Code

Diagonalizer multi gpu.v2#14

Code

Open

glaurung24 wants to merge 53 commits intodafer45:master.v2dafer45/TBTK:master.v2from glaurung24:DiagonalizerMultiGPU.v2glaurung24/TBTK:DiagonalizerMultiGPU.v2Copy head branch name to clipboard

+26,630-12,495

Commits

Commits on Jan 25, 2023

• Added cuda routine to diagonalizer

  glaurung24

  committed
  5faeace

  Copy full SHA for 5faeace

Commits on Jan 26, 2023

• Fixed bad memory access bug

  glaurung24

  committed
  1256e4b

  Copy full SHA for 1256e4b

• Fixed bug in Diagonalizer.cpp causing wrong result

  glaurung24

  committed
  c6b53df

  Copy full SHA for c6b53df

Commits on Jan 27, 2023

• Switched to CUDA unified memory for buffer

  glaurung24

  committed
  34c5ca3

  Copy full SHA for 34c5ca3

Commits on Feb 7, 2023

• Added test cases+support for nocuda install

  glaurung24

  committed
  e1d4251

  Copy full SHA for e1d4251

• Tended to compiler warnings including updated json

  glaurung24

  committed
  c957862

  Copy full SHA for c957862

• Fixed CMake template to include cusolver library

  glaurung24

  committed
  7fb1adc

  Copy full SHA for 7fb1adc

Commits on Feb 9, 2023

• Added GPU specific basis transformation

  glaurung24

  committed
  2acf5b0

  Copy full SHA for 2acf5b0

Commits on Feb 13, 2023

• Added tests for GPU Diagonalizer

  glaurung24

  committed
  25556a0

  Copy full SHA for 25556a0

• Removed compiler warning

  glaurung24

  committed
  e88a872

  Copy full SHA for e88a872

• Set default for gpu acceleration to false

  glaurung24

  committed
  3b08c8b

  Copy full SHA for 3b08c8b

• Changed device allocation to unified memory

  glaurung24

  committed
  f00e445

  Copy full SHA for f00e445

Commits on Feb 20, 2023

• Added device memory prefetch prevening page faults

  glaurung24

  committed
  cf310e8

  Copy full SHA for cf310e8

Commits on Feb 21, 2023

• Added python 3.10

  glaurung24

  committed
  85c7d1f

  Copy full SHA for 85c7d1f

Commits on Apr 26, 2023

• Added changable number of significant digits

  glaurung24

  committed
  28ec73f

  Copy full SHA for 28ec73f

Commits on Jul 17, 2023

• Added GPU support to BlockDiagonalizer, untested

  glaurung24

  committed
  7ef8af9

  Copy full SHA for 7ef8af9

• Removed uneccesary use of CArray hamiltonian

  glaurung24

  committed
  11ad6ef

  Copy full SHA for 11ad6ef

Commits on Sep 20, 2023

• Added test for gpu implementation

  glaurung24

  committed
  57e1bf0

  Copy full SHA for 57e1bf0

• Fixed bug caused by wrong block offsets

  glaurung24

  committed
  9a00863

  Copy full SHA for 9a00863

• Changed to using device memory exclusively

  glaurung24

  committed
  542f046

  Copy full SHA for 542f046

• Fixed error in memory calculation

  glaurung24

  committed
  915e938

  Copy full SHA for 915e938

• Fixed exessive memory use of buffer

  glaurung24

  committed
  7d3673b

  Copy full SHA for 7d3673b

• Fixed exessive memory use of buffer

  glaurung24

  committed
  5ec475e

  Copy full SHA for 5ec475e

Commits on Sep 21, 2023

• Fixed unnecessary output

  glaurung24

  committed
  9a5628d

  Copy full SHA for 9a5628d

• Enabled parallel execution while using gpu

  glaurung24

  committed
  892134d

  Copy full SHA for 892134d

• Allow to run several instances on one gpu

  glaurung24

  committed
  de4be07

  Copy full SHA for de4be07

Commits on Sep 27, 2023

• Fixed bug by reseting ssin

  glaurung24

  committed
  ba03e48

  Copy full SHA for ba03e48

Commits on Jan 22, 2024

• Added multiGPU support to Diagonalizer (experimental)

  

  Andreas Theiler

  committed
  2a70e2d

  Copy full SHA for 2a70e2d

Commits on Jan 23, 2024

• Implemented tests for MultiGPU diagonalizer and blockDiagonalizer

  glaurung24

  committed
  2de1fc0

  Copy full SHA for 2de1fc0

• Fixed compilation issues

  glaurung24

  committed
  687f2f1

  Copy full SHA for 687f2f1

• Removed bug where only lower part of mat was filled

  glaurung24

  committed
  dda3c60

  Copy full SHA for dda3c60

• Moved NVIDIA *util.h files to External and adjusted build

  glaurung24

  committed
  1d35727

  Copy full SHA for 1d35727

Commits on Jan 24, 2024

• Added CalculationMode to diagonalizer

  glaurung24

  committed
  0c140fa

  Copy full SHA for 0c140fa

Commits on Jan 26, 2024

• Put repeated blas calls into one function

  glaurung24

  committed
  1f6aed5

  Copy full SHA for 1f6aed5

• Added safeguard in case only eigenvalues are calculated

  glaurung24

  committed
  6242ebc

  Copy full SHA for 6242ebc

Commits on Jan 30, 2024

• Moving to use lower triangular matrix, test fails

  glaurung24

  committed
  d7891a6

  Copy full SHA for d7891a6

• Changed Solver::Diagonalizer to use lower triangular matrix, tests passed

  glaurung24

  committed
  309ff78

  Copy full SHA for 309ff78

• Switched to use Solver::Diagonalizers solveCPU instead of Blas zhpev_

  glaurung24

  committed
  03bb68e

  Copy full SHA for 03bb68e

• Fixed bug that would cause crash in cuSolverMg library

  glaurung24

  committed
  04a237d

  Copy full SHA for 04a237d

• Added GPU tests

  glaurung24

  committed
  285d648

  Copy full SHA for 285d648

Commits on Jan 31, 2024

• Improved error handling

  glaurung24

  committed
  702ff9d

  Copy full SHA for 702ff9d

• Temporary fix for memory leak in suSolverMg library

  glaurung24

  committed
  0e9bbb1

  Copy full SHA for 0e9bbb1

• Changed to async cudamemcpy for sligthly better performance

  glaurung24

  committed
  088b6da

  Copy full SHA for 088b6da

• Added gpu tests

  glaurung24

  committed
  0542232

  Copy full SHA for 0542232

Commits on Feb 5, 2024

• Matrix descriptor not static anymore, allowing to solve different problem sizes

  glaurung24

  committed
  a10209b

  Copy full SHA for a10209b

Commits on Feb 7, 2024

• Changed to lower triangle

  glaurung24

  committed
  7dd92df

  Copy full SHA for 7dd92df

Commits on Feb 9, 2024

• Implemented more robust check to see if eigenvectors are available

  glaurung24

  committed
  d6c2cdd

  Copy full SHA for d6c2cdd

Commits on Feb 23, 2024

• Remove installation problem with cusolverMg

  glaurung24

  committed
  fc8317c

  Copy full SHA for fc8317c

• Made changes that allows copying solver data to new solver

  glaurung24

  committed
  a2df55b

  Copy full SHA for a2df55b

Commits on Feb 27, 2024

• Clean up and extra comments

  glaurung24

  committed
  0962706

  Copy full SHA for 0962706

Commits on Jun 12, 2024

• Fixed bug were plotter would use arguments not recognized by matplotlib.

  glaurung24

  committed
  2dedaaa

  Copy full SHA for 2dedaaa

• Added check to test if the input hamiltonian is hermitian

  glaurung24

  committed
  f9164ca

  Copy full SHA for f9164ca

Commits on Sep 6, 2024

• Added a recovery option for out of device memory problems

  glaurung24

  committed
  21c4558

  Copy full SHA for 21c4558