Code

Feature/multi rhs#1318

Code

Merged

maddyscientist merged 77 commits intodeveloplattice/quda:developfrom feature/multi-rhslattice/quda:feature/multi-rhsCopy head branch name to clipboardOct 25, 2022

+3,482-3,864

Commits

Commits on Aug 12, 2022

• Add initial multi-RHS interface for Dirac and DiracMatrix classes

  maddyscientist

  committed
  1da7761

  Copy full SHA for 1da7761

• tunecache is not auto-dumped if the time since last tuning exceeds 120 seconds. This allows us to clean up various calls to saveTuneCache()

  maddyscientist

  committed
  9c83b1a

  Copy full SHA for 9c83b1a

• Some initial cleanup of the eigen solver

  maddyscientist

  committed
  ab149d5

  Copy full SHA for ab149d5

Commits on Aug 14, 2022

• Minor cleanup of multi_reduce.cu

  maddyscientist

  committed
  f95ec63

  Copy full SHA for f95ec63

Commits on Aug 20, 2022

• Add support for non-sum reductions in blas reduce functors. Add two new function max and max_deviation

  maddyscientist

  committed
  854f184

  Copy full SHA for 854f184

• Update is_enabled precision helper to use function arg instead of template - make it more useful

  maddyscientist

  committed
  64c4cd3

  Copy full SHA for 64c4cd3

Commits on Aug 21, 2022

• Add support to make_set to handle std::vector<T*> to std::vector<reference_wrapper> conversion

  maddyscientist

  committed
  587f57e

  Copy full SHA for 587f57e

• Some dilution test cleanup

  maddyscientist

  committed
  fd979fb

  Copy full SHA for fd979fb

• Update VectorIO::load/save functions to internally use vector of ref_wrappers with a generic multi-rhs interface and a cleanup of the internals

  maddyscientist

  committed
  bbbbd7c

  Copy full SHA for bbbbd7c

Commits on Aug 22, 2022

• math_helper should be included in blas_helper to ensure it is included

  maddyscientist

  committed
  ce6380a

  Copy full SHA for ce6380a

• Fix compilation warning

  maddyscientist

  committed
  32a389d

  Copy full SHA for 32a389d

• Move some QUDA feature enablement to quda_define.h.in instead of requiring compiler flags

  maddyscientist

  committed
  75a8034

  Copy full SHA for 75a8034

• Added new test file io_test: this does unit testing on i/o functionlity for both gauge and color-spinor fields

  maddyscientist

  committed
  6615c1d

  Copy full SHA for 6615c1d

• deflation.cpp should use VectorIO class for load/saving vectors

  maddyscientist

  committed
  0602fe7

  Copy full SHA for 0602fe7

• Simplify eigensolver.cpp

  maddyscientist

  committed
  4350fc0

  Copy full SHA for 4350fc0

Commits on Aug 23, 2022

• Replace the blas ax_ functor with axy_ which allows for reading from and writing to distinct fields. blas::ax is now just a wrapper around this, and introduce blas::axy which exposes this functiona…

  Show description for 450b359
  maddyscientist

  committed
  450b359

  Copy full SHA for 450b359

Commits on Aug 24, 2022

• Improve error verbosity in multi-BLAS _U/_L functions

  maddyscientist

  committed
  66ecbe7

  Copy full SHA for 66ecbe7

• Add ColroSpinorField_cref alias for const fields

  maddyscientist

  committed
  f1da382

  Copy full SHA for f1da382

• Update multi-blas interface function to accept vectors of ColorSpinorField_cref instead of ColorSpinorField_ref where the vectors are read only

  maddyscientist

  committed
  5bf5c3f

  Copy full SHA for 5bf5c3f

• Mark ColorSpinorField/LatticeField copy and move constructors to be noexcept - this allows for move assignment instead of copy assignment when std::vector::resize is called

  maddyscientist

  committed
  1a0d4d4

  Copy full SHA for 1a0d4d4

Commits on Aug 25, 2022

• Fix issue with 5bf5c3f, where for functions that who require a vector of crefs, we would endup self recursing if passing a vector of refs: added explicit make_cset helper, to create explicit cref s…

  Show description for 250cb83
  maddyscientist

  committed
  250cb83

  Copy full SHA for 250cb83

• Add custom resize functions for std::vector<ColorSpinorField> instances that allow us to avoid the copying that would occur if we used std::vector::resize

  maddyscientist

  committed
  4c88d29

  Copy full SHA for 4c88d29

• Use custom resize function in blas_test and dslash_test: this significantly boosts the performance of the blas_test, removing a lot of copies

  maddyscientist

  committed
  e6d40a7

  Copy full SHA for e6d40a7

Commits on Aug 27, 2022

• Rewrite of eigensolvers to utilize vector of fields instead of vector of pointers. As part of this change, the multi-blas function interfaces have been overhauled to utilized a new specialized type…

  Show description for 96a2cb1
  maddyscientist

  committed
  96a2cb1

  Copy full SHA for 96a2cb1

Commits on Aug 28, 2022

• Merge branch 'hotfix/arpack' of github.com:lattice/quda into feature/multi-rhs

  maddyscientist

  committed
  d0e9855

  Copy full SHA for d0e9855

Commits on Aug 29, 2022

• Add SVD computation to ARPACK path

  cpviolator

  committed
  6a06844

  Copy full SHA for 6a06844

• Merge branch 'feature/multi-rhs' of https://github.com/lattice/quda into feature/multi-rhs

  cpviolator

  committed
  29b49ac

  Copy full SHA for 29b49ac

• Fix resize bug

  maddyscientist

  committed
  06885a4

  Copy full SHA for 06885a4

• Add function to allow for ARPACK sorting, use 'A' param to return Ritz vectors in ARPACK, not Schur vectors

  cpviolator

  committed
  53b0edb

  Copy full SHA for 53b0edb

• Merge branch 'feature/multi-rhs' of https://github.com/lattice/quda into feature/multi-rhs

  cpviolator

  committed
  0ce2afd

  Copy full SHA for 0ce2afd

Commits on Aug 30, 2022

• Promote block axpy functions to be templated on coefficient type: this allows their use in generic code that deals with real or complex-valued arithmetic

  maddyscientist

  committed
  dc740a3

  Copy full SHA for dc740a3

• The member functions rotateVecs and blockRotate are now generic, and work with real or complex arithmetic

  maddyscientist

  committed
  68929f8

  Copy full SHA for 68929f8

• Minor clean up of ARPACK eigen/singular value computation.

  cpviolator

  committed
  6a5aacd

  Copy full SHA for 6a5aacd

• Minor clean up of ARPACK eigen/singular value computation.

  cpviolator

  committed
  b20749c

  Copy full SHA for b20749c

Commits on Aug 31, 2022

• Add init member to LatticeFieldParam, which is used to check if the parameter structure is trivially constructed or not. Use this to simplify the ColorSpinorField::create_alias interface

  maddyscientist

  committed
  95bfc5c

  Copy full SHA for 95bfc5c

• Generalizer pair constructor for reference_wrapper to a variadic constructor

  maddyscientist

  committed
  1b02e05

  Copy full SHA for 1b02e05

• Remove all use of std::vector<ColorSpinorField*> from regular solvers

  maddyscientist

  committed
  260d4ec

  Copy full SHA for 260d4ec

• Merge branch 'feature/multi-rhs' of github.com:lattice/quda into feature/multi-rhs

  maddyscientist

  committed
  46039b5

  Copy full SHA for 46039b5

• Delete accidental inclusion

  maddyscientist

  committed
  ed01e57

  Copy full SHA for ed01e57

• Minor fix for max_deviation functor

  maddyscientist

  committed
  ab28f80

  Copy full SHA for ab28f80

Commits on Sep 1, 2022

• Added FieldTmp class for caching temporaries fields. Removed all explicit management of temporaries relating to the Dirac operators (significant cleanup)

  maddyscientist

  committed
  efa1bc8

  Copy full SHA for efa1bc8

• Add missing header from last commit

  maddyscientist

  committed
  429ba11

  Copy full SHA for 429ba11

• Remove ColorSpinorField_ref type: replaced all usage using vector_ref

  maddyscientist

  committed
  41d4b6b

  Copy full SHA for 41d4b6b

• Update multi-RHS Dirac interface to use vector_ref

  maddyscientist

  committed
  7c0066a

  Copy full SHA for 7c0066a

Commits on Sep 2, 2022

• Cleanup for blas functions: added missing doxygen, all blas function arguments are now const where they should be, axpyCGNorm now returns a double2 instead of Complex

  maddyscientist

  committed
  7ceb35b

  Copy full SHA for 7ceb35b

• Remove now unnecessary const_cast usage when calling blas functions - enabled by last commit

  maddyscientist

  committed
  f18c843

  Copy full SHA for f18c843

• Initial multi-RHS awareness added to block Lanczos

  maddyscientist

  committed
  83ed9b5

  Copy full SHA for 83ed9b5

• Add deflation to bicgstab

  cpviolator

  committed
  cd85a7a

  Copy full SHA for cd85a7a

• Merge branch 'feature/multi-rhs' of https://github.com/lattice/quda into feature/multi-rhs

  cpviolator

  committed
  228473d

  Copy full SHA for 228473d

• Remove 'mat' argument from eigesolver member functions, use 'mat' reference instead

  cpviolator

  committed
  b78a87c

  Copy full SHA for b78a87c

• Make blas::zero function multi-RHS aware. Fix bug in EigenSolver::rotateVecs function

  maddyscientist

  committed
  f3576c8

  Copy full SHA for f3576c8

Commits on Sep 3, 2022

• Fix PCG when no preconditioner is used

  maddyscientist

  committed
  5cb7fc6

  Copy full SHA for 5cb7fc6

Commits on Sep 7, 2022

• Merge branch 'develop' of github.com:lattice/quda into feature/multi-rhs

  maddyscientist

  committed
  2a7fc1d

  Copy full SHA for 2a7fc1d

• Fix issue with field cache: needs to be nuked if the comm partitioning is changed

  maddyscientist

  committed
  deb42f0

  Copy full SHA for deb42f0

Commits on Sep 8, 2022

• Remove last remnants of Dirac::tmp1 / tmp2

  maddyscientist

  committed
  add330a

  Copy full SHA for add330a

• Fix for VectorIO interface, and add WAR for multigrid's use of VectorIO (will be fixed when MG uses vector of fields, instead of vector of pointers)

  maddyscientist

  committed
  b40818a

  Copy full SHA for b40818a

• Add missing string header inclusion

  maddyscientist

  committed
  1359142

  Copy full SHA for 1359142

• Fix Jenkins error with BiCGStab

  maddyscientist

  committed
  a9b6a0e

  Copy full SHA for a9b6a0e

• Add PCG to the invert_test list

  maddyscientist

  committed
  2e87cb7

  Copy full SHA for 2e87cb7

Commits on Sep 13, 2022

• Upate multi-RHS interface functions to use const vector_ref lvalue instead of vector_ref rvalue. Add array accessor for vector_ref, and apply the resulting simplifications.

  maddyscientist

  committed
  77018ba

  Copy full SHA for 77018ba

Commits on Sep 16, 2022

• Created cvector_ref alias for const vector_ref

  maddyscientist

  committed
  f7c2b15

  Copy full SHA for f7c2b15

• Fix some long-standing issues in PCG, and apply some cleanup

  maddyscientist

  committed
  ca7035f

  Copy full SHA for ca7035f

• Tweak default parameter max_res_increase to ensure convergence for all ctest solver test

  maddyscientist

  committed
  0790fd0

  Copy full SHA for 0790fd0

Commits on Sep 20, 2022

• Fix bug when running io_test with multiple processes

  maddyscientist

  committed
  27c5b07

  Copy full SHA for 27c5b07

• Fix bug with dslash_ctest: test_split_grid is not yet initialized when checking if kernel will skip (fixes occasional hangs when running on multiple processes)

  maddyscientist

  committed
  284f951

  Copy full SHA for 284f951

Commits on Sep 21, 2022

• Address Mathias' review comments

  maddyscientist

  committed
  b37ed33

  Copy full SHA for b37ed33

• Add Schwarz solvers to the invert_test gtest list. Change default CA basis used in preconditioners to Chebyshev

  maddyscientist

  committed
  dbf647e

  Copy full SHA for dbf647e

Commits on Sep 22, 2022

• fix a few typos

  mathiaswagner

  committed
  c2ea87f

  Copy full SHA for c2ea87f

Commits on Sep 27, 2022

• Fix bug in invert_test, introduced when adding Schwarz preconditioner support

  maddyscientist

  committed
  5ca4bc2

  Copy full SHA for 5ca4bc2

• Delete dead code

  maddyscientist

  committed
  6df7059

  Copy full SHA for 6df7059

• Merge branch 'feature/multi-rhs' of github.com:lattice/quda into feature/multi-rhs

  maddyscientist

  committed
  327d369

  Copy full SHA for 327d369

Commits on Sep 28, 2022

• clean up headers, fix typo

  mathiaswagner

  committed
  c0c9457

  Copy full SHA for c0c9457

• Remove commented code

  maddyscientist

  committed
  cdc47f9

  Copy full SHA for cdc47f9

• Merge branch 'feature/multi-rhs' of github.com:lattice/quda into feature/multi-rhs

  maddyscientist

  committed
  d46a748

  Copy full SHA for d46a748

Commits on Oct 5, 2022

• Fix compiler warnings when QUDA_INVERFACE_NVTX=ON

  maddyscientist

  committed
  92a770c

  Copy full SHA for 92a770c

• Remove some legacy std::vector<std::reference_wrapper<ColorSpinorField>>

  maddyscientist

  committed
  b607f03

  Copy full SHA for b607f03

• Merge branch 'develop' of github.com:lattice/quda into feature/multi-rhs

  maddyscientist

  committed
  a595b02

  Copy full SHA for a595b02