by @evaleev in #367
• update vg_cmake_kit by @evaleev in #368
• enable cuda unit tests by @evaleev in #354
• cutt -> librett by @evaleev in #353
• TILEDARRAY_REVISION is now a runtime constant returned by TA::revision() by @evaleev in #370
• switch from git revision to description by @evaleev in #371
• Evaleev/fix/umpire allocators lifetime by @evaleev in #372
• Evaleev/feature/tensor memory profile and trace by @evaleev in #373
• fixes for C++20 by @evaleev in #374
• Expr::set_shape is a no-op with null shape by @evaleev in #375
• bump MAD tag to reduce C++20 noise by @evaleev in #376
• Evaleev/fixup/tensor empty by @evaleev in #377
• Asadchev/feature/einsum ta dot by @asadchev in #369
• enable scalapack for github actions by @evaleev in #378
• relax constraints on allowed characters in annotations, can use unico… by @evaleev in #379
• *TsrExpr manages lifetime of the array object referred by the bound variable by @evaleev in #380
• [cmake] use CMAKE_{C,CXX}COMPILER_LAUNCHER instead of RULE_LAUNCH{COMPILE,LINK} by @evaleev in #381
• umpire allocator fixes by @evaleev in #382
• update FindOrFetchBoost by @evaleev in #383
• TA Tensor memory trace by @evaleev in #384
• [Umpire] thread-safety provided by umpire_allocator_impl by @evaleev in #385
• [cmake] globalize Boost targets imported by TA only by @evaleev in #386
• [python] can load with initialized MPI by @evaleev in #388
• Evaleev/feature/concat by @evaleev in #389
• fix sparse shape threshold by @evaleev in #392
• [WIP] some einsum tests fail by @evaleev in #391
• reentrant TA::rand() by @evaleev in #394
• disambiguate rank 1 index vs ordinal accessors by @evaleev in #393
• Kmp5/feature/cp by @kmp5VT in #335
• TiledRange1::make_uniform uses @kmp5's implementation by @evaleev in #395
• complex ta dense asymm by @evaleev in #397
• pull in MADWorld fixes to control #threads when running over PaRSEC by @evaleev in #398
• cleanup complex API by @evaleev in #399
• extend device API for complex types by @evaleev in #400
• paper demo by @evaleev in #401
• bumps MADNESS tag to pull in PR 471 by @evaleev in #403
• bump MAD tag to use master PaRSEC backend + associated TTG bump by @evaleev in #406
• misc CUDA fixes/improvements by @evaleev in #409
• ToT*T: round 1 by @evaleev in #405
• [cmake] bump VG cmake kit to bump lapackpp tags by @evaleev in #411
• Fix make_array when target shape is rank-1 by @wavefunction91 in #414
• Various CMake + ENABLE_CUDA Fixes by @wavefunction91 in #415
• Fix CUDA compilation with Cray Wrappers by @wavefunction91 in #416
• CMake: Use GNUInstallDirs variables instead of hard-coded paths by @topazus in #404
• implements initial HIP/ROCm support by @evaleev in #418
• loosen equality tolerance in um_expressions_suite/dot_permute by @evaleev in #424
• fix stream handling in multi-op device tasks by @evaleev in #421
• [ci] update path to OneAPI MKL vars.sh script by @evaleev in #425
• Fix INSTALL paths by @wavefunction91 in #426
• [unit] set MAD_NUM_THREADS when running w >1 rank by @evaleev in #431
• [cmake] bump VG kit, BTAS, and MADNESS tags by @evaleev in #429
• tiny step towards supporting T*ToT in expr by @bimalgaudel in #433
• disable throw tests unless assert policy throws by @evaleev in #434
• bump pybind11 version to VG/v2.11 by @evaleev in #436
• [cmake] for cmake v3.28 set policy CMP0146 to OLD by @evaleev in #439
• Create proper target when installed Umpire is provided by @devreal in #440
• einsum support for generalized product involving tensor-of-tensor and regular tensor by @bimalgaudel in #437
• patch Umpire to be able to shut down its' I/O cleanly by @evaleev in #441
• modularized boost by @evaleev in #443
• bump MAD tag to pull in m-a-d-n-e-s-s/madness#520 ... by @evaleev in #444
• fix Tensor(range, elemop) ctor to use placement-new instead of (move) assignment by @evaleev in #446
• better support for zero-volume ranges by @evaleev in #447
• allows to use fair dispatch in Intel MKL by @evaleev in #448
• cleanup gemm examples by @evaleev in #449
• ExternalProject_Add avoids touching install directory at configure time by @evaleev in #450
• This branch implements tensor contraction between inner tensors and their reduction along an outer tensor's mode. by @bimalgaudel in #442
• Implements support for more corner cases involving ToT times T. by @bimalgaudel in #451
• solver adaptors for eigen matrix block by @evaleev in #453
• bump MADNESS tag to pull in m-a-d-n-e-s-s/madness#539 by @evaleev in #454
• [ci] send most of the CI jobs to SaaS runners by @evaleev in #456
• C++20 build fixes by @evaleev in #455
• Update the volume(DistArray<Tile,Policy>) function. by @bimalgaudel in #457
• upgrade umpire to v2024.02.1 by @evaleev in #458
• Generalize TA::squared_norm by @bimalgaudel in #459
• Gaudel/feature/tot support for linalg func by @bimalgaudel in #461
• Tests and fixes one more corner case of ToT x ToT evaluation. by @bimalgaudel in #460
• gemm examples support nonuniform tiling by @evaleev in #462
• SVD computes full sets of vectors, not partial by @evaleev in #464
• concat(arrays) can handle zero-volume arrays by @evaleev in #465
• can change DistArray's trange (retile + more) by @evaleev in #466
• singleToDoublePrecPerfRatio hip device property does not exist before… by @powellsr in #467
• [cmake] pull in most recent {blas,lapack}pp by @evaleev in #468
• introduced make_uniform(Range1,tilesize) by @evaleev in #469
• misc mixups for Frontier by @evaleev in #470
• device::Env::initialize: use correct page sizes for Umpire allocators by @evaleev in #472
• array<->eigen conversions + assignment to block expressions work with arrays/expressions with nonzero lobound/base by @evaleev in #471
• host allocator is serializable by @evaleev in #476
• better btas::Tensor interoperation with TA tensorials by @evaleev in #477
• use ccache correctly by @evaleev in #478
• [ci] build ta_test as part of "Build" step by @evaleev in #479
• BTAS pr 179 by @evaleev in #480
• [cmake] Umpire #913 by @evaleev in #481
• Update CMakeLists.txt by @JonathonMisiewicz in #482
• Update umpire.cmake by @JonathonMisiewicz in #483
• bump MAD tag to pull in #550 by @evaleev in #485
• fix synchronization in collective DistArray initializations/transformations by @evaleev in #484
• nvToolsExt -> nvtx3 by @evaleev in #487
• availability of CUDA/HIP does not mean they should be used by @evaleev in #489
• Make sure namespace device is always closed by @devreal in #490
• TA::retile support for DistArray with tensor-of-tensors tiles by @bimalgaudel in #474
• Hush compiler warning and fix typos by @ajay-mk in #493
• Support non-zero ToTs with some zero inner Ts by @bimalgaudel in #492
• Powellsr/fix/debug attach handling by @powellsr in #473

New Contributors

• @powellsr made their first contribution in #228
• @bimalgaudel made their first contribution in #257
• @keceli made their first contribution in #293
• @therault made their first contribution in #301
• @awild82 made their first contribution in #315
• @victor-anisimov made their first contribution in #352
• @topazus made their first contribution in #404
• @devreal made their first contribution in #440
• @JonathonMisiewicz made their first contribution in #482
• @ajay-mk made their first contribution in #493

Full Changelog: v1.0.0...v1.1.0